text stringlengths 1.23k 293k | tokens float64 290 66.5k | created stringdate 1-01-01 00:00:00 2024-12-01 00:00:00 | fields listlengths 1 6 |
|---|---|---|---|
A Fast K-prototypes Algorithm Using Partial Distance Computation
: The k-means is one of the most popular and widely used clustering algorithm; however, it is limited to numerical data only. The k-prototypes algorithm is an algorithm famous for dealing with both numerical and categorical data. However, there have been no studies to accelerate it. In this paper, we propose a new, fast k-prototypes algorithm that provides the same answers as those of the original k-prototypes algorithm. The proposed algorithm avoids distance computations using partial distance computation. Our k-prototypes algorithm finds minimum distance without distance computations of all attributes between an object and a cluster center, which allows it to reduce time complexity. A partial distance computation uses a fact that a value of the maximum difference between two categorical attributes is 1 during distance computations. If data objects have m categorical attributes, the maximum difference of categorical attributes between an object and a cluster center is m. Our algorithm first computes distance with numerical attributes only. If a difference of the minimum distance and the second smallest with numerical attributes is higher than m, we can find the minimum distance between an object and a cluster center without distance computations of categorical attributes. The experimental results show that the computational performance of the proposed k-prototypes algorithm is superior to the original k-prototypes algorithm in our dataset.
Introduction
K-means algorithm is one of the simplest clustering algorithm as unsupervised learning, so that is very widely used [1].As it is a partitioning-based clustering method in cluster analysis, a dataset is partitioned into several groups according to a similarity measure as a distance to the average of a group.A K-means algorithm minimizes the objective function known as squared error function iteratively by finding a new set of cluster centers.In each iteration, the value of the objective function becomes lower.In a k-means algorithm, the objective function is defined by the sum of square distances between an object and a cluster center.
The purpose of using k-means is to find clusters that minimize the sum of square distances between each cluster center and all objects in each cluster.Even though the number of clusters is small, the problem of finding an optimal k-means algorithm solution is NP-hard [2,3].For this reason, a k-means algorithm adapts heuristics and finds local minimum as approximate optimal solutions.The time complexity of a k-means algorithm is O(i*k*n*d), where i iterations, k centers, and n points in d dimensions.
K-means algorithm spends a lot of processing time for computing the distances between each of the k cluster centers and the n objects.So far, many researchers have worked on accelerating k-means algorithms by avoiding unnecessary distance computations between an object and cluster centers.Because objects usually remain in the same clusters after a certain number of iterations, much of the repetitive distance computation is unnecessary.So far, a number of studies on accelerating k-means algorithms to avoid unnecessary distance calculations have been carried out [4][5][6].
The K-means algorithm is efficient for clustering large datasets, but it only works on numerical data.However, the real-world data is a mixture of both numerical and categorical features, so a k-means algorithm has a limitation of applying cluster analysis.To overcome this problem, several algorithms have been developed to cluster large datasets that contain both numerical and categorical values, and one well-known algorithm is the k-prototypes algorithm by Huang [7].The time complexity of the k-prototypes algorithm is also O(i*k*n*d): one of the k-means.In case of large datasets, time cost of distance calculation between all data objects and the centers is high.To the best of our knowledge, however, there have been no studies that reduce the time complexity of the k-prototypes algorithm.
Recently, big data has become a big issue for academia and various industries.Therefore, there is a growing interest in the technology to process big data quickly from the viewpoint of computer science.The research on the fast processing of big data in clustering has been limited to numerical data.However, big data deals with numerical data as well as categorical data.Because numerical data and categorical data are processed differently, it is difficult to improve the performance of clustering algorithms that deal with categorical data in the existing way of improving performance.
In this paper, we propose a fast k-prototypes algorithm for mixed data (FKPT).The FKPT reduces distance calculation using partial distance computation.The contributions of this study are summarized as follows.
1.
Reduction: computational cost is reduced without an additional data structure and memory spaces.
2.
Simplicity: it is simple to implement because it does not require a complex data structure.
3.
Convergence: it can be applied to other fast k-means algorithms to compute the distance between each cluster center and an object for numerical attributes.
4.
Speed: it is faster than the conventional k-prototypes.
This study presents a new method of accelerating k-prototypes algorithm using partial distance computation by avoiding unnecessary distance computations between an object and cluster centers.As a result, we believe the algorithm proposed in this paper will become the algorithm of choice for fast k-prototypes clustering.
The organization of the rest of this paper is as follows.In Section 2, various methods of accelerating k-means and traditional k-prototypes algorithm are described, for the proposed k-prototypes are defined.A fast k-prototypes algorithm proposed in this paper is explained and its time complexity is analyzed in Section 3. In Section 4, experimental results demonstrate the scalability and effectiveness of the FKPT using partial distance computation by comparison with traditional k-prototypes algorithm.Section 5 concludes the paper.
Related Works
In this section, we briefly describe various methods of accelerating k-means and traditional k-prototypes algorithm.
K-means
The k-means is one of the most popular clustering algorithm due to its simplicity and scalability for large datasets.The k-means algorithm is to partition n data objects into k clusters while minimizing the Euclidean distance between each data object and the cluster center it belongs to [1].The fundamental concept of k-means clustering is as follows.
1.
It chooses k cluster centers in some manner.The final result of the algorithm is sensitive to the initial selection of k initial centers, and many efficient initialization methods have been proposed to calculate better final k centers.
2.
The k-means repeats the process of assigning individual objects to their nearest centers and updating each k center as the average of a value of object's vector assigned to the centers until no further changes occur on the k centers.
K-means algorithm spend most of the time computing distance between an object and current cluster centers.However, much of these distance computations are unnecessary, because objects usually remain in the same clusters after a few iterations [6].Thus, k-means is popular and easy to implement, but it is wasting processing time on redundant and unnecessary distance computations.
The reason why the k-means are inefficient is because, in each iteration, all objects must identify the closest center.In one iteration, all nk distance computations is needed between the n objects and the k centers.After the end of one iteration, the centers are changed and the nk distance computations occur again in the next iteration.
Pelleg and Moore (1999) and Kanungo et al. (2002) adapted a k-d tree to store datasets for accelerating k-means [8,9].These algorithms are effective for large datasets, but not effective for high (d > 10) dimensions.For low dimensional data, indexing the large data to be clustered is an effective way for fast k-means.These results can be explained as the curse of dimensionality, namely a distance between points and centers tend to be far from one another.Therefore, the performance of pruning is degraded for many dimensions.These algorithms must consider the overhead costs of constructing the k-d tree.A time complexity of constructing the k-d tree is O(nlog(n)) for n data points.
One way to accelerate algorithms is to search using partial distance [10,11] in a processing to identify the closest points.The reason for calculating distance is to identify the closest center point from an object point.Therefore, we do not need to exact distances if we can confirm the minimum distance.In general, the distance calculation is the sum of the square for all point attributes.If we calculate x − c 2 , the distance between a point x and a center c can be calculated by summing squared distances in each dimension.In a distance calculation between point x and another center c', if the sum exceeds x − c 2 , the distance x − c' 2 cannot be the minimum distance, so the distance calculation stops before all attribute calculations.The cost of a partial distance search is usually effective in high dimension.
Our proposed idea of this study was inspired by this partial distance search [10,11].We extend this partial distance method to the pruning technique of k-prototypes algorithm.
K-prototypes
K-prototypes algorithm integrates the k-means and k-modes algorithms to deal with the mixed data types [7].The k-prototypes algorithm is more useful practically because data collected in the real world are mixed type objects.Assume a set n objects, where C i is an i-th cluster center.The distance d X i , C j between X i and C j can be calculated as follows: where d r X i , C j is the distance between numerical attributes, d c X i , C j is the distance between categorical attributes, and γ is a weight for categorical attributes.
Symmetry 2017, 9, 58 In Equation (2), d r X i , C j is the squared Euclidean distance measure between cluster centers and an object on the numerical attributes.d c X i , C j is the simple matching dissimilarity measure on the categorical attributes, where δ x il , c jl = 0 for x il = c jl and δ x il , c jl = 1 for x il = c jl .x il and c jl , 1 ≤ l ≤ p, are values of numerical attributes, whereas x il and c jl , p + 1 ≤ l ≤ m are values of categorical attributes for object i and the cluster center j. p is the numbers of numerical attributes and m − p is the numbers of categorical attributes.
K-prototypes Using Partial Distance Computation
The existing k-prototypes algorithm allocates objects to the cluster with the smallest distance by calculating the distance between each cluster center and a new object to be allocated to the cluster.Distance is calculated by comparing all attributes of an object with all attributes of each cluster center using the brute force method.Figure 1 illustrates how k-prototypes algorithm organizes clusters with a target object.In this figure, an object consists of two numerical attributes and two categorical attributes.The entire dataset is divided into three clusters, C = {C 1 , C 2 , C 3 }.The center of each cluster is C 1 = (3, 3, C, D), C 2 = (6, 6, A, B), and C 3 = (9, 4, A, B).The traditional k-prototypes algorithm calculates distance with each cluster center to find the cluster to which X i = (5, 3, A, B) is assigned.The distance about the numerical attribute of X i and respectively.The total distance of X i and C 1 , C 2 , and C 3 is 6, 10, and 17, respectively, and the cluster closest to X i is X i .Thus, the traditional k-prototypes algorithm computes the distance as a brute force method that compares both numerical and categorical properties.
tegorical attributes, where , = 0 for = and , = 1 for ≠ .and ≤ , are values of numerical attributes, whereas and , + 1 ≤ ≤ are value rical attributes for object i and the cluster center j. is the numbers of numerical attrib − is the numbers of categorical attributes.
rototypes Using Partial Distance Computation he existing k-prototypes algorithm allocates objects to the cluster with the smallest distanc ating the distance between each cluster center and a new object to be allocated to the clu nce is calculated by comparing all attributes of an object with all attributes of each cluster ce the brute force method.Figure 1 he purpose of distance computation is to find a cluster center closest to an object.How is an unnecessary distance calculation in the traditional k-prototypes algorithm.Accordin ion (4), the maximum value that can be obtained is 1 when comparing a single catego ute.In Figure 1, an object has two categorical properties, so the maximum value that ca ted from the distance comparing the categorical property is 2. In Figure 1, when usi rical attribute, the closest cluster with the object is and a distance of 4, the second cl r is , a distance of 4. Since the difference between these two values is greater tha The purpose of distance computation is to find a cluster center closest to an object.However, there is an unnecessary distance calculation in the traditional k-prototypes algorithm.According to Equation (4), the maximum value that can be obtained is 1 when comparing a single categorical attribute.In Figure 1, an object has two categorical properties, so the maximum value that can be extracted from the distance comparing the categorical property is 2. In Figure 1, when using a numerical attribute, the closest cluster with the object is C 1 and a distance of 4, the second closest Symmetry 2017, 9, 58 5 of 10 cluster is C 2 , a distance of 4. Since the difference between these two values is greater than 2, comparing the numerical property, nevertheless the measured minimum distance, 4 added to the categorical property comparison maximum value 2, it does not exceed 10.In such a case, the cluster center closest to an object can be determined by calculating the numerical attribute value without calculating the category curl attribute in the distance calculation.Of course, the minimum distance cannot be obtained by comparing numerical properties for all cases only.In Figure 1, at X j = (0, 0, 0, 0), the distance of the numerical attribute of X j and This paper studies a method to find the closest center to an object without comparison for all attributes in a distance computation.We prove that the closest center to an object can be found without comparison for all attributes.For a proof, we define a computable max difference value as follows.
Definition 1.The computable max difference value means that the maximum difference value can be calculated in the distance measured between an object and cluster centers for one attribute.
According to Equation (4), the distance for a single categorical attribute between cluster centers and an object is either 0 or 1.Therefore, a computable max difference value for a categorical attribute, according to Definition 1, becomes 1 without taking the value of the attribute into account.If an object in the dataset consists of m categorical attributes, then a computable max difference value between a cluster and object is m.A computable max difference value of a numerical attribute is a difference of a maximum value and a minimum value of the attribute.Thus, to know a computable max difference value of a numerical attribute, we have to scan full datasets so that maximum and minimum values are obtained.
The proposed k-prototypes algorithm finds a minimum distance without distance computations of all attributes between an object and a cluster center using the computable max difference value of the object.The k-prototypes algorithm updates a cluster center after an object is assigned to the cluster of the closest center by the distance measure.By Equation (1), the distance d X i , C j between an object and a cluster center is computed by adding the distance of numerical attributes and the distance of categorical attributes.If a difference of the first and the second minimum distance on numerical attributes is higher than m, we can find a minimum distance between an object and a cluster center using only distance computation of numerical attributes without distance computations of categorical attributes.Lemma 1.For a set of objects with m categorical attribute, if d r ( According to Definition 1, the categorical distance between an object and a cluster center with m categorical attributes can be 0 We introduce a way to determine the minimum distance between an object and each cluster center with only computation of numerical attribute by an example. Example 1.We assume that k = 3.Each current cluster center is as follows: C 1 = (A, A, A, 5), C 2 = (B, B, B, 7), and C 3 = (B, B, B, 8).As shown by Figure 2, we have to compute the distance between X i = (B, B, B, 4) and each of cluster centers (C 1 , C 2 , and C 3 ) for assigning X i to the cluster of the closest center.Firstly, we compute the distance of numerical attributes, d r X i , C j is 1, 9, and 16, respectively.X i is the closest to C 1 only with numerical attributes.In this example, objects consisted of three categorical attributes; the minimum value of possible distance is 0, and the maximum value is 3.The difference of numerical distance between d r (X i , C 1 ) and d r (X i , C 2 ) is 8. Thus, X i continues to be the closest to C 1 , even if d c (X i , C 1 ) is calculated by 3 as the computable max difference value.
roposed Algorithm
In this section, we describe our proposed algorithm.The proposed k-prototypes algorithm in this paper is similar to traditional k-prototypes ence between the proposed k-prototypes and traditional k-prototypes is that the dis een an object and cluster centers on the numerical attributes, , , is calculated firstly In Line 4, firstly, you calculate the distance for a numerical attribute.You obtain the cl nce and the second closest distance value while calculating the distance.Using these two v he number of the categorical attributes, m, the discriminant is performed.If the result o iminant is true, the distance to the categorical property is calculated, and then the result o distance is derived by adding the distance of the numerical property.If the result o iminant is false, the final distance is measured by the numerical attribute result only ding in the cluster measured at the smallest distance, the value of the corresponding cl r is updated.Definition 1 is a function that determines whether to compare the categorical attribute wit ithm that implements it.Returns the true value if the difference between the second sm nce and the first smallest distance is less than m in the distance measured only by a num erty comparison between an object and cluster center.If a true value is returned, Algorit a function that compares the distance of the categorical attribute to calculate the final dist lse value is returned, the distance measured by only the numerical property comparison e final results value without comparing the categorical property.The larger the diffe een the two distances, the greater the number of categorical attributes that need no ared.
Proposed Algorithm
In this section, we describe our proposed algorithm.The proposed k-prototypes algorithm in this paper is similar to traditional k-prototypes.The difference between the proposed k-prototypes and traditional k-prototypes is that the distance between an object and cluster centers on the numerical attributes, d r X i , C j , is calculated firstly.
In Line 4, firstly, you calculate the distance for a numerical attribute.You obtain the closest distance and the second closest distance value while calculating the distance.Using these two values and the number of the categorical attributes, m, the discriminant is performed.If the result of the discriminant is true, the distance to the categorical property is calculated, and then the result of the final distance is derived by adding the distance of the numerical property.If the result of the discriminant is false, the final distance is measured by the numerical attribute result only.By including X i in the cluster measured at the smallest distance, the value of the corresponding cluster center is updated.
Definition 1 is a function that determines whether to compare the categorical attribute with the algorithm that implements it.Returns the true value if the difference between the second smallest distance and the first smallest distance is less than m in the distance measured only by a numerical property comparison between an object and cluster center.If a true value is returned, Algorithm 1 calls a function that compares the distance of the categorical attribute to calculate the final distance.If a false value is returned, the distance measured by only the numerical property comparison is set as the final results value without comparing the categorical property.The larger the difference between the two distances, the greater the number of categorical attributes that need not be compared.
Algorithm 1 Proposed k-prototypes algorithm
Input: n: the number of objects, k: the number of cluster, p: the number of numeric attribute, q: the number of categorical attribute Output: k cluster 01: INITIALIZE // Randomly choosing k object, and assigning it to C j .02: While not converged do 03: for i = 1 to n do 04: dist_n[] = DIST-COMPUTE-NUM(X i , C, k, p) // distance computation only numeric numerical attributes 05: first_min = DIST-COMPUTE.first_min// first minimum value among d r X i , C j 06: second_min = DIST-COMPUTE.second_min// second minimum value among d r X i , C j 07: 12: In Algorithm 2, DIST-COMPUTE-NUM() calculates a distance between an object and cluster centers for numerical attributes and returns all distances for each cluster.In this algorithm, first_min and second_min is calculated to determine whether calculation of categorical data in a distance computation.In Algorithm 3, a distance between an object and cluster centers is calculated for categorical attributes of each cluster.In Algorithm 4, the center vector of a cluster is assigned to new center vector.The center vectors consisted of two parts: numerical and categorical attributes.The numerical part of the center vector is calculated by an average value of each numerical attributes and the categorical part of the center vector is calculated by the value of the highest frequency in each categorical attribute.
Time Complexity
The time complexity of traditional k-prototypes is O(I * k * n * m), where I is the number of iterations, k is the number of clusters, n is the number of data objects, and m is the number of attributes.The best-case complexity of the proposed k-prototypes has a lower bound of Ω(I * k * n * p), where p is the number of numerical attributes and p < m.The best case is that the difference of the first and the second minimum distance between an object and cluster centers for all objects in a given dataset on numerical attributes is less than m.The worst-case complexity has an upper bound of O(I * k * n * m).The worst case is that the difference of the first and the second minimum distance between an object and cluster centers for all objects in a given dataset on numerical attributes is higher than m.
Experimental Results
All experiments are conducted on an Intel(R) Pentium(R) 3558U 1.70 GHz, 4GB RAM.All programs are written in Java.We generate several independent, uniform distribution mix typed datasets.A distribution of numerical attributes is from 0 to 100, and one of the categorical attributes is from A to Z.
Effect of Cardinality
We set |X| (number of objects) = {500000, 800000, 1000000}, numerical attributes = 2, categorical attributes = 16 and k = 3. Figure 3 shows the CPU time versus cardinality in different datasets.In the figure, there are two lines.In general, the CPU time increases linearly when the cardinality increases linearly.The experimental shows proposed k-prototypes algorithm improves computational performance than original k-prototypes algorithm in our dataset.We set |X| (number of objects) = {500000, 800000, 1000000}, numerical attributes = 2, categorical attributes = 16 and k = 3. Figure 3 shows the CPU time versus cardinality in different datasets.In the figure, there are two lines.In general, the CPU time increases linearly when the cardinality increases linearly.The experimental shows proposed k-prototypes algorithm improves computational performance than original k-prototypes algorithm in our dataset.To analyze the difference CPU time for each dataset, 114,844, 85,755, and 5,753,212 computation decreased in 500k, 800k and 1000k datasets, respectively.These computation reductions means that the number of calculation categorical attribute in distance calculation between a point and a center decreases.Decreasing the computation in distance calculation between a point and a center seems to reduce CPU time.The final clustering results of our proposed algorithm is the same as the clustering results of original k-prototypes algorithm.These results lead us to conclude that the proposed algorithm has better performance than the original k-prototypes algorithm.To analyze the difference CPU time for each dataset, 114,844, 85,755, and 5,753,212 computation decreased in 500k, 800k and 1000k datasets, respectively.These computation reductions means that the number of calculation categorical attribute in distance calculation between a point and a center decreases.Decreasing the computation in distance calculation between a point and a center seems to reduce CPU time.The final clustering results of our proposed algorithm is the same as the clustering results of original k-prototypes algorithm.These results lead us to conclude that the proposed algorithm has better performance than the original k-prototypes algorithm.
Conclusions
In this paper, we have proposed a fast k-prototypes algorithm for clustering mixed datasets.Experimental results show that our algorithm is fast than the original algorithm.Previous fast k-means algorithm focused on reducing candidate objects for computing distance to cluster centers.Our k-prototypes algorithm reduces unnecessary distance computation using partial distance computation without distance computations of all attributes between an object and a cluster center, which allows it to reduce time complexity.The experimental shows proposed k-prototypes algorithm improves computational performance than the original k-prototypes algorithm in our dataset.
However, our k-prototypes algorithm does not guarantee that the computational performance will be improved in all cases.If the difference of the first and the second minimum distance between an object and cluster centers for all objects in a given dataset on numerical attributes is less than m, then the performance of our k-prototypes is the same as the original k-prototypes.Our k-prototypes algorithm is influenced by the variance of the numerical data values.The larger variance of the numerical data values, the higher probability that the difference of the first and the second minimum distance between an object and cluster centers is large.
The k-prototypes algorithm proposed in this paper simply reduces the computational cost without using additional data structures and memories.Our algorithm is faster than the original k-prototypes algorithm.The goal of the existing k-means acceleration algorithm is to reduce the number of dimensions to be compared when calculating the distance between center and object, in order to reduce the number of objects compared with the center of the cluster.K-means, which deals only with numerical data, is the most widely used algorithm among clustering algorithms.Various acceleration algorithms have been developed to improve the speed of processing large data.However, real-world data is mostly a mixture of numerical data and categorical data.In this paper, we propose a method to speed up the k-prototypes algorithm for clustering mixed data.The method proposed in this paper is illustrates how k-prototypes algorithm organizes clusters et object.In this figure, an object consists of two numerical attributes and two catego utes.The entire dataset is divided into three clusters, = , , .The center of each cl = (3,3, , ) , = (6,6, , ) , and = (9,4, , ) .The traditional k-prototypes algor ates distance with each cluster center to find the cluster to which = (5, 3, A, B) is assig istance about the numerical attribute of and , , is (3 − 5) + (3 − 3) = 4, (6 − 3) = 10 , (9 − 5) + (4 − 3) = 17 , respectively.The distance about the catego ute of and,,is 1 + 1 = 2 ∵ C ≠ A, ≠ B, 0 + 0 = 0 ∵ A = A, B = B, respecti tal distance ofand , ,, and is 6, 10, and 17, respectively, and the cluster close .Thus, the traditional k-prototypes algorithm computes the distance as a brute force me ompares both numerical and categorical properties.
igure 1 .
A process of assigning an object Xi to a cluster of which the center is the closest to the objects.
Figure 1 .
Figure 1.A process of assigning an object X i to a cluster of which the center is the closest to the objects.
Figure 2 .
Figure 2. Finding the closest cluster center without computing categorical attributes.
Algorithm 3
DIST-COMPUTE-CATE()Input: X i : an object vector, C: cluster center vectors, k: the number of clusters, p: the number of numeric attribute Output: dist_c[] 01: for i = 1 to k do 02:for j = p + 1 to m do 03: if (X[j] = C i [j]) Return dist_c[]
Figure 3 .
Figure 3.Effect of cardinality.FKPT (fast k-prototypes) is the result of our propose k-prototypes algorithm and TKPT (traditional k-prototypes) is the result of original k-prototypes algorithm.
Figure 3 .
Figure 3.Effect of cardinality.FKPT (fast k-prototypes) is the result of our propose k-prototypes algorithm and TKPT (traditional k-prototypes) is the result of original k-prototypes algorithm. | 6,787.2 | 2017-04-21T00:00:00.000 | [
"Computer Science"
] |
A Bibliometric Study : Recommendation based on Artificial Intelligence for iLearning Education
Bibliometric Study : Recommendation based on Artificial Intelligence for iLearning Education. Aptisi Transactions on Technopreneurship (ATT)
Research Method
The phases and disciplines of the Rational Unified Process (RUP) approach were appropriate for developing this project [10] [11] [12]. Hence it was chosen for it. An agile software development methodology called the Rational Unified Process (RUP) divides a project or software development lifecycle into four phases. Creation, development, building, and migration Figure 1. Shows an iterative RUP lifecycle. The project's launch is the primary goal of the commencement phase. We created the UML diagrams using Microsoft Visio and the Unified Modeling Language (UML) as the modeling language. According to the evolving requirements of the iterative process, UML diagrams might be changed [13] [14] [15]. By completing this phase, we know the project's status and whether it should move forward. The system architecture of the project is also specified. This phase results in a defined project goal and area, precise requirements for the project, an initial overview of the LMS use cases and domain model that are only ten complete, and a choice of technologies to use. It must be developed in the application [16] [17] [18].
• Construction phase
The primary subjects of the engineering phase are the comprehensive creation and implementation of the system design. This step extends the refining stage by establishing a working system in the actual world [19] [20] [21]. Moodle currently has a working system that is available. The module has been tested and put to use. After this stage, the system was beta tested, and the subsequent phase could start. The project objectives have been achieved, the requirements and design have been completed, and as a result of this phase, the AI chatbot module has been tested on an open Moodle platform and received user approval. There were tests run, and project documentation and reports were produced [22] [23] [24].
• Transition phase
During the migration phase, the focus was on meeting end-user needs. This result suggests that user feedback is essential for system reviews that address issues such as usability and security. At the end of this phase, you can release your module to your students. At this stage, all bugs were fixed, and a complete user guide for students and teachers was created [25] [26] [27]. Figure 2. Keyword co-occurrence in the author. The maximum is five appearances in a single version. Based on keyword usage, node size is determined. Additionally, the spacing between nodes reveals how closely related co-occurring keywords are. Generated with VOSviewer and shown as lines connecting nodes. Figure 2 In a categorized fashion, list the words that are used the most frequently. Geographic distance classifies phrases used in conjunction into colored clusters when used together. Keywords and lines represent connections. There are seven clusters known. Such as the singular and plural forms of the term "smart contract," groups of almost identical terms are produced. This shows how current the research environment is today. For instance, the IoT, AI, and sharing economy (green) clusters combine supply chain tracking and logistics (red). The placement of the two clusters near one another is expected. Other clusters focus on security, fairness, access control (yellow), privacy related to underlying assets, and game theory (blue). Purple clusters contain keywords associated with cryptocurrencies.
Literature Review • Moodle learning management system
Moodle is an open distribution network founded in 2001 with the help of Dr. Martin Dougiamas. Established. As Internet software, Moodle has limited support and extensions. This place is nothing out of the ordinary. This is due to the Moodle implementation. Users interacting with your version of Moodle will perform more clicks than usual, and Moodle will generate multiple SQL queries when developing pages. This is where Moodle excels. However, what it does is very complicated [28] [29] [30]. In other words, as a developer, I want to know what architecture my extension is most likely to be deployed on. It is also helpful if you still remember the impact your coding has on overall performance. Picture. 1 Below are rare stains. Moodle setup in production. Like any other basic PHP software framework, Moodle has many parts. Databases are essential for scalability and can be easily moved to another physical server. Then you can use a load balancer to load the Front Quit Net server. Developers also need a shared garage for Moodle records while using multiple network servers. Session recordings can be both Moodle recordings and databases [
• Dialogflow and Communicate integration
The primary tool used in this project is Dialog Flow. It is a platform for natural language comprehension that makes it simple to develop conversational user interfaces and incorporate them into gadgets, interactive voice response systems, bots, mobile apps, and more. Give your customers fresh and exciting ways to interact with your product using Dialog Flow (Google Cloud, date unknown). In order to give students who have questions about their curriculum access to online material, Dialog Flow has integrated Google Chatbot into his Moodle platform [37] [38] [39]. The goal of the proposed integration is to assist/support higher education institutions in making informed selections and providing prompt (real-time) responses to inquiries and requests of interest. Pay attention to student needs, respond to inquiries, and discuss research, admissions, and student life. Communicate wants to make it possible for companies to develop enduring customer relationships that lead to expansion. Developers can use Communicate to centrally manage customer conversations and develop and integrate chatbots, website chats, support agents, team meetings, and services that increase customer happiness. Increase. I can manage it. To win. Four key activities were involved in integrating a chatbot through a cloud platform. The cloud workflow is displayed in
Problem
The primary technological risks encountered frequently in work are the focus of this phase. Approximately half of the user requirements for an LMS system and those contained in the sequence diagram and activity diagram can be finished in this phase. Similarly, roughly half of the plugin modules can be finished, including those for planning and job completion and use cases and domain models [43] [44].
Research Implementation
This section covers implementing and testing an AI chatbot, a cloud-integrated application that disseminates knowledge on academic subjects. The steps of implementation and testing are critical to the overall process of system development.
Students can inquire about this project using intents and entities that developers have trained. Below is an example process that allows students to download their academic Recommendation based on Artificial Intelligence for iLearning Education.... ■ 117 calendar directly by clicking a link provided by a chatbot. Chatbots provide answers based on user input, but chatbots answer first. Greet the user before starting the conversation.
Conclusion
The LMS platform is now widely used because most students start their studies online. Many of its LMS platforms used worldwide depend on University/College implementation. Therefore, new features are developed daily and uploaded to the LMS platform. As long as students have an internet connection and a student ID card to access the dashboard, they can communicate with the chatbot anytime, anywhere, to meet their learning needs. They can also explain the results or temptations of the task objectives and the task execution plan. Task designs can be identified, learning objectives and scope can be determined, task-related literature is surveyed, system analysis and development methodologies are determined and discussed, and system implementation and testing are carried out. | 1,738.8 | 2022-11-30T00:00:00.000 | [
"Computer Science",
"Education"
] |
Amyloid β 1-42 induces hypometabolism in human stem cell-derived neuron and astrocyte networks
Alzheimer's disease (AD) is the most common form of dementia, affecting more than 35 million people worldwide. Brain hypometabolism is a major feature of AD, appearing decades before cognitive decline and pathologic lesions. To date, the majority of studies on hypometabolism in AD have used transgenic animal models or imaging studies of the human brain. As it is almost impossible to validate these findings using human tissue, alternative models are required. In this study, we show that human stem cell-derived neuron and astrocyte cultures treated with oligomers of amyloid beta 1-42 (Aβ1-42) also display a clear hypometabolism, particularly with regard to utilization of substrates such as glucose, pyruvate, lactate, and glutamate. In addition, a significant increase in the glycogen content of cells was also observed. These changes were accompanied by changes in NAD+/NADH, ATP, and glutathione levels, suggesting a disruption in the energy-redox axis within these cultures. The high energy demands associated with neuronal functions such as memory formation and protection from oxidative stress put these cells at particular risk from Aβ-induced hypometabolism. Further research using this model may elucidate the mechanisms associated with Aβ-induced hypometabolism.
INTRODUCTION
It is now widely accepted that Alzheimer's disease (AD) is accompanied by hypometabolism, of differing severity in different regions of the brain. Crucially, signs of hypometbolism such as a reduction in central nervous system glucose utilization as well as mitochondrial function, begin decades before any symptoms or histopathologic changes appear, making such events useful biomarkers of AD risk. 1 In addition, reductions in key mitochondrial enzyme complex activities such as the α-ketoglutarate dehydrogenase complex have been observed. 2 One obvious explanation for the observed reduction in glucose utilization could be neuronal loss, which is observed in AD patients. However, many studies have shown a reduction in the cerebral metabolic rate of glucose before the onset of the disease and subsequent cell loss. This has been seen in individuals at risk of developing AD such as APOE4 carriers, those carrying autosomal dominant mutations linked with familial AD or patients with mild cognitive impairment. 3 As hypometabolism is the earliest significant event linked with AD, it suggests that changes in energy metabolism precede neuronal loss and may actually contribute to the development and progression of the disease.
To investigate these changes, a number of in vivo studies have attempted to model changes in cerebral glucose utilization in transgenic mice. However, conflicting results have been reported and were dependent on the model used. Studies using models that overexpress the Swedish and Indiana mutations of human amyloid precursor protein 4 as well as the 3xTG model of AD, which overexpresses human amyloid precursor protein, PS1, and tau mutations 5 have showed reductions in cerebral glucose uptake. While studies using the Tg2576 mouse model that over expresses the Swedish mutation of amyloid precursor protein, 6 as well as a model that overexpresses the Swedish and London mutations of amyloid precursor protein, along with PS1 7 have all displayed an increase in brain glucose uptake.
Furthermore, reductions in glucose uptake in response to Aβ have also been reported in vitro using primary rat hippocampal neurons, 8,9 while in contrast, an increase in glucose uptake has been reported in primary mouse astrocytes. 10 Difficulties in obtaining viable human tissue from living patients have made it almost impossible to validate these findings in human neurons. It is clearly important to be able to model glucose metabolic changes in tissues and experimental models that are directly relevant to the disease. While rodent and human brain share a number of common features, rodent models do not naturally develop AD; indeed, it is important note the differences between the human and rodent central nervous system. Human cortical astrocytes have been shown to be both larger and structurally more complex and diverse than those of rodents. 11 These findings highlight fundamental differences between species used to study human diseases and show the importance of developing functional human models that can be used in comparative studies.
In this way, it may be possible not only to replicate the findings observed in human patients but also to test novel hypotheses that may predict changes in human brain tissue. As such, human stem cell-derived models may provide the most simplistic and relevant platforms to study cell-cell interaction in AD. To this end, we have used human NT2.D1 stem cell-derived neuron and astrocyte cocultures to determine the effects of exposure to oligomeric Aβ1-42 on cellular metabolism and oxidative stress in comparison with primary rat hippocampal cultures.
The NT2.D1 embryocarcinoma cell line is well characterized and generates neuronal (NT2.N) cells containing heterogeneous subpopulations of dopaminergic, cholinergic, GABAergic, and glutamatergic neurons [12][13][14][15] as well as astrocytic (NT2.A) cells. 16,17 These cells have been shown to generate action potentials on depolarization 18 and form functional spontaneously active neuronal networks. 19 We have shown recently that these cells show metabolic coupling and can link neuronal activity to changes in astrocytic metabolism. 20 Critically, increasing numbers of studies have highlighted the important role of astrocytes in AD. 21 Indeed, astrocytes in the healthy brain enhance neuronal survival, axonal growth, synaptogenesis, as well as neurovascular and neurometabolic coupling; hence, their inclusion confers structural and functional relevance in any model of brain function.
Crucially, recent studies have also showed the role of human astrocytes early in AD 22 and have reported metabolic changes after treatment of primary mouse astrocyte cultures with Aβ oligomers 10 thus supporting the inclusion of astrocytes into in vitro models of AD.
As changes in brain metabolism occur decades before any other symptoms, the development of relevant human models of the earliest significant metabolic foundations of the disease may provide important future insights into disease process milestones, that may one day be therapeutic targets. Therefore, this study focused on using cocultures of human NT2-derived neurons and astrocytes to test the hypothesis that Aβ alters cellular metabolism in these cells.
Aggregation Protocol
Hexafluoroisopropanol (HFIP) treated Aβ1-42 (Anaspec, Freemont, CA, USA) was resuspended in 200 mmol/L HEPES, pH 8.5 to a concentration of 100 μmol/L. Treatment with HFIP has previously been shown to dissolve higher aggregates, eliminating the 'nucleating seeds' and removing any secondary or tertiary structures. 23 The aliquots were stored at − 80°C and used at working concentrations of 20, 2, and 0.2 μmol/L. These concentrations were selected based on the previous research demonstrating the toxicity of Amyloid oligomers in primary cultures. 23,24 In addition, these concentrations are within the micromolar concentration range of soluble and insoluble Aβ1-42 levels reported in AD patient brain tissue. 25 Cell Culture Human teratocarcinoma NT2.D1 cells used in this study were kindly donated by Professor Andrews (University of Sheffield, UK). The cells were cultured in DMEM (Dulbecco's Modified Eagle Medium) high glucose, containing, Glutamax, (Life Technologies, Paisley, UK), 10% heat inactivated fetal bovine serum (Life Technologies), 100 units/mL penicillin, and 100 μg/mL streptomycin. NT2.D1 cells were differentiated according to the previously published methods. 26 Briefly, NT2.D1 cells were differentiated using DMEM Glutamax high glucose medium with pyruvate, supplemented with 10% (v/v) fetal bovine serum, 100 units/mL penicillin, 100 μg/mL streptomycin (pen/strep), and 1 × 10 − 5 M all-trans retinoic acid (RA) (Sigma-Aldrich, Dorset, UK) for 4 weeks. Differentiated cells were seeded at a lower density, then dislodged and re-seeded onto CellBIND 12-well plates (Corning, New York, USA) at 1.25 × 10 6 cells/well and treated with medium containing antiproliferative agents (MI) for 28 days; 0.1 μmol/L cytosine arabinoside (for the first 7 days only), 3 μmol/L fluorodeoxyuridine, and 5 μmol/L uridine. Pure astrocytic cultures were produced as previously described. 26 Briefly, NT2.N/A cells were dissociated into a singlecell suspension using Accutase (PAA Laboratories, Yeovil, UK) and subsequently seeded onto fresh CellBIND 12-well plates in MI-free medium. After incubation for 2 hours, the plates were shaken briefly using a plate shaker (400 r.p.m.) to dissociate loose NT2.N cells from the more adherent NT2.A cells leaving an astrocytic mono-layer after rinsing. All cells were maintained by incubation at 37°C in a humidified atmosphere of 5% CO 2 . Unless otherwise stated, all experiments were performed at 37°C in a humidified atmosphere of 5% CO 2 . The proportion of cell types produced by this method in this study were in agreement with previously published values (33 ± 4% neurons and 63 ± 4% astrocytes). 26 Primary Cortical Cultures All procedures were approved by Aston University Bioethics committee, and performed in accordance with the UK Animals (Scientific Procedures) Act 1986 and associated procedures. Cortical cultures were prepared from 2-to 3-day-old male Wistar rats. Animals were anesthetized with isoflurane and euthanized by cervical dislocation. After removal, the brain was placed in ice-cold Gey's salt solution (Sigma-Aldrich) containing 20 μg/mL gentamycin (Life Technologies). The cortex was dissected and placed in icecold Gey's salt solution containing 20 μg/mL gentamycin. The tissue was then minced using a scalpel and placed in Ca 2+ -free and Mg 2+ -free Hanks' buffered saline solution (Life Technologies), containing 0.1% trypsin (Life Technologies) for 30 minutes at 37°C. The trypsin was inactivated by adding Neurobasal medium (Life Technologies) containing B27 (Life Technologies), 100 units/mL penicillin and 100 μg/mL streptomycin and 10% horse serum (Life Technologies). Cells were then centrifuged at 258 g for 5 minutes and then medium was replaced with 5 mL of fresh Neurobasal medium. Cells were dissociated by trituration with a glass Pasteur pipette with a flamerounded tip and passed through 70 μm filter (BD Biosciences, Oxford, UK). Cells were counted using a hemocytometer and plated onto poly-D-lysine coated 12-well plates at a final concentration of 5 × 10 5 cells/mL. Cells were maintained at 37°C and 5% CO 2 and fed twice a week, cells were used after 5 days. Plates for primary cortical cultures were coated with poly-D-lysine (Sigma-Aldrich) at a concentration of 50 μg/mL. Briefly, poly-D-lysine was resuspended in sterile H 2 O and filtered through 0.22 μm filter, the wells were coated with 2 mL of the solution and incubated at 37°C overnight. The poly-D-lysine was aspirated and plates were rinsed with sterile H 2 O and dried.
Cell Viability
The viability of the cultures after treatment with Aβ1-42 was determined using the Cell-titre Blue assay (Promega, Southampton, UK). After experimental treatment, medium was removed from the wells of the 12well cell-culture plate. Subsequently, the plate was washed with 500 μL phenol red-free DMEM media (Life Technologies), supplemented with 10% heat inactivated fetal bovine serum (NT2.N/A, NT2.A) or 10% horse serum (primary cortical cultures), 100 units/mL penicillin and 100 μg/mL streptomycin and 2 mmol/L L-glutamine. In all, 1 mL of Cell-titre blue reagent was mixed with 10 mL of the DMEM. In all, 500 μL of this solution was added to each well of the plate. The plate was incubated for 3 hours at 37°C. After incubation, medium was transferred to a 96-well plate and the absorbance was measured at 590 nm using a Thermo multiscan EX 96-well plate reader (Thermofisher, Loughborough, UK).
Determination of Carbohydrate Levels
The levels of glycogen in biological samples were determined using the method described by Nahorski and Rogers. 27 Lactate was measured using the Fluorescent Lactate Assay Kit (Abcam, Cambridge, UK) according to the manufacturer's instructions. Pyruvate was measured using the Pyruvate assay kit (Abcam) according to the manufacturer's instructions. Glucose levels were measured using Glucose (HK) Assay Kit (Sigma-Aldrich) according to the manufacturer's instructions.
Statistics
Results were expressed as the mean of three samples ± standard error of the mean (s.e.m.). Comparisons between treatments were performed using analysis of variance (ANOVA) followed by Dunnett's or Tukey's post test or where appropriate Student's T-test using GraphPad Prism Software (GraphPad Prism software, La Jolla, CA, USA). Differences were considered as significant for P values o0.05.
Characterization of Cells
After differentiation, the cultures began to display distinct neuronal and astrocytic morphology. Neurons extended axons and dendrites, and astrocytes with projections appeared in close proximity to aggregations of neuronal perikarya and neurites throughout the culture. Under the microscope astrocytes were identified by their flat phase dark appearance, while neurons were typically phase bright and often seen on top of the astrocytic monolayer. Identification was confirmed using immunohistochemistry for the specific markers GFAP and β-tubulin ( Figures 1A and 1B).
Primary mixed glial and neuronal cultures were prepared from cortices of Wistar rat pups and maintained in Neurobasal media.
These cultures produced a mixed culture of astrocytes and neurons ( Figures 1C and 1D).
Amyloid Aggregation
To assess the composition of each preparation, freshly prepared samples of Aβ at a concentration of 20 μmol/L were separated using SDS and Native polyacrylamide gel electrophoresis and visualized using western blotting ( Figure 2A). Aβ(1-42) prepared in 100 mmol/L HEPES at pH 8.5 showed a large proportion of monomers at around~4 kDa and high n-oligomers at the top of the gel (~130 to 250 kDa) ( Figure 2A).
The Effects of Amyloid on Cell Viability
To determine the temporal and concentration-dependent effects of Aβ on NT2.N/A cocultures, NT2.A cultures, and mixed rat glial and neuronal cultures, cells were treated with different concentrations of Aβ1-42 oligomers (0.2, 2, and 20 μmol/L) for 6, 24, 48, 72, and 96 hours. After treatment, the viability of the cells was measured indirectly using Cell-titre blue assay (Promega).
In cocultures the only significant change was seen at the highest concentration of Aβ1-42 when compared with the untreated control ( Figure 2B). Application of 20 μmol/L Aβ caused an increase in reduction of resazurin at 6 hours (112 ± 4.2%, P o0.05) and a decrease at 48 hours (89 ± 4.3%, P o 0.05) compared with control. In the pure astrocytic cultures, there was a significant increase in the reduction of resazurin in the cells treated with 0.2 μmol/L Aβ for 72 hours (106 ± 2.1%, P o 0.05) ( Figure 2C). Similarly the primary cortical cultures did not show any significant cell death over time ( Figure 2D). After 24 hours treatment with 20 and 0.2 μmol/L the cultures showed an increase in the reduction of resazurin (20 μmol/L: 107 ± 1.7, Po 0.05; 0.2 μmol/L: 108 ± 1.3%, P o0.05). The same increase in the reduction of resazurin was seen after 72 hours treatment with the highest concentration of Aβ (106 ± 0.99%, P o0.05).
Glucose Uptake Is Decreased After Treatment with Amyloid Beta
To determine the effect of amyloid on glucose uptake, the glucose concentration of media from Aβ-treated cells was measured using the Glucose (Hexokinase) Assay Kit (Sigma-Aldrich). NT2.N/A, NT2.A and primary cortical cultures all showed a significant decrease in glucose uptake after treatment with 2 and 0.2 μmol/L Aβ at all time points investigated. Glucose levels in the medium from cocultures were significantly increased (P o0.001) at all time points ( Figure 3A). Similar increases in the glucose content of culture media were also seen in primary cortical cultures (P o 0.001) at all time points with the exception of 2 μmol/L Aβ treatment at 6 hours ( Figure 3C). Astrocytic cultures also showed a decrease in glucose uptake, though to a lesser extent than neuronal and astrocytic cocultures ( Figure 3B). At 24 hours, the glucose levels in the media were significantly increased after the treatment with Aβ (control: 10.2 ± 0.29 mmol/L; 2 μmol/L: 12.3 ± 0.09 mmol/L, Po 0.001; 0.2 μmol/L: 12.5 ± 0.25 mmol/L, P o0.001). The decrease in glucose uptake became less significant over time in astrocytes. At 72 hours, the increase in glucose content in culture media is at P o 0.05 for both concentrations and at 96 hours only 0.2 μmol/L Aβ treatment had any substantial impact (P o 0.05).
The glucose uptake over time differed between the cultures. In all cases the starting concentration of glucose in the medium was 25 mmol/L. The NT2.N/A cultures used up over 50% of the available glucose in the first 6 hours (control: 10.1 ± 0.19 mmol/L). Pure astrocytic cultures took up less glucose than cocultures (control: 11.5 ± 0.36 mmol/L; P o 0.05) while uptake in primary cultures was even slower (control: 16.4 ± 0.31 mmol/L; P o 0.0001).
Intracellular Glucose and Glucose-6-Phosphate Is Increased After Treatment with Amyloid Beta Intracellular glucose and glucose-6-phosphate levels were measured after the treatment with Aβ. As in previous experiments the changes in glucose levels were restricted to the two lower concentrations of Aβ, 2 and 0.2 μmol/L. The NT2.N/A cocultures showed an accumulation of glucose and glucose-6-phosphate at all time points ( Figure 3G). The increase in glucose content was most significant at 6 hours (control: 116 ± 5.61 nmol/mg protein; 0.2 μmol/L: 172 ± 9.06 nmol/mg protein; P o0.01) and 24 hours time point (control: 84.8 ± 3.9 nmol/mg protein; 0.2 μmol/L: Figure 3H). Similarly primary cortical cultures also showed a very significant (P o 0.001) accumulation at all time points after treatment with Aβ ( Figure 3I).
Intracellular Glycogen Is Increased after Treatment with Amyloid Beta
The levels of glycogen within the control cultures were stable throughout the incubation period ( Figure 3D). After the treatment of the cultures with Aβ, glycogen levels inside the cells were measured at 6, 24, 48, 72, and 96 hours. NT2.N/A cocultures showed an initial decrease in glycogen levels at 6 hours compared with control (0.2 μmol/L: 96.1 ± 0.47%, P o 0.001; 2 μmol/L: 96.6 ± 0.15%, P o 0.001) ( Figure 3D). Glycogen levels increased to control levels at 24 hours, with significant (Po 0.001) increases at 72 hours to 129 ± 5.57% when treated with 2 μmol/L Aβ and to 123 ± 3.94% with 0.2 μmol/L Aβ. At 96 hours, the differences were less apparent with only 0.2 μmol/L Aβ having a significant effect (113 ± 4.32%, P o 0.05) ( Figure 3D). Primary cortical cultures showed a similar pattern, with glycogen levels increasing at 48 hours. However, the increase in glycogen was more significant and at 96 hours levels of glycogen reached 277 ± 16.5% (P o 0.001) after treatment with 2 μmol/L Aβ and 259 ± 19.9% (P o 0.001) when treated with 0.2 μmol/L Aβ ( Figure 3F). Additionally, from 48 hours there was a decrease in glycogen levels in cells treated with 20 μmol/L Aβ which was significant at 48 hours (P o 0.001) and 72 hours (P o 0.001).
Pure astrocytes also showed an increase in glycogen levels from 6 hours (2 μmol/L: 130 ± 7.1%, P o0.01) ( Figure 3E). Increases were observed at all time points, but were only significant at 24 and 96 hours.
Lactate was found to accumulate over time in all cultures. In NT2.N/A cultures and primary cultures, the lactate levels were much higher than in pure astrocytes. Additionally, in NT2.N/A cocultures the accumulation of lactate was slower than in primary cultures. NT2.N/A lactate reached 19.95 ± 0.18 mmol/L levels at 96 hours while primary cultures attained 20.6 ± 0.8 mmol/L levels at 48 hours after which time the levels remained stable.
Pyruvate Levels in the Cell Conditioned Media Are Significantly Increased After Treatment with Amyloid Beta
The levels of pyruvate in the cell culture media were measured after treatment with Aβ. The initial pyruvate concentration in NT2. N/A and NT2.A medium was 1 mmol/L while in primary cortical cultures the initial pyruvate concentration was 200 μmol/L.
Glutamate Levels in the Cell Conditioned Media Are Significantly Decreased after Treatment with Amyloid Beta Treatment with Aβ was also found to have an effect on glutamate levels in the media. In NT2.N/A cocultures, there was a significant decrease in glutamate levels from 24 hours with the largest decrease at 72 hours and 96 hours (control: 22.3 ± 1.19 μmol/L; 2 μmol/L: 13.3 ± 0.41 μmol/L, P o0.001; 0.2 μmol/L: 13.05 ± 1.52 μmol/L, Po 0.001) ( Figure 4G). This decrease became more significant with time, as the control and 20 μmol/L Aβ-treated cells accumulated more glutamate in the media. In pure astrocytes, the decrease was also present but to a lesser extent ( Figure 4H). Overall, the levels of glutamate in the media samples were markedly lower in pure astrocytic cultures than cocultures of neurons and astrocytes. Primary cortical cultures also showed a significant decrease in glutamate levels after treatment with 2 and 0.2 μmol/L Aβ. The decrease was significant from 24 hours and most significant at 96 hours (control: 9.08 ± 0.34 μmol/L; 2 μmol/L: 5.51 ± 0.37 μmol/L, P o 0.001; 0.2 μmol/L: 5.35 ± 0.43 μmol/L, Po 0.001) ( Figure 4I).
The Cellular GSH/GSSG Ratio Is Significantly Decreased after Treatment with Amyloid Beta
The total GSH and GSSG levels of cultures varied dependent on the length of incubation and treatment (Supplementary material). NT2.N/A cocultures treated with all concentrations of Aβ showed a significant decrease in the GSH/GSSG ratio at 6 and 24 hours ( Figure 5A). Pure astrocytes also showed a decrease in GSH/GSSG ratio from the 6-hour time point ( Figure 5B). This decrease was significant up to 48 hours time point. After 72 hours, the GSH/GSSG ratio increased to control ratio (control: 18.4 ± 1.67; 20 μmol/L: 18.7 ± 1.8; 2 μmol/L: 23.1 ± 1.8; 0.2 μmol/L: 21.8 ± 1.22). At 96 hours, there was a further increase in the ratio ( Figure 5B), which was significant at all three concentrations (P o0.05). In NT2.N/A after treatment with Aβ GSSG levels varied from 2.74% to 5.71% of the total GSH-GSSG levels, compared with the controls 1.75% to 2.89%. In NT2.A after treatments GSSG levels were higher varying from 3.67% to 18.2%, with controls 4.76% to 10.7%.
The Cellular NAD+/NADH Ratios and ATP Levels Are Significantly Decreased after Treatment with Amyloid Beta NAD + /NADH ratios and ATP levels were measured in NT2.N/A cocultures after treatment with 2 μmol/L Aβ. The NAD + /NADH ratio increased at 24 hours and was followed by a steady decrease at 48, 72, and 96 hours ( Figure 6A). Both Aβ treated and control cells followed this trend. There was an initial increase in NAD + / NADH ratio after the treatment with Aβ at 24 hours (control: 0.66 ± 0.06; 2 μmol/L: 0.91 ± 0.02, P o 0.05). However, after this time point the ratio was much lower after the Aβ treatment and became significantly lower at 96 hours (control: 0.52 ± 0.03; 2 μmol/L: 0.18 ± 0.01, P o0.001). At 96 hours NAD + levels in control were 820 ± 17.5 pmol/mg of protein, while after treatment with 2 μmol/L, Aβ levels decreased to 348 ± 20.2 pmol/mg of protein. ATP levels after treatment with 2 μmol/L Aβ decreased from 6 hours, becoming significant at 24 hours ( Figure 6B). ATP levels reached their lowest levels at 72 and 96 hours (control: 8.34 ± 0.86 nmol/mg protein; 2 μmol/L: 3.29 ± 0.20 nmol/mg protein, P o 0.01).
DISCUSSION
Difficulties in studying the cellular mechanisms of AD in living patients combined with shortcomings in the relevance of animal models have confined much AD research to the end-stage pathologies (Aβ plaques and Tau tangles). Clinical imaging approaches in mild cognitive impairment or early-stage AD have provided intriguing insights into the early stages of AD but have been unable to explore these changes at the cellular level.
In this study, we provide the first description of the impact of Aβ on glucose homeostasis in neurons and astrocytes derived from human stem cells. Furthermore, we show changes in the utilization of metabolites such as lactate, pyruvate, and glutamate. In addition, we show an increase in the storage of glycogen as well as changes in the NAD/NADH ratio as well as ATP levels within cells. Since hypometabolism is a key feature of AD, this model may provide a complex functional in vitro system for studying early changes in disease pathology.
Human stem cell-derived cocultures of neurons and astrocytes successfully modelled the earliest metabolic changes induced by Aβ. In addition, results obtained from these cells were compared with a well-characterized primary rat cortical tissue culture preparation. As the NT2 line is a form of human stem cell, our protocol concept in this report should inform the design of future studies using patient-derived induced pluripotent stem cells. NT2 cultures treated acutely with exogenous Aβ in this study show 'metabolic toxicity' over a period of hours/days. As such changes in metabolic homeostasis are observed before the development of symptoms in patients, it would be of great interest to model the long-term effects of Aβ on cellular viability in culture. However, caution should be taken with the interpretation of this response when modelling a disease that may take years/decades to develop in vivo. In addition, the concentration of glucose in the media used (25 mmol/L) may impact upon the metabolism of cells and therefore their survival after Aβ treatment. Future studies will determine the effect of low glucose concentrations on cell survival.
After exposure to Aβ, the cultures showed a decrease in the uptake and/or utilization of glucose. A decrease in glucose uptake upon treatment with 2 and 0.2 μmol/L Aβ1-42 could be explained by the accumulation of glucose and glucose-6-phosphate that was observed intracellularly. This could lead to a reduction in the flow of glucose through the hexokinase pathway due to allosteric feedback inhibition of hexokinase. Indeed amyloid itself may decrease hexokinase activity via interference with hexokinase activity and subcellular localization. 29 The increase in glycogen levels after the treatment with 2 and 0.2 μmol/L Aβ1-42 could also be explained by accumulation of glucose-6-phosphate, which instead of passing through glycolysis, is directed through glycogen synthesis pathways. However, very little is understood about the storage and metabolism of glycogen during AD and further research is required to understand the impacts of these changes on cellular processes.
In this report, glutamate production in both rat cortical and human NT2.N/A cocultures was shown to be affected after treatment with Aβ1-42. In addition, overall glutamate levels in pure NT2-A astrocytic cultures were lower compared with those of cocultures. This is not unexpected as the majority of glutamate is produced by neurons, while astrocytes take up the glutamate released by neurons and convert it to glutamine. Previous studies of AD patient brains have showed a decrease in glutamine synthase, as well as a marked decrease in phosphate-activated glutaminase levels and α-ketoglutarate dehydrogenase complex. 2,[30][31][32][33] Such changes could conceivably lead to disturbances in glutamate production and recycling in vivo. Indeed, studies using magnetic resonance spectroscopy have showed that patients with AD have lower levels of brain glutamate compared to controls with mild cognitive impairmen patients demonstrating intermediate levels. 34 After exposure to Aβ1-42, lactate levels were significantly decreased in both coculture systems. Interestingly, this effect is not seen in pure astrocytes. The results shown herein strongly suggest that Aβ1-42 leads to inhibition of the glycolytic pathway, which results in the accumulation of glucose/glucose-6-phosphate, decreased uptake of glucose and therefore possibly lactate production. The pathway involved in the production of lactate by astrocytes in the brain is unclear. According to the Astrocyte Neuron Lactate Shuttle hypothesis, lactate production in astrocytes is triggered in response to glutamate uptake. 35 Uptake of pyruvate, which has been shown to be an important energy source in isolated mitochondria from brain, 37 was also decreased after treatment with low levels of Aβ1-42 in NT2.N/A cocultures and primary cortical cultures, further limiting availability of substrates for the TCA cycle. However, the relevance of this finding to the whole brain is unclear with brain cells relying almost exclusively on glucose to meet its energy demands.
Interestingly, higher concentrations of amyloid (20 μmol/L) did not have the same effects on metabolism as lower concentrations (2 and 0.2 μmol/L). The kinetics of protofibril formation are greatly dependent on ambient conditions such as temperature, buffer composition, and peptide concentration. 38 It is possible that higher concentrations of amyloid used in this study may aggregate faster and thus produce amyloid species with altered toxic effects. As such, it is conceivable that such species may not initiate the metabolic effects observed in these cultures after treatment with lower concentrations.
As the treatment of cells with Aβ1-42 leads to a decrease in the substrates available for entry into the TCA cycle, changes in the ratio of NAD + /NADH were investigated. Cocultures showed an initial increase in NAD + /NADH ratio in both control and Aβ-treated cells and this was followed by a significant decrease, particularly in Aβ-treated cells. Increases in the NAD + /NADH ratio seen in the control cultures is possibly a result of increased glycolysis and subsequent production of lactate after the application of fresh media.
During glycolysis NAD + is reduced to NADH. Normally regeneration of NAD + takes place in mitochondria via the malate/aspartate shuttle. However, it has been shown that astrocytes express low levels of a key component of this shuttle and may instead use the conversion of pyruvate to lactate by lactate dehydrogenase to regenerate NAD. 39,40 A decrease in the availability of substrates (glucose, pyruvate) and an increase in the requirement for NAD + can lead to NAD + depletion. In addition, decrease in NAD + stores leads to a heavy demand on ATP stores for synthesis of new NAD + , 41 which in turn could provoke an energy crisis in the cell. 42 In this study, treatment of NT2.D1-derived cocultures with 0.2 μmol/L Aβ1-42 caused a significant decrease in ATP levels from 6 hours onwards in comparison with the control, further highlighting the negative impact of the peptide on cellular energy metabolism.
Dringen et al 43 have showed that detoxification of peroxide by neurons is less efficient than astrocytes, indicating an increase in the susceptibility of neurons to oxidative stress compared with astrocytes. As oxidative stress has been associated with AD, NT2.N/A and NT2.A cultures were investigated for increases in GSH oxidation in response to oxidative stress after treatment with Aβ1-42. Pure astrocytic cultures showed a decrease in GSH/GSSG ratio. After 48 hours, the ratio increased suggesting a recovery from the initial insult. A similar decrease in the GSH/GSSG ratio was seen in cocultures at 6 and 24 hours. A study by Abramov et al 44 performed on rat mixed hippocampal neuronal and glial cultures as well as on pure cortical astrocytes has linked sporadic Ca 2+ signals caused by Aβ to GSH depletion. Such Ca 2+ fluctuations were only seen in astrocytes and were associated with a decrease in GSH in those cells. Astrocytes in these cultures were found to withstand this insult and were resistant to cell death, while neurons were found to die within 24 hours after exposure, highlighting the importance of cocultures. 44 Results described by Dringen et al 43 and Abramov et al 44 also suggest that a compromised astrocytic glutathione system may contribute to a lower defence capacity of the brain against reactive oxygen species. As hypoglycaemic conditions may deplete NADPH levels, this may in turn increase cells susceptibility to oxidative damage due to a limited ability to reduce GSSG to GSH.
As numerous cell types interact during the progression of AD, it is important that any in vitro model of the disease reflects this arrangement. Data presented here provide evidence for a detrimental effect of Aβ on carbohydrate metabolism in both neurons and astrocytes. Replication of results using a wellestablished primary cortical culture model system shows the utility of stem cell-derived neurons and astrocytes for studying hypometabolism observed in AD. As a purely in vitro system, human stem cell models can be readily manipulated and maintained in culture for a period of months without the use of animals. In our laboratory, the NT2 model can be maintained in culture for over 6 months thus providing the opportunity to study the consequences of these changes over extended periods of time relevant to aspects of the disease progression timeframe in vivo. In addition, their human origin provides a more realistic in vitro model as well as informing other human in vitro models such as patient-derived induced pluripotent stem cells. Future studies will attempt to investigate the earliest cell-specific changes that lead to hypometabolism using this model as well as the effects of abnormal metabolism on neuronal and astrocytic activity.
DISCLOSURE/CONFLICT OF INTEREST
The authors declare no conflict of interest. | 7,477.4 | 2015-04-08T00:00:00.000 | [
"Biology"
] |
Copper Utilization, Regulation, and Acquisition by Aspergillus fumigatus
Copper is an essential micronutrient for the opportunistic human pathogen, Aspergillus fumigatus. Maintaining copper homeostasis is critical for survival and pathogenesis. Copper-responsive transcription factors, AceA and MacA, coordinate a complex network responsible for responding to copper in the environment and determining which response is necessary to maintain homeostasis. For example, A. fumigatus uses copper exporters to mitigate the toxic effects of copper while simultaneously encoding copper importers and small molecules to ensure proper supply of the metal for copper-dependent processes such a nitrogen acquisition and respiration. Small molecules called isocyanides recently found to be produced by A. fumigatus may bind copper and partake in copper homeostasis similarly to isocyanide copper chelators in bacteria. Considering that the host uses copper as a microbial toxin and copper availability fluctuates in various environmental niches, understanding how A. fumigatus maintains copper homeostasis will give insights into mechanisms that facilitate the development of invasive aspergillosis and its survival in nature.
Introduction
All living organisms have evolved to maintain metal homeostasis as many metal ions are essential for certain biological processes, but high concentrations of these very same ions are toxic. Transition elements such as iron, copper, nickel, and zinc are required as cofactors for processes such as electron transfer or for maintaining protein structure. Estimates show that approximately 50% of all proteins require some metal cofactor to function [1,2]. However, these elements have a darker side. If microbes are not able to carefully regulate homeostasis of the metal, it can quickly become toxic, damaging macromolecules, affecting homeostasis of other metals, and ultimately killing the cell.
Metal homeostasis is also critical in host/pathogen interactions. The best studied transition metal in that regard is iron [3,4]. However, several recent studies have highlighted the role of copper in these interactions. Copper is a transitional metal that has two biologically relevant oxidation states, Cu + and the more soluble Cu +2 . Of all the cellular transition metals, copper establishes the most stable ligand complexes according to the Irving-Williams series [5], resulting in a high tendency to compete with and displace other metal cofactors [6]. Copper also participates in the generation of reactive oxygen species (ROS) through Fenton chemistry, a process host cells use to kill pathogenic microbes [7]. In turn, microbes have evolved several mechanisms to counteract host copper-mounted attacks [8].
Here we seek to provide a comprehensive synopsis of the current state of knowledge of how the opportunistic human pathogen Aspergillus fumigatus regulates copper homeostasis and the role of copper in host-fungal interactions. This pathogen causes invasive aspergillosis (IA) in immunocompromised hosts, with mortality rates as high as 90% [9]. The study of copper homeostasis in fungi was initiated in the model yeast Saccharomyces cerevisiae [10] and Schizosaccharomyces pombe [11]. Subsequent work has mainly focused on copper regulation in the pathogenic fungi Cryptococcus neoformans, Candida albicans, and recently A. fumigatus [11,12]. We compare and contrast the system in these fungi and provide a view of the potential small molecule copper biology of A. fumigatus.
Copper Metabolic Processes
All living organisms require copper for several biological processes ranging from cofactor requirements for enzyme activity to transcription factor functionality. Delivery of the copper cofactor to specific proteins is a function of copper chaperones, although compared to the number of copper-containing proteins there have been only a few chaperones demonstrated to be responsible for delivery of the cofactor.
Enzymes Using Copper as A Cofactor
Possibly the most important role of copper is its requirement for heme-copper oxidases such as cytochrome c oxidase [13]. Cytochrome c oxidase, CycA in A. fumigatus, is a ubiquitous protein in eukaryotes that is essential for facilitating energy generation by catalyzing the final step in the electron transport chain [14]. Inhibition of the cytochrome c oxidase complex assembly in mammalian cells or loss of Rcf1 and Rcf2, yeast proteins supporting the cytochrome c oxidase complex, results in defects in aerobic respiration [15] indicating the importance of the complex in energy generation [14]. Superoxide, a toxic ROS by-product of respiration [16,17] is detoxified by copper-containing superoxide dismutases (SODs). Fungi use these copper SODs to convert the superoxide into hydrogen peroxide, which is subsequently converted into water and oxygen by an iron-dependent catalase [18,19].
Copper is also essential for the uptake and utilization of other nutrients. The metabolism of a variety of nitrogen sources requires a copper-dependent process. Primary amines, a source of nitrogen, can only be utilized as a nitrogen source by an amine oxidase, a protein requiring copper as a cofactor. The A. fumigatus genome contains five putative copper-binding amine oxidases, none of which have been thoroughly investigated. Some nitrite reductases utilize copper as a cofactor and reduce nitrate to nitrite, another step in utilizing nitrogen for amino acid biosynthesis [20]. A. fumigatus contains a known nitrite reductase, NiiA, that has not been shown to bind copper, but contains another putative nitrite reductase ( Table 1) that contains a copper-binding motif. Nitrite reductase also plays an important role in cell communication and development as nitrous oxide is a major signaling molecule described in bacteria, plants, fungi, and mammals [21,22]. Reductive iron uptake, the process by which A. fumigatus reduces ferric iron (Fe +3 ) to the more soluble ferrous iron (Fe +2 ) is most likely catalyzed by the copper-utilizing ferroxidase, FetC [23]. Cu + oxidation is coupled to the reduction of iron, which is then transported into the cell via the iron permease FtrA [23]. The exact mechanism by which copper and iron are involved in iron uptake has not been experimentally determined in A. fumigatus.
Copper is required for the biosynthesis of some secondary metabolites, bioactive molecules that are not required for growth but confer a fitness advantage to the producing organism. Certain pigments, including melanins, commonly require laccases (also called phenoloxidases) for synthesis. For instance, 1-8-dihydroxynapthalene (DHN), the melanin found in A. fumigatus spores, requires two copper-dependent phenoloxidases, Abr1 and Abr2, for the final step of its production [24]. Copper-dependent laccases also participate in the degradation of lignocellulose [25]. Several other enzymes use copper as a cofactor or contain copper-binding sites that have yet to be further investigated (Table 1).
Copper Transporters and Copper-dependent Transcription Factors
Copper needs to be transported into the cell and delivered to copper-dependent enzymes. A. fumigatus possesses both high-affinity (CtrC and CtrA2) and one low-affinity copper importer (Ctr2), which are expressed to ensure copper sufficiency under conditions when copper is limited [28]. It is unknown whether the copper transporter CtrA1 is high affinity or low affinity. The copper-specific exporter CrpA is used to pump excess copper out of the cell [26]. The copper transporters are regulated by the copper-dependent transcription factors, AceA and MacA, which are also responsible for regulating other genes associated with copper homeostasis [26,27]. Details on copper transport and regulation in fungi will be described in Section 4 of this review.
Copper Toxicity
While it is an essential cofactor for many proteins, copper also has toxic characteristics. The element participates in Fenton chemistry, generating hydroxyl radicals by reacting with hydrogen peroxide, a natural by-product of aerobic respiration [33,34]. Hydroxyl radicals damage DNA by inducing strand breaks, oxidizing nucleoside bases and inactivating iron-sulfur cluster-containing enzymes, a process shared with hydrogen peroxide [35,36]. ROS inactivation of iron-sulfur cluster proteins in yeast has been demonstrated by Murakami and Yoshino by treating yeast with paraquat, an ROS generator. This resulted in the inactivation of aconitase, an enzyme containing an iron-sulfur cluster [37].
Copper can also cause mismetallation, where it replaces metal cofactors in proteins, rendering them inactive [38]. Proteins that depend on solvent-exposed iron-sulfur clusters for single-electron transfer, such as fumarase A and aconitase in the citric acid cycle, or dehydratases that are responsible for branched-chain amino acid biosynthesis, are particularly vulnerable [36]. The toxic effects of copper on iron-sulfur clusters has been investigated in the context of the Yah1 protein in yeast. Yah1, a mitochondrial ferredoxin containing an iron-sulfur cluster and required for electron transfer, is vulnerable to copper-mediated damage and when YAH1 is overexpressed, it results in a strain with increased resistance to copper toxicity [39]. A. fumigatus contains several iron-sulfur cluster enzymes, but it is unknown if copper-mediated mismetallation damage exists in this fungus. Copper can specifically affect the homeostasis of iron since copper is needed for FetC, a ferroxidase that reduces iron prior to import [23]. Potentially implicating ROS generated by copper, a study by De Freitas et al. demonstrated that S. cerevisiae lacking the superoxide dismutase, sod1, resulted in an induction of genes involved in iron acquisition. They hypothesized that this is most likely due to the need for iron-sulfur cluster biogenesis to replace those that have been inactivated due to ROS [40].
Infection Biology and Copper
During infection, pathogenic fungi typically encounter elevated toxic copper levels as a result of a protective host response. Macrophages are activated and express high levels of the copper transporter Ctr1p, raising intracellular copper levels. The Cu + -transporting P-type ATPase ATP7Ap is transported from the Golgi to the phagolysosomal membrane, pumping copper to toxic levels within the phagolysosome, killing the ingested pathogen [41,42]. Although host copper sequestration is not a widespread response to infection, it has been reported for C. albicans and C. neoformans infecting the kidneys and brain, respectively, suggesting that fungi may encounter copper starvation in specific body niches [43][44][45].
Fungal Response to Copper Limitation
Copper homeostasis in fungi is mediated by the transcriptional regulation of genes involved in copper uptake, sequestration, and removal ( Figure 1). In response to low cellular copper concentrations, fungal transcription factors regulate the expression of genes responsible for the uptake and absorption of copper. In S. cerevisiae, the nuclear-localized transcriptional regulator Mac1 is activated under low copper conditions. Two conserved cysteine-rich motifs in the carboxy-terminal of Mac1 bind up to eight Cu + atoms, resulting in an inhibitory intramolecular interaction with the amino-terminal DNA-binding domain. Under low copper, the bound Cu + atoms dissociate, the inhibitory interaction is released, and Mac1p binds conserved copper responsive elements (CuREs) in the promoters of target genes. These include the plasma membrane Cu/Fe reductase encoded by FRE1 and the high-affinity Cu + transporters encoded by CTR1 and CTR3 [46][47][48]. The cell-surface reductase Fre1 reduces exogenous Cu +2 to Cu + , which can then be transported into the cell by Ctr1 and Ctr3. These high-affinity Cu + transporters contain three transmembrane domains that form a trimeric channel and an extracellular methionine-or cysteine-rich region that funnels Cu + into the pore. Inside the yeast cell, excess Cu + is sequestered by the metallothioneins Cup1 and Crs5, and by glutathione [49][50][51]. Cu + is also directed to the desired cellular compartments by the dedicated Cu + chaperones Atx1 and Ccs1, which respectively transport copper to the Ccc2p Cu + transporter in the secretory compartment and the superoxide dismutase Sod1p that degrades oxygen radicals. An as yet unidentified chaperone directs Cu + to the mitochondrial Cox17 chaperone that transfers it to the mitochondrial cytochrome C proteins Cox1 and Cox2, involved in the electron-transport chain [29,52,53]. to Cu+ (red circles) for uptake by transporters Ctr1 and Ctr3. Inside the cell, Cu+ is bound by chaperone proteins Cox17, Atx1, and Ccs1 that transport Cu+ to the mitochondrial cytochrome oxidase Cox1, ER-localized Fet3 ferric reductase, and Sod1 superoxide dismutase, respectively. Low intracellular Cu+ levels are sensed by transcription factor Mac1 to activate genes encoding the copper transporters Ctr1, Ctr3, and Fre1 reductase. High intracellular Cu+ levels are sensed by Ace1 transcription factor to activate the genes encoding metallothioneins (Mts) Cup1 and Crs5 and superoxide dismutase Sod1. Cup1 and Crs5 bind excess intracellular Cu+ and Sod1 oxidizes oxygen radicals formed under excess Cu+. High intracellular Cu+ levels also inhibit Mac1 activation to downregulate Ctr1 and Ctr3 expression. (B) In C. neoformans under low Cu+, Fre1 reduces Cu+2 to Cu+ (red circles) for uptake by transporters Ctr1 and Ctr4. Inside the cell, Cu+ is bound by chaperone proteins Atx1 and possibly a Ccs1 homolog. Atx1 transports Cu+ to the ER-localized laccase involved in melanin biosynthesis. A Ccs1 homolog is predicted to transport Cu+ to Sod1 superoxide dismutase. Low intracellular Cu+ levels are sensed by the transcription factor Cuf1 to activate the genes encoding the copper transporters Ctr1 and Ctr4. High intracellular Cu+ levels are also sensed by Cuf1 to activate the genes encoding metallothioneins (Mts) Cmt1 and Cmt2 to bind excess Cu+ and downregulate copper transporters Ctr1 and Ctr4. (C) In C. albicans under low Cu+, ferric reductase Fre7 reduces Cu+2 to Cu+ (red circles) for uptake by Cu+ transporter Ctr1. Inside the cell, Cu+ is bound by chaperone protein ccs1 that provides Cu+ to superoxide dismutase sod1. A putative Atx1 homolog is proposed to transfer Cu+ to the ER-Cu+ transporter Ccc2, providing copper for the Fet3 ferric reductase involved in iron uptake. Low intracellular Cu+ levels are sensed by transcription factor Mac1 to activate Ctr1 encoding copper transporters. High intracellular Cu+ levels are sensed by transcription factor Cup2 to activate the genes encoding Crp1 copper exporter, Cup1 and Crd2 metallothioneins to respectively remove or bind excess Cu+. (D) In A. fumigatus under low Cu+, an unknown ferric reductase (Fre?) reduces Cu+2 to Cu+ (red circles) for uptake by transporters CtrA2 and CtrC. Inside the cell, Cu+ presumably binds uncharacterized chaperone proteins homologous to yeast Atx1 and Ccs1. The ER-Cu+ transporter CtpA provides copper for the conidial laccases Abr1 and Abr2 that generate melanin. Low intracellular Cu+ levels are sensed by transcription factor MacA to activate genes encoding the copper transporters CtrA2 and CtrC. High intracellular Cu+ levels are sensed by AceA to activate CrpA encoding a copper exporter. Induced overexpression of the metallothionein CmtA also partially protects against Cu+ excess. A. fumigatus gene designations are provided. Genes whose deletion reduces virulence are marked by an asterisk *. Cu + (red circles) for uptake by transporters Ctr1 and Ctr3. Inside the cell, Cu + is bound by chaperone proteins Cox17, Atx1, and Ccs1 that transport Cu + to the mitochondrial cytochrome oxidase Cox1, ER-localized Fet3 ferric reductase, and Sod1 superoxide dismutase, respectively. Low intracellular Cu + levels are sensed by transcription factor Mac1 to activate genes encoding the copper transporters Ctr1, Ctr3, and Fre1 reductase. High intracellular Cu + levels are sensed by Ace1 transcription factor to activate the genes encoding metallothioneins (Mts) Cup1 and Crs5 and superoxide dismutase Sod1. Cup1 and Crs5 bind excess intracellular Cu + and Sod1 oxidizes oxygen radicals formed under excess Cu + . High intracellular Cu + levels also inhibit Mac1 activation to downregulate Ctr1 and Ctr3 expression. (B) In C. neoformans under low Cu + , Fre1 reduces Cu +2 to Cu + (red circles) for uptake by transporters Ctr1 and Ctr4. Inside the cell, Cu + is bound by chaperone proteins Atx1 and possibly a Ccs1 homolog. Atx1 transports Cu + to the ER-localized laccase involved in melanin biosynthesis. A Ccs1 homolog is predicted to transport Cu + to Sod1 superoxide dismutase. Low intracellular Cu + levels are sensed by the transcription factor Cuf1 to activate the genes encoding the copper transporters Ctr1 and Ctr4. High intracellular Cu + levels are also sensed by Cuf1 to activate the genes encoding metallothioneins (Mts) Cmt1 and Cmt2 to bind excess Cu + and downregulate copper transporters Ctr1 and Ctr4. (C) In C. albicans under low Cu + , ferric reductase Fre7 reduces Cu +2 to Cu + (red circles) for uptake by Cu + transporter Ctr1. Inside the cell, Cu + is bound by chaperone protein ccs1 that provides Cu + to superoxide dismutase sod1. A putative Atx1 homolog is proposed to transfer Cu + to the ER-Cu + transporter Ccc2, providing copper for the Fet3 ferric reductase involved in iron uptake. Low intracellular Cu + levels are sensed by transcription factor Mac1 to activate Ctr1 encoding copper transporters. High intracellular Cu + levels are sensed by transcription factor Cup2 to activate the genes encoding Crp1 copper exporter, Cup1 and Crd2 metallothioneins to respectively remove or bind excess Cu + . (D) In A. fumigatus under low Cu + , an unknown ferric reductase (Fre?) reduces Cu +2 to Cu + (red circles) for uptake by transporters CtrA2 and CtrC. Inside the cell, Cu + presumably binds uncharacterized chaperone proteins homologous to yeast Atx1 and Ccs1. The ER-Cu + transporter CtpA provides copper for the conidial laccases Abr1 and Abr2 that generate melanin. Low intracellular Cu + levels are sensed by transcription factor MacA to activate genes encoding the copper transporters CtrA2 and CtrC. High intracellular Cu + levels are sensed by AceA to activate CrpA encoding a copper exporter. Induced overexpression of the metallothionein CmtA also partially protects against Cu + excess. A. fumigatus gene designations are provided. Genes whose deletion reduces virulence are marked by an asterisk *.
Fungal Response to Copper Excess
As in yeast, C. albicans counters low copper levels by activating Mac1 to transcribe FRE7 copper reductase and CTR1 high-affinity Cu + transporter [54] (Figure 1C). In contrast, C. neoformans expresses a single dual-function transcription factor, Cuf1, that controls the response to both low and high copper to induce expression of the high-affinity Cu + transporters Ctr1/Ctr2 or the metallothioneins Cmt1/Cmt2, respectively [45,55,56] (Figure 1B). The effect of MAC1 deletion on C. albicans virulence has not been tested. Deletion of C. neoformans Cuf1 or overexpression of Ctr1/Ctr2 results in increased lung fungal load in infected mice [45,55].
Deletion of macA in A. fumigatus leads to reduced growth and conidiation under low copper [26,57,58]. Intracellular Cu + levels are strongly diminished. As a result, the activity of both SOD and laccase, which use Cu + as a cofactor, is impaired, leading to reduced resistance to oxygen radicals and reduced conidial pigmentation [59]. A. fumigatus MacA binds conserved 5 -TGTGCTCA-3 motifs in the promoters of the high-affinity Cu + transporters CtrA1, CtrA2, and CtrC, leading to their transcriptional activation [26,57,58]. Interestingly, overexpression of CtrA2 or CtrC rescues the copper-sensitive phenotype of the macA-null mutant, while deletion of both transporters phenocopies the macA-null mutant. This implies that the principal targets regulated by MacA are CtrA2 and CtrC [28,58].
RNA-seq analysis reveals that the main cellular functions affected by A. fumigatus macA deletion are oxidation-reduction, metabolism, and transmembrane transport, including strong downregulation of the high-affinity Cu + transporters CtrA1, CtrA2, and CtrC and differential expression of numerous metal, ATP-binding cassette (ABC), and major facilitator superfamily (MFS) transporters [58]. Park et al. [59] showed by microarray and northern blot analysis that following macA deletion, genes involved in iron siderophore synthesis (sidA, sidD), siderophore transport (mirB, mirD, and sit1), reductive iron transport (ftrA, fetC), and iron response regulator hapX are downregulated. ChIP-seq and EMSA were used to identify the MacA-binding motif in the promoter region of these genes. MacA localized to the nucleus under iron-or copper-depleted conditions and was mostly detected in the cytoplasm under iron-or copper-replete conditions [60]. Together, these results suggest that MacA may function as a bifunctional transcription factor of copper and iron metabolism in A. fumigatus, a situation that is unexpected and distinct from that found in other organisms.
The involvement of MacA in A. fumigatus virulence is contradictory, despite the use of apparently identical strains and mouse models. Wiemann et al. [26] showed that the macA-null strain was unaltered in virulence and in susceptibility to killing by mouse alveolar macrophages. In contrast, Cai et al. [27] and Park et al. [61] showed attenuated virulence, reduced fungal load, and increased susceptibility to macrophage killing of the macA-null strain. These differences could be associated with the heterogeneity of fungal isolates [62].
Fungal Response to Copper Excess
In response to high extracellular copper concentrations, fungal transcription factors regulate the expression of genes that are responsible for sequestration and efflux of excess copper (Figure 1). In S. cerevisiae, the transcriptional regulator Ace1 is activated under high copper conditions. Binding of four Cu + atoms to the single Ace1 cysteine-rich Cu-binding domain triggers intramolecular conformational changes within the adjacent amino-terminal DNA-binding domain. Ace1 then binds to conserved motifs in target genes that include CUP1 and CRS5, encoding metallothioneins and SOD1, encoding superoxide dismutase [49,[63][64][65][66]. Cup1, Crs5, and Sod1 suppress copper toxicity by sequestering the excess metal ( Figure 1A). Interestingly, the protective role of SOD1 in copper buffering seems unrelated to its superoxide scavenging activity, as the enzyme protects against copper toxicity under anaerobic as well as aerobic conditions [67].
C. albicans and species of Aspergillus also sense elevated copper through Ace1 homologs. Ace1/AceA contains a conserved zinc-binding domain and (R/K)GRP ((Arg/Lys)-Gly-Arg-Pro) sequence motif essential for DNA minor groove site-specific binding and function, and eight cysteine-rich residues that form a polycopper cluster that binds four Cu + ions cooperatively [58]. Unlike in S. cerevisiae and C. neoformans, the copper-buffering system in C. albicans and species of Aspergillus relies primarily on the AceA-dependent transcriptional activation of CRP1/crpA encoding a Cu + P-type ATPase ( Figure 1C,D). CRP1/CrpA actively transports excess copper from the cytoplasm to the extracellular environment [26,58,[68][69][70]. CRP1/CrpA contains eight transmembrane domains, a conserved CPC (Cys-Pro-Cys) copper translocation motif in the sixth transmembrane segment and cysteine-rich metal-binding motifs in the cytoplasmic N-terminal and is apparently localized in the endoplasmic reticulum and plasma membrane [58,68,69]. Aspergillus flavus, which is both a plant and animal pathogen, contains two redundant AceA-activated crpA homologs, crpA and crpB. Deletion of both homologs is necessary for induction of copper sensitivity. The crpA/crpB-null mutant exhibits attenuated virulence in infected mice but retains full activity in infecting corn seeds [70]. Interestingly, in S. cerevisiae, the CRP1 homolog Ccc2 plays a completely different role in transporting Cu + from the cytosolic chaperone protein Atx1 into the ER where it is incorporated into Fet3 Cu-oxidase required for reductive iron uptake. It is not known whether Crp1/CrpA also accepts Cu + from a dedicated cytosolic chaperone protein similar to yeast Atx1 or whether they directly transport excess cytosolic Cu + .
In C. albicans, Crp1 is responsible for the high resistance to copper, whereas the metallothionein Cup1 is responsible for the residual copper resistance [68]. CRP1 deletion attenuates C. albicans virulence in infected mice, as assessed by kidney fungal load [44].
A. fumigatus deletion mutants of aceA and crpA are hypersensitive to both elevated extracellular copper and ROS in vitro, suggesting these two stresses are inextricably connected. Both strains accumulate higher copper levels and show greater susceptibility to killing by macrophages. In a mouse model of infection, these mutants display reduced growth and virulence [26,58]. Overexpression of CrpA in the aceA-null background reestablishes a wild-type phenotype, confirming that CrpA is the major effector target gene of AceA. Deletion of the single A. fumigatus metallothionein-encoding gene cmtA does not affect growth on elevated copper conditions or resistance to macrophage challenge [58,71]. However, overexpression of CmtA in the crpA-null background provides partial protection against high Cu + indicating that metallothioneins may play a minor role in protecting this fungus against toxic levels of copper [58]. Interestingly, Cai et al. [58] also demonstrated increased sensitivity to zinc following deletion of aceA and crpA, but we were unable to replicate these results in our strains possibly due to differences in genetic backgrounds.
The A. fumigatus genome contains two additional putative Cu + -transporting P-type ATPase genes, ctpA and pcaA [24,31]. CtpA is important for conidial melanization under copper limitation, most likely by supplying the metal to the conidial laccases Abr1/2 [24]. PcaA is not a copper transporter and has been recently shown to provide protection against cadmium, apparently by efflux of this toxic metal [31].
Copper-Binding Secondary Metabolites
As described above, fungi contain several copper transporters that import Cu +2 from the environment [26,27]. This process can be likened to iron import, where FtrA transports Fe +3 into the cell [23]. However, many fungi (and bacteria) also synthesize specialized secondary metabolites that acquire iron from the environment; these Fe chelating metabolites are commonly known as siderophores [72]. Unlike other human pathogenic fungi, A. fumigatus contains over 30 biosynthetic gene clusters (BGC) that produce secondary metabolites [73]. Two of these BGCs encode for the production of siderophores, the extracellular siderophore triacetylfusarinine, shown to be critical for iron uptake and virulence in murine models [23] and the intracellular siderophore ferricrocin [74]. A third BGC encodes for an iron-binding metabolite, hexadehydroastechrome, which is important for iron homeostasis and enhances virulence in murine models of IA when overproduced [71,75].
Like fungi, bacterial utilize siderophores for iron uptake and iron homeostasis but bacteria also have been found to synthesize small molecules for analogous functions in copper biology. Chalkophores, the copper analog of siderophores, are used for copper uptake and as a mechanism to mitigate copper-mediated damage in bacteria [76]. For example, coproporphyrin III is used by the denitrifying bacteria Paracoccus denitrificans to ensure adequate copper supply for the denitrifying process ( Figure 2B) [20,77]. Supplementing copper-depleted growth media with coproporphyrin III remediates the copper growth defect of P. denitrificans suggesting a role for coproporphyrin III in copper uptake [78]. Methanobactin is a chalkophore produced by methylotrophic bacteria and is secreted to acquire copper for methane monooxygenase, an enzyme used to convert methane to methanol (Figure 2A). Yersiniabactin, originally described as a siderophore and associated with the virulence of members of the Enterobacteriaceae, has recently been shown to bind copper in addition to iron [79]. Interestingly, the yersiniabactin metallophore system has been implicated in both detoxification of excess copper and the uptake of copper ( Figure 2D) [80]. Isolates that produce yersiniabactin are more resistant to copper toxicity suggesting a role for its protective effect. In addition, isolates that do not produce yersiniabactin but are supplemented with purified yersiniabactin regain resistance to toxic levels of copper [81]. Yersiniabactin has also been shown to increase the bioavailability of copper, leading to an adequate supply for functionality of the amine oxidase, TynA [82]. Another chalkophore is the Streptomyces thioluteus isocyanide compound SF2768 that has been shown to chelate Cu + and is important for copper uptake ( Figure 2C) [83].
Isocyanides in particular have metal-chelating capabilities including copper [84], and certain bacteria produce isocyanides to promote metal homeostasis or inhibit copper-dependent enzymes [85,86]. Recently, isocyanide synthase BGCs have been found in A. fumigatus, and one of these BGCs was shown to synthesize a series of isocyanides including xanthocillin ( Figure 2E) [87]. Deletion of the isocyanide synthase XanB eliminated xanthocillin production in A. fumigatus. Additionally, a second isocyanide synthase-containing protein, CrmA, was located in another BGC termed the crm (copper responsive metabolite) BGC although the encoded metabolite was not characterized. The crm cluster genes are upregulated during copper starvation and the xan cluster genes during copper excess [87]. Furthermore, MacA positively regulates the crm genes whereas AceA positively regulates xan gene expression, thus firmly tying regulation of both metabolites with the copper regulon. Current studies are aimed at determining whether A. fumigatus isocyanides are important in the copper homeostasis and virulence of this organism. Copper-binding small molecules. (A) Methane-oxidizing bacteria such as Methylosinus trichosporium OB3b produce methanobactins to acquire copper for the particulate methane monooxygenase (pMMO) enzyme. Methanobactin is secreted via an unidentified mechanism, binds copper in the environment, and is transported into the cell via MbnT and MbnE for delivery to pMMO for the oxidation of methane to methanol. (B) Paracoccus denitrificans produces the chalkophore coproporphyrin III that is transported via an unknown mechanism to bind copper in the environment and deliver it into the cell for use in nitrous oxide reductase. (C) Yersiniabactin is produced by bacteria containing the Yersinia high pathogenicity island (HPI) and is implicated as a virulence factor for enteropathogenic Escherichia coli, binding copper and preventing it from damaging the pathogen. The compound also acts as a chalkophore, being required for copper sufficiency, binding copper in the environment and transporting it into the cell for the amine oxidase, TynA. (D) Streptomyces thioluteus has been shown to produce the isocyanide chalkophore SF2768 that is required for copper uptake. The chalkophore is produced and transported via putative transporter proteins, binds copper, and transports it back into the cell. (E) The isocyanide xanthocillin and xanthocillin-like derivatives produced by A. fumigatus are proposed to act as chalkophores, where they are secreted by an unknown mechanism, bind copper in the environment, and transport it back into the cell for use in copper-dependent enzymes such as cytochrome c oxidase, nitrite reductase, amine oxidase(s), superoxide dismutase (SOD1), and laccases(Abr1/Abr2). (F) The insect pathogen Xenorhabdus nematophila produces the isocyanide, virulence factor rhabduscin, which inhibits the copper- Copper-binding small molecules. (A) Methane-oxidizing bacteria such as Methylosinus trichosporium OB3b produce methanobactins to acquire copper for the particulate methane monooxygenase (pMMO) enzyme. Methanobactin is secreted via an unidentified mechanism, binds copper in the environment, and is transported into the cell via MbnT and MbnE for delivery to pMMO for the oxidation of methane to methanol. (B) Paracoccus denitrificans produces the chalkophore coproporphyrin III that is transported via an unknown mechanism to bind copper in the environment and deliver it into the cell for use in nitrous oxide reductase. (C) Yersiniabactin is produced by bacteria containing the Yersinia high pathogenicity island (HPI) and is implicated as a virulence factor for enteropathogenic Escherichia coli, binding copper and preventing it from damaging the pathogen. The compound also acts as a chalkophore, being required for copper sufficiency, binding copper in the environment and transporting it into the cell for the amine oxidase, TynA. (D) Streptomyces thioluteus has been shown to produce the isocyanide chalkophore SF2768 that is required for copper uptake. The chalkophore is produced and transported via putative transporter proteins, binds copper, and transports it back into the cell. (E) The isocyanide xanthocillin and xanthocillin-like derivatives produced by A. fumigatus are proposed to act as chalkophores, where they are secreted by an unknown mechanism, bind copper in the environment, and transport it back into the cell for use in copper-dependent enzymes such as cytochrome c oxidase, nitrite reductase, amine oxidase(s), superoxide dismutase (SOD1), and laccases (Abr1/Abr2). (F) The insect pathogen Xenorhabdus nematophila produces the isocyanide, virulence factor rhabduscin, which inhibits the copper-dependent insect laccase. The laccase is essential for producing melanin, a component of the insect immune response.
Future Directions
There are several gaps in the knowledge of copper homeostasis in A. fumigatus and how the host immune system manipulates the metal in response to infection. Eukaryotes need to carefully manage the import, compartmentalization, delivery, and export of copper. For most fungi, including A. fumigatus, global knowledge of all the triggers that induce or repress the copper regulon is incomplete, and it is unknown whether small molecules, such as fungal isocyanides, play a role in copper homeostasis. Intracellular copper storage and chaperoning as well as several putative copper-binding proteins have not yet been fully investigated in A. fumigatus. Gaining a better understanding of copper homeostasis in this opportunistic human pathogen has the potential to lead to a better understanding of how A. fumigatus is able to manage copper availability in the environment as well as provide novel treatment or prophylactic insights in the context of invasive aspergillosis. | 7,221.2 | 2019-04-01T00:00:00.000 | [
"Biology",
"Environmental Science"
] |
ANALYSIS OF ELASTICITY IN WOODS SUBMITTED TO THE STATIC BENDING TEST USING THE PARTICLE IMAGE VELOCIMETRY ( PIV ) TECHNIQUE
The most important parameter in terms of material mechanical behavior knowledge is the modulus of elasticity, being traditionally obtained through destructive tests. The objective of this study is the verification of the potential use of the particle image velocimetry technique (PIV) as a tool to obtain the modulus of elasticity in sawed wood samples (Pinus oocarpa and Eucaliptus grandis) and wood panels (Plywood, LVL and OSB). The PIV technique has as characteristics the low cost of equipment, fast results, no need for contact with the object tested, accuracy and possibility of application in the field. The application of the PIV technique occurred during the static bending tests where the deformations were also measured with a dial indicator (conventional method), thus obtaining comparative measurements. From the load values applied by a universal test machine and deformation values obtained by the dial indicator and PIV techniques, it was possible to calculate the modulus of elasticity through both methods. With the “Student’s t” statistical test application with significance level of 1%, it was verified that the modulus of elasticity found by the PIV technique and the dial indicator were statistically equal. Average values for the modulus of elasticity found were respectively for the use of the conventional method and for the PIV method the values of: 13,077 and 13,027 MPA for Eucalyptus grandis; 6,171.6 and 6,418.8 MPa for Pinus oocarpa; 10,481.2 and 11,094.3 MPa for plywood; 8,687.4 and 10,261.0 MPa for the LVL; and 2,480.1 and 2,899 MPa for the OSB. It was concluded that the PIV technique is capable of measuring modulus of elasticity values with similar precision to the test techniques traditionally used for this purpose.
INTRODUCTION
The uses of different types of materials in civ il construction demand the detailed knowledge of its characteristics and properties, main ly in loading situation.
The parameter most used when it co mes to evaluation of the mechanical behavior is the modulus of elasticity (E).According to Matos (1997), this parameter provides informat ion regarding the rig idity of the material and can be understood as the effort required doubling the size of a 1 cm² body.The higher the modulus of elasticity, the higher the resistance and the lower the deformity of the wood.Therefore, low values of this parameter will lead to poorer quality timber (Servolo Filho, 2013).
However, the conventional methodologies analysis traditionally employed nowaday are time-consuming and require specific equip ment, besides a large number of samples (Mendes et al., 2012).
Non-destructive testing techniques are options for characterizat ion of materials in comparison with conventional techniques, since this type of methodology does not detract the use of the objects after the analysis, it can be applied to structural parts in use, it has fast results and in general does not demand high-cost equipment.
One of the great advantages of the non-destructive test methods is the possibility of performing the in loco test, that is, without the need for laboratory procedures.This is of great value especially when it co mes to materials that are already being used and need some sort of evaluation.
In view of the increasing use of sawn wood and reconstituted panels of wood as structural parts in construction, it is necessary to search for new test methodologies capable of providing reliab le and accurate results for a better application of these materials in the structures of buildings.
In this context, the particle image velocimetry (PIV) technique, which is an optical technique based on image analysis, appears as an option for the detailed evaluation of the mechanical behavior of the materials subjected to stress.
The objective of this research was the evaluation and validation of the PIV technique as a tool capable of characterizing lu mber (Eucalyptus grandis and Pinus oocarpa) and reconstituted wood panels (Plywood, LVL and OSB) submitted to loading and providing their respective modules of elasticity.
MATERIAL AND MET HODS
The research was carried out at the Federal University of Lavras, and the tests were carried out at the Materials Resistance and Structural Mechanics Laboratory at the Department of Engineering and the manufacture of all samp les at the Experimental Unit in Wood Panels (UEPAM) at the Depart ment of Forestry Sciences (DCF).
Five types of materials were used: sawn wood of the species Pinus oocarpa and Eucalyptus grandis, LVL wood panels, plywood and OSB.The number of test bodies for each type of material can be verified in Table 1.Source: The Author.
The sawed woods of Pinus oocarpa and Eucalyptus grandis were obtained from trees of experimental forest plantations on the campus of the Federal University of Lavras.The samp les for the static bending tests on the sawn wood were made by means of a circular sawing cutter according to ASTM D143-94 (ASTM, 1994), dimensions 2.5 x 2.5 x 41 cm, and were then conditioned at 22 ± 2º C and relat ive air humidity of 65 ± 5%, as used by several authors (Avila Delucis et al., 2016;Cezaro et al., 2016).
For the preparation of the samples test of Plywood and LVL it was necessary to realize the production of the panels.For the elaboration of the panels of Plywood and LVL were used the Pinus oocarpa wood, with 25 years of age.The logs were sectioned using a chainsaw, becoming two logs, 60 centimeters long.They were peeled and heated in water at 66ºC for a period of 24 hours, as recommended by Iwakiri (2005).With the use of a lathe mill the logs were processed, obtaining sheets with a thickness of 2 mm.The sheets were guillotine at 55 x 55 cm and oven dried at 60 ° C until a moisture content of 5 to 6% was reached (Gu imarães Júnior et al., 2015).
For the production of the panels the phenolformaldehyde adhesive was used, with a solids content of 50.5%, at pH of 12.05, a timer gel of 5.30 minutes and a viscosity of 659cP.The adhesive formulation fo r the application was as follows (in parts by weight): adhesive FF = 100; wheat flour = 10 and water = 10.The sheets were glued with weights of 180 g.m -² (single line).The pressing cycle to obtain the plywood will occur with a temperature of 150 ° C, a t ime o f 15 minutes and a pressing pressure of 11 kgf.cm-².
The plywoods were produced with seven sheets crossed among each other, while the LVL panels were produced with seven sheets positioned in the same direction.
The sheet quality classification (A, B, C and D) was performed with reference to NBR 9531 (ABNT, 1986), and it was possible to identify the classification B for the layers and the core of the panels.
The static bending test samples for the ply wood and LVL panels were made using a circular saw, wh ich was later air-conditioned at 22 ± 2ºC and 65 ± 5% relat ive humid ity.The modulus of elasticity (E) was evaluated in a static bending test according to EN 310 (EN, 1993).
The Oriented Strand Board (OSB) panels were obtained through purchase in Lavras -Minas Gerais market.The OSB panels purchased are produced with phenol-formaldehyde adhesive, density of 0.65 g.cm -3 and dimensions of 244 x 122 x 15 cm (length, width and thickness).The static bending test samples for the OSB panels were prepared and stored in accordance with ASTM D1037 (ASTM, 2006) andDIN 52362 (1982).The modulus of elasticity (E) was calculated according to ASTM D1037 (A STM, 2006).
The samples were submitted to static bending in a Universal Testing Machine with load capacity of 30 tonsforce, in wh ich the PIV technique was applied to determine the Modulus of elasticity.
The Universal Testing Machine was instrumented with dial indicator (one in the center of the samples and two in the middle of the distance between the supports and the point of application of the load) to measure the displacement values of the beams.To capture the images, a professional digital camera (CANON EOS Rebel T3) was positioned perpendicular to the surface of the sample (25 cm away).The camera was equipped with a set of lenses to better adjust the focus to the surface of the samples.The capture of the images occurred with the use of a remote control to avoid any disturbance in the camera.The equipment used the configuration of the test, the test samples tested and the instrumentation of the universal test mach ine occurred according to Figure 1.Prior to the capture of the images the surface of all the samples were marked with points made with a finetipped brush.The points were distributed throughout the surface of the samples with a density of points equal to 9.4 points.cm-².The mean diameter of the points was 1.5 mm, according to Figure 2. The images were captured during static bending tests at a regular interval of time.For the test samples of Eucalyptus grandis and Pinus oocarpa the interval between images was 30 seconds and for the LVL, Plywood OSB Panels the interval between images was 5 seconds.
Considering that on average the static bending tests for Eucalyptus grandis and Pinus oocarpa samples had a duration of approximately 600 seconds (loading speed of 1.3 mm.min) and that the duration of the tests with the samples of The LVL, OSB and Plywood rotated around 90 seconds (loading speed of 5 mm.min), each samples obtained between 18 and 20 images, enough to verify the behavior of these materials in the loading situation through the PIV technique.
For the application of the PIV technique, the first image was captured before loading began (t = 0, d = 0) and the others were captured according to the preestablished time interval.Thus the first image (t = 0, d = 0) serves as a comparison parameter for subsequent images.
After the tests were finished, the captured images were man ipulated in an image processing software (Image J) where the images were converted to the 8 bit format and the number of pixels of the images was reduced to 25% of the original with the intention of decrease its storage size.This procedure is important in order to make it possible to process them through the PIV algorith m in the GNU Octave free software.For the processing of the PIV technique, it was used the interrogation window of 32x32 pixel, step size of 1 pixel, search arm around the analysis region of 50 pixels and similarity threshold was used for correlation of the interrogation windows of 0.82.
After the images were processed in the PIV algorith m, the values of deformation were obtained for the places chosen for analysis.In this case, the chosen regions were those close to the three positioned dial indicator (left, center and right).
Based on ASTM D143-94 standard (ASTM, 1994) for sawn wood and in the EN 310 standards (EN, 1993) for Plywood and LVL panels and ASTM D1037 (ASTM, 2006) for OSB panels, it was used the values of deformation of the central part of the samples for calculating the respective modulus of elasticity (E).
For each test sample, two graphs "Load x Deformat ion" were made, one with the deformat ions obtained by the PIV technique and the other with the deformations from the dial indicator.The load values were provided by the Univers al Testing Machine.In this way, the modulus of elasticity in each test sample of each material was calculated by means of the PIV technique and the conventional method (Dial indicator).
The statistical analysis of the data had the objective of comparing the modulus of elasticity calculated with the results of the PIV technique and with the values provided by the dial indicator.
The statistical procedure was delineated with the calculation of the modulus of elasticity in all the samp les.Thus, it was possible at the end of the calculations to obtain two mean values of the modulus of elasticity (E P IV and E Dial indicator ) for each material tested.
The statistical comparison between the values of the modulus of elasticity by the PIV (E P IV ) technique and the modulus of elasticity by the dial indicator (E Dial indicator ) was performed by comparing averages by applying the "Student' s t" test.
For statistical verification of the equality between the module of elasticity by the PIV technique and by the dial indicator, the fo llowing hypothesis was H o : μ P IV = μ Dial indicator, that is, if the means of the modulus of elasticity by the two methods are equal.
With the application of the "Student's t" test with a confidence interval of 99%, the acceptance or reject ion of the H o hypothesis was determined, thus verifying the equality or statistical difference between the average modulus of elasticity found in each material by the two analysis methods.
Modulus of elasticity Comparati on
Fro m the static bending tests the deformat ion values of the samples of the tested materials were obtained.Each test sample generated two sets of values containing the deformations during the test, one by the dial indicator and the other by the PIV technique.
Based on the specific standards for each type of material and fro m the "Strength x Deformation" graph of each test sample, the modulus of elasticity (E) was calculated by both methodologies.The comparison of these values is presented in Figure 3.
It can be seen from Figure 3 that in all tested materials the PIV technique presented values of modulus of elasticity (E) very close to those found by means of measurements with dial indicator.
In order to evaluate the variation between the values of modulus of elasticity (E) among samples of the same material, such as Pinus oocarpa lumber (FIGURE 3
Statistical anal ysis
Fro m the modulus of elasticity of each test sample, it was possible to calculate two mean values for each material, one for the values calculated with the dial indicator and the other with the PIV technique, comparing the two methodologies.
Statistical analysis was performed using a "Student's t" test for comparison of means.The data concerning the type of test, standard used, mean values of modulus of elasticity in each material and the statistical test are checked according to Table 2. Fro m the statistical comparison (Table 2), it was verified that in all materials tested, the mean modulus of elasticity calculated by the two methods was the same, according to the "Student' s t" test with 1% confidence to means comparation.Th is result indicates the accuracy and reliability of the PIV technique versus a conventionally used method.
Other authors such as Ribeiro et al. (2016) and Melo & Menezzi (2016) studied the modulus of elasticity by means of non-destructive test techniques being, Stress wave timer and Ultrasound, respectively.They observed that these techniques were efficient in the inference of the Modulus of elasticity.Ho wever, techniques such as Stress wave timer and Ult rasound rely on physical principles such as density, humid ity, fiber discontinuity and presence of imperfections inside the materials tested.
According to Stangerlin et al. (2011), this behavior is due to the fact that the voltage induced during the dynamic tests is small, that is, the dynamic measurements are based on the mechanical properties only at the elastic limit .
In this sense, the PIV technique has the advantage of evaluating only the surface of the tested material, not depending on the physical properties and characteristics of the interior of the material.Thus, it was able to follow the displacements of the samples fro m the beginning to the end of the test, even at the moment of rupture.
The values found in this study can be compared with values obtained by other authors.Trianoski et al. (2014) found a modulus of elasticity for Pinus oocarpa of 7,993.0MPa.The values of modulus of elasticity found in this study, which were between 6,171.6 MPa (conventional method) and 6,418.8MPa (PIV).This variation between the values can be caused by the environmental conditions where the forest individuals grew, as well as the place of removal of the samples fro m the trunk of the tree, near the base or the crown, or with a larger amount of heart wood or sapwood.
In relation to Eucalyptus grandis, several authors determined the mechanical propert ies of this material.Missio et al. (2013), for example, found mean modulus of elasticity values of 7,813.0 and 9,103.0MPa for destructive testing and ultrasonic testing, respectively.These values of modulus of elasticity, when compared to the ones found in this research (13,077.0MPa (conventional method) and 13,027.0MPa (PIV)) show a certain variation, however similar values are still considered for this parameter.Iwakiri et al. (2002) wo rking with different adhesive phenol formaldehyde formu lations on Pinus oocarpa plywood panels, found mean modulus values of 7,548.5 to 10,366.3MPa.These values are consistent with those found in this research 10,481.2M Pa (conventional method) and 11,094.3M Pa (PIV).Lima et al. (2013), studying the mechanical properties of LVL panels produced with different species and sheet configurations, found an average "E" value of 5,338.9MPa in LVL panels of Pinus oocarpa.However, Müller et al. (2015) found an average modulus of elasticity (E) of 15,270.0MPa in LVL panels of Pinus taeda.In this research the values of "E" found were 8,687.4MPa using the conventional method and 10,261.0MPa by the PIV technique, that is, intermediate values those found by the aforementioned authors.This can be explained by the number of sheets used to make the panels LVL, Lima et al. (2013) used 9 sheets, Müller et al. (2015) used 5 sheets and the panels of this study were made with 7 sheets.
Saldanha & Iwakiri (2009) found for OSB panels of Pinus taeda values of parallel modulus of elasticity (E) of 6,069.0MPa.Mendes (2010) found a parallel (E) value of 8,222.0M Pa.These values are higher than those found in this study.It should be considered that the panels used in this research are commercial; however the values found through the conventional methodology were statistically the same as those found by the PIV technique, regardless of the value.
CONCLUS IONS
Based on the results of this research it was possible to conclude that: By the statistical analyzes, the mean values of the modulus of elasticity found by the PIV technique did not present a statistically significant difference in comparison with the means of the modulus of elasticity obtained by the dial indicator in none of the materials.
The mean values of modulus of elasticity found with the conventional method and the PIV technique, respectively, were for Eucalyptus grandis of 13,077 and 13,027 MPA, for Pinus oocarpa of 6,171.6 and 6,418.8MPa, for the plywood of 10,481.2 and 11,094.3MPa, for the LVL of 8,687.4 and 10,261.0MPa and for the OS B of 2480.1 and 2899 MPa.
The PIV technique was able to characterize all the materials tested in this study by means of their respective modules of elasticity with precision.In this way, the particle image velocimet ry (PIV) technique can be used to characterize materials in loading situations, with the possibility of apply ing in loco in structural parts in use.
General View of the Universal Testing Machine and the instrumentation of the static bending test for the application of the PIV technique (a) and the samples used on the experiment (b).Subtitle: 1-Sawed wood Pinnus oocarpa; 2-Sawed wood Eucalyptus grandis; 3-LVL Panel; 4-Plywood Panel; 5-OSB Panel.Source: The Author.
FIGURE 2. Markers insert on the surface of the materials tested.Subtitle: Samples painted with markers and random pattern.(a) Eucalyptus grandis.(b) Pinus oocarpa.(c) Plywood Panel.(d) LVL Panel.(e) O SB Panel.Source: The Author.
FIGURE 3. Co mparation of modulus of elasticity obtained with the PIV technique and for the dial indicator.Subtitle: M odulus of elasticity of samples of (a) Pinus oocarpa.(b) Eucalyptus grandis.(c) Plywood.(d) LVL.(e) OSB.
TABLE 1 .
Nu mber of samples of each material tested.
TABLE 2 .
Static co mparat ion of the modulus of elasticity obtained by the conventional method and by the PIV technique. | 4,635.4 | 2018-04-01T00:00:00.000 | [
"Materials Science"
] |
Hierarchical Population Game Models of Coevolution in Multi-Criteria Optimization Problems under Uncertainty
: The article develops hierarchical population game models of co-evolutionary algorithms for solving the problem of multi-criteria optimization under uncertainty. The principles of vector minimax and vector minimax risk are used as the basic principles of optimality for the problem of multi-criteria optimization under uncertainty. The concept of equilibrium of a hierarchical population game with the right of the first move is defined. The necessary conditions are formulated under which the equilibrium solution of a hierarchical population game is a discrete approximation of the set of optimal solutions to the multi-criteria optimization problem under uncertainty.
Introduction
One of the main problems that arise in the development of modern control systems is the problem of ensuring the required quality of their functioning in a wide range of changes in operating conditions. The effectiveness of control systems design methods is determined by the possibilities of taking into account uncertain factors, such as the multicriteria nature of control goals and the uncertainty of environmental conditions. Thus, formally, the problem of control systems design is a problem of multi-criteria optimization under uncertainty (MOU).
The analysis of numerous bibliographies shows that currently the approaches that generalize the guaranteed result principle of Germeyer [1] for the class of MOU problems are actively developing and are the most promising. In [2,3], the principles of vector minimax and vector minimax risk which are multi-criteria generalizations of the well-known Wald and Savage principles, respectively, are developed. Generalizations of the principles of vector minimax and vector minimax risk for models of binary preference relations in the form of convex dominance cones are also considered. Mathematic methods for solving dynamical MOU problems based on vector minimax principle and its generalizations are being developed in [4]. In [5][6][7] a more general concept of operator minimax is introduced. The necessary and sufficient conditions for its existence in functional spaces are investigated. In [3,8,9], the interpretation of the vector minimax principle from the standpoint of game theory is given, and the relationship between the concepts of vector minimax and the saddle point is investigated.
However, the application of these approaches to solving applied multi-criteria problems of control optimization under uncertainty faces a number of problems. The frequently occurring need to implement control algorithms in real time requires the representation of control actions in the general case in the form of parameterized program-corrected control laws. Such cases are characterized by a high dimension of the criterion space and the space of control parameters, non-linearity, non-convexity, and the presence of discontinuous points in the components of vector performance indicators. These features of the problem statements, combined with the problem of global optimization, make it difficult or impossible to use known methods and algorithms for solving MOU problems.
Thus, there is a need to develop a new, more efficient computing technology that combines the advantages of global and local multi-criteria search and allows the implementation of control algorithms in real time. The developed computing technology should be compatible with promising architectures of distributed computing systems [10,11], models and methods of distributed computing [12][13][14].
Currently, a qualitatively new approach to solving optimization problems with high computational and structural complexity is being intensively developed, based on the development of co-evolutionary algorithms. In [13], models, forms of coevolution, types of interaction of populations, and models of the distribution of computing resources between subpopulations are discussed. Depending on the nature of the interaction of coevolving populations, two forms of coevolution are studied: cooperative coevolution and competitive coevolution.
Cooperative coevolution involves the decomposition of a set of parameters and/or an objective function of the optimization problem being solved. The most widespread are the following types of cooperative coevolution. Soft Grouping Cooperative Coevolution (SGCC) implements a "soft" distribution of variables across several groups with control of the degree of belonging of variables to groups using the probability distribution function [15]. Differential Grouping Cooperative Coevolution (DGCC) implements a decomposition strategy that minimizes the interdependence between groups of variables [16]. Multi-Level Cooperative Coevolution (MLCC) [17] uses the size of a group of parameters as an optimized parameter. Hierarchical Coevolution Model, a model of coevolution of symbiotic species [18], takes into account homogeneous and heterogeneous aspects of coevolution to maintain diversity, accelerates convergence, preserves diversity and prevents premature convergence of the process of finding optimal solutions.
Competitive coevolution uses the following types of interactions of subpopulations: interaction according to the "host-parasite" scheme; interaction of subpopulations with different search areas; and interaction of subpopulations that differ in search strategies (algorithms or algorithm settings). The latter type of coevolution is used to adapt the parameter settings of search algorithms that ensure the dominance of algorithms with the best settings. In particular, [19] considers a co-evolutionary "cultural" particle swarm algorithm that implements the concept of improving population algorithms based on taking into account the experience gained during solving the problem. Examples of using co-evolutionary particle swarm algorithms for solving optimal design problems with constraints and minimax problems are considered in [20,21]. In addition, competitive evolution can be used as a tool for effective dynamic distribution of computing resources between subpopulations [22,23] in the process of solving the problem.
In [24][25][26], co-evolutionary technologies for solving multi-criteria problems are developed using cooperative, competitive and combined coevolution schemes. It is shown that the combined coevolution schemes look more preferable, since they allow solving quite complex problems of multi-criteria optimization, providing a representative approximation of the Pareto set and adaptive configuration of the algorithm for a specific task. A co-evolutionary algorithm for solving the multi-criteria optimization problem under uncertainty is considered in [27]. However, the assumption of the probabilistic nature of the uncertainty limits the possibilities of using this algorithm.
Thus, co-evolutionary optimization technologies make it possible, in general, to solve the problem of finding a set of globally optimal solutions under multimodality and multicriteriality quite effectively.
At the same time, the complexity of the MOU problem is that when solving it, it is fundamentally necessary to take into account the presence and conflicting nature of the interaction of two types of uncertain factors: the uncertainty of the goal (interpreted as multi-criteria) and the uncertainty of the environment. It is assumed that the uncertainty is known only that it belongs to a certain area, and there are no statistical characteristics. Therefore, the spread of co-evolutionary technology to the MOU tasks requires the develop-ment of new models of coevolution that take into account the conflict nature of the problem being solved and, as a result, the conflict nature of the interaction of subpopulations.
The purpose of this article is to develop hierarchical game models of coevolution that take into account the conflict nature of the MOU problem, as well as the structure of the optimality principles used to find a set of optimal solutions.
A Problem Statement
Consider the problem of multi-criteria optimization under uncertainty (MOU) in the form In the problem (1) U ⊂ E r -the set of valid solutions, u ∈ U; Z ⊂ E k -the set of possible values of an undefined factor, z ∈ Z; J(u, z) ∈ E m -vector efficiency indicator defined on the Descartes product, U × Z; Ω ⊂ E m -a convex dominance cone of that defines a binary strict preference relation on the set of achievable vector estimates, It is necessary to determine the set of optimal solutions to the problem (1) under uncertainty Z. It should be noted that the guaranteeing properties of optimal solutions depend on the optimality principle used to solve problem (1).
The most well-known optimality principles used to solve problem (1) are the principles of vector minimax, vector minimax risk, and their generalizations for models of binary relations in the form of convex dominance cones [2,3].
The Ω-Minimax Principle
Definition 1. Ref. [2] Vector estimation V Ω (G) ∈ E m is called the point of extreme pessimism with respect to the dominance cone Ω (extreme Ω-pessimism), on a set G, if it has the following properties:
The Ω-Minimax Risk Principle
Definition 3. Ref. [2]. The vector estimate P Ω (G) ∈ E m is called the "ideal" point (the "utopia" point») with respect to the dominance cone Ω (Ω-ideal point) on the set G, if it has the following properties: for anyP = P Ω (G) such that G ⊂P − Ω, there is an inclusion of Definition 4. Ref. [3]. A vector function defined on U × Z, is called the vector risk function, and the value R(u, z) under given {u, z} is called the vector risk when choosing an alternative x ∈ X and implementing uncertainty z ∈ Z.
We formulate an auxiliary MOU problem: R(u, z), Ω where U, Z have the same meaning as in problem (1), R(u, z)-a vector risk function of the form (7), Ω ⊂ E m -a convex dominance cone that sets a binary relation of strict preference on the set of achievable vector estimates R(U, Z) = ∪ u ∈ U z ∈ Z R(u, z). Definition 5. Ref. [3]. The Ω -minimax solution u * ∈ U of the MOU problem (8) is called the solution that guarantees the Ω-minimax risk (R Ω -minimax) in the MOU problem of the form (1).
Hierarchical Population Game Model for Finding a Set of Optimal Solutions to the MOU Problem
Based on the statement of the problem (1), we will form a hierarchical population game with the right of the first move It is assumed that two players take part in the game (9) In relation to the hierarchical population game (9), the well-known multi-stage mechanism for forming an equilibrium solution can be implemented on the basis of hierarchical coevolution algorithms. In this case, the equilibrium solution can be interpreted as Ωminimax or R Ω -minimax of the MOU problem (1), depending on the type of functions
Algorithm of Hierarchical Coevolution Search for Set of -Minimax Solutions to the MOU Problem
The proposed algorithm includes the following main steps.
Step 1. In the hierarchical game (9), the first move is made by the coordinating center-it tells the lower-level players its population strategyŨ ⊂ U.
Step 2. With a fixed population strategyŨ ⊂ U the lower-level player solves the problem where In problem (11) Step 3. The coordinating center evaluates the effectiveness of the population strategỹ U ⊂ U by calculating the value of the function F 0 Ũ . To do this, a function is calculated for each oneũ i ∈Ũ: where ψ is a free parameter that determines the selection rules in the evolutionary algorithm; b i -the number of pointsũ j ∈Ũ, j = i, for which the condition is met. After that, the function F 0 Ũ is calculated in the form Step 4. The coordinating center solves the problem The optimal solutionŨ max to problem (17) (3)Ũ max is the optimal solution of the problem (17), where the objective function F 0 Ũ is calculated in accordance with the rules (14)- (16).
Then the population strategyŨ max ⊂ U Ω , where U Ω is the set of Ω-minimax solutions of the MOU problem of the form (1).
Algorithm of Hierarchical Coevolutionary Search for the Set of R Ω -Optimal Solutions to the MOU Problem
Step 1. Formulate an auxiliary MOU problem of the form (8).
Step 2. Form a hierarchical population game (10). The first move is made by the coordinating center C 0 -it tells the lower-level player its population strategyŨ ⊂ U.
Step 3. With a fixed population strategyŨ ⊂ U, the lower-level player solves problem (10), where In (18), the player's C vector criterion is given as The optimal solution to problem (18) is a population strategyZ i Λ ⊂ Z, that maximizes the components of the vector criterion (19).
Step 4. The coordinating center evaluates the effectiveness of the population strategỹ U ⊂ U by calculating the value of the function F 0 Ũ in accordance with rules (14)- (16).
Step 5. The coordinating center solves the problem (17). An optimal solution to the problem (17)Ũ max is called the R Ω -guaranteeing population strategy of the coordinating center. (3)Ũ max is the optimal solution to the problem (17), where the objective function F 0 Ũ is calculated in accordance with the rules (14)- (16).
Then the population strategyŨ max ⊂ U R Ω , whereU R Ω is the set of R Ω -minimax solutions to the MOU problem of the form (1).
The peculiarity of algorithm 2 is that the vector risk function R(u, z) is used to solve problem (18). In this case, the calculation of the function R(u, z) is a separate problem, for the solution of which the following coevolution algorithm is proposed.
Coevolution Algorithm for Calculating the Vector Risk Function
Calculations are performed at the player C level with a fixed population strategỹ U ⊂ U. For fixedũ i ∈Ũ solve the problem of calculating the vector risk function R ũi , z .
The algorithm includes the following basic steps.
Step 3. For each onez j ∈Z i , the problem of constructing an ideal point is solved The optimal solution to problem (21) is a population strategyÛ
Discussion
The formalization of a control system design problem in the form of a MOU problem is relevant because it reflects the conflicting nature of the design task, which manifests itself in the need to take into account several types of uncertain factors: the uncertainty of the goal and the uncertainty of the environment. The application of the principles of vector Ω-minimax and vector Ω-minimax risk allows us to find solutions to the MOU problem that have guaranteeing properties.
The developed hierarchical population game models of co-evolutionary algorithms represent a new type of mathematical models of co-evolutionary algorithms that take into account the conflicting nature of populations interaction, as well as the structure of the principles of optimality in the MOU problem.
Hierarchical population game models of co-evolutionary algorithms have a universal character and allow implementing practically all the principles of optimality used to solve the MOU problems on a single methodological basis. The formed hierarchical algorithmic structures are an effective tool for building parallel architectures of co-evolutionary algorithms for solving high-dimensional MOU problems. The main functional blocks of the developed co-evolutionary algorithms are implemented on the basis of libraries of evolutionary algorithms for multi-criteria optimization in conditions of conflict and uncertainty [28,29].
In the near future, an article will be published in which the applied problem of multicriteria synthesis parameters of an unmanned aerial vehicle neuro-stabilization system under extreme environmental changes is considered. In this case, the problem of training an artificial neural network is formalized in the form of an MOU problem, for which a parallel version of the hierarchical co-evolutionary algorithm for finding the equilibrium of a hierarchical population game with the right of the first move is used.
In the near future, an article will be published in which the applied problem of multicriteria synthesis parameters of an unmanned aerial vehicle neurostabilization system under extreme environmental changes is considered. In this case, the problem of training an artificial neural network is formalized in the form of an MOU problem, for which a parallel version of the hierarchical co-evolutionary algorithm for finding the equilibrium of a hierarchical population game with the right of the first move is used.
Conclusions
The MOU problem statement was formalized in the form of a hierarchical population game with the right of the first move. The concept of a population strategy was defined, and methods for evaluating the effectiveness of population strategies using functions of the form (12) and (16) were proposed. The definitions of Ω-equilibrium and R-equilibrium of a hierarchical population game were formulated.
A hierarchical co-evolutionary algorithm for solving a hierarchical population game with the right of the first move on the basis of Ω-equilibrium was developed.
The necessary conditions were formulated under which the Ω-equilibrium of a hierarchical population game with the right of the first move is a discrete approximation of the set of Ω-minimax solutions of the original MOU problem.
A hierarchical co-evolutionary algorithm for solving a hierarchical population game with the right of the first move based on R-equilibrium and a co-evolutionary algorithm for calculating the vector risk function was developed.
The necessary conditions were formulated under which the R-equilibrium of a hierarchical population game with the right of the first move is a discrete approximation of the set of R Ω -minimax solutions of the original MOU problem. | 3,851.4 | 2021-07-16T00:00:00.000 | [
"Mathematics",
"Computer Science"
] |
Detecting and Mitigating Hallucinations in Multilingual Summarisation
Hallucinations pose a significant challenge to the reliability of neural models for abstractive summarisation. While automatically generated summaries may be fluent, they often lack faithfulness to the original document. This issue becomes even more pronounced in low-resource settings, such as cross-lingual transfer. With the existing faithful metrics focusing on English, even measuring the extent of this phenomenon in cross-lingual settings is hard. To address this, we first develop a novel metric, mFACT, evaluating the faithfulness of non-English summaries, leveraging translation-based transfer from multiple English faithfulness metrics. We then propose a simple but effective method to reduce hallucinations with a cross-lingual transfer, which weighs the loss of each training example by its faithfulness score. Through extensive experiments in multiple languages, we demonstrate that mFACT is the metric that is most suited to detect hallucinations. Moreover, we find that our proposed loss weighting method drastically increases both performance and faithfulness according to both automatic and human evaluation when compared to strong baselines for cross-lingual transfer such as MAD-X. Our code and dataset are available at https://github.com/yfqiu-nlp/mfact-summ.
In addition, current summarisation models, opensource or proprietary, struggle in low-resource settings (Parida and Motlicek, 2019;Hasan et al., 2021;Bai et al., 2021;Urlana et al., 2023), when the target language is under-represented (e.g., Vietnamese and Urdu).Fortunately, cross-lingual transfer methods (Pfeiffer et al., 2020b;Xue et al., 2021;Hu et al., 2020) leverage task-specific knowledge learned from a resource-rich source language to summarise documents in many resource-poor target languages, in a zero-shot fashion or only with few annotated examples.Nevertheless, it remains unclear to what extent cross-lingual summarisation suffers from the problem of hallucination, compared to monolingual systems where English is the only language.
The main challenge in addressing this question is that most faithfulness evaluation metrics are available only for English and do not support lowresource languages.Hence, our first contribution (Section 2) is a model-based metric (mFACT) that measures the factual consistency of multilingual conditional generation, obtained from four diverse English faithfulness metrics (Goyal and Durrett, 2021;Fabbri et al., 2022;Cao et al., 2022a) via 'translate train' knowledge transfer (Artetxe et al., 2020).As illustrated in Figure 1, we use existing faithfulness metrics to label the English documentsummary pairs as positive (i.e., faithful) or negative (i.e., hallucinated) and translate them into each target language.We then train a classifier in each We average the score of four English metrics to rank the training samples in XSum.We then translate the most faithful and hallucinated samples into each target language and train a classifier to distinguish them.
target language to predict the faithfulness scores for the translated document-summary pairs.We verify the reliability of mFACT on the translated test set and, most importantly, with human evaluation.These confirm the effectiveness of mFACT in capturing hallucinations in target languages.
Equipped with this new metric, we conduct extensive cross-lingual transfer experiments on XL-Sum (Hasan et al., 2021) for abstractive summarisation in six typologically diverse languages: Chinese, Spanish, French, Hindi, Turkish and Vietnamese.We find that state-of-the-art cross-lingual transfer methods increase summarisation performance in the target languages, but also introduce more hallucinations compared to English monolingual models in comparable experimental settings, thus further exacerbating this tendency (Section 6).
We also employ the mFACT metric to assess the faithfulness of some recently released multilingual large language models (LLMs), including Phoenix, BLOOMZ, and Vicuna (Chen et al., 2023;Muennighoff et al., 2022;Chiang et al., 2023;Le Scao et al., 2022).We show that LLMs that use multilingual data for pre-training or conversational finetuning fail to ensure faithfulness in summarisation in various languages, producing more hallucinations in low-resource ones.
To overcome this limitation and promote faithful summarisation in multiple languages, we adapt a series of existing methods for reducing hallucinations originally devised for monolingual summarisation (Section 3.2).In addition, we introduce a novel, simple but effective method (Section 3.3): we weigh the loss for each training example according to their faithful scores.We evaluate our loss-weighting method with automated metrics and human judgements.We observe significant gains in both summarisation performance and faithfulness over a series of strong baselines (Section 8).In a nutshell, our main contributions are the following: • We propose mFACT, a multilingual faithful metric developed from four English faithfulness metrics.This enables detecting hallucinated summaries in languages other than English.• To the best of our knowledge, we are the first to study hallucination in a cross-lingual transfer setting.We show that state-of-the-art methods like MAD-X (Pfeiffer et al., 2020b) can improve the performance for low-resource summarisation, but also amplify hallucinations.• We apply mFACT to study the faithfulness in summarisation of the recent multilingual Large Language Models.We observe that despite their scale, these models are still struggling to reduce hallucinations for languages other than English.• We propose a novel method to enhance faithfulness and performance in cross-lingual transfer for summarisation, which consists of weighting training samples' loss based on their faithfulness score.Both automatic and human evaluations validate the superiority of our method over existing baselines.
mFACT: A Multilingual Metric for Faithfulness
metrics into any target language, given the availability of a machine translation model.
Translation-based Transfer for Faithfulness Metrics
One straightforward way to implement a faithful metric in any target language is by implementing it from scratch following the design of monolingual English metrics.However, these often rely on data annotated with auxiliary language-specific tools.For instance, Dependency Arc Entailment (DAE; Goyal and Durrett 2021) requires an external dependency parser to label fine-grained hallucinated segments.This is impractical due to the lack of annotated data and auxiliary tools in most languages.Another strategy relies on "translate test" knowledge transfer (Artetxe et al., 2020), where test documents and their corresponding generated summaries are translated from the target language to English.Then, English metrics can measure faithfulness; however, this introduces noise from translation and is costly at inference time, which makes this unsuitable for model development.For instance, model selection is commonly based on early stopping according to validation faithful scores (Choubey et al., 2021;Aharoni et al., 2022), which necessitates translating all generated summaries at each validation step.
Our solution instead is to formulate faithfulness evaluation as a binary classification problem, i.e., to predict whether a given document-summary pair is faithful or hallucinated.In other terms, our proposed approach aims to distil knowledge from multiple teacher models, i.e., existing English model-based metrics, into a target-language classifier as a student model.Specifically, we use multiple English faithful metrics to assign the pseudo labels of "faithful" or "hallucinated" for English document-summary pairs, then translate them to create a faithfulness binary classification dataset in target languages.We then train the target-language classifier on the resulting silver dataset.Formally, we aim to obtain a faithfulness scoring model g(•) in target language tgt that predicts the faithfulness for a given document-summary pair (x, y).Hence g (tgt) (x (tgt) , y (tgt) ) ≜ p(z = 1 | x (tgt) , y (tgt) ) where z = 1 and z = 0 represent whether the pair is faithful or hallucinated, respectively.
The pipeline for creating mFACT is presented in Figure 1.We start with four diverse English faith-fulness metrics1 , and use them to score the training samples from the English XSum summarisation dataset (Narayan et al., 2018).Following Maynez et al. (2020), we select the metrics based on two categories of hallucinations generated by the model: 1) intrinsic hallucinations where the summary distorts the information present in the document; 2) extrinsic hallucinations where the model adds information that cannot be directly supported by the document.We select two model-based metrics capturing intrinsic hallucinations: • DAE (Goyal and Durrett, 2021) which consists in an entailment classifier trained with annotation at a fine-grained dependency level; • QAFactEval (Fabbri et al., 2022) which focuses on generating questions whose answer is a span of the summary, and attempts to answer them based on the document alone; and two metrics for extrinsic hallucinations: • ENFS% is a simple rule-based measurement presented by Cao et al. (2022a), which counts the proportion of entities which appear in a summary but not in its corresponding document.• EntFA (Cao et al., 2022a) which estimates the posterior and prior probabilities of generated entities with language models conditioned (or not conditioned, respectively) on the source document.Using these probabilities as features, a KNN classifier detects token-level hallucination.We chose XSum as the source English dataset because 1) all our selected faithfulness metrics are trained on XSum, which allows us to maximise the reliability of these metrics.2) XSum has been shown to include abundant and diverse hallucinations (Maynez et al., 2020;Pagnoni et al., 2021), which allows our metrics to capture as many types of hallucinations as possible.
We then normalise the scores from the abovementioned four metrics between [0, 1] and average them for each training sample.We rank the samples from the most faithful to the most hallucinated according to the resulting faithfulness scores.The k top-ranked and k bottom-ranked documentsummary pairs are then considered positive and negative examples, respectively.We translate these into a series of target languages with the Google Translation API2 and create our silver faithfulness dataset splitting its examples with a proportion of 95/2.5/2.5 as the training/validation/testing sets.
Finally, a multilingual BERT-based classifier is fine-tuned on our dataset.We follow the sentencepair classification setting from (Devlin et al., 2019) to concatenate the document-summary pairs as the input.A classifier receives the last-layer representation for the [CLS] special token and returns a score between 0 (hallucinated) and 1 (faithful).
Reducing Hallucination in Cross-lingual Transfer
We first provide some background on cross-lingual transfer.Then, we show how to adapt several methods promoting faithfulness in monolingual summarisation to cross-lingual transfer settings.Finally, we describe a new approach based on loss weighting.
Cross-lingual Transfer with MAD-X
We adopt the Multiple ADapters framework (MAD-X; Pfeiffer et al. 2020b), which constitutes a stateof-the-art method for cross-lingual transfer.MAD-X learns independent language and task adapters (i.e., parameter-efficient model fine-tunings), and then combines them.Specifically, to transfer the ability to summarise documents from a source language to a target language, we follow these steps: 1) We train two separate language adapters on the Wikipedia corpora for both the source and target languages.2) We stack the (frozen) source language adapter with a randomly initialised task adapter and train the latter with annotated data in the source language.3) We stack the trained task adapter with the target language adapter and then perform zero-shot inference in the target language.
Expert and Anti-Expert Approaches
The majority of strategies to reduce hallucinations in monolingual settings rely on creating experts or anti-experts that steer the model towards positive behaviour or away from negative behaviour.As a by-product of the pipeline to create our metric, mFACT (Section 2), we obtained two separate subsets of faithful and hallucinated samples in both source and target languages.These subsets can serve as training data for experts/anti-experts in multiple languages, thus making them suitable for cross-lingual transfer.We explore three methods in this family.In all in stances, we first train a base adapter with the source summarisation dataset.Then, we further tune it with the faithful (hallucinated) subset to obtain an expert (anti-expert) adapter.
Task Vector Negation (TVN; Ilharco et al. 2022).Task vector negation mitigates hallucinated generation by subtracting the task vector of the antiexpert model from the fine-tuned model.Formally, given a fine-tuned model with parameter θ 0 and an anti-expert model θ − , the interpolated model parameters θ ⋆ are obtained as where λ is the importance hyperparameter that controls the degree of fusion between the fine-tuned model and the anti-expert.
Contrastive Parameter Ensembling (CAPE; Choubey et al. 2021).To compensate for the potential loss of summarisation ability by only subtracting the anti-expert task vector from the base model, CAPE proposes to also add the expert parameters.Formally, the interpolated model parameters θ ⋆ are obtained as: where λ again is the importance hyperparameter.
DExpert Decoding (Liu et al., 2021).Contrary to Task Vector Negation and CAPE, which directly manipulate the model parameters, DExpert uses expert and anti-expert models to modify the predicted logits at each decoding step.Given the base model f θ and a pair of expert f θ + and anti-expert f θ − models, the scores for the next token at each decoding step t are: where z t , z + t , z − t are the outputs from f θ , f θ + , f θ − at time step t, respectively.Again, an importance hyper-parameter λ controls the degree of fusion during decoding.
Weighted Loss Approach
We also introduce a simple but effective approach to reduce hallucination during cross-lingual transfer.Previous works have shown that controlling the quality of the training samples can improve the model's faithfulness (Kang and Hashimoto, 2020;Aharoni et al., 2022).However, simply filtering out hallucinated training data may sacrifice the summarisation performance (Dziri et al., 2022).
We thus propose a "soft" data filtering approach where we weigh the training loss according to each sample's faithfulness score.More formally, we rely Model Acc.
on a faithfulness metric for the source language, which outputs a score z (i) for the i th documentsummary pair's faithfulness.Then the update rule of training parameters for each batch becomes ) where θ is the vector of trainable model parameters, α is the learning rate, m is the batch size, J(•; θ) is the loss function for a single training example (x (i) , y (i) ), and ∇ θ J(•) is the gradient of the loss function wrt. the model parameters.
Experimental Setup
Evaluation Metrics.We use ROUGE-1/2/L scores (Lin and Och, 2004) to evaluate the task of abstractive summarisation.We use the four metrics mentioned in Section 2 to evaluate the faithfulness of English summaries and our mFACT metric for summaries in other languages.Dataset.We conduct our experiments on XL-Sum, which is a large-scale multilingual summarisation dataset (Hasan et al., 2021).XL-Sum provides a large collection of annotated document-summary pairs in 45 languages in addition to English.We test our approach on six target languages: Chinese, Spanish, French, Hindi, Turkish and Vietnamese.Table 7 shows the dataset statistics.
Classification Results
Firstly, we verify the reliability of mFACT by using our translated test sets in multiple languages to benchmark mFACT and several baselines for faithfulness classification.Baselines.Previous works (Maynez et al., 2020;Kryscinski et al., 2020) showed that models train for natural language inference (NLI), a related task for which more annotated data is readily available, can be used also for assessing faithfulness for English summarisation.We thus include a baseline, namely XNLI, which consists in fine-tuning multilingual BERT with the corresponding language split in the XNLI dataset (Conneau et al., 2018).As an alternative, we further fine-tune the XNLI baseline with our translated data (XNLI-mFACT), thus verifying whether combining the supervision signal from both sources boosts the performance.Finally, we incorporate an ablation study for using zeroshot multilingual transfer instead of "translate train" (Artetxe et al., 2020).In mFACT-Transfer, we train a multilingual encoder on our English faithfulness classification dataset without translating it, then deploy it directly on examples in other languages.
Results and Discussion.We report the classification performance in Table 1.We find that NLI classifiers do not achieve a level of performance on par with classifiers trained on our faithfulness classification dataset.This demonstrates that evaluating faithfulness is indeed distinct from NLI, which is consistent with previous findings in assessing faithfulness in English (Kryscinski et al., 2020;Maynez et al., 2020).Comparing mFACT and mFACT-Transfer, we also observe the positive effects of translation-based transfer, which achieves a much higher recall rate than zero-shot cross-lingual transfer.Hence, mFACT is more likely to identify faithful document-summary pairs as such.
External Evaluation by Inverse Transfer
Finally, we conduct an evaluation based on inverse cross-lingual transfer (i.e., from other languages to English) as a downstream task with our newly intro-duced approach (Section 3.3).This setting allows us to compare the impact of using different multilingual faithfulness metrics, among those listed in Section 5.1, to weigh the training samples in target languages.The logic behind this experiment is that if the scorer captures the model's faithfulness in target languages, the English summaries generated by the corresponding model should be more faithful according to the four English metrics from Section 2.1.
The results are shown in Table 2. Unsurprisingly, we observe that in general weighting the training samples in target languages with faithfulness metrics can achieve considerable improvements over the MAD-X baseline on English faithfulness scores.This suggests that these metrics are well aligned with the actual faithfulness of generated summaries.Specifically, comparing mFACT-Ours and mFACT-Transfer methods with XNLI and XNLI+mFACT, we find that our constructed dataset is much more effective in improving faithfulness than NLI signal, which again verifies our previous assumption that faithfulness classification and NLI are only vaguely related.Finally mFACT-Transfer performs worse than mFACT in ROUGE, which can be caused by the much lower recall rate of mFACT-Transfer in faithfulness classification (see Table 1).
Cross-lingual Transfer Introduces Additional Hallucinations
The second analysis of this paper aims to corroborate our observation that cross-lingual transfer can introduce additional hallucinations over monolingual fine-tuning, though it improves the task performance for summarisation in the target language.
Transfer Setup.We compare two data scenarios and two styles of fine-tuning.To begin, we investigate the impact of initial training on source data, followed by applying few-shot learning techniques on target data (cross-lingual transfer) instead of direct application.We attribute the difference in faithfulness scores to the additional hallucinations introduced by the training phase in the source language.Taking Chinese as an example of few-shot cross-lingual transfer, we train the summarisation model first on XL-Sum (Chinese) and then with 1K randomly sampled XSum (English) examples.Secondly, we compare fine-tuning the full model, where all parameters are updated, with parameterefficient fine-tuning, where only the adapters are updated.This allows us to study the effect of dif- ferent transfer methods on faithfulness.
Results and Discussion
In Figure 3, we observe that cross-lingual transfer improves ROUGE scores for both full-model fine-tuning and MAD-X, outperforming monolingual fine-tuning.This underscores its effectiveness in transferring task-specific knowledge from source to target languages in lowresource scenarios.However, it's important to note that leveraging source language data can also increase hallucination in both cases.
Hallucinations in Multilingual Large Language Models
We also assess the summarisation performance of recent multilingual large language models (LLMs) on XL-Sum in Table 4: Automatic evaluation for zero-shot crosslingual transfer performance from English to other languages when selecting the checkpoint with the best validation mFACT.Numbers represent the average of three runs with different random seeds.mF stands for mFACT and mF-T stands for mFACT-Transfer.bi% and tr% stand for the percentages of novel bigrams and trigrams.
BLOOMZ with an additional 267K and 189K instances of multilingual instructions and conversation rounds.
We select three languages, aside from English, which are present in the pre-training data for BLOOMZ and the conversational tuning data for Vicuna and Phoenix.We also report the percentage of examples in each of these languages that these models have been exposed to during their Lang.
Phoenix
multilingual training.Table 5 demonstrates that current LLMs display notable faithfulness limitations in cross-lingual transfer contexts for languages beyond English, including well-resourced languages like French and Spanish.Furthermore, a noticeable trend emerges: LLM faithfulness across languages tends to correlate highly to the number of samples from target languages observed during their training.These observations align with recent findings (Lai et al., 2023;Laskar et al., 2023) which highlight the challenges in maintaining faithfulness while generating content in low-resource languages.
Reducing Hallucinations
In this section, we test different methods for crosslingual transfer of summarisation to multiple languages and for promoting faithfulness.We compare our new method of loss weighting based on mFACT with MAD-X, as well as with a series of approaches for reducing hallucinations (Section 4).We evaluate these methods with automated metrics for performance, faithfulness, and abstractiveness (i.e., the ability to rephrase the document instead of copy-pasting spans of text).We also conduct human evaluations to corroborate these results.Automatic Evaluation.We report ROUGE scores for performance, faithfulness (mFACT), and abstractiveness (novel bigrams and trigrams in the summary) for the test set of each target language in Table 4.We first observe that the expert/anti-expert methods adapted from monolingual summarisation are partly effective for improving ROUGEs and mFACT score in cross-lingual transfer over MAD-X; however, no clear winner emerges among them, as their gains are marginal or inconsistent.For example, TVN produces the most faithful summaries for Hindi and Vietnamese, CAPE for Turkish, and DExpert for French.All three models, however, display a similar trend of sacrificing ROUGE scores to improve faithfulness.Instead, as Table 4 demonstrates, our proposed weighted-loss approach (WL) improves the performance across the board while achieving a comparable mFACT score with the most faithful expert models.In particular, WL achieves the best faithfulness in Chinese and Spanish and the best ROUGE scores for all languages except Hindi.These results suggest that our weightedloss method strikes the best balance between summarisation abilities and faithfulness.
Abstractiveness.We also measure the levels of ab-stractiveness of different methods, which is known to be inversely correlated with faithfulness (Ladhak et al., 2022;Daheim et al., 2023).In fact, reducing hallucinations has the side effect of encouraging the model to copy-paste spans of the document (i.e., acquiring an extractive behaviour).Following Cao et al. (2022a) and See et al. (2017), we use the percentage of novel n-grams in summaries compared with the document as a measure of abstractiveness.Figure 3 illustrates the distributions of abstractiveness and faithfulness for all models in six XL-Sum datasets.Both positive and negative predictions of mFACT scatter with different levels of abstractiveness.We also observe that summaries generated by the weighted loss method generally have a higher level of abstractiveness when they are similarly faithful compared with other baselines.Table 4 shows most expert/anti-expert models sacrifice abstractiveness to improve faithfulness score.In contrast, the weighted loss approach produces more novel n-grams.These findings show that our method does not improve faithfulness by simply favouring extractive summaries.Human Evaluation.Finally, we recruited human annotators from the Prolific platform3 for a blind comparison between MAD-X and our weighted-loss model.We randomly sampled nine documents for each language and paired them with the summaries generated by the two models.We asked the human participants to evaluate the summaries via A/B testing in two aspects, Informativeness: An informative summary should cover as much information from the document as possible, while it should convey the main idea of the document.Faithfulness: A faithful summary should only contain information already present in the document 4 and should not contain information contradicting the document.Participants will first read the document, then select the better summary (or both, if they are similar) in terms of informativeness and faithfulness (see Appendix A.5).We require participants to be native speakers of the language they evaluate and have obtained at least a bachelor's degree.Each document and its paired summaries are evaluated by 3 participants.These settings allow us to achieve a fair inter-rater agreement of 0.28 in terms of Fleiss' κ (Landis and Koch, 1977).
The results in Figure 2 indicate that human evaluators prefer the summaries generated by our weighted loss method rather than MAD-X, demonstrating that our weighted loss approach improves faithfulness and informativeness for all six languages.
Finally, we study the correlation between the human preferences from Figure 2 and various faithfulness metrics presented in Section 5.1.From Table 6, it emerges that mFACT achieves the strongest correlation with human judgements (0.45 Pearson ρ and 0.34 Spearman ρ), which is statistically significant.In comparison with XNLI and XNLI-mF, we reconfirm that metrics designed for faithfulness classification, rather than natural language inference, more effectively align with human preferences.
Conclusion
We investigate how to measure and mitigate hallucinations of summarisation models in cross-lingual transfer scenarios.We first propose a multilingual metric, mFACT, to facilitate the evaluation of faithfulness in low-resource languages.By virtue of this new metric, we find empirical evidence that while common cross-lingual transfer methods benefit summarisation performance, they amplify hallucinations compared to monolingual counterparts.We also point out that faithfulness in summarisation for languages other than English is still challenging for multilingual large language models.Finally, with the aim of reducing these hallucinations, we adapt several monolingual methods to crosslingual transfer and propose a new method based on weighting the loss according to the mFACT score of each training example.Based on both automated metrics and human evaluation, we demonstrate that mFACT is the most reliable metric in detecting hallucinations in multiple languages.Moreover, compared to a series of state-of-the-art baselines, we find that summaries produced by loss weighting achieve higher performance and abstractiveness, competitive faithfulness, and a higher alignment with human preferences.We hope that this work will attract more attention from the community to the phenomenon of hallucination in languages different from English and facilitate future research by establishing evaluation metrics and baselines.
Limitations
We use machine translation to construct the faithfulness classification dataset for training the faithfulness metrics in target languages.The required resources may constrain the feasibility of extending mFACT to other languages.The quality of the learned metrics may also be limited by the propagation of errors during translation, especially for languages with poor translation performance.Additionally, although the weighted-loss approach is effective in a diverse sample of languages, we note that its gains in faithfulness are not consistent for all languages, as we discussed in Section 8. Finding a method that is equally effective in reducing hallucinations across all languages is still an open research question for future work.
Ethical Consideration
All human workers participating in our evaluation are informed of the intended use of the provided assessments of summary quality and comply with the terms and conditions of the experiment, as specified by Prolific.In regards to payment, workers from different regions are paid on the same high scale with a wage of £13.5 hourly.This work (and specifically, the human evaluation) has also passed an ethical review by the ethical panel in our institute.
A.1 Dataset Statistics
We show the dataset statistics for all six used subsets of XL-Sum in table 7.
mFACT Classifiers We implement mFACT with the transformers package (Wolf et al., 2020).We train the multilingual BERT model for two epochs, with a batch size of 32 and a learning rate of 5e-5.We set the max input length to 512 and apply truncation to the input article if necessary.The same hyper-parameter settings are applied to all the languages we test.Weighted Loss Summarisation Models We implement our weighted loss model for cross-lingual transfer with adapter-transformers package (Pfeiffer et al., 2020a).We use the officially released mBART-50 checkpoint as the base model for equipping language and task adapters.
To train the language adapters, we follow the same adapter architecture and training settings in (Pfeiffer et al., 2020b).We use the batch size of 64, and a learning rate of 1e-4.We train each adapter with 48K update steps.Task Adapters To train the task adapters for summarisation, we set the batch size to 32, the learning rate to 1e-4, label smoothing factor to 0.1.We use the polynomial scheduler for adjusting the learning rate during training, with weighted decay at 0.01 and maximum gradient norm at 0.1.The model is trained for ten epochs, and we set the first 500 update steps as the warm-up stage.We select the best checkpoint following either the best validation ROUGE or the best mFACT score, respectively.During the decoding step for zero-shot cross-lingual transfer, we follow most settings of (Hasan et al., 2021).We apply the beam search with a size of 6, and the minimum/maximum decoding steps are set to 30/84, respectively.The length penalty is applied at 0.6, and we block all repeated tri-grams.
A.3 Sanity Check for English Faithfulness Metrics
We perform a sanity check experiment and report the results in Table 9 to verify the reliability of these model-based hallucination metrics.We randomly shuffle the alignments of document-summary pairs predicted by the mBART model and the reference.We then feed these misaligned document-summary pairs into the evaluation models and test their performance.We observe that all hallucination metrics drop considerably, showing that these metrics are indeed sensitive to random summaries and reliable to some extent.
A.4 Translation Quality Check
Our first experiment is to confirm the effectiveness of mFACT in capturing hallucinations in target languages.To support our method, we conduct a quality check for translation outputs, a comparison of different metrics on our translated faithfulness classification dataset, and an external evaluation of downstream tasks.
Machine translation (MT)-based transfer can arguably suffer from error propagation, where MT tools introduce hallucinations into their outputs.This issue is even more serious in our setting where translating faithful samples is necessary to create the mFACT metric as training with false positives might significantly degrade its quality.To ensure the feasibility of our pipeline to develop mFACT, we first check the translation quality manually.We randomly pick 100 samples from the Chinese positive set and label their faithfulness.Through this sanity check, we found 13 hallucinated samples; however, only 4 of them are caused by poor translation, while the other 9 are due to an incorrect ranking based on the four English metrics.This shows that MT-based transfer is mostly reliable: only a small amount of noise is introduced by MT.
A.5 Extended Results for Faithfulness Classification
To gain a deeper comprehension of the averaged faithfulness classification results presented in add a reference to Table 1, we analyse the individual language-specific outcomes (Table 10).Across the six language experiments, we consistently observe a significant performance gap between the models trained on the NLI task and those trained on the faithfulness classification task.
The following is the guide for annotators to indicate whether a summary is informative and faithful.
A.6 Full-model transfer vs. MAD-X transfer
We conduct a comparative study on the performance of summarisation and faithfulness in two cross-lingual transfer approaches: MAD-X style and full-model transfer.
For both MAD-X style and full-model crosslingual transfer, we observe that cross-lingual trans-
A.8 Prompts Used for Multilingual LLM's Summarisation
We show the prompt templates used for all languages in our LLM's summarisation experiments in Figure 6.
A.9 Assembling Metrics for mFACT does better than Single Metric We conducted an additional experiment to support our assembling design of mFACT.Rather than averaging four metrics, we individually apply single English metric -DAE, QAFE, ENFS, and EntFA -to rank the XSum dataset and train a multilingual classifier similar to mFACT-Transfer without translation, denoted as DAE-T, QAFE-T, ENFS-T, and EntFA-T.
To examine mFACT with other metrics originating from each single metric, we extend the human evaluation results in Table 6.We compare these four metrics with mFACT-Transfer, and again we measure the Pearson and Spearman correlations to human annotations.
In Table 11, we find mFACT consistently emerges with the highest human correlation when compared to other four metrics.This observation underscores mFACT's better correlation with human evaluations.The reason could be relying on a single metric can introduce biased preference in models and a lack of diversity for captured hallucinations.In general, multiple teacher models lead to a robust, unbiased process (Wu et al., 2021;Ilichev et al., 2021)
A.10 Strategy for Selecting Best Model Checkpoint
Table 12 compares the summarisation model performance when we select the model checkpoint with the best ROUGE-1 or the best mFACT score.We find that under both strategies, the weighted loss model can achieve better ROUGE and faithfulness scores in most languages.However, similar to other works (Choubey et al., 2021;Aharoni et al., 2022), selecting the model checkpoint with the best validation faithfulness score has a higher positive contribution to model's faithfulness.
A.11 Distributions of Faithfulness and Abstractiveness for All Languages
We show the distributions for the percentage of novel 2-grams and mFACT scores for all six languages in Figure 7.
Figure 1 :
Figure1: Pipeline of mFACT for transferring English faithfulness metrics to target languages via machine translation.We average the score of four English metrics to rank the training samples in XSum.We then translate the most faithful and hallucinated samples into each target language and train a classifier to distinguish them.
Figure 3 :
Figure 3: Distributions for Novel 2-gram% and mFACT scores for all five hallucination reduction methods in cross-lingual transfer for the datasets of 6 languages in XL-Sum.
Figure 5 :
Figure 5: Validation mFACT scores curve for each model's training dynamics.Weighted loss consistently outperforms MAD-X in terms of faithfulness during the whole training period.
Table 3 :
Performance and faithfulness scores for fewshot cross-lingual transfer (CLTF) and monolingual finetuning (MFT) on abstractive summarisation.CLTF generally improves the model's performance but decreases its faithfulness.↑ and ↓ indicate higher or lower values are better, respectively.
Table 6 :
Correlation between several faithfulness metrics and human preferences.mF and mF-T stand for mFACT and mFACT-Transfer, respectively.We calculate both Pearson and Spearman statistics on documentsummary pairs from all six languages to ensure that the sample size is significant.
Table 10 :
Classification performance on our translated faithfulness dataset for all target languages.
Figure 4: Comparison of Full-model and MAD-X crosslingual transfer in ROUGE and faithfulness.The left column is the zero-shot performance, and the right column is the few-shot performance.We provide the average scores over all six languages.
Table 11 :
. Using diverse metrics in mFACT's training helps the classifier detect various hallucination types -our inverse transfer experiments (Table2) also show mFACT's promising correlations with both intrinsic and extrinsic hallucination metrics.Correlation with human preferences for mFACT and four transferred metrics developing from single metric.We again calculate both Pearson and Spearman statistics on document-summary pairs from all six languages to ensure that the sample size is significant. | 7,646.8 | 2023-05-23T00:00:00.000 | [
"Computer Science",
"Linguistics"
] |
Moment Explosions in the Rough Heston Model
We show that the moment explosion time in the rough Heston model [El Euch, Rosenbaum 2016, arxiv:1609.02108] is finite if and only if it is finite for the classical Heston model. Upper and lower bounds for the explosion time are established, as well as an algorithm to compute the explosion time (under some restrictions). We show that the critical moments are finite for all maturities. For negative correlation, we apply our algorithm for the moment explosion time to compute the lower critical moment.
Introduction
It has long been known that the marginal distributions of a realistic asset price model should not feature tails that are too thin (as, e.g., in the Black-Scholes model). In many models that have been proposed, the tails are of power law type. Consequently, not all moments of the asset price are finite. Existence of the moments has been thoroughly investigated for classical models; in particular, we mention here Keller-Ressel's work [20] on affine stochastic volatility models. Precise information on the critical moments -the exponents where the stock price ceases to be integrable, depending on maturity -is of interest for several reasons. It allows to approximate the wing behavior of the volatility smile, to assess the convergence rate of some numerical procedures, and to identify models that would assign infinite prices to certain financial products. We refer to [3,20] and the article Moment Explosions in [8] for further details and references on these motivations. Moreover, when using the Fourier representation to price options, choosing a good integration path (equivalently, a good damping parameter) to avoid highly oscillatory integrands requires knowing the strip of analyticity of the characteristic function. Its boundaries are described by the critical moments [24,26].
In recent years, attention has shifted in financial modeling from the classical (jump-)diffusion and Lévy models to rough volatility models. Since the pioneering work by Gatheral et al. [15], the literature on these non-Markovian stochastic volatility models, inspired by fractional Brownian motion, has grown rapidly. We refer, e.g., to Bayer et al. [4] for many references. In the present paper we provide some results on the explosion time and the critical moments of the rough Heston model. While there are several "rough" variants of the Heston model, we work with the one proposed by El Euch and Rosenbaum [9]. The dynamics of this model are Date: April 5, 2018. 2010 Mathematics Subject Classification. 91G20,45D05. Financial support from the Austrian Science Fund (FWF) under grants P 24880 and P 30750 is gratefully acknowledged. We thank Omar El Euch, Antoine Jacquier, and Martin Keller-Ressel for helpful comments.
given by where W and Z are correlated Brownian motions, ρ ∈ (−1, 1), and λ, ξ,v are positive parameters. The smoothness parameter α is in ( 1 2 , 1). (For α = 1, the model clearly reduces to the classical Heston model.) Besides having a microstructural foundation, this model features a characteristic function that can be evaluated numerically in an efficient way, by solving a fractional Riccati equation (equivalently, a non-linear Volterra integral equation; see Section 2). Its tractability makes the rough Heston model attractive for practical implementation, and at the same time facilitates our analysis.
We first analyze the explosion time, i.e., the maturity at which a fixed moment explodes. While the explosion time of the classical Heston model has an explicit formula, for rough Heston we arrive at a well-known hard problem: Computing the explosion time of the solution of a non-linear Volterra integral equation (VIE) of the second kind. There is no general algorithm known, and in most cases that have been studied in the literature, only bounds are available. See Roberts [30] for an overview. Using the specific structure of our case, we show that the explosion time is finite if and only if it is finite for the classical Heston model, and we provide a lower and an upper bound (Sections 3-5). As a byproduct, the validity of the fractional Riccati equation, respectively the VIE, for all moments is established, which culminates in Section 6. In Section 7 we derive an algorithm to compute the explosion time, under some restrictions on the parameters. The critical moments are finite for all maturities (Section 8) and can be computed by numerical root finding (Section 9). Our approach has two limitations: First, to compute the critical moments, maturity must not be too high. Second, our algorithm can compute the upper critical moment only for ρ > 0, and the lower critical moment for ρ < 0. As the latter is the more important case in practice, we focus on the left wing of implied volatility when recalling the relation between critical moments and strike asymptotics (Lee's moment formula; see Section 10).
Corollary 3.1 in [10] is related to our results. For each maturity, it gives explicit lower and upper bounds for the critical moments. Inverting them yields a lower bound for the explosion time; the latter is not comparable to our bounds.
Preliminaries
El Euch and Rosenbaum [9] established a semi-explicit representation of the moment generating function (mgf) of the log-price X t = log(S t /S 0 ) in the rough Heston model. The mgf is given by where ψ satisfies a fractional Riccati differential equation (see below). The constraint ρ ∈ (−1/ √ 2, 1/ √ 2] from [9] was removed recently in [10]. The paper [2] contains an alternative derivation of the fractional Riccati equation, and [1] has more general results, embedding the rough Heston model into the new class of affine Volterra processes. Recall the following definition (see e.g. [21]): Definition 2.1. The (left-sided) Riemann-Liouville fractional integral I α t of order α ∈ (0, ∞) of a function f is given by whenever the integral exists, and the (left-sided) Riemann-Liouville fractional derivative D α t of order α ∈ [0, 1) of f is given by whenever this expression exists.
(The fractional derivative D α t can be defined for α > 1 as well, but this is not needed in our context.) The function ψ from (2.1) is the unique continuous solution of the fractional Riccati initial value problem where R is defined as with coefficients For α = 1, this becomes a standard Riccati differential equation, which admits a well-known explicit solution [14,Chapter 2]. The roots of R(u, ·) are located at the points 1 c3 (−e 0 (u) ± e 1 (u)) with e 0 (u) := 1 2 c 2 (u) = 1 2 (ρξu − λ), (2.7) The following result, relating fractional differential equations and Volterra integral equations, is a special case of Theorem 3.10 in [21]. Theorem 2.2. Let α ∈ (0, 1), T > 0 and suppose that ψ ∈ C[0, T ] and H ∈ C(R). Then ψ satisfies the fractional differential equation Using Theorem 2.2, the Riccati differential equation (2.4) with initial value (2.5) can be transformed into the non-linear Volterra integral equation This integral equation was used in [9] to compute ψ numerically. The function (2.10) where e 0 (u) and e 1 (u) are defined in (2.7) and (2.8). Equation (2.9) is a nonlinear Volterra integral equation with weakly singular kernel; it will be used to analyze the blow-up behavior of f (and thus of ψ) in Section 3. We quote the following standard existence and uniqueness result for equations of this kind. Theorem 2.3. Let α ∈ (0, 1), g ∈ C[0, ∞), and suppose that H : R → R is locally Lipschitz continuous. Then there is T * ∈ (0, ∞] such that the Volterra integral equation Proof. For existence and uniqueness on a sufficiently small interval [0, T 0 ] with T 0 > 0 see, e.g., Theorem 3.1.4 in Brunner's recent monograph [6]. The continuation to a maximal right-open interval is discussed there as well (p. 107; see also Section 12 of Gripenberg et al. [19]).
Note that cases (A) and (B) combined are exactly the cases in which the moment explosion time T * 1 (u) in the classical Heston model is finite, by (2.13). We can now state our first main result. The proof of Theorem 2.4 consists of two main parts. First, Propositions 3.2, 3.4, 3.6, and 3.7 discuss the blow-up behavior of the solution of (2.9) in cases (A)-(D), and Lemma 3.8 shows that blow-up of f leads (unsurprisingly) to blow-up of the right-hand side of (2.1). Second, we show in Section 5 that the explosion time of f (u, ·) (the solution of (2.9)) agrees with T * α (u) (the explosion time of the rough Heston model) for all u ∈ R. As mentioned after Theorem 2.3, this is not obvious from the results in the existing literature.
Explosion time of the Volterra integral equation
We begin by citing a result from Brunner and Yang [7] which characterizes the blow-up behavior of non-linear Volterra integral equations defined by positive and increasing functions. We note that some arguments in our subsequent proofs (from Proposition 3.2 onwards) are similar to arguments used in [7]. Alternatively, it should be possible to extend the arguments in Appendix A of [16]; there, u is in [0, 1].
Proof. This is a special case of Corollary 2.22 in Brunner and Yang [7], with G not depending on time.
In case (A), all assumptions of Proposition 3.1 are satisfied and only the integrability condition (3.1) has to be checked to determine whether the solution f of (2.9) blows up in finite time. Proof. Fix u ∈ R such that c 1 (u) > 0 and e 0 (u) ≥ 0. Note that e 2 0 − e 1 > 0 in this case. (Here and in the following, we will often suppress u in the notation.) If we write the Volterra integral equation (2.9) in the form with non-linearityḠ(w) = w 2 + 2e 0 w and φ(t) = [7], or from Lemma 3.2.11 in [6].) It is easy to check that all the assumptions (G1), (G2), (P) and (K) of Proposition 3.1 are satisfied. Moreover, lim t→∞ φ(t) = ∞ and for all U > 0. By Proposition 3.1, the solution f blows up in finite time.
In case (B), Proposition 3.1 cannot be applied directly to the solution f of (2.9). Hence, the Volterra integral equation has to be modified in order to satisfy the assumptions of Proposition 3.1 in a way that f is still a subsolution of the modified equation, i.e. f satisfies (2.9) with "≥" instead of "=". First, we provide a comparison lemma for solutions and subsolutions.
be a strictly increasing, continuous function and T > 0. Suppose that g is the unique continuous solution of the Volterra integral equation where k satisfies condition (K) from Proposition 3.1. If f is a continuous subsolution, Since c ∈ (0, T ) was arbitrary, the result follows easily. Proof. Fix u ∈ R such that c 1 (u) > 0, e 0 (u) < 0 and e 1 (u) < 0. Note that in this case, the non-linearity G is obviously positive by (2.10). However, G is strictly decreasing on [0, −e 1 ]. To deal with this problem, let 0 < a < −e 1 and define the modified non-linearityḠ a as ThenḠ a is a positive, strictly increasing, Lipschitz continuous function that starts at a andḠ a ≤ G. Letf be the unique continuous solution (recall Theorem 2.3) of the Volterra integral equation Note that the second equality in (3.3) follows from (7.3). Due to the positivity of φ andḠ on (0, ∞), the solutionf is positive on (0, ∞) as well. The functions φ,Ḡ and k satisfy the assumptions (G1), (G2), (P) and (K) in Proposition 3.1. Furthermore, lim t→∞ φ(t) = ∞ andḠ satisfies (3.1). By Proposition 3.1,f blows up in finite time. Because f satisfies (2.9) andḠ a ≤ G, it follows that f is a subsolution of the modified Volterra integral equation, i.e., Since G is non-negative and k is decreasing, which is a contradiction. Therefore, f satisfies 0 ≤ f (t) ≤ a for all t ≥ 0. Proposition 3.6. In case (C), the solution f of (2.9) is non-negative and bounded, and exists globally.
Proof. Fix u ∈ R such that c 1 (u) > 0, e 0 (u) < 0 and e 1 (u) ≥ 0. Note that the inequality 0 ≤ e 1 = e 2 0 − c 1 c 3 < e 2 0 implies a := −e 0 − √ e 1 > 0. Moreover, from (2.10), it follows that a is the smallest positive root of G. Define the non-linearityḠ asḠ ThenḠ is a non-negative, Lipschitz continuous function that starts at e 2 0 − e 1 > 0. Therefore, Lemma 3.5 yields that the unique continuous solutionf of , the functionf solves the original Volterra integral equation and from the uniqueness of the solution we obtain f =f .
Proposition 3.7. In case (D), the solution f of (2.9) is non-positive and bounded, and exists globally.
If we define the non-linearityḠ as thenḠ is a non-negative, Lipschitz continuous function that starts at e 1 − e 2 0 > 0. With Lemma 3.5 we obtain that the unique continuous solutionf of The uniqueness of the solution yields We have shown that (A) and (B) are exactly the cases in which the solution f of the Volterra integral equation (2.9), and thus the solution ψ of the fractional Riccati differential equation (2.4) with initial value (2.5), blows up in finite time.
The following lemma shows that blow-up of ψ is equivalent to blow-up of the righthand side of (2.1).
Proof. First, suppose that the non-negative, continuous function f explodes at T and let M > 0. Then we can find ε ∈ (0,T /2) such that f (t) ≥ M for all t ∈ (T − ε,T ). Hence, for all t ∈ (T − ε,T ). For the second assertion, suppose that f is continuous and bounded with M > 0. Then we have
Bounds for the explosion time
We now establish lower and upper bounds forT α (u), valid whenever it is finite (cases (A) and (B)). We denote byT α (u) the explosion time of the solution f (u, ·) of (2.9). As we will see later, it agrees with T * α (u), and so both bounds of this section hold for the explosion time of the rough Heston model. We prove them first, because we will apply the lower bound in the proof of T * α (u) =T α (u). of (2.9) satisfies where a(u) = 0 in case (A) and a(u) = −e 0 (u) > 0 in case (B).
Proof. Fix u satisfying the requirements of case (A) or (B). It follows from Propositions 3.2 and 3.4 that in either case the solution f is non-negative, starts at 0 and lim t↑Tα f (t) = ∞. For any n ∈ N 0 choose Thus, we obtain for n ∈ N Finally,T Maximization over c > 0, then r > 1, and the substitution w = s α + a yield the inequality (4.1).
For α ↑ 1, the right-hand side of (4.1) simplifies to In case (A), the lower bound (4.1) is sharp in the limit α ↑ 1: We have a(u) = 0 then, and therefore Thus, we obtain for n ∈ N Therefore,T Note that from the definition of t 0 , it depends on c > 0 and r > 1. The fact that f is only zero at t = 0 implies that t 0 → 0 as c ↓ 0. Taking the limit c ↓ 0, then minimizing over r > 1 and substitution w = s α yieldŝ In case (A), we are finished. In case (B), we haveḠ =Ḡ a . Then the dominated convergence theorem for a ↑ −e 1 yields the inequality (4.3).
See Figures 1-3 for numerical examples of these bounds.
Explosion time in the rough Heston model
In Section 3, we established that the right-hand side of (2.1), defined using the solution f of the VIE (2.9), explodes if and only if u satisfies the conditions of cases (A) or (B). As before, we writeT α (u) for the explosion time of f (u, ·). Recall that T * α (u) denotes the explosion time of the rough Heston model, as defined in (2.11). The goal of the present section is to show thatT α (u) = T * α (u), and that (2.1) holds for all u ∈ R and 0 < t < T * α (u). The following result from [10] was already mentioned at the end of the introduction.
Proof. We check the requirements of Theorem 13.1.2 in Gripenberg et al. [19]. The polynomial G(u, w) is differentiable. The kernel (t − s) α−1 /Γ(α) =: k(t − s) is of continuous type in the sense of [19]; see the remark to Theorem 12.1.1 there, which states local integrability of k as a sufficient condition for this property. Proof. We only discuss the case u < 0, because u > 0 is analogous.
(i) Note that (5.1) is a "linear VIE" that can be written as where we define to bring the notation close to that of Section 6.1.2 in [5]. Clearly, (5.2) is not really a linear VIE, because the unknown function f appears in g and K (u) . But as our aim is not to solve it, but to control the sign of ∂ 1 f , this viewpoint is good enough.
(ii) Recall that we assume that u < 0, because u > 0 is analogous. We have to show that satisfies τ (u) <T α (u). We use the following facts: ∂ 1 G(u, w) < 0 for w large, ∂ 2 G(u, w) > 0 for w large, and f (u, t) explodes as t ↑T α (u). Thus, g from (5.3) satisfies (5.9) lim t↑Tα(u) and K (u) satisfies lim t↑Tα(u) K (u) (t) = +∞. We can therefore pick ε > 0 such that For z ∈ [0, 1] and any s, t <T α (u) satisfyingT α (u) − ε ≤ s ≤ t, we have Using this observation in (5.7), we see from a straightforward induction proof that The same then holds for the resolvent kernel (5.6), By (5.5), we obtain where the right-hand side is positive. Indeed, (5.12) follows from (5.9) and (5.10), as g(s) on the left-hand side of (5.12) is O(1). Thus, letting t ↑T α (u), we find that the negative terms g(t) + t t−ε on the right-hand side of (5.11) dominate. This completes the proof.
Proof. According to Section 3.1.1 in [6], the solution can be constructed by successive iteration and continuation. We just show that the first iteration step leads to an analytic function, because the finitely many further steps needed to arrive at arbitrary t <T α (u) can be dealt with analogously. Define the iterates f 0 = 0 and On a sufficiently small time interval, f n (v, ·) converges uniformly to f (v, ·), and the solution can then be continued by solving an updated integral equation and so on (see [6]), until we hitT α (v). Now fix u and t as in the statement of the lemma. For a sufficiently small open complex neighborhood U u, it is easy to see that t <T α (v) holds for v ∈ U . Define γ := 1 ∨ sup v∈U |v| and η := 1 ∨ t α Γ(α + 1) .
Then there is c ≥ 1 such that, for arbitrary v ∈ U and w ∈ C, By the definition of f n , a trivial inductive proof then shows that By a standard result on parameter integrals (Theorem IV.5.8 in [11]), the bound (5.13) implies that each function f n (·, t) is analytic in U . From the bounds in Section 3.1.1 of [6], it is very easy to see that the convergence f n (v, t) → f (v, t) is locally uniform w.r.t. v for fixed t. It is well known (see Theorem 3.5.1 in [18]) that this implies that the limit function f (·, t) is analytic.
Lemma 5.5. The function u →T α (u) increases for u ≤ 0 and decreases for u ≥ 1.
Proof. We assume that u < 0, as u ≥ 0 is handled analogously. By Lemma 5.5, u →T α (u) increases. In this proof, we write M (u, t) for the right-hand side of (2.1), andM (u, t) = E[e uXt ] for the mgf. Now fix u < 0 and 0 < t <T α (u) such that (u, t) has positive distance from the graph of the increasing functionT α (·). Clearly, it suffices to consider pairs (u, t) with this property. By Lemma 5.1, there are v − < v + such that We now show that (5.15) extends to u ≤ v ≤ v + by analytic continuation. From general results on characteristic functions (Theorems II.5a and II.5b in [34]), v → M (v, t) is analytic in a vertical strip w − < Re(v) < w + of the complex plane, and has a singularity at v = w − . If we suppose that w − > u, then Lemma 5.4 leads to a contradiction: The left-hand side of (5.15) would then be analytic at v = w − , and the right-hand side singular. This shows that (5.15) can be extended to the left up to u by analytic continuation.
The following theorem completes the proof of Theorem 2.4.
Proof. In the light of Lemma 5.6, it only remains to show thatT α (u) ≥ T * α (u). (Obviously, Lemma 5.6 implies thatT α (u) ≤ T * α (u).) But this is clear from the continuity of the map t →M (u, t) = E[e uXt ] on the interval (0, T * α (u)). This continuity follows from the continuity of t → X t , Doob's submartingale inequality, and dominated convergence.
Validity of the fractional Riccati equation for complex u
Although the focus of this paper is on real u, the mgf needs to be evaluated at complex arguments when used for option pricing. The following result fully justifies using the fractional Riccati equation (2.4), respectively the VIE (2.9), to do so. As above, we write T * α (u) for the moment explosion time of S, andT α (u) for the explosion time of the VIE (2.9). Theorem 6.1. Let u ∈ C. Then T * α (u) = T * α (Re(u)), and (2.1) holds for 0 < t < T * α (u).
Proof of Theorem 6.1. The first statement is clear from |e uXt | = e Re(u)Xt . Now let t > 0 be arbitrary. As above, we writeM for the mgf and M for the right-hand side of (2.1). By Theorem 5.7, we have M (v, t) =M (v, t) for v in the real interval The functionM (·, t) is analytic on the strip (6.1) {v ∈ C : Re(v) ∈ I} = {v ∈ C : T * α (v) ≥ t}. By the same argument as in Lemma 5.4, the function M (·, t) is analytic on the set {v ∈ C :T α (v) ≥ t}, which contains the strip (6.1) by Lemma 6.2. Therefore, M (·, t) andM (·, t) agree on (6.1) by analytic continuation. This implies the assertion.
Computing the explosion time
Recall that, for fixed u ∈ R, the explosion time T * α (u) of the rough Heston model is the blow-up time of f (t) = f (u, t) = c 3 ψ(u, t), where ψ solves the fractional Riccati initial value problem (2.4)-(2.5). We know from Theorem 2.4 that T * α (u) < ∞ exactly in the cases (A) and (B), defined in Section 2. We now develop a method (Algorithm 7.5) to compute T * α (u) for u satisfying the conditions of case (A). In case (B), a lower bound can be computed, which is sometimes sharper than the explicit bound (4.1). The function f satisfies the fractional Riccati equation where d 1 (u) := c 1 (u)c 3 and d 2 (u) := c 2 (u), with initial condition I 1−α f (0) = 0.
(Recall that we often suppress the dependence on u in the notation.) We try a fractional power series ansatz a n (u)t αn with unknown coefficients a n = a n (u).
Note that v n is an increasing sequence; this follows easily from the fact that log • Γ is convex (see Example 11.14 in [32]). By Stirling's formula, v n ∼ (αn) α for n → ∞. From (7.5), we obtain a convolution recurrence for a n = a n (u): a k (u)a n−k (u) , n ≥ 1.
The function f can thus be expressed as f (u, t) = F (u, t α ), where F (u, z) := ∞ n=1 a n (u)z n . Lemma 7.2. Let u ∈ R, satisfying case (A) (recall the definition in Section 2). Then F (u, ·) is analytic at zero, with a positive and finite radius of convergence R(u).
Proof. To see that the radius of convergence is positive, we show that there is A = A(u) > 0 such that (7.8) |a n | ≤ A n n α−1 , n ≥ 1.
(Adding the factor n α−1 to this geometric bound facilitates the inductive proof.) We have Choose n 0 such that the left-hand side is ≤ 3α −α Γ(α) 2 /Γ(2α) for all n ≥ n 0 , and such that 2v n ≥ (αn) α for all n ≥ n 0 . The latter is possible because v n ∼ (αn) α . Fix a number A with A ≥ 3α −α Γ(α) 2 /Γ(2α) and such that A n n α−1 ≥ |a n | holds for 1 ≤ n ≤ n 0 . Let n ≥ n 0 and assume, inductively, that |a k | ≤ A k k α−1 holds for 1 ≤ k ≤ n. From the recurrence (7.7), we then obtain Since x α−1 (n − x) α−1 is a strictly convex function of x on (0, n) with minimum at n/2, it is easy to see that where the last equality follows from the well-known representation of the beta function in terms of the gamma function (see 12.41 in [33]). We conclude This completes the inductive proof of (7.8).
The finiteness of the radius of convergence will follow from the existence of a number B = B(u) > 0 such that (7.9) a n ≥ B n , n ≥ 1.
To this end, define By Stirling's formula, we have r n /r n−1 = 1 + (1 − α)/n + O(n −2 ) as n → ∞, and so r n eventually increases. Let n 0 ≥ 2 be such that r n increases for n ≥ n 0 , and define B := min{r n0 , a 1 , a 1/2 2 , . . . , a 1/n0 n0 }. This number satisfies a n ≥ B n for n ≤ n 0 by definition. Let us fix some n ≥ n 0 and assume, inductively, that a k ≥ B k holds for 1 ≤ k ≤ n. By (7.7) Thus, (7.9) is proved by induction.
From the estimates in Lemma 7.2, it is clear that termwise fractional derivation of the series (7.2) is allowed, and so the right-hand side of (7.2) really represents the solution f of (7.1) with initial condition I 1−α f (0) = 0, as long as t satisfies 0 ≤ t < R(u) 1/α . We proceed to show how the explosion time T * α (u) can be computed from the coefficients a n (u). The essential fact is that there is no gap between R(u) 1/α and T * α (u). For this, we require the following classical result from complex analysis ( [29], p. 235). Theorem 7.3 (Pringsheim's theorem, 1894). Suppose that the power series F (z) = ∞ n=0 a n z n has positive finite radius of convergence R, and that all the coefficients are non-negative real numbers. Then F has a singularity at R. Theorem 7.4. Suppose that u ∈ R satisfies case (A). Define the sequence a n (u) by the recurrence (7.7) with initial value (7.6). Then we have (7.10) lim sup n→∞ a n (u) −1/(αn) = T * α (u).
Note that, in case (B), we can argue similarly as in the preceding proof. However, the coefficients a n are no longer positive, and so Pringsheim's theorem is not applicable. Then, the inequality R(u) 1/α ≤ T * α (u) need not be an equality. Still, we can compute a lower bound for the explosion time: (7.11) lim sup n→∞ |a n (u)| −1/(αn) ≤ T * α (u).
Now assume that we are in case (A) again. We now discuss how to speed up the convergence in (7.10). Roberts and Olmstead [31] studied the blow-up behavior of solutions of nonlinear Volterra integral equations with (asymptotically) fractional kernel. Their arguments hinge on the asymptotic behavior of the nonlinearity for large argument. In particular, in our situation, with G(u, w) from (2.10) satisfying G(u, w) ∼ w 2 for w → ∞, formula (3.2) in [31] yields We write (?) ∼ for two reasons: First, our integral equation (2.9) does not quite satisfy the technical assumptions in [31]. Second, not all steps in [31] are rigorous. We proceed, heuristically, to infer refined asymptotics of a n (u) from (7.12). Define Φ(z) := ∞ n=1 a n (u)R(u) n z n , a power series with radius of convergence 1, by the definition of R(u) in Lemma 7.2. Its asymptotics for z ↑ 1 can be derived from (7.12). Recall that the explosion time and the radius of convergence of F are related by T * The method of singularity analysis (see Section VI in [12]) allows to transfer the asymptotics of Φ to asymptotics of its Taylor coefficients a n R n . Sweeping some analytic conditions under the rug, we arrive at and thus (7.13) a n (u) Numerical tests confirm (7.13), and we have little doubt that it is true (in case (A)). Summing up, T * α (u) can be computed by the following algorithm, which converges much faster than the simpler approximation lim sup n→∞ a −1/(αn) n : Algorithm 7.5. Let u be a real number satisfying case (A).
We stress that, while the arguments leading to (7.13) are heuristic, we have rigorously shown in Theorem 7.4 that T * α (u) is the lim sup of the left-hand side of (7.14). The heuristic part is that the subexponential factor n 1−α × const improves the relative error of the approximation from O( log n n ) to O( 1 n 2 ). Note that our approach to compute the blow-up time can of course be extended to more general fractional Riccati equations. Finally, as mentioned above (see (7.11)), we can compute a lower bound for T * α (u) if it is finite, but u is outside of case (A): Algorithm 7.6. Let u be a real number satisfying case (B).
Remark 7.7. As for the applicability of Algorithm 7.5, suppose that ρ < 0 (with analogous comments applying to the less common case ρ > 0). From (2.7), we have e 0 (u) ∼ 1 2 ρξu > 0 for u ↓ −∞, and so we are in case (A) for large enough |u|. More precisely, case (A) corresponds to the interval u ∈ (−∞, λ/(ξρ)]. For u from that interval, the explosion time can be computed by Algorithm 7.5. To the right of u = λ/(ξρ), there is a (possibly empty) interval corresponding to case (B), where T * α (u) is still finite, but Algorithm 7.5 cannot be applied. Still, a lower bound can be computed by (7.11), and we have the bounds from Theorems 4.1 and 4.2, which can be easily evaluated numerically. Proceeding further to the right on the u-axis, we encounter an interval containing [0, 1], on which T * α (u) = ∞ (cases (C) and (D)). Afterwards, T * α (u) becomes finite again, but these u belong to case (B), leaving us with bounds for T * α (u) only. To conclude this section, we note that f can be approximated by replacing the coefficients in (7.2) by the right-hand side of (7.13). Let us write b n (u) for the latter. Retaining the first N exact coefficients, this leads to the approximation a n (u) − b n (u) t αn , (7.15) where Li ν (z) := ∞ n=1 z n /n ν denotes the polylogarithm. While this approximation seems to be very accurate even for small N (see [17]), it is limited to real u satisfying case (A), and thus not applicable to option pricing. are of interest. Using the upper bound for the moment explosion time T * α in Theorem 4.2, we will now show the finiteness of the critical moments for every maturity T > 0. Computing u + (T ) and u − (T ) is discussed in Section 9. Proof. Only the finiteness of u + (T ) is proven, as the proof for u − (T ) is very similar. Denote the upper bound of T * α (u) in (4.3) by B(u) for all u ∈ R in the cases (A) and (B). First, we show that for sufficiently large u, we are always in case (A) or (B), depending on the sign of the correlation parameter ρ. From (2.7) and (2.8), it is easy to see that (8.2) e 0 (u) ∼ 1 2 ξρu and e 1 (u) ∼ − 1 4 ξ 2ρ2 u 2 as u → ∞, whereρ 2 = 1 − ρ 2 . Thus, eventually e 1 (u) < 0 for sufficiently large u. In the next step, we show that the upper bound B(u) converges to 0 as u → ∞. Indeed, in for all T > 0, which suffices for numerical computations (under the above restriction on T ). The validity of (9.2) and (9.3) is clear from (5.16): If T * α (·) is constant on some interval, lying to the left of zero, say, then the mgf blows up as u approaches the interval's right endpoint from the right.
Application to asymptotics
In the introduction we mentioned several potential applications of our work. In this section, we give some details on one of them: Knowing the critical moments gives first order asymptotics for the implied volatility for large and small strikes. We writeσ(k) for the implied volatility, where k = log(K/S 0 ) is the log-moneyness. According to Lee's moment formula [23], the left wing of implied volatility satisfies We focus on negative log-moneyness, because then the slope depends on the lower critical moment, which Algorithm 7.5 computes in the important case ρ < 0. As in any model with finite critical moments, the marginal densities of the rough Heston model have power-law tails. More precisely, if we write f T for the density of S T , then f T (x) = x −u + (T )−1+o(1) , x → ∞, and Our approach (see Section 9) allows to evaluate the right-hand sides of (10.1) and (10.2) numerically for the rough Heston model, if T is not too large.
In [13], (10.1)-(10.2) were considerably sharpened for the classical Heston model. We expect that such a refined smile expansion can be done for rough Heston, too, with density asymptotics of the form f T (x) ∼ c 1 x −u + (T )−1 e c2(log x) 1−1/(2α) (log x) c3 , x → ∞, where the c i depend on T and α. In the classical Heston model, the factor e c2(log x) 1−1/(2α) becomes e c2 √ log x , in line with [13]. Extending the analysis of [13] to 1 2 < α < 1 will require a detailed study of the blow-up behavior of the Volterra integral equation (2.9). Among other things, (a special case of) the heuristic analysis in [31], which we already mentioned in Section 7, would have to be made rigorous, and extended to ensure uniformity w.r.t. the parameter u. We postpone this to future work. Note that the approximation (7.15) might be useful in this context. | 8,584.8 | 2018-01-29T00:00:00.000 | [
"Economics",
"Mathematics"
] |
CpG-C ODN M362 as an immunoadjuvant for HBV therapeutic vaccine reverses the systemic tolerance against HBV
Chronic Hepatitis B virus (CHB) infection is a global public health problem. Oligodeoxynucleotides (ODNs) containing class C unmethylated cytosine-guanine dinucleotide (CpG-C) motifs may provide potential adjuvants for the immunotherapeutic strategy against CHB, since CpG-C ODNs stimulate both B cell and dendritic cell (DC) activation. However, the efficacy of CpG-C ODN as an anti-HBV vaccine adjuvant remains unclear. In this study, we demonstrated that CpG M362 (CpG-C ODN) as an adjuvant in anti-HBV vaccine (cHBV-vaccine) successfully and safely eliminated the virus in HBV-carrier mice. The cHBV-vaccine enhanced DC maturation both in vivo and in vitro, overcame immune tolerance, and recovered exhausted T cells in HBV-carrier mice. Furthermore, the cHBV-vaccine elicited robust hepatic HBV-specific CD8+ and CD4+ T cell responses, with increased cellular proliferation and IFN-γ secretion. Additionally, the cHBV-vaccine invoked a long-lasting follicular CXCR5+ CD8+ T cell response following HBV re-challenge. Taken together, CpG M362 in combination with rHBVvac cleared persistent HBV and achieved long-term virological control, making it a promising candidate for treating CHB.
Introduction
Chronic Hepatitis B virus (CHB) infection currently affects approximately 240 million people worldwide [1]. Patients with CHB are at higher risk for developing cirrhosis and hepatocellular carcinoma (HCC) [2]. Current antiviral therapies, such as pegylated interferon alpha2a (PegIFN) and nucleoside/nucleotide analogues (NAs), are unable to achieve efficient hepatitis B surface antigen (HBsAg) loss [3]. Therefore, a novel strategy involving immunomodulation needs to be developed to achieve long-term virological control in CHB.
The pathological basis of CHB infection is the persistent presence of HBV covalently closed circular DNA (cccDNA) in the nuclei of infected hepatocytes, as well as an immunosuppressive environment in the infected liver [4,5]. Both the innate and adaptive immune responses are involved in the pathogenesis of CHB infection, and influence its clinical outcome [5][6][7]. However, HBV-specific immune responses are characteristically weak, transient and nearly undetectable in CHB patients. Furthermore, neither the prophylactic HBV vaccine alone nor its combination with other antiviral compounds has successfully eliminated HBV-infected cells in CHB.
Since the ideal HBV therapy should reverse host immune tolerance to the virus and recover the function of HBV-specific T cells, an effective adjuvant may improve the success of therapeutic HBV vaccines.
CpG oligodeoxynucleotide (CpG ODN) are synthetic single-stranded DNA molecules containing unmethylated cytosine-guanine dinucleotide (CpG) Ivyspring International Publisher motifs. As an agonist for Toll-like receptor 9 (TLR9), immuno-stimulatory CpG activates antigenpresenting cells (APCs) to produce inflammatory cytokines, thus inducing the T helper type 1 (Th1)-type immune response via TLR9 activation [8][9][10]. At least three major classes of CpG-ODNs have been characterized according to their backbone, sequence, and immuno-stimulatory properties: class A (D-type), class B (K-type), and class C [11]. CpG-A ODNs such as CpG 1585, CpG 2216, and CpG 2336 activate plasmacytoid dendritic cells (pDCs) to produce interferon-α (IFN-α), but fail to induce B cell activation [8,9]. In contrast, CpG-B ODNs such as CpG 1826, CpG 2006, and CpG 7909 strongly induce B cells to produce interleukin-6 (IL-6), but promote pDC maturation with absence of IFN-α secretion [8,12]. Finally, CpG-C ODNs combine the features of classes A and B, activating both pDCs and B cells [13]. CpG ODNs have been widely used as adjuvants for various antiviral vaccines against hepatitis C virus (HCV), human immunodeficiency virus (HIV), and HBV, as well as cancer cells [14][15][16]. The prophylactic HBV vaccine Engerix-B combined with CpG 7909 (CpG-B ODN) resulted in higher HBs antibody (anti-HBs) titers and enhanced affinity maturation to improve the avidity of anti-HBs [16]. In addition, a two-dose schedule of HEPLISAV-B, comprised of recombinant HBsAg and 1018 ISS (CpG-B ODN), demonstrated a significantly higher rate of protection (95%) compared to that observed with Engerix-B (81%) in a phase III trial [17,18]. Systemic administration of CpG 1826 (CpG-B ODN) inhibited HBV replication by inducing type I IFNs in HBV transgenic mice [19,20]. In another study, administration of nanoparticles containing unmethylated CpG-A ODNs (HBV-CpG) exerted a strong immuno-stimulatory effect on DCs, NK cells, and T cells in vivo, and led to viral clearance in HBV-carrier mice [21]. Taken together, these studies indicate that CpG-A and -B agonists may function as potent immunomodulatory agents against CHB infection by augmenting HBV-specific T or B cell responses. In a phase 1b multicenter trial, CpG 10101 (CpG-C ODN) activated the immune response along with secretion of IFN-α to reduce HCV RNA levels in a dose-dependent manner [22]. Furthermore, intra-tumoral SD-101 (CpG-C ODN) administration in combination with low-dose radiation in a phase 1/2 trial promoted the generation of tumor-specific CD8 + and CD4 + effector T-cells, and reduced the abundance of T regulatory cells (Tregs) in the tumor microenvironment, which in turn led to complete tumor regression in both treated and untreated tumor sites [23]. However, the therapeutic utility of CpG-C ODNs as HBV vaccine adjuvants remains unclear.
Exhausted CD8 + T cells play a critical role in the development of CHB infection. Recent studies have reported that CXCR5-expressing CD8 + T cells are partially exhausted with strong antiviral activity [24][25][26][27], producing higher levels of IFN-γ, TNF-α, IL-21, and granzymes during lymphocytic choriomeningitis virus (LCMV), HIV, and other chronic infections than CXCR5 -CD8 + T cells. Additionally, CXCR5 + CD8 + T cells can migrate into B cell follicles, thereby supporting B cell activation, affinity maturation, and antibody production [26][27][28]. CXC chemokine ligand 13 (CXCL13) exclusively binds to chemokine receptor CXCR5 expressed on CD8 + T cells to help recruit CXCR5 + T cells to the inflammatory site, thus coordinating both humoral and cellular immune responses [29,30]. Moreover, Li et al. reported that elevated expression of CXCL13 facilitated the recruitment of CXCR5 + CD8 + T cells in the liver, which in turn inhibited HBV replication and regulated production of B cell antibodies in patients with CHB [27]. However, whether CpG-C ODNs can sustain HBV control by inducing the follicular CXCR5 + CD8 + T cell response is unknown.
Previous studies have demonstrated that an HBV-carrier mouse model of persistent HBV infection can be generated via injection of an AAV-HBV vector [31][32][33][34]. These mice do not mount the specific immune response to conventional HBV vaccines, thus mimicking the immune tolerance exhibited in human CHB [32,33,35]. Therefore, the goal of the current study was to evaluate the therapeutic feasibility of using CpG-C ODN M362 as an HBV vaccine adjuvant using AAV/HBV-transduced HBV-carrier mice.
Animals and reagents
Five-to six-week-old male C57BL/6J mice were purchased from the Beijing HFK Bioscience Co. Ltd (Beijing, China). All animals were treated in accordance with the Guidelines for the Care and Use of Laboratory Animals of the Ethical Committee of Shandong University and the protocol was approved by the Institutional Animal Care and Use Committee of Shandong University. rHBVvac (Hansenula polymorpha) was purchased from Dalian Hissen Bio-pharm. Co., Ltd. (Dalian, China) and CpG M362 was obtained from Invivogen (San Diego, CA, USA).
Generation and stimulation of bone marrow-derived dendritic cells (BMDCs) in vitro
Murine BMDCs were generated as previously described [37], and CD11c + BMDCs were identified and enriched using fluorescence-activated cell sorting (FACS) (> 90%). To assess their antigen-presenting ability, the isolated BMDCs were incubated with CpG M362 for 12 h, and surface expression of CD86 and MHC-II was analyzed by flow cytometry.
HBV DNA detection
HBV DNA was extracted from 50 µL mouse serum using the HBV DNA quantitation kit according to the manufacturer's instructions (Daan Gene, Guangzhou, China) and measured by quantitative PCR using UltraSYBR Mixture (CW Biotech, Beijing, China) with a Lightcycler® 96 (Roche, Basel, Switzerland).
Cell isolation
Single-cell suspensions from the liver, spleen, and draining lymph nodes (dLNs) were isolated as previously described [36]. Briefly, the PBS-perfused liver was passed through a 200-μm nylon cell strainer to obtain the single-cell suspension, which was centrifuged at 100 × rcf for 1 min to remove hepatocytes. Then, the supernatant was centrifuged at 400 × rcf for 10 min to collect residual cells, which were layered over 40% Percoll (GE Healthcare, Uppsala, Sweden). Hepatic mononuclear cells (MNCs) were harvested after centrifugation at 400 × rcf for 10 min, followed by red blood cell (RBC) lysis and washing. The spleens and dLNs were passed through a 200-μm nylon cell strainer, single cells were harvested, followed by RBC lysis, and washing.
cHBV-vaccine efficiently eliminated HBV in carrier mice
HBV-carrier mice were used to evaluate the efficiency and safety of the cHBV-vaccine (Fig. 1A). The results confirmed that rHBVvac alone did not eliminate HBV (Fig. S1A-C). Compared to serum HBsAg and HBeAg levels in untreated mice, those in cHBV-vaccinated mice decreased significantly and remained at low levels after the third immunization ( Fig. 1B, 1C). Furthermore, serum HBV DNA, and intrahepatic HBsAg and HBcAg were nearly undetectable in cHBV-vaccinated mice (Fig. 1D-F). However, none of these treatments induced the generation of anti-HBs (Fig. S1C). In addition, the serum ALT remained at baseline levels, indicating that cHBV vaccination did not result in liver injury (Fig. 1G). Taken together, these results indicated that CpG M362 is a promising and safe adjuvant for use in HBV therapeutic vaccines.
CpG M362 promoted the maturation and antigen-presenting ability of DCs
As professional APCs, DCs play a vital role in generating antigen-specific T-cell responses against chronic HBV infection [38]. As shown in Fig. 2A-C, the spleen of cHBV-vaccinated mice exhibited higher proportions of myeloid DCs (mDCs) (Lin -/-MHCII + CD317 -CD11c + ) and pDCs (Lin -/-MHCII + CD317 + CD11c int ) compared to those in untreated mice. Moreover, the expression of co-stimulatory molecules CD80, CD86, and CD40 was upregulated on mDCs and pDCs in spleen following treatment with the cHBV-vaccine compared to that in untreated mice (Fig. 2D, 2E). A similar phenomenon was observed on mDCs and pDCs in dLNs (Fig. S2A-D). However, treatment with rHBVvac alone did not efficiently enhance DC activation ( Fig. 2B-E, Fig. S2A-D). To further confirm the stimulatory effect of CpG M362 on DCs, BMDCs were generated in vitro and treated with different doses of CpG M362. Consistently, CpG M362 increased the surface expression of MHC-II and CD86 on BMDCs compared to that in the unstimulated controls (Fig. 2F). These results confirmed that the CpG M362 adjuvant promoted the maturation and antigen-presenting ability of DCs in vivo and in vitro.
cHBV-vaccine amplified robust Ag-specific CD8 + and CD4 + T cell responses
Since HBV-specific CD8 + and CD4 + T cells play a crucial role in controlling HBV progression [35,39], we evaluated whether vaccination enhanced the magnitude and quality of HBV-specific CD8 + and CD4 + T cell responses. As shown in Fig. 3A and 3B, cHBV-vaccinated mice exhibited a significantly higher proportion of HBV-specific CD11a hi CD8α lo cells [35,36,40] in their peripheral blood, liver, and spleen, compared to untreated mice. Furthermore, cHBV-vaccination resulted in a marked increase in the expression of CD69 and CD107a on CD8 + T cells (Fig. 3C), along with increased IFN-γ secretion (Fig. 3D). In addition, the abundance of HBV-specific CD4 + CD11a hi CD49d hi CD4 + T cells [35,41] was also significantly higher in the spleens of cHBV-vaccinated mice than that in untreated mice (Fig. 3E, 3F), along with increased expression levels of the activation antigen CD69 on CD4 + T cells (Fig. 3G). Taken together, these results indicated that CpG M362, as a vaccine adjuvant, enhanced HBV-specific cellular responses.
cHBV-vaccine alleviated immunosuppression and restored the exhausted HBV-specific CXCR5 + CD8 + T cells
The immunosuppressive environment and exhaustion of CD8 + T cells in the HBV-infected liver are the major causes underlying the refractoriness of CHB [5,39,42]. cHBV-vaccination downregulated the expression of PD-L1 on hepatocytes of HBV-carrier mice (Fig. 4A), reduced the proportion of Tregs (Fig. 4B), and decreased TGF-β1 levels in liver tissues and serum (Fig. 4C). Furthermore, HBV-specific CD11a hi CD8α lo T cells in cHBV-vaccinated mice displayed significantly reduced expression of LAG-3, CTLA-4, TIGIT, and PD-1 compared to that in untreated mice (Fig. 4D), accompanied by upregulation of pro-proliferative nuclear antigen Ki-67 (Fig. 4E). Follicular CXCR5 + CD8 + T cells have been recently found to play a pivotal role in controlling viral replication during chronic infections such as HIV, LCMV, and HBV [24,27].
Interestingly, approximately 5% of the HBV-specific CD11a hi CD8α lo cells in the spleens of HBV-carrier mice abundantly expressed CXCR5. However, the subset of CXCR5 + HBV-specific CD11a hi CD8α lo cells was more exhausted than the predominant CXCR5subset ( Figure S3). Although the frequency and abundance of CXCR5 + CD11a hi CD8α lo cells were not significantly altered by cHBV vaccination (Fig. 4F, 4G), the expression of multiple co-inhibitory receptors such as LAG-3, PD-1, TIGIT, and TIM-3 was significantly downregulated (Fig. 4H). Meanwhile, cHBV vaccination did not significantly affect the frequency of CXCR5 + CD4 + T cells, but decreased the expression of TIGIT and PD-1 on CXCR5 + CD11a hi CD4 + T cells (Fig. S4). These results indicated that CpG M362 as a vaccine adjuvant was able to overcome the mechanisms that impaired the functional immune responses in HBV-carrier mice.
cHBV-vaccine induced long-term immune memory against HBV re-challenge
The major challenge of clinical HBV therapy is to prolong the immunological memory against the recurrence of HBV infection [43,44]. To this end, HBV-carrier mice were re-challenged with HBV on day 59 after cHBV vaccination. Both HBsAg and HBV DNA were nearly undetectable in the serum of cHBV-vaccinated mice (Fig. 5A, 5B) compared to that in untreated mice, accompanied by higher levels of protective anti-HBs (Fig. 5C). Furthermore, the proportion of CXCR5 + PD-1 + follicular helper T cells increased upon HBV re-challenge (Fig. 5D). Meanwhile, serum ALT remained at the baseline levels in cHBV-vaccinated mice after HBV re-challenge (Fig. S5). Taken together, the results suggested that CpG M362 as a vaccine adjuvant induced long-term immune memory against HBV re-infection. The proportion of CD4 + CD11a hi CD49d hi T cells (E, F) and CD4 + CD69 + T cells (G) in the spleen on day 21 post-immunization. All data are expressed as mean ± SEM (n ≥ 5). * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001 versus untreated mice.
cHBV-vaccine induced long-lasting CXCR5 + CD8 + T cell response during HBV re-challenge
An effective vaccine should generate appreciable numbers of high-quality memory CD8 + T cells that can be immediately activated upon re-exposure to the pathogen [45][46][47]. Compared to untreated mice, the abundance of HBV-specific CD11a hi CD8α lo cells in the liver and spleen of cHBV-vaccinated mice increased significantly after HBV re-challenge (Fig. 6A), accompanied by downregulated expression of PD-1 and LAG-3, indicating an effective immune response (Fig. 6B). Furthermore, cHBV vaccination significantly increased the frequency of splenic CXCR5 + CD11a hi CD8α lo cells (Fig. 6C) and decreased the expression of multiple co-inhibitory receptors (Fig. 6D). Although the frequency of splenic CXCR5 + CD11a hi CD4 + T cells increased in response to HBV re-challenge, cHBV vaccination only decreased PD-1 levels on CXCR5 + CD11a hi CD4 + T cells (Fig. S6). These data indicated that the adjuvant properties of CpG M362 induced a long-lasting antiviral CXCR5 + CD8 + T cell response against HBV.
Discussion
CHB infection results from a dynamic balance between viral replication and the host immune response. Liver-induced systemic immune tolerance is the basis of CHB, which is characterized by high viral load and impaired HBV-specific adaptive T cell responses [5,39]. The ideal endpoints of HBV treatment are HBsAg loss, seroconversion to anti-HBs, and sustained inhibition of HBV DNA [48]. Since HBV antigens are weakly immunogenic, adjuvants are needed to accelerate, prolong, and enhance the immune response against CHB. Among the TLR9 agonists tested as adjuvants in vaccines against cancer cells and intracellular pathogens, CpG ODNs induce activation of both the Th1-polarized immune response, including NK cells, macrophages, and APCs, and the humoral immune response to control local tumor growth [22,49,50]. Studies have shown that CHB impairs the host innate immune system by downregulating TLR expression and inhibiting downstream signaling pathways [51,52]. Meanwhile, treatment with synthetic TLR9 agonists, such as class A and B CpG ODNs, reportedly enhanced anti-HBV immunity and HBV elimination during CHB therapy [10,19,21]. Therefore, the current study investigated the effectiveness of CpG M362, a class C CpG ODN that combines the features of both class A and B CpG ODNs, as an adjuvant with rHBVvac in HBV-carrier mice. DCs expressing TLRs play an important role in triggering adaptive immunity to pathogens [53]. Recent studies have demonstrated that administering a TLR3 agonist enhanced the maturation of CD8α + DCs, which in turn promoted cross-presentation in the tumor microenvironment and tumor regression [54,55]. Concurring with these findings, cHBV vaccination upregulated the expression of CD40, CD80, and CD86 on mDCs and pDCs. Moreover, pDCs exhibited stronger activation than mDCs upon treatment with the cHBV-vaccine, mainly due to the presence of CpG M362. In addition, cHBV vaccination decreased PD-L1 expression on hepatocytes, which is pivotal to the progressive loss of HBV-specific T cell function during CHB infection and has been shown to affect treatment response in later clinical therapy [5,56].
Furthermore, mice administered the cHBV-vaccine exhibited significantly reduced Treg cell frequency and lower serum TGF-β1 levels compared to untreated mice. Meanwhile, both HBsAg and HBV DNA levels were nearly undetectable in the vaccinated mice during the long-term memory immune response assay on day 66, accompanied with higher levels of protective anti-HBs. Taken together, these results indicate that cHBV vaccination inhibits CHB-induced immune tolerance and triggers long-term anti-HBV-specific immunity in HBV-carrier mice. Furthermore, the results support that treatment with rHBVvac alone does not efficiently eliminate HBV, nor induce the same immunological effects as those induced by the cHBV-vaccine.
Antigen-experienced CD8 + T cells after infection or vaccination exhibit upregulated expression of CD11a and downregulated expression of CD8α on CD8 + T cells, whereas inflammatory stimulation alone, such as CpG administration, does not drive these changes [40]. During CHB infection, HBV-specific CD8 + T cells are known to gradually lose effector functions, proliferative capacity, and cytolytic activity, in that order [5,56]. In the current study, cHBV vaccination increased the abundance of CD11a hi CD8α lo cells, as well as the proliferative capacity and activation of HBV-specific CD8 + T cells. Furthermore, cHBV-vaccinated mice exhibited significantly increased serum levels of IFN-γ without elevated ALT levels, which concurred with previous studies reporting that IFN-γ mediates non-cytolytic clearance of HBV from hepatocytes without liver damage [5,60]. Therefore, the therapeutic effects of the cHBV-vaccine are likely mediated via a non-cytolytic HBV CD8 + T cell effect.
Follicular CXCR5-expressing CD8 + T cells are a major reservoir of long-term immunity [24,61]. Compared to the CXCR5-subset, CXCR5 + CD8 + T cells are localized to the splenic B cell follicles and tend to display an exhausted phenotype, expressing intermediate levels of PD-1 and TIGIT. In addition, they have enhanced effector potential, self-renewal capacity, and are negatively correlated with HBV progression [27,61]. HBV-specific CD11a hi CD8α lo cells in the spleens of HBV-carrier mice abundantly expressed CXCR5, and were the predominant subset among exhausted cells. However, cHBV vaccination significantly reduced the expression of multiple co-inhibitory receptors on CXCR5 + CD11a hi CD8α lo cells, indicating that HBV was eliminated by restoring exhausted CXCR5 + HBV-specific CD8 + T cells.
Effector CD4 + T cells also participate in humoral and cellular immune responses by generating and maintaining both neutralizing antibodies and CD8 + T cells to facilitate HBV clearance [62,63]. Moreover, antigen-specific CD4 + T cells reportedly display CD11a hi CD49d + surface marker expression [41]. Thus, cHBV vaccination increased the generation and activation of HBV-specific CD4 + T cells, indicating that CD4 + T cells might contribute to the humoral and CD8 + T cell response. Combined, these results suggest that augmented HBV-specific CD8 + and CD4 + T-cell responses induced by the cHBV-vaccine contribute to the elimination of HBV. Moreover, effective vaccines against viruses should generate a stable high-quality population of memory CD8 + T cells to confer long-term protective immunity [43,46,47,64]. The abundance of HBV-specific CD11a hi CD8α lo cells, which were predominantly CXCR5 + CD11a hi CD8α lo T cells, increased remarkably after HBV re-challenge in cHBV-vaccinated mice, suggesting the cHBV-vaccine promoted a long-term memory response against the virus in HBV-carrier mice, accompanied by higher levels of protective anti-HBs.
In conclusion, CpG ODNs have been widely used as adjuvants for cancer and antiviral vaccines, such as CpG 1018 (CpG-B ODN) in HEPLISAV-B [14][15][16][17][18]. Moreover, a dose of spike (S)-protein (S-Trimer) combined with CpG 1018/Alum adjuvants induced robust humoral and cellular immune responses against SARS-CoV-2 in a phase II/III trial [65,66]. In the current study, the CpG M362-based HBV vaccine overcame systemic immune tolerance in HBV-carrier mice and restored exhausted HBV-specific CD8 + T cells, which were predominantly CXCR5 + CD11a hi CD8α lo T cells. More importantly, this approach effectively induced long-term immune memory against HBV recurrence. Considering that CpG M362 combines the features of both class A and B CpG ODNs, the cHBV-vaccine provides a promising candidate for anti-HBV immunotherapy and the prevention of HBV. | 4,887.4 | 2022-01-01T00:00:00.000 | [
"Biology",
"Medicine"
] |
Numerical Analysis on the Thermal Performance in an Excavating Roadway with Auxiliary Ventilation System
A steady and proper thermal environment in deep underground is imperative to ensure worker health and production safety. Understanding the thermal performance in the roadway is the premise of temperature prediction; ventilation design; and improvement in cooling efficiency. A full coupled model incorporated with a moving mesh method was adopted; reflecting the dynamic condition of roadway construction. This study revealed the characteristics of the thermal performance and its evolution law in an excavating roadway. Several scenarios were performed to examine the designs of the auxiliary ventilation system on thermal performance in the roadway. The results show that there is a limitation in the cooling effect by continuously increasing the ventilation volume. Reducing the diameter of the air duct or distances between the duct outlet and the working face will aggravate the heat hazard in the roadway. The heat release from the roadway wall increases with the increase of the advance rate of the working face or roadway section size. Furthermore; an orthogonal experiment was conducted to investigate the effect of major factors on the average air temperature and local heat accumulation in the roadway
Introduction
With the increase in decreasing of shallow resources, deeper resource exploitation in high-temperature circumstances has become a necessity [1,2]. The thermal stresses resulted from high-temperature surrounding rock become a major challenge for the deep construction and safe operation [3,4]. The high-temperature environment not only threatens the health of workers, reduces the work efficiency, but also shortens the mechanical lifetime limit [5][6][7]. The heat hazard occurs frequently in the development zone of roadway owing to the high original rock temperature and poor ventilation [8]. The auxiliary ventilation is the major and common method for providing cooling energy and controlling heat hazards in the construction of roadway [9,10]. It is a challenge that designing an appropriate ventilation system and improving the cooling performance to control heat hazard in the high temperature roadway.
The mechanical ventilation is a crucial factor to eliminate pollutants and ensure safe production in the roadway. Therefore, a series of studies have been conducted by scholars to design an appropriate ventilation pattern aiming at different mining environment. Parra et al. established three types of ventilation models to identify the dead zone and the risk of methane explosion under different criteria respectively [11]. Hasheminasab et al. examined the distribution of methane in the development zones of underground coal mines at different ventilation scenarios [12]. Chang et al. [13] and Huang et al. [14] analyzed the dispersion and accumulation of CO in the roadway and determined areas with lower discharge. Wang et al. performed a long roadway model and revealed the relationships among the pressure difference, the air leakage rate and air quantity [15]. The airflow conditions have a significant relationship with dust movement, the extensive studies also have been carried out to examine the diffusion rules of dust and the optimal dust-removal airflow rate [16][17][18]. These subjects have been widely studied, but there are few studies focused on the problem of heat hazard control by auxiliary ventilation in the mine roadway, and the auxiliary ventilation plays an important role in regulating air temperature [19].
The thermal performance in the roadway can be classified into two types: operation period and excavated period. In an operation roadway, the airflow velocity near the wall of the roadway is small, and the heat transfer is relatively stable and slow [20,21]. The local temperature difference in airflow is low. The fluctuation of airflow temperature inside the roadway is mainly determined by the ventilation temperature, the length of the roadway and ventilation time [22,23]. While in an excavating roadway, the temperature of airflow and surrounding rock change more dramatically and rapidly [24,25]. Especially in the development zone, the airflow is a turbulent state, and there are significant differences in airflow velocity in the wall of roadway, which results in obvious differences in heat flux of the surrounding rock [26].
Maintaining a steady and comfortable thermal environment in underground spaces has received more attention in recent years due to the more severe heat hazard situation. Various cooling systems such as split-type vapor compression refrigerators [27], high temperature exchange machinery system (HEMS) [28,29], liquid carbon dioxide cycle refrigeration systems [30] et al. have been proposed for removing heat and improving working conditions. Although these approaches can provide a huge amount of cryogenic energy for heat hazard control in the underground, an improper application of cold air will increase the cost of cooling energy and reduce the thermal comfort for workers. Therefore, understanding the cooling characteristic of air temperature in the roadway and the influence of ventilation pattern is absolutely critical for heat hazard control.
Several studies investigated the characteristics of airflow and heat transfer in the development zones to predict the thermal performance and design the schemes of heat hazard control. The thermal performance in the roadway is affected by variety of factors, such as the temperature of rock and ventilation, state of airflow, mining situation, production time, et al. [31,32] Zhang et al. developed a physical simulation test system and explored the evolution law of temperature for surrounding rock [33]. Wang et al. calculated the heat release source from the roadway based on ignoring the local temperature difference, and proposed the prediction formula for airflow temperature in the roadway [34]. Besides, some of numerical models have been developed to describe the heat transfer in rock mass [35][36][37]. Habibi et al. measured the rock thermal conductivity and developed a numerical model investigating the heat transfer between air and rock [38]. In addition, Ji et al. theoretically analyzed the characteristics of heat transfer at working face under a jet flow [39].
Although a considerable quantity of studies focused on the cooling system and heat transfer in a roadway, few research revealed the thermal performance in an excavating roadway, considering the convective heat transfer between surrounding rock and airflow, unsteady-state heat transfer in rock, non-isothermal flow in the roadway, and the advance of the working face. To reflect the dynamic condition of roadway construction, a fully coupled mathematical model incorporated with a moving mesh method is adopted. The aim is to understand the cause of high temperature in the excavating roadway and efficiently control the heat damage. The characteristics of the thermal performance in an excavating roadway, and its evolution law are generally obtained and analyzed. The crucial factors such as the ventilation volume, the diameter of air duct, the distance between duct outlet and working face and the roadway section size affecting the air temperature in the roadway will be investigated in detail. An orthogonal test is carried out to comprehensively analyze the experiment and find out the influence rule of each factor on the thermal performance in the excavating roadway. This study can provide a robust theoretical basis for saving cooling energy in heat hazard control and improving the thermal comfort in roadway construction.
Numerical Methods
In construction of a roadway, the heat hazard is most severe in the development zones as the high-temperature of the original rock. The thermal performance inside the roadway is affected by the airflow and temperature of the rock, which is a fluid-solid coupling process. The airflow state is crucial to the heat transfer in the wall of roadway, which is determined by solving the continuity, momentum, and turbulence model based on the Navier-Stokes equation. The realizable k-ε turbulent model is selected as the governing equation for the turbulent kinetic energy, and it has great advantages in simulation for isothermal flow, especially near-wall flow it has good accuracy and efficiency [40,41]. The energy balance equation is used to control the heat transfer in rock and fluid. The established geometric model is shown in Figure 1, and it is similar to that of the actual situation, including an excavating roadway equipped with an air duct, unexcavated rock and surrounding rock. The shape of the roadway section adopts the conventional arch roadway with a width of 5 m, a straight wall of 1.5 m, an arch height of 0.75 m. The initial length of the excavated roadway is 18 m, as well as the length of the air duct is 12 m. The auxiliary ventilation duct with a diameter of 0.6 m is arranged on the left side of the roadway, and the duct outlet is 6 m away from the working face. The original temperature of the surrounding rock and air in the roadway is 45 • C, and the advance rate of the working face is 0.2 m/h. The main parameters of the model are set in Table 1, which are substituted from the contemporary literature [42,43].
Numerical Methods
In construction of a roadway, the heat hazard is most severe in th zones as the high-temperature of the original rock. The thermal perform roadway is affected by the airflow and temperature of the rock, which is a pling process. The airflow state is crucial to the heat transfer in the wall of is determined by solving the continuity, momentum, and turbulence mod Navier-Stokes equation. The realizable k-ε turbulent model is selected a equation for the turbulent kinetic energy, and it has great advantages in isothermal flow, especially near-wall flow it has good accuracy and efficie energy balance equation is used to control the heat transfer in rock and fl The established geometric model is shown in Figure 1, and it is simi actual situation, including an excavating roadway equipped with an air du rock and surrounding rock. The shape of the roadway section adopts t arch roadway with a width of 5 m, a straight wall of 1.5 m, an arch heigh initial length of the excavated roadway is 18 m, as well as the length of t m. The auxiliary ventilation duct with a diameter of 0.6 m is arranged on the roadway, and the duct outlet is 6 m away from the working face. The o ature of the surrounding rock and air in the roadway is 45 °C, and the adv working face is 0.2 m/h. The main parameters of the model are set in Ta substituted from the contemporary literature [42,43]. A moving mesh method is incorporated into the mathematical model to perform the dynamic excavation of the roadway. The advance of the working face represents the excavation of the roadway. In the case that the region of roadway and rock are transformed as the advance of working face, the mapped mesh will deform with the movement of the boundary. Avoiding the mesh distortion arisen from the deformation of mapped mesh, the Laplace smoothing method is adopted to modify the deformation of the mesh by global convergence. Figure 2 depicts the grid movement of the physical model. In the process of roadway excavation, the working face of the roadway gradually advances to the unexcavated rock, and the length of the roadway increases correspondingly. The distance between the duct outlet and the working face mains the same. All the governing equations are implemented and numerically solved using the COMSOL Multiphysics software based on the finite element method. Simultaneously, as the temperature of surrounding rock and heat transfer are related to time, the transient solver is selected.
Parameters Value
Density of rock, ρ s (kg·m −3 ) 2600 Specific heat capacity of rock, C ps (J·(kg·K) −1 ) 1300 Heat conduction coefficient of rock, K ps (W·(m·K) −1 ) 3.5 Density of gas, ρ g (kg·m −3 ) 1.213 Gas dynamic viscosity, µ g (Pa·s) 1.84 × 10 −5 Heat conduction coefficient of gas, K pg (W·(m·K) −1 ) 0.259 Specific heat capacity of gas, C pg (J·(kg·K) −1 ) 1012 The working face advance rate, R a (m/h) 0.2 The diameter of air duct, d a (m) 0.6 The distance between duct outlet and working face, D s (m) 6 The initial rock temperature, T r ( • C) 45 The initial temperature of ventilation airflow, T a ( • C) 25 A moving mesh method is incorporated into the mathematical mod dynamic excavation of the roadway. The advance of the working face r cavation of the roadway. In the case that the region of roadway and rock as the advance of working face, the mapped mesh will deform with the boundary. Avoiding the mesh distortion arisen from the deformation o the Laplace smoothing method is adopted to modify the deformation global convergence. Figure 2 depicts the grid movement of the physical m cess of roadway excavation, the working face of the roadway gradually unexcavated rock, and the length of the roadway increases correspondin between the duct outlet and the working face mains the same. All the gov are implemented and numerically solved using the COMSOL Multi based on the finite element method. Simultaneously, as the temperatur rock and heat transfer are related to time, the transient solver is selected
Numerical Simulation of the Airflow Field
The characteristic of airflow in the development zone of the roadw cavation is shown in Figure 3. The condition of the airflow state near th very complicated. The cold fresh air is released from the air duct to the ro of it impinges on the working face and then turns and flows to the road the working face, the airflow velocity on the two sides of the walls is larg fields can be distinguished in the roadway: Jet zone, Backflow zone, and
Numerical Simulation of the Airflow Field
The characteristic of airflow in the development zone of the roadway during the excavation is shown in Figure 3. The condition of the airflow state near the working face is very complicated. The cold fresh air is released from the air duct to the roadway, and most of it impinges on the working face and then turns and flows to the roadway outlet. Near the working face, the airflow velocity on the two sides of the walls is larger. Three airflow fields can be distinguished in the roadway: Jet zone, Backflow zone, and Vortex zone. The Jet zone is located in the space between the duct outlet and the working face, the airflow jetted from the air duct to the working face, and the air along the path is continuously sucked in, which leads to an increase and diffusion of air along the path. As the limitation of space, the airflow collided with the left-hand side wall of the roadway and the working face generates an opposite direction airflow in the Backflow zone (right-hand side wall), and then it flows toward the roadway outlet. A swirling vortex flow formed in the middle of the roadway near the working face and exhibited a triangle pattern, and the reason for the formation is the entrainment action of the jet airflow in Jet zone and return airflow in the Backflow zone. The velocity of airflow is the highest in the jet region and the lowest in the vortex region.
Numerical Simulation of the Temperature Field
The cold fresh air supplied by auxiliary ventilation flows in the roadway, causing simultaneous cooling of the surrounding rock and heating of the airflow due to convective heat transfer. The evolution of the temperature field in the development of roadway is evident in Figure 4. At the beginning stages of auxiliary ventilation, the temperature decreases signally in the most zone of the roadway, and a local high-temperature region is presented. Compared Figure 4 with the airflow in Figure 3, it can be found that the temperature characteristics in the roadway are associated with the airflow state. The lowest temperature region is located in the Jet zone due to the cold fresh air supplied by auxiliary ventilation. In the Backflow Zone, the temperature of airflow increases significantly, especially in the area near the wall. It is evident that the heat transfer between the airflow and the roadway wall is larger here. When the airflow flows toward the roadway outlet, only a small part of the airflow is sucked into the Vortex zone. Consequently, the air in the Vortex zone is hard to refresh, and its temperature is higher than in the surrounding region. The local high-temperature zone moves forward as the advance of the working face and keeps a constant distance (4-5 m) from the working face.
Numerical Simulation of the Temperature Field
The cold fresh air supplied by auxiliary ventilation flows in the roadway, causing simultaneous cooling of the surrounding rock and heating of the airflow due to convective heat transfer. The evolution of the temperature field in the development of roadway is evident in Figure 4. At the beginning stages of auxiliary ventilation, the temperature decreases signally in the most zone of the roadway, and a local high-temperature region is presented. Compared Figure 4 with the airflow in Figure 3, it can be found that the temperature characteristics in the roadway are associated with the airflow state. The lowest temperature region is located in the Jet zone due to the cold fresh air supplied by auxiliary ventilation. In the Backflow Zone, the temperature of airflow increases significantly, especially in the area near the wall. It is evident that the heat transfer between the airflow and the roadway wall is larger here. When the airflow flows toward the roadway outlet, only a small part of the airflow is sucked into the Vortex zone. Consequently, the air in the Vortex zone is hard to refresh, and its temperature is higher than in the surrounding region. The local high-temperature zone moves forward as the advance of the working face and keeps a constant distance (4-5 m) from the working face.
With the increase of ventilation time, the surrounding rock is continuously cooled, and the heat exchange between the airflow and the wall of the roadway decreased, so the air temperature in the roadway decreased. As the ventilation progresses, the heat transfer between the rock and airflow achieved a thermal equilibrium, and the temperature field in the roadway tends to stabilize. Figure 5 presents the air temperature in the central axis of the roadway (Z = 0.5 m) at different times. The peak of air temperature does not exist near the working face. On the contrary, the temperature near the working face is low, and the air temperature within 1 m from the working face is between 27.8-28.2 • C. The air temperature in the roadway rises first and then decreases with the increase of distance from the working face. The air temperature peak existed 5 m away from the working face. After 1 h of ventilation, the air temperature peak on the curve is 31.6 • C, and 3.3 • C higher than the air temperature at the exit of the roadway. The air temperature peak gradually decreased with the ventilation time, and finally stabilized at about 30.5 • C. The air temperature at the exit of the roadway decreased from 28.2 • C in the 1 h to 27.7 • C in the 29 h. creases signally in the most zone of the roadway, and a local high-temperature region is presented. Compared Figure 4 with the airflow in Figure 3, it can be found that the temperature characteristics in the roadway are associated with the airflow state. The lowest temperature region is located in the Jet zone due to the cold fresh air supplied by auxiliary ventilation. In the Backflow Zone, the temperature of airflow increases significantly, especially in the area near the wall. It is evident that the heat transfer between the airflow and the roadway wall is larger here. When the airflow flows toward the roadway outlet, only a small part of the airflow is sucked into the Vortex zone. Consequently, the air in the Vortex zone is hard to refresh, and its temperature is higher than in the surrounding region. The local high-temperature zone moves forward as the advance of the working face and keeps a constant distance (4-5 m) from the working face. With the increase of ventilation time, the surrounding rock is cont and the heat exchange between the airflow and the wall of the roadway d air temperature in the roadway decreased. As the ventilation progresses, between the rock and airflow achieved a thermal equilibrium, and the t in the roadway tends to stabilize. Figure 5 presents the air temperature i of the roadway (Z = 0.5 m) at different times. The peak of air temperatu near the working face. On the contrary, the temperature near the working the air temperature within 1 m from the working face is between 27.8temperature in the roadway rises first and then decreases with the incr from the working face. The air temperature peak existed 5 m away from t After 1 h of ventilation, the air temperature peak on the curve is 31.6 °C, a than the air temperature at the exit of the roadway. The air temperature decreased with the ventilation time, and finally stabilized at about 30.5 perature at the exit of the roadway decreased from 28.2 °C in the 1 h to 2 hour.
Sensitivity Analysis for Single Factor
There is no doubt that the air temperature in roadway will be directl auxiliary ventilation system and mining situation. So a number of scenari to examine the thermal performance under different auxiliary ventilation cavated condition.
Sensitivity Analysis for Single Factor
There is no doubt that the air temperature in roadway will be directly affected by the auxiliary ventilation system and mining situation. So a number of scenarios are simulated to examine the thermal performance under different auxiliary ventilation pattern and excavated condition.
The Effect of Ventilation Volume
The ventilation volume had a greater impact on the airflow field and the convective heat transfer between airflow and surrounding rock, so it is an essential factor for the thermal performance in the roadway. Five cases with different air duct ventilation velocity are performed in this section, i.e., u = 4, 5, 6, 7 and 8 m/s, and the other parameters, D s = 6 m, d a = 0.6 m, T r = 45 • C, R a = 0.2 m/h, remain unchanged. Figure 6 depicts the air temperature distribution in the roadway under different ventilation velocity. A local high-temperature zone still existed in the middle of the roadway regardless of the ventilation volume, and it expands with decreasing ventilation volume. The ventilation volume has an obvious influence on the temperature at the place near the working face and the right-hand wall of the roadway, where the air temperature decreases significantly with increasing ventilation volume. Figure 7 shows that the air temperature in the central axis of roadway decreases with the increase of ventilation volume, and the peak of air temperature decreases more quickly as the ventilation time. The larger the ventilation volume decreases the air temperature in most zone of the roadway, especially in high airflow velocity areas as the quick heat diffuses. Compared to the curves in Figure 7, the air temperature difference between the curves is gradually reduced with increases in ventilation volume, so there is a limit to adjust the air temperature in roadway by increasing ventilation volume. decreases with the increase of ventilation volume, and the peak of air temperature decreases more quickly as the ventilation time. The larger the ventilation volume decreases the air temperature in most zone of the roadway, especially in high airflow velocity areas as the quick heat diffuses. Compared to the curves in Figure 7, the air temperature difference between the curves is gradually reduced with increases in ventilation volume, so there is a limit to adjust the air temperature in roadway by increasing ventilation volume.
The Effects of the Diameter of Air Duct
The diameter of the air duct directly affects the released airflow velocity, so it is a critical factor to the thermal performance in the roadway. In this section, the airflow and thermal performance in the roadway with a constant ventilation volume of Q = 102 m 3 /min and diameter of air duct sizes of d m = 0.5, d m = 0.6, d m = 0.7 and d m = 0.8 m are investigated. Figure 8 shows that when the diameter of the air duct decreases, the airflow velocity increases in the roadway especially near the roadway wall and the working face, which increases the convective heat transfer between the surrounding rock and airflow. Therefore, the smaller the air duct, the higher the air temperature in the roadway and the larger the area of the local high-temperature zone, as shown in Figure 9. When the diameters of the air duct are 0.5, 0.6, 0.7 and 0.8 m respectively, the air temperatures in the exit of the roadway are 29.16 • C, 27.93 • C, 27.28 • C, 26.81 • C respectively, the maximum difference in the air temperatures is 2.35 • C. When the diameter of the air duct is 0.8 m, the maximum temperature in the vortex zone is higher than that when the diameter of the air duct is 0.7 m. This is because although the reduction in duct diameter reduces the area of the vortex zone, the low airflow velocity results in a more serious accumulation of heat in the vortex zone. Figure 10 shows that there is an obvious variation in air temperature near the working face as the air duct diameter, which results from the change in airflow velocity. In general, increasing the diameter of the air duct has a benefit to improving the cooling effect in the roadway by weakening convection heat transfer in the working face. However, the selection of air duct diameter is limited by the roadway construction space.
The Effects of the Diameter of Air Duct
The diameter of the air duct directly affects the released airflow velocity, so it is a critical factor to the thermal performance in the roadway. In this section, the airflow and thermal performance in the roadway with a constant ventilation volume of Q = 102 m 3 /min and diameter of air duct sizes of dm = 0.5, dm = 0.6, dm = 0.7 and dm = 0.8 m are investigated. Figure 8 shows that when the diameter of the air duct decreases, the airflow velocity increases in the roadway especially near the roadway wall and the working face, which increases the convective heat transfer between the surrounding rock and airflow. Therefore, the smaller the air duct, the higher the air temperature in the roadway and the larger the area of the local high-temperature zone, as shown in Figure 9. When the diameters of the air duct are 0.5, 0.6, 0.7 and 0.8 m respectively, the air temperatures in the exit of the roadway are 29.16 °C, 27.93 °C, 27.28 °C, 26.81 °C respectively, the maximum difference in the air temperatures is 2.35 °C. When the diameter of the air duct is 0.8 m, the maximum temperature in the vortex zone is higher than that when the diameter of the air duct is 0.7 m. This is because although the reduction in duct diameter reduces the area of the vortex zone, the low airflow velocity results in a more serious accumulation of heat in the vortex zone. Figure 10 shows that there is an obvious variation in air temperature near the working face as the air duct diameter, which results from the change in airflow velocity. In general, increasing the diameter of the air duct has a benefit to improving the cooling effect in the roadway by weakening convection heat transfer in the working face. However, the selection of air duct diameter is limited by the roadway construction space. Figure 11 depicts the airflow performance with a constant ventilation volum 102 m 3 /min and different distances between the duct outlet and working face. Inc the distance between the duct outlet and the working face, the center of the vortex further away from the working face. Meanwhile, the area of the local high-temp zone reduces with the increase of distance between the duct outlet and working shown in Figure 12. So the increase of the distance between the duct outlet and w face can prevent the accumulation of heat in the roadway to some extent. As we from Figure 13a, when the distance between the duct outlet and working face is 8 m is no obvious peak of air temperature on the center line. In general, the longer the d between the duct outlet and working face, the lower the air temperature in the ro According to Figure 13b, it can be concluded that increasing the distance between duct and the working face will reduce the airflow velocity near the working face, so the convective heat transfer between the surrounding rock and airflow decreases results in a reduction of heat release from the working face. Therefore, the coolin in the roadway can be appropriately enhanced by increasing the distance between Figure 11 depicts the airflow performance with a constant ventilation volume of Q = 102 m 3 /min and different distances between the duct outlet and working face. Increasing the distance between the duct outlet and the working face, the center of the vortex zone is further away from the working face. Meanwhile, the area of the local high-temperature zone reduces with the increase of distance between the duct outlet and working face, as shown in Figure 12. So the increase of the distance between the duct outlet and working face can prevent the accumulation of heat in the roadway to some extent. As we can see from Figure 13a, when the distance between the duct outlet and working face is 8 m, there is no obvious peak of air temperature on the center line. In general, the longer the distance between the duct outlet and working face, the lower the air temperature in the roadway. According to Figure 13b, it can be concluded that increasing the distance between the air duct and the working face will reduce the airflow velocity near the working face, so where the convective heat transfer between the surrounding rock and airflow decreases, which results in a reduction of heat release from the working face. Therefore, the cooling effect in the roadway can be appropriately enhanced by increasing the distance between the air duct and the working face. Nonetheless, it should pay attention to the accumulation of toxic and harmful gases near the working face caused by the reduction of airflow near the working face as the increase of the distance between the air duct and the working face.
The Effect of the Distance between Duct Outlet and Working Face
According to Figure 13b, it can be concluded that increasing the distance b duct and the working face will reduce the airflow velocity near the working the convective heat transfer between the surrounding rock and airflow de results in a reduction of heat release from the working face. Therefore, the in the roadway can be appropriately enhanced by increasing the distance b duct and the working face. Nonetheless, it should pay attention to the ac toxic and harmful gases near the working face caused by the reduction of a working face as the increase of the distance between the air duct and the w Figure 14 depicts that the air temperature in the Z = 0.5 m plane under a different advance rate of the working face. The adjustment of the advance rate of the working face has no significant effect on the airflow field. At a different advance rate of the working face, the local high-temperature zone maintains the same distance from the working face and the air temperature in the local high-temperature zone is higher. The air temperature in the central axis of roadway increases with the mining speed as shown in Figure 15. As the advance rate of the working face is increased from 0.1 to 0.5 m/s, the peak of air temperature on the central axis of roadway raises from 30.48 to 30.96 °C. The main reason is that the faster the advance rate of the working face is, the higher the temperature of the working face, and more heat release from the working face. Besides, the rapid growth of the length of roadway also increases the heat release from the surrounding rock. Therefore, decreasing the advance rate of the working face can improve the thermal environment for the workers. Figure 14 depicts that the air temperature in the Z = 0.5 m plane under a different advance rate of the working face. The adjustment of the advance rate of the working face has no significant effect on the airflow field. At a different advance rate of the working face, the local high-temperature zone maintains the same distance from the working face and the air temperature in the local high-temperature zone is higher. The air temperature in the central axis of roadway increases with the mining speed as shown in Figure 15. As the advance rate of the working face is increased from 0.1 to 0.5 m/s, the peak of air temperature on the central axis of roadway raises from 30.48 to 30.96 • C. The main reason is that the faster the advance rate of the working face is, the higher the temperature of the working face, and more heat release from the working face. Besides, the rapid growth of the length of roadway also increases the heat release from the surrounding rock. Therefore, decreasing the advance rate of the working face can improve the thermal environment for the workers.
The Effects of the Roadway Section Size
At a constant ventilation volume, the variation of the size of the roadway section will affect the movement of airflow in the roadway. The thermal performance in the roadway under different roadway section sizes of U = 5.276, U = 6.075, U = 6.886 and U = 7.696 m 2 are investigated. When the roadway section size is 5.276 m 2 , the airflow field has an obvious change in the roadway and there are two cycle flow areas in the Z section as shown in Figure 16. The airflow jetted from the air duct does not flow directly along the wall to the working face, but first moves to the other side of the roadway and then flows to the working face. In that scenario, in the range of 0~3 m from the working face, the air temperature decreases obviously, but the air temperature rises elsewhere, which can be seen in Figure 17. When the roadway section size is large, the airflow directly flows to the working face and then back to the exit of the roadway, the low temperature zone is very close to the working face and road wall. This implies that it is essential to control the movement of airflow, and the airflow should first cool down where people and machines are, which can promote the utilization of cool energy. As the roadway section size increases, the surface area of the surrounding rock of roadway increases, and the air temperature increases due to the increase in the heat release from the surrounding rock. The main region of temperature rise is in the Backflow zone of the right wall and Vortex zone.
The Effects of the Roadway Section Size
At a constant ventilation volume, the variation of the size of the roadway section will affect the movement of airflow in the roadway. The thermal performance in the roadway under different roadway section sizes of U = 5.276, U = 6.075, U = 6.886 and U = 7.696 m 2 are investigated. When the roadway section size is 5.276 m 2 , the airflow field has an obvious change in the roadway and there are two cycle flow areas in the Z section as shown in Figure 16. The airflow jetted from the air duct does not flow directly along the wall to the working face, but first moves to the other side of the roadway and then flows to the working face. In that scenario, in the range of 0~3 m from the working face, the air temperature decreases obviously, but the air temperature rises elsewhere, which can be seen in Figure 17. When the roadway section size is large, the airflow directly flows to the working face and then back to the exit of the roadway, the low temperature zone is very close to the working face and road wall. This implies that it is essential to control the movement of airflow, and the airflow should first cool down where people and machines are, which can promote the utilization of cool energy. As the roadway section size increases, the surface area of the surrounding rock of roadway increases, and the air temperature increases due to the increase in the heat release from the surrounding rock. The main region of temperature rise is in the Backflow zone of the right wall and Vortex zone. When the section area of roadway increases from 6.075 to 7.696 m, the air temperature at the exit of roadway rises from 27.9 to 30.1 • C, as shown in Figure 18. Thus, once the roadway section size increases during the construction of roadway, the cooling approaches should be strengthened correspondingly. When the section area of roadway increases from 6.075 to 7.696 m, the air temperature at the exit of roadway rises from 27.9 to 30.1 °C, as shown in Figure 18. Thus, once the roadway section size increases during the construction of roadway, the cooling approaches should be strengthened correspondingly.
Orthogonal Test
The orthogonal test method is to select representative tests from a large number of tests according to the mathematical statistics and principle of orthogonality. It can not only greatly reduce the number of tests, but also can comprehensively analyze the experiment and find out the influence rule of each factor on the evaluation index of the experiment.
In this paper, a seven-factor and eighteen-level orthogonal (U 18 *(3 7 )) test is designed to evaluate the influence of factors such as the ventilation volume, initial temperature of ventilation airflow, advance rate of working face, distance between duct outlet and working face, roadway section size, temperature of surrounding rock and diameter of the air duct. The average temperature at the exit of the roadway (ATE) and the difference between the peak of air temperature and the air temperature in the exit of the roadway on the center line (DPE) is selected for the list analysis. The index of ATE can reflect the influence of the factors on the average air temperature in the roadway, and the DPE index depicts the degree of local heat accumulation. The orthogonal experiment table for the simulation is shown in Table 2. The maximum difference of test results for three groups of each influencing factor is defined as the Range, and the sensitivity of each factor on the roadway temperature field is determined by the Range, in other words, the greater the value, the larger the influence of the factor. Table 3 shows the sensitivity of factors for each index. The orthogonal test of the ATT presents that the Range of the initial temperature of ventilation airflow is the largest, indicating that the initial temperature of ventilation airflow has the greatest influence on the average air temperature in the roadway. The influence of the surrounding rock temperature is next in line for the average air temperature and the roadway section size is minimal. The factors impacting the average air temperature in the roadway in a descending sequence are as follows, the initial temperature of ventilation airflow > temperature of surrounding rock > ventilation volume > diameter of air duct > distance between duct outlet and working face > advance rate of working face > roadway section size. These factors have different performances on the local heat accumulation in the roadway. The temperature of the surrounding rock has the greatest influence on the DPE, which means that the higher the temperature of surrounding rock is, the easier it is to accumulate heat and form a local high-temperature zone in the roadway. The influence of the advance rate of working face to the local heat accumulation is the least. The sensitive comparison of the factors on the indexes of ATE and DPE is shown in Figure 19.
In general, in the construction of a roadway, the temperature of surrounding rock and initial temperature of ventilation airflow have the greatest influence on the average air temperature and local heat accumulation in the roadway. Therefore, reducing the temperature of surrounding rock or the initial temperature of ventilation airflow is the most powerful way to control the thermal hazard in the roadway. However, these two approaches are both costly for improving the cooling effect in practice. Increasing the ventilation volume is the most common method to promote thermal comfort, but there is a limit to adjust air temperature. Adjusting the size of the air duct and the distance between the duct and the working face are often neglected in the past, but in fact, it also can appropriately regulate the air temperature in the roadway, especially in local heat accumulation in the roadway. propriately regulate the air temperature in the roadway, especially in local lation in the roadway.
Conclusions
The thermal stresses of the working environment in the roadway significantly affect worker health, labor productivity, and the failure likelihood of both equipment and workers. The thermal performance in an excavating mine roadway equipped with auxiliary ventilation was investigated by three-dimensional numerical simulations. The airflow pattern and temperature field in the excavating mine roadway under different ventilation times were obtained and analyzed. The critical factors, such as the ventilation volume, diameter of the air duct, distance between duct outlet and working face, advance rate of the working face and roadway section size, were investigated to determine how to change the thermal performance in the roadway. An orthogonal experiment was performed on the average temperature at the exit of the roadway and the difference between the peak of air temperature and the temperature of exit of the roadway on the center line, for examining the effect of different critical factors on the average air temperature and local heat accumulation in the roadway. The main conclusions are as follows: (1) The airflow field is distinguished into three parts: Jet zone, Backflow zone, and Vortex zone. The triangular swirling vortex exists in the middle of the roadway and 4-5 m away from the working face where the heat is easily accumulated and the air temperature is high. (2) Under the condition of continuous ventilation and excavation of roadway, the air temperature in the roadway decreases first and then go stabilized. The local hightemperature zone in the roadway moves forward with the advance of the working face.
(3) Increasing the ventilation volume can promote the thermal environment in the roadway, but there is a limit to adjust air temperature. Reducing the diameter of air duct or distance between the duct outlet and the working face will increase the airflow velocity near the working face and enhance the convective heat in the working face, which leads to an increase in air temperature in the roadway. (4) The temperature of surrounding rock, the initial temperature of ventilation airflow and ventilation volume have a significant influence on the air temperature and local heat accumulation in the roadway, and decreasing the initial temperature of ventilation airflow and the temperature of surrounding rock is the key to control the heat hazard in an excavating roadway. The priority of control measures for heat hazard can be determined by referring to the sensitivity degree of factors. | 10,145.8 | 2021-01-29T00:00:00.000 | [
"Engineering"
] |
Hydrated Salt/Graphite/Polyelectrolyte Organic-Inorganic Hybrids for Efficient Thermochemical Storage
Hydrated salt thermochemical energy storage (TES) is a promising technology for high density energy storage, in principle opening the way for applications in seasonal storage. However, severe limitations are affecting large scale applications, related to their poor thermal and mechanical stability on hydration/dehydration cycling. In this paper, we report the preparation and characterization of composite materials manufactured with a wet impregnation method using strontium bromide hexahydrate (SBH) as a thermochemical storage material, combined with expanded natural graphite (G). In addition to these fully inorganic formulations, an organic polyelectrolyte (PDAC, polydiallyldimethylammonium chloride) was exploited in the structure, with the aim to stabilize the salt, while contributing to the sorption/desorption process. Different formulations were prepared with varying PDAC concentration to study its contribution to material morphology, by electron microscopy and X-ray diffraction, as well as water sorption/desorption properties, by thermogravimetry and differential calorimetry. Furthermore, the SBH/G/PDAC powder mixture was pressed to form tabs that were analyzed in a climatic chamber, which is evidence for an active role of PDAC in the improvement of water sorption, coupled with a significant enhancement of mechanical resistance upon hydration/dehydration cycling. Therefore, the addition of the polyelectrolyte is proposed as an innovative approach in the fabrication of efficient and durable TES devices.
Introduction
Fighting climate change is one of the biggest challenges, attracting research efforts from all over the world. Smart heat management is presently a central topic in greenhouse gas mitigation and the approach of thermal energy storage (TES) has a key role in achieving this goal [1]. The simple and fundamental concept at the base of TES is to identify heat sources with low environmental and economic impact, store their energy when it is not needed, and use it in later times instead of producing it with other traditional alternatives. These technologies have been proven to have a potential reduction in CO 2 emissions up to 5.5% compared to 1990 levels [2]. In particular, low temperature heat sources (up to 150 • C) are considered among the greatest opportunities in this field. These sources can be identified mainly in two scenarios: Waste heat reuse and smart heat management in buildings. In the first case, low grade heat that is commonly released to the environment may in principle be conveniently recovered and stored for later use. Several sources were identified in different industrial fields such as oil, chemicals, steel, glass [3], and food [4] industries, as well as municipal solid waste, mining wastes [5], data centers [6], and car engines [7]. On the other hand, one of the main issues in energy management in the building sector is the mismatch between heat supply and demand. In this field, two main areas are identified: Short-term storage (e.g., from night to day), and long-term storage (e.g., from summer to winter). Following these needs, many TES devices were successfully developed and implemented in both residential and commercial structures [8]. In addition to this, the increasing development of renewable energy sources creates the need to reinvent energy demand management resulting in integrated solutions where TES technologies are coupled with photovoltaic panels and/or solar thermal technologies in order to manage the peak of electricity demand and reduce the costs related to electricity consumption [9,10].
The most used classification of TES materials takes into account the form in which heat is stored. Sensible heat storage is the most widely adopted and well-known techniques, because it is based on cheap and highly available materials (e.g., water or concrete) with high specific heat. The second approach uses the latent heat of phase change materials (PCMs), such as paraffins, to obtain higher energy storage densities with respect to sensible heat [11]. The third class is thermochemical TES and it includes materials showing a high-enthalpy reversible gas/solid reaction. One of the most promising thermochemical materials (TCMs) class is inorganic salt hydrates (M n A m ·XH 2 O) [12], in which the storage reaction is: When heat is transferred from a selected source to the TCM, in the so-defined charging step, dehydration occurs. As long as the salt is maintained in the dehydrated state, latent heat is stored. When water is made available to the salt (discharging step), hydration occurs and hydration heat is released. The possibility to control the heat release by controlling the water feed to the dehydrated salt, is one of the main advantages of this technique, making the heat discharge controllable on demand [13]. The second important advantage of TCMs over PCMs is the higher (one order of magnitude) energy storage density associated with the employed materials [14]. Due to this great potential, many efforts were made to identify the best performing salt hydrates with both experimental [15,16], and theoretical methods [17,18]. Despite the selection of the hydrated salt primarily depending on the available source temperature, one of the most promising TCMs discussed in literature for low temperature TES applications is SrBr 2 ·6H 2 O (SBH) owing to its effective combination of a relatively high storage density (798 kJ/kg) and low dehydration temperature ( 100 • C) as reviewed in a detailed study [19]. Unfortunately, first efforts to implement this salt in a working device also showed some severe limitations, in terms of low thermal conductivity, low chemical stability over hydration/dehydration cycles, and slow mass/heat transfer [20]. Research efforts were mostly aimed at overcoming these limits by design improvement, while few studies dealt with new material concepts in order to fully exploit the potentialities of SBH. One practical approach is to include the TCMs in a porous matrix to overcome the drawbacks of solid/gas reactions [21]. In particular, expanded natural graphite (G) was proposed for combination with TCMs, based on its low price and density, coupled with high thermal conductivity and surface area. Recently, hybrid salt/graphite materials were prepared and tested [22][23][24], but in most cases an inherent incompatibility between the structures of the two materials resulted in big salt aggregates, thus minimizing water adsorption kinetics, and the rate of heat transfer between salt and graphite layers. In this manuscript, we aim at overcoming these limitations by producing graphite composites encompassing a polyelectrolyte binder, PDAC (polydiallyldimethylammonium chloride), to enhance the compatibility between salt and matrix. Indeed, PDAC is known to have a strong interaction with graphite layers [25], and it is expected to show a good affinity with ionic materials due to its high charge density. The aim is to obtain a better distribution of salt on the pores surfaces, maximizing the area of the air/salt and salt/graphite interfaces. In addition, PDAC also shows good moisture sorption ability, thus potentially improving the hydration kinetics of the salt [26].
Materials
PDAC (Mw = 400,000-500,000 g/mol) was purchased from Sigma-Aldrich ® as 20% wt/wt water solution. In order to obtain solid PDAC samples for the analyses, the solution was dried in an oven at 120 • C until constant weight was reached and hydrated in a climatic chamber at 23 • C and 50% relative humidity (RH) overnight. Expanded natural graphite (G), with 28.4 m 2 /g surface area (as reported in the material datasheet) was purchased by TIMCAL (Bodio, Switzerland), commercial grade TIMREX ® BNB90. SrBr 2 ·6H 2 O (S) with >95% purity in powder form was purchased from Alfa Aesar ® (Haverhill, MA, US). All reagents were used as received for preparing stable water dispersions using deionized water supplied by a Direct-Q ® 3 UV Millipore System (Milano, Italy).
TCM Composite Manufacturing
The main steps in the manufacturing process are depicted in Figure 1.
Materials
PDAC (Mw = 400,000-500,000 g/mol) was purchased from Sigma-Aldrich ® as 20% wt/wt water solution. In order to obtain solid PDAC samples for the analyses, the solution was dried in an oven at 120 °C until constant weight was reached and hydrated in a climatic chamber at 23 °C and 50% relative humidity (RH) overnight. Expanded natural graphite (G), with 28.4 m 2 /g surface area (as reported in the material datasheet) was purchased by TIMCAL (Bodio, Switzerland), commercial grade TIMREX ® BNB90. SrBr2•6H2O (S) with >95% purity in powder form was purchased from Alfa Aesar ® .(Haverhill, MA, US) All reagents were used as received for preparing stable water dispersions using deionized water supplied by a Direct-Q ® 3 UV Millipore System (Milano, Italy).
TCM Composite Manufacturing
The main steps in the manufacturing process are depicted in Figure 1. PDAC water solution was diluted with 30 mL of water with subsequent additions of G and SBH. Four samples were prepared varying the amount of polymer while keeping the G/SBH ratio constant (their compositions are shown in Table 1). The suspension was stirred overnight to obtain a homogeneous dispersion and then heated at 100 °C on a plate while stirring for around 5 h to remove water via evaporation. In this step, SBH and PDAC started precipitating on the graphite matrix. When the dispersion viscosity was too high to allow any further stirring, the wet material was placed in a vacuum oven at 50 °C overnight to complete the process. After these steps, the prepared mixtures were subjected to a complete dehydration and hydration cycle prior to the subsequent charge/discharge cycles [27]. The samples were dehydrated in a ventilated oven at 120 °C until they reached constant weight and rehydrated in a climatic chamber at 23 °C and 50% RH. After that the samples were tableted using a stainless steel mold in a hydraulic press under the pressure of 1 t. The nominal tab size was 30 mm in diameter and 3 mm in height. PDAC water solution was diluted with 30 mL of water with subsequent additions of G and SBH. Four samples were prepared varying the amount of polymer while keeping the G/SBH ratio constant (their compositions are shown in Table 1).
The suspension was stirred overnight to obtain a homogeneous dispersion and then heated at 100 • C on a plate while stirring for around 5 h to remove water via evaporation. In this step, SBH and PDAC started precipitating on the graphite matrix. When the dispersion viscosity was too high to allow any further stirring, the wet material was placed in a vacuum oven at 50 • C overnight to complete the process. After these steps, the prepared mixtures were subjected to a complete dehydration and hydration cycle prior to the subsequent charge/discharge cycles [27]. The samples were dehydrated in a ventilated oven at 120 • C until they reached constant weight and rehydrated in a climatic chamber at 23 • C and 50% RH. After that the samples were tableted using a stainless steel mold in a hydraulic press under the pressure of 1 t. The nominal tab size was 30 mm in diameter and 3 mm in height.
Characterization
The materials morphology was investigated with a LEO-1450VP (Zeiss, Oberkochen, Germany) scanning electron microscope (SEM) with a 15 kV accelerating voltage, on the cross section of tabs, obtained by fragile fracture upon bending. Surfaces were gold-coated prior to SEM observations. XRD analyses were performed on a Philips/Panalytical X´Pert Pro (Malvern, Milano, Italy) using a Philips PW3040/60 X-ray generator with a Cu anode using a Kα wavelength. A broad interval of 2θ angles of 10-70 were chosen to identify the SBH structure using a 0.026 • 2θ as scan step and nominal time per step of 100 s, using a scanning PixCell 1d detector. Intensity of reported diffractograms was normalized on an SBH (110) peak.
The performance of the composite materials was investigated with both differential scanning calorimetry (DSC) on a TA Instruments Q20 system (TA Instruments, Milano, Italy) using open aluminum pans and thermogravimetric analysis (TGA) on a TA Instruments Discovery gravimetric balance using open platinum pans. Both experiments were performed with the same temperature program: An equilibration at 35 • C, a heating ramp to 90 • C at 10 • C/min, and an isotherm for 90 min with a dry nitrogen flux of 50 mL/min for DSC and 25 mL/min for TGA. Only for the PDAC sample, was the isotherm time set to 10 h to assure complete dehydration. Samples weight was ≈7 ± 0.5 mg.
Thermal conductivity tests of the prepared tabs were carried out on a TPS 2500S by Hot Disk AB (Göteborg, Sweden) with a Kapton sensor (radius 6.4 mm) using the slab method [28]. Before each measurement, specimens were stored in a constant climate chamber (Binder KBF 240, Tuttlingen, Germany) at 23.0 ± 0.1 • C and 50.0 ± 0.1% RH for at least 48 h before tests. The test temperature (23.00 ± 0.01 • C) was controlled by a silicon oil bath (Haake A40, Thermo Scientific Inc., Waltham, MA USA) equipped with a temperature controller (Haake AC200, Thermo Scientific Inc., Waltham, MA, USA).
A custom setup was assembled to observe the hydration of the prepared composites. The tabs were vertically held to maximize the surface area exposed to the environment. They were dehydrated in an oven at 120 • C until they reached constant weight. After that they were placed in a climatic chamber (Binder KBF 240, D) at 23.0 ± 0.1 • C, and 50.0 ± 0.1% RH, and weighted on an analytical balance (Radwag AS 220.R2, PL) with an accuracy of ±0.5 mg to record the hydration over time. An experimental deviation of ±10% on the normalized mass gain during rehydration was estimated, after having performed several tests.
Morphology Analysis
As the microstructure of the TCM may have affected the kinetic of hydration/dehydration, SEM was employed to investigate the influence of PDAC concentration on the morphology and microstructure of the prepared samples.
The micrographs (Figure 2) unveil the effect of PDAC in the salt distribution. In particular, in the absence of PDAC, salt aggregates in globular shapes with dimensions in the order of few µm were observed between flakes of expanded graphite. In the presence of PDAC, the shape of salt aggregation changes evolved with polyelectrolyte concentration. Indeed, the average size of globular salt agglomerates were reduced in the sample with a PDAC/G ratio of 0.1 (PDAC content of 2% w/w, Figure 2b), and completely absent in the samples with ratio 0.5 (PDAC content of 10% w/w of PDAC) and 1 (PDAC content of 18% w/w of PDAC), as shown in Figure 2c,d respectively. It appears that the PDAC acted as a binder between salt crystals as well as an adhesion promoter at the salt/graphite interface, as schematized in Figure 2e. XRD analysis was used to identify the crystal structure of the employed salt, in particular, it was used to investigate possible anion exchange reactions between SBH and the polyelectrolyte during the water dissolution and recrystallization process. Diffraction patterns for SBH/G and the counterparts with different PDAC concentrations are reported in of Figure 3, while the collected diffractograms for purchased SBH and G are reported in Figure S1. Using the Joint Committee on Powder Diffraction Standards-International Centre for Diffraction Data database (JCPDS-ICDD) [29], both SrBr2•6H2O and graphite crystal planes were identified, their Miller indices being reported on the diffractogram in black and red, respectively. The absence of additional peaks excluded the formation of crystalline byproducts derived from ion exchanges between PDAC and SBH during the manufacturing process. Nonetheless, differences in relative intensities of selected peaks (i.e., signals at 24.7° and 39.8°) were observable in the diffractogram between samples with and without polyelectrolyte addition. This could be ascribed to the influence of the polyelectrolyte in the growth of salt hydrates crystals, as previously reported in the literature [30]. In addition, a limited broadening of the SBH main peaks was observed at the highest PDAC concentration, confirming the binder role in the aggregation of SBH crystals. Finally, the absence of extra peaks in PDAC-containing composites confirmed the amorphous nature of the polyelectrolyte.
Thermal Properties
DSC and TGA analyses were first employed to study the dehydration of the composites in dry conditions. The temperature program was chosen to simulate a 90 °C heat source charging the thermochemical system. DSC results are reported in Figure 4, showing heat flow plots for different samples, characterized by an endothermic peak corresponding to the dehydration reaction. Using the Joint Committee on Powder Diffraction Standards-International Centre for Diffraction Data database (JCPDS-ICDD) [29], both SrBr2•6H2O and graphite crystal planes were identified, their Miller indices being reported on the diffractogram in black and red, respectively. The absence of additional peaks excluded the formation of crystalline byproducts derived from ion exchanges between PDAC and SBH during the manufacturing process. Nonetheless, differences in relative intensities of selected peaks (i.e., signals at 24.7° and 39.8°) were observable in the diffractogram between samples with and without polyelectrolyte addition. This could be ascribed to the influence of the polyelectrolyte in the growth of salt hydrates crystals, as previously reported in the literature [30]. In addition, a limited broadening of the SBH main peaks was observed at the highest PDAC concentration, confirming the binder role in the aggregation of SBH crystals. Finally, the absence of extra peaks in PDAC-containing composites confirmed the amorphous nature of the polyelectrolyte.
Thermal Properties
DSC and TGA analyses were first employed to study the dehydration of the composites in dry conditions. The temperature program was chosen to simulate a 90 °C heat source charging the thermochemical system. DSC results are reported in Figure 4, showing heat flow plots for different samples, characterized by an endothermic peak corresponding to the dehydration reaction. Using the Joint Committee on Powder Diffraction Standards-International Centre for Diffraction Data database (JCPDS-ICDD) [29], both SrBr 2 ·6H 2 O and graphite crystal planes were identified, their Miller indices being reported on the diffractogram in black and red, respectively. The absence of additional peaks excluded the formation of crystalline byproducts derived from ion exchanges between PDAC and SBH during the manufacturing process. Nonetheless, differences in relative intensities of selected peaks (i.e., signals at 24.7 • and 39.8 • ) were observable in the diffractogram between samples with and without polyelectrolyte addition. This could be ascribed to the influence of the polyelectrolyte in the growth of salt hydrates crystals, as previously reported in the literature [30]. In addition, a limited broadening of the SBH main peaks was observed at the highest PDAC concentration, confirming the binder role in the aggregation of SBH crystals. Finally, the absence of extra peaks in PDAC-containing composites confirmed the amorphous nature of the polyelectrolyte.
Thermal Properties
DSC and TGA analyses were first employed to study the dehydration of the composites in dry conditions. The temperature program was chosen to simulate a 90 • C heat source charging the thermochemical system. DSC results are reported in Figure 4, showing heat flow plots for different samples, characterized by an endothermic peak corresponding to the dehydration reaction. (1) where Ec is the energy density of the composite materials with SBH, G, and PDAC, Ep is the energy density of PDAC, Es is the energy density of SBH/G and xp and xs are the weight fractions of PDAC and SBH/G in the final composites, respectively. As reported in Figure 4b, the experimental values were corresponding to the expected ones, within experimental deviations; this points out that the system acts as an ideal mixture of the two components, with no synergic nor antagonist interaction between PDAC and SBH, in terms of total energy stored. By increasing the content of PDAC, a lowering of the total energy storage density was obtained, diminishing the efficiency of the composite for thermal storage applications by approximately 15% at the highest PDAC concentration. This was ascribed to the great difference in energy storage density between PDAC and the SBH, which was related to the different hydration mechanisms of the two substances. The dehydration of prepared samples were also evaluated by TGA measurements allowing for the assessment of the amount and kinetics of water removal upon heating, as a function of the polyelectrolyte concentration ( Figure 5).
where E c is the energy density of the composite materials with SBH, G, and PDAC, E p is the energy density of PDAC, E s is the energy density of SBH/G and x p and x s are the weight fractions of PDAC and SBH/G in the final composites, respectively. As reported in Figure 4b, the experimental values were corresponding to the expected ones, within experimental deviations; this points out that the system acts as an ideal mixture of the two components, with no synergic nor antagonist interaction between PDAC and SBH, in terms of total energy stored. By increasing the content of PDAC, a lowering of the total energy storage density was obtained, diminishing the efficiency of the composite for thermal storage applications by approximately 15% at the highest PDAC concentration. This was ascribed to the great difference in energy storage density between PDAC and the SBH, which was related to the different hydration mechanisms of the two substances. The dehydration of prepared samples were also evaluated by TGA measurements allowing for the assessment of the amount and kinetics of water removal upon heating, as a function of the polyelectrolyte concentration ( Figure 5). At low PDAC concentrations (0.1 and 0.5 PDAC/G weight ratios), the polyelectrolyte did not significantly alter the dehydration kinetics with respect to the SBH/G composite. On the other hand, SBH/G/P(1) exhibited a delayed weight loss compared to SBH/G, reflecting the slow dehydration kinetic observed for PDAC. To further investigate the effect of PDAC, the theoretical weight loss curves (Wth) for the composites were calculated by applying a rule of mixture between the neat PDAC and SBH/G (Equation (2)). (2) where Wp and xp are the weight and mass fraction of PDAC in the composite, while Ws and xs are the weight and mass fraction of SBH in the composite, respectively. As reported in Figure 5b,c, the samples with 0.1 or 0.5 weight ratios showed limited differences between the theoretical and experimental plots, thus suggesting two dehydration processes, from PDAC and SBH, to proceed independently. For SBH/G/P(1), a significant deviation was observed between theoretical and experimental plots, suggesting that kinetics of dehydration were controlled by the interaction between the two phases. This is consistent with the polyelectrolyte binding action between the salt crystals, observed by SEM and ascribed to the delayed diffusion of the water, released by the salt, through the polyelectrolyte. Indeed, while the release of water from crystalline hydrated SBH was simply triggered by the temperature, the amorphous structure of PDAC, with its high free volume, broadened the water release in time, through a series of absorption/desorption steps, eventually reduced the overall dehydration rate. The above results suggest that a high concentration of polyelectrolyte binder can partially reduce the efficiency of the thermochemical system under study by both decreasing the heat storage density (Figure 4b) and slowing down water dehydration kinetics (Figure 5d). Thus, only composites with a PDAC/G weight ratio of 0.1 and 0.5 were selected for tablet preparation and hydration kinetics characterization. At low PDAC concentrations (0.1 and 0.5 PDAC/G weight ratios), the polyelectrolyte did not significantly alter the dehydration kinetics with respect to the SBH/G composite. On the other hand, SBH/G/P(1) exhibited a delayed weight loss compared to SBH/G, reflecting the slow dehydration kinetic observed for PDAC. To further investigate the effect of PDAC, the theoretical weight loss curves (W th ) for the composites were calculated by applying a rule of mixture between the neat PDAC and SBH/G (Equation (2)).
W c (t) = x p W p (t) + x s W s (t) (2) where W p and x p are the weight and mass fraction of PDAC in the composite, while W s and x s are the weight and mass fraction of SBH in the composite, respectively. As reported in Figure 5b,c, the samples with 0.1 or 0.5 weight ratios showed limited differences between the theoretical and experimental plots, thus suggesting two dehydration processes, from PDAC and SBH, to proceed independently. For SBH/G/P(1), a significant deviation was observed between theoretical and experimental plots, suggesting that kinetics of dehydration were controlled by the interaction between the two phases. This is consistent with the polyelectrolyte binding action between the salt crystals, observed by SEM and ascribed to the delayed diffusion of the water, released by the salt, through the polyelectrolyte. Indeed, while the release of water from crystalline hydrated SBH was simply triggered by the temperature, the amorphous structure of PDAC, with its high free volume, broadened the water release in time, through a series of absorption/desorption steps, eventually reduced the overall dehydration rate. The above results suggest that a high concentration of polyelectrolyte binder can partially reduce the efficiency of the thermochemical system under study by both decreasing the heat storage density (Figure 4b) and slowing down water dehydration kinetics (Figure 5d). Thus, only composites with a PDAC/G weight ratio of 0.1 and 0.5 were selected for tablet preparation and hydration kinetics characterization.
Composite Tabs Hydration
It is known that the surface/volume ratio can strongly influence the heat and mass transfer phenomena in a thermochemical storage system [23]. For this reason, while values collected with dry TGA and DSC analyses may be used to compare the performance of different composites, these are not representative of a real application, both in terms of mass and moisture effects. In order to obtain more realistic results, hydration kinetics have been evaluated in a climatic chamber on composites tabs having dimensions suitable for real applications. Hydration curves, calculated as the measured weight gain normalized over the weight of the dry tabs, are reported in Figure 6a. Nanomaterials 2019, 9, x FOR PEER REVIEW 8 of 11
Composite Tabs Hydration
It is known that the surface/volume ratio can strongly influence the heat and mass transfer phenomena in a thermochemical storage system [23]. For this reason, while values collected with dry TGA and DSC analyses may be used to compare the performance of different composites, these are not representative of a real application, both in terms of mass and moisture effects. In order to obtain more realistic results, hydration kinetics have been evaluated in a climatic chamber on composites tabs having dimensions suitable for real applications. Hydration curves, calculated as the measured weight gain normalized over the weight of the dry tabs, are reported in Figure 6a. The plots clearly show a monotonic weight gain for both SBH/G and counterparts including PDAC. However, dramatic differences in moisture absorption kinetics were proven as a function of PDAC concentration. Indeed, while limited differences existed between SBH/G/P(0.1) and SBH/G, the hydration rate of SBH/G/P(0.5) was much higher, especially within the first hours of the test, reaching the full hydration of both phases in the mixture (equivalent to 0.26 gwater/gmixture calculated on the basis of Equation (2)) within approximately 10 h, whereas hydration of the other samples was still ongoing after 45 h (Figure 6a). Another important aspect to evaluate when considering a real application was represented by the durability of the prepared composite. This had a potentially strong impact on the effectiveness and practicability of the thermochemical storage solution. In the present study we observed severe damage in the form of cracks to the SBH/G tabs immediately after the first hydration/dehydration cycle (Figure 6b), thus proving this aspect to be a significant weakness of the graphite/salt hydrate composite approach. The mechanical stress in the samples may have been caused by the volume change of the salt during the hydration process. In fact, it is reported that the density of strontium bromide changed between 3.5 g/cm 3 and 2.4 g/cm 3 from hexahydrate to monohydrate form [31]. On the other hand, the presence of a polymeric binder appeared to reduce The plots clearly show a monotonic weight gain for both SBH/G and counterparts including PDAC. However, dramatic differences in moisture absorption kinetics were proven as a function of PDAC concentration. Indeed, while limited differences existed between SBH/G/P(0.1) and SBH/G, the hydration rate of SBH/G/P(0.5) was much higher, especially within the first hours of the test, reaching the full hydration of both phases in the mixture (equivalent to 0.26 g water /g mixture calculated on the basis of Equation (2)) within approximately 10 h, whereas hydration of the other samples was still ongoing after 45 h (Figure 6a). Another important aspect to evaluate when considering a real application was represented by the durability of the prepared composite. This had a potentially strong impact on the effectiveness and practicability of the thermochemical storage solution. In the present study we observed severe damage in the form of cracks to the SBH/G tabs immediately after the first hydration/dehydration cycle (Figure 6b), thus proving this aspect to be a significant weakness of the graphite/salt hydrate composite approach. The mechanical stress in the samples may have been caused by the volume change of the salt during the hydration process. In fact, it is reported that the density of strontium bromide changed between 3.5 g/cm 3 and 2.4 g/cm 3 from hexahydrate to monohydrate form [31]. On the other hand, the presence of a polymeric binder appeared to reduce crack formation at a polymer/G ratio of 0.1, and completely prevent it a 0.5 ratio, thus maintaining the structural integrity of the composites.
In addition, the thermal conductivity of prepared composite tabs was also evaluated, as heat exchange was obviously crucial for the efficiency of heat storage devices. Results reported in Figure S2 show that the thermal conductivity of the prepared composites remained constant within the experimental error, in the 16-16.5 W/mK range, demonstrating no detrimental effects related to the presence of the polyelectrolyte. Furthermore, thermal conductivity values obtained in this work were significantly higher than previously reported values for similar graphite SBH composites. [24,32]. As clearly depicted by the characterization reported in Figure 6, the composite with a PDAC/G ratio of 0.5 was capable of achieving superior water adsorption kinetics while maintaining high thermal conductivity values thus proving that the inclusion of a polymer binder was a successful strategy for the design of an efficient thermochemical storage solution.
Conclusions
This work was focused on the production of composites comprising of strontium bromide hexahydrate, expanded natural graphite, and polydiallyldimethylammonium chloride for thermochemical energy storage applications, using a simple and environmentally sustainable, water-based process.
Morphological analysis performed by SEM showed a stabilizing effect of the polymer binder on the salt particles, while XRD data confirmed the presence of SrBr 2 ·6H 2 O in the final material without undesired byproducts. The materials were characterized with different thermal analysis techniques to understand their performance in terms of energy storage density and capability of heat and mass transfer. The prepared composites were further molded in centimeter scale tabs suitable for exploitation in a modular-design reactor, in order to analyze their hydration kinetics and thermal conductivity properties. High contents of PDAC polyelectrolyte (SBH/G/P(1)) resulted in slightly limited dehydration kinetics and energy storage densities.
On the other hand, lower PDAC contents (SBH/G/P(0.1) or SBH/G/P(0.5)) did not affect dehydration kinetics and caused minimal reduction if energy storage density. Tabs prepared with SBH/G/P(0.5) were found to have significantly higher hydration rates in ambient conditions (23 • C and 50% RH) with respect to the conventional SBH/G composites. Indeed, the polyelectrolyte-containing formulation allowed us to reach complete hydration of the tabs in~10 h, while the samples with no PDAC reached only~35% of total hydration in the same time. These results relate to the physical action of the organic polyelectrolyte, acting as a binder between salt crystals, controlling moisture diffusion, and mechanically stabilizing the structure against stress-cracking, which is typical of pristine salt and salt/graphite formulations. Furthermore, a state of the art value of 16 W/mK thermal conductivity was obtained for the tabs, almost independent of the presence of the polyelectrolyte.
The results collected in this paper clearly demonstrate the proposed approach as a promising strategy for the design of efficient thermochemical storage solutions. Future studies should aim to investigate the cyclability of multiple hydration/dehydration cycles of the composite tabs, under controlled air flow, as well as their engineering in order to exploit processing conditions and geometries capable of reducing charge/discharge cycles and improving the efficiency of the system. Mechanical characterizations of the tabs might also prove important to better understand the stabilizing effect of the polyelectrolyte in the composite structure. All these characteristics will help in the fabrication of a TCM suitable for low grade heat reuse and ready for a scale up in prototypes focused on specific applications. | 7,525 | 2019-03-01T00:00:00.000 | [
"Materials Science"
] |
Phosphorylation of the Amyloid-Beta Peptide Inhibits Zinc-Dependent Aggregation, Prevents Na,K-ATPase Inhibition, and Reduces Cerebral Plaque Deposition
The triggers of late-onset sporadic Alzheimer’s disease (AD) are still poorly understood. Impairment of protein phosphorylation with age is well-known; however, the role of the phosphorylation in β-amyloid peptide (Aβ) is not studied sufficiently. Zinc-induced oligomerization of Aβ represents a potential seeding mechanism for the formation of neurotoxic Aβ oligomers and aggregates. Phosphorylation of Aβ by Ser8 (pS8-Aβ), localized inside the zinc-binding domain of the peptide, may significantly alter its zinc-induced oligomerization. Indeed, using dynamic light scattering, we have shown that phosphorylation by Ser8 dramatically reduces zinc-induced aggregation of Aβ, and moreover pS8-Aβ suppresses zinc-driven aggregation of non-modified Aβ in an equimolar mixture. We have further analyzed the effect of pS8-Aβ on the progression of cerebral amyloidosis with serial retro-orbital injections of the peptide in APPSwe/PSEN1dE9 murine model of AD, followed by histological analysis of amyloid burden in hippocampus. Unlike the non-modified Aβ that has no influence on the amyloidosis progression in murine models of AD, pS8-Aβ injections reduced the number of amyloid plaques in the hippocampus of mice by one-third. Recently shown inhibition of Na+,K+-ATPase activity by Aβ, which is thought to be a major contributor to neuronal dysfunction in AD, is completely reversed by phosphorylation of the peptide. Thus, several AD-associated pathogenic properties of Aβ are neutralized by its phosphorylation.
INTRODUCTION
The amyloid-beta peptide (Aβ) is a normal subnanomolar component of biological fluids (Masters and Selkoe, 2012); however, its deposition in the form of amyloid plaques is one of the hallmarks of AD (Selkoe and Hardy, 2016). Amyloid plaques are associated with neuronal loss and cognitive impairment (Musiek and Holtzman, 2015), and they enhance the tau pathology (He et al., 2017). The trigger of the pathologic Aβ aggregation in AD is unknown (Musiek and Holtzman, 2015); however, in animal models of AD, the conversion of a monomeric Aβ into fibrillar aggregates, through neurotoxic oligomers, is triggered by chemically and structurally modified Aβ species (Meyer-Luehmann et al., 2006;Prusiner, 2012;Jucker and Walker, 2013). Amyloid plaques are abnormally rich in Fe, Cu, and Zn ions (Cummings, 2004), and data from animal models suggest that the formation of amyloid plaques is zinc dependent (Friedlich et al., 2004;Frederickson et al., 2005;DeGrado et al., 2016;James et al., 2017). It has been assumed that zinc ions promote Aβ aggregation via a population shift of polymorphic states (Miller et al., 2010). Zinc-induced aggregation of Aβ is governed by its metal-binding domain (Aβ 16 ), which comprises the N-terminal region (residues 1-16) of Aβ (Istrate et al., 2016). It has recently been shown that the metal-binding domain of Aβ containing isomerized Asp7 (isoD7-Aβ 16 ) is more prone to zinc-induced oligomerization (Istrate et al., 2016), suggesting a role for Asp7 isomerization in the initiation of the pathological aggregation process. Indeed, a full-length isoD7-Aβ 42 peptide is more neurotoxic than the unmodified peptide (Mitkevich et al., 2013) and is able to trigger cerebral amyloidosis in vivo (Kozin et al., 2013). Thus, the ability of Aβ to form zinc-induced aggregates may correlate with its amyloidogenic and neurotoxic properties.
Recently, it has been shown that Aβ undergoes phosphorylation at Ser8 both in vitro and in vivo (Kumar et al., 2011(Kumar et al., , 2013. In the absence of zinc ions, phosphorylated Aβ (pS8-Aβ) was found to be a fast-aggregating peptide species, producing stable fibrillar aggregates and neurotoxic oligomers in vitro (Rezaei-Ghaleh et al., 2016;Jamasbi et al., 2017). In the presence of Zn 2+ , the metal-binding domain of pS8-Aβ (pS8-Aβ 16 ) forms stable dimers (Kulikova et al., 2014); however, in contrast to the zinc-mediated behavior of similar domains in several other Aβ isoforms (including native Aβ), the zinc-bound dimers of pS8-Aβ 16 do not give rise to larger oligomers and aggregates (Istrate et al., 2016). Moreover, the zinc-driven heterodimers formed between pS8-Aβ 16 and isoD7-Aβ 16 or Aβ 16 also cannot oligomerize. We hypothesized that pS8-Aβ 42 does not aggregate in the presence of zinc ions and that pS8-Aβ 42 can prevent the aggregation of native Aβ through the formation of non-propagating heterodimers between pS8-Aβ 42 and Aβ (Mezentsev et al., 2016). By contrast, phosphorylation at Ser8 may change the interaction of Aβ with other proteins, such as Na + ,K + -ATPase. Previously, it was shown that Na + ,K + -ATPase activity was inhibited in post-mortem tissues of AD patients and in amyloid-containing hippocampi of transgenic mice (but not in the amyloid-free cerebellum) (Dickey et al., 2005;Kreutz et al., 2013;Zhang et al., 2013). The latest studies demonstrated that Aβ 42 in form of monomers or oligomers directly binds to Na + ,K + -ATPase, which results in the inhibition of the enzyme as well as the triggering of intracellular signaling cascades (Ohnishi et al., 2015;Petrushanko et al., 2016). Therefore, phosphorylation at Ser8 might neutralize some pathogenic properties of Aβ. Furthermore, the concentration of pS8-Aβ is very likely to change with age since brain aging is associated with changes in the activity of kinases and phosphatases in nerve tissue (Jin and Saitoh, 1995;Norris et al., 1998). An age-related shift in the neuronal protein phosphorylation has recently been shown in Drosophila models (Thomas and Haberman, 2016). Thus, phosphorylated Aβ species could be significant for the development of late-onset AD. However, there are no published data on the zinc-dependent oligomerization and related properties of a full-length pS8-Aβ peptide. In the present study, we investigated the role of pS8-Aβ 42 as a potential quencher of zinc-induced oligomerization of endogenous Aβ species and of pathological effects associated with this process, such as the inhibition of Na + ,K + -ATPase and induction of cerebral amyloidosis, in AD model mice.
Host Mice
B6C3-Tg(APPswe,PSEN1dE9)85Dbo mice were used at the age of 2-8 months (weight of 24-26 g). Mice were housed in the Pushchino Animal Breeding Facility (branch of the Shemyakin and Ovchinnikov Institute of Bioorganic Chemistry, Russian Academy of Sciences), under specific pathogen-free conditions. Housing and use of laboratory animals were approved by the commission IACUC,protocol No. 479 The amino acid sequence of the peptide was confirmed with an ultra-high resolution Fourier transform ion cyclotron resonance mass-spectrometer 7T Apex Qe BRUKER (Bruker Daltonics, Bellerica, MA, United States) utilizing a de novo sequencing approach based on a CID fragmentation technique as described earlier .
Synthetic pS8-Aβ 42 Preparations for Injections
Two thousand micrograms of peptide pS8-Aβ 42 were dissolved in 2000 µL of sterile physiological saline (PS), then the prepared solution was filtered through a 0.22µm filter (Millex-GV, Millipore), aliquoted to 125 µL and frozen. For injection, one aliquot was thawed, sterile PS was added to obtain 1500 µL of solution with a peptide concentration of 0.08333 µg/µL ("administration solution"). Then, 150 µL of "administration solution" were withdrawn and 125 µL of this solution were injected into one animal. Thus, with a single injection, 10 µg of the peptide were ingested.
The aggregation state of the synthetic peptide pS8-Aβ 42 in the samples used for the injections, characterized by us using a standard test based on thioflavin T (Ban et al., 2003), did not change during the time of storage (1-8 months), and did not differ from that of the freshly prepared solutions of the corresponding peptides.
Synthetic Aβ 42 and pS8-Aβ 42 Preparations for Na,K-ATPase Studies To prepare working solutions of Aβ 42 and pS8-Aβ 42 peptides, cold hexafluoroisopropanol ("Fluka") was added to dry Aβ to a concentration of 1 mM and incubated for 60 min at room temperature. The resulting solution was then placed in ice for 10 min and transferred to non-siliconized microcentrifuge tubes (0.56 mg peptide per tube). The peptides in the tubes were vacuum-dried with Eppendorf Concentrator 5301. The resulting dry peptides were stored at −80 • C. A fresh 2.5 mM Aβ solution was prepared by adding 20 µl of 100% anhydrous dimethyl sulfoxide ("Sigma") to 0.56 mg of the peptide, followed by incubation for 1 h at room temperature.
Synthetic Aβ 42 and pS8-Aβ 42 Preparations for DLS Measurements
To measure the average diameter of Aβ aggregates, synthetic peptides Aβ 42 and pS8-Aβ 42 were treated with hexafluoroisopropanol, dried, and dissolved in 10 mM NaOH at concentration of 0.5 mM. Peptides were brought to required concentrations in 10 mM HEPES (pH 7.4) supplemented with 150 mM NaCl, using appropriate buffers.
Dynamic Light Scattering
Dynamic light scattering measurements were carried out using a Zetasizer Nano ZS apparatus (Malvern Instruments, Ltd., United Kingdom) at 25 • C in accordance with the manufacturer instruction. The instrument is equipped with a He-Ne laser source (λ = 632.8 nm) and operates in the back-scatting mode, measuring particle size in the range between 0.6 nm and 10 µm. The particle size distribution is estimated with a spherical approximation of particles, employing a CONTIN data analysis utility available as a part of the instrument software, and used to calculate the average particle diameter (D). The aggregation of Aβ 42 and pS8-Aβ 42 peptides was triggered by diluting peptide solutions with a ZnCl 2 -containing buffer so as to provide a twofold molar excess of zinc ions over peptides.
Turbidimetry
A peptide/zinc mixture was placed in a BRAND UV microcuvette (BRAND GMBH, Germany) and its turbidity was monitored in time (at room temperature) as optical density at 405 nm, using an Agilent 8453E spectrophotometer (Agilent Technologies, United States). Turbidity measurements were started in 0.5 min after triggering the zinc-induced Aβ aggregation as described for the DLS experiments. The initial (zero time) points were measured in the absence of zinc ions, using 25-µM Aβ solutions.
Intravenous Injections
Retro-orbital injections of the venous sinus in mice were performed according to Yardeni et al. (2011). Mice received one retro-orbital injection with 1-month intervals between injections. The contents of injections for each group of mice are described in Table 1. The mice were assigned to the various groups randomly.
Histology and Immunohistochemistry
Euthanasia procedure was applied to 8-month-old mice. Mouse euthanasia was carried out by CO 2 according to the IACUCapproved protocol with the use of automated CO 2 -box Bioscape (Germany). Mice were transcardially perfused with 50 mL of PBS, followed by 50 mL of 4% paraformaldehyde (PFA). Mouse brains were fixed in 10% formalin. Process for paraffin embedding was scheduled as follow: 75% ethanol overnight, 96% ethanol 5 min, 96% ethanol 10 min, 100% ethanol 10 min (two changes), ethanol-chloroform (1:1) 30 min, chloroform overnight. Paraffin embedding was performed at 60 • C for 3 h (three changes). The embedding of tissues into paraffin blocks was done using Leica EG1160 device. Serial brain sections (8 um) were cut using microtome Leica RM2265 mounted onto slides. For deparaffinization, hydration and staining of the sections the following steps were done: slides were consistently put in xylene three changes (10 min each), 96% ethanol (5 min), 90% ethanol (2 min), 75% ethanol (2 min), H 2 O three changes (5 min each), Congo Red solution (5 min), potassium alkali solution, and water. The Immu-Mount medium (Thermo Scientific) was used for mounting.
Immunostaining was carried out as described elsewhere (Kozin et al., 2013). Briefly, sections were deparaffinized and after antigen retrieval by microwaving in citrate buffer washed in PBS and blocked with 10% goat serum in 0.04% Tween20 in PBS (T-PBS). Sections were incubated with primary antibody for 2 h at room temperature, washed thrice in T-PBS, incubated with secondary antibody for 1 h at room temperature followed by washing in T-PBS. The mouse anti-human Aβ monoclonal antibodies 6E10 (Covance, Dallas, TX, United States) diluted in the block solution 1:1000 were used as the primary antibodies, Alexa Fluor 488 goat anti-mouse antibodies (Invitrogen, Grand Island, NY, United States) were used as the secondary antibodies for immunofluorescence staining (dilution 1:1000 in T-PBS). The images were captured with Leica DFS490 digital camera (Leica Microsystems, Wetzlar, Germany) at 100x magnification using Leica DMI 4000 fluorescent microscope (Leica Microsystems, Wetzlar, Germany).
Quantitative Assessment of Cerebral β-Amyloidosis
The sections spanning brain from 0.48 to 1.92 mm relative to the midline in lateral stereotaxic coordinates were used to quantify congophilic amyloid plaques in the hippocampus. Every 15th section was analyzed, yielding 10 sections per animal. Amyloid plaques were identified by Congo Red staining and manually counted as described previously (Ninkina et al., 2009;Bachurin et al., 2012) using Zeiss Axiovert 200 M microscope with 10×, 20×, and 40× objectives (Carl Zeiss, Oberkochen, Germany), with examination under bright-field and between crossed polarizers. Amyloid plaques of all sizes were accepted for counting if they were visible and met the following requirements: red coloring under bright-field and the green birefringence in polarized light. Analyses were undertaken by two researchers independently (EB, SK). To determine the reproducibility of the plaques counts, an intra-class correlation (ICC) was calculated yielding good inter-rater reliability between the two researchers (ICC > 0.85). For each group of mice, the average values ( ± SD) of the plaques number per section were calculated.
Hydrolytic Activity of Na + /K + -ATPase The purified preparation of Na + /K + -ATPase (α1β1 isozyme) was obtained from duck salt glands as described elsewhere (Petrushanko et al., 2016). The purity grade of Na + /K + -ATPase was 99% and specific activity of the enzyme reached ∼2400 µmol of Pi (mg of protein × h) −1 at 37 • C.
Hydrolytic activity of Na + /K + -ATPase in the purified preparation was measured as ouabain-sensitive (1 mM) ATP cleavage in the reaction medium containing 130 mM NaCl, 20 mM KCl, 3 mM MgCl 2 , 3 mM ATP, and 30 mM imidazole, pH 7.4, 37 • C as elsewhere (Petrushanko et al., 2016). Stock solutions containing Aβ 42 and pS8-Aβ 42 peptides, prepared as described above, were added to the reaction medium (without ATP) up to a concentration of 40 µM. An equivalent amount of DMSO was added to the control samples. Following 3-60 min of preincubation with amyloid peptides, the enzymatic reaction was started by adding ATP.
Modeling of the Structure of pS8-Aβ42:Na +/ K + -ATPase Complex Model of the pS8-Aβ 42 peptide was constructed using as templates the Aβ 42 structure modeled in Petrushanko et al. (2016). The resulting model of pS8-Aβ 42 was minimized in the AMBER99 force field with the AutoDockTools program. Modeling of the pS8-Aβ 42 :Na + /K + -ATPase complex was performed using the structure of Na + /K + -ATPase from shark glands 2zxe (PDB id) solved at 2.4 Å resolution (Shinoda et al., 2009), and the modeled structure of pS8-Aβ 42 . Docking has been carried out with VinaAutoDock program (Trott and Olson, 2010), and the docking was constrained to cover only the extracellular part of the protein.
Isothermal Titration Calorimetry (ITC)
The thermodynamic parameters of amyloid peptides binding to Na + /K + -ATPase were measured using a MicroCal iTC200 instrument, as described elsewhere (Mitkevich et al., 2012;Petrushanko et al., 2014). Experiments were carried out at 25 • C in 10 mM imidazole buffer (pH 7.5), containing 130 mM NaCl, 30 mM KCl, 3 mM MgCl 2 . Aliquots (2.6 µl) of ligands were injected into a 0.2-ml cell containing protein solution to achieve a complete binding isotherm. Protein concentration in the cell ranged from 10 to 20 µM, and ligand concentration in the syringe ranged from 100 to 200 µM. The resulting titration curves were fitted using the MicroCal Origin software, assuming one set of binding sites. Affinity constants (Ka), enthalpy variations ( H) and stoichiometry of binding (N) were determined and the Gibbs energy ( G) and entropy variations ( S) were calculated from the equation: G = −RTlnKa = H-T S.
Statistical Methods Used for Data Analysis
Data are presented as means of at least three independent experiments ± SD. The Mann-Whitney test was used for pairwise comparison between examined groups of mice; P < 0.05 was considered significant. The comparison of Na + /K + -ATPase data groups was performed using one-way ANOVA with post hoc testing (using paired samples Student's t-test with Bonferroni correction); after a Bonferroni correction a P-value < 0.016 was considered as statistically significant. Statistical analysis was performed using STATISTICA 8.0 (StatSoft, Inc., Tulsa, OK, United States).
pS8-Aβ 42 Suppresses Zinc-Dependent Aggregation of Aβ 42
Using DLS, we observed a time-dependent formation of Aβ 42 or pS8-Aβ 42 in the presence of Zn 2+ (Figure 1A). Prior to zinc addition, only oligomers 20-30 nm in size were detected. In the absence of zinc ions, there were no appreciable changes in the characteristic size of both Aβ 42 and pS8-Aβ 42 oligomers during the 90 min incubation period (at 25 • C and quiescent conditions). After 10 min of incubation with Zn 2+ , the characteristic diameter of Aβ 42 aggregates reached 700-800 nm, and it became more than 2,000 nm by the end of the observation period (100 min). In contrast, pS8-Aβ 42 did not form aggregates larger than 50 nm in the presence of zinc ions during the entire observation period. The results of the turbidity measurements support those of DLS experiments ( Figure 1B): Ser8 phosphorylation substantially suppressed the propensity of Aβ peptides for zinctriggered aggregation, which was manifested in the remarkably higher turbidity of the Aβ 42 /Zn 2+ mixture than that of the pS8-Aβ 42 /Zn 2+ mixture. The effect of Ser8 phosphorylation on zinc-induced Aβ aggregation was the opposite of that on spontaneous Aβ aggregation: as revealed in the ThT assay, in the absence of zinc ions, pS8-Aβ 42 aggregates much faster than Aβ 42 (Supplementary Figure 1). This result is consistent with those reported by Kumar et al. (2011Kumar et al. ( , 2013. To study a possible effect of pS8-Aβ 42 on the zinc-induced aggregation of Aβ 42 , we determined the characteristic size of zincinduced aggregates in a series of pS8-Aβ 42 /Aβ 42 mixtures with different molar ratios of the peptides (Figure 1C) at the 70-min time point (the time point at which D reaches a plateau at a zinc/Aβ 42 molar ratio of 1:3; Supplementary Figure 2). When equimolar mixtures of pS8-Aβ 42 and Aβ 42 (12.5 µM each) were tested in the presence of 50 µM Zn 2+ , the aggregate diameter decreased to (45 ± 10) nm. Thus, mixing of pS8-Aβ 42 peptides with the unmodified peptide strongly suppressed the zincdependent aggregation of the latter. The half-maximal inhibitory concentration (IC 50 ) for pS8-Aβ 42 was estimated, under the conditions tested, to be approximately 9 µM ( Figure 1C); this value corresponds to about one pS8-Aβ 42 peptide per three Aβ 42 peptides.
Injection of pS8-Aβ 42 Reduces the Amyloid Load in the Hippocampus of Transgenic AD Model Mice
We examined the ability of the synthetic pS8-Aβ 42 peptide to reduce the cerebral amyloidogenesis in an APP/PS1 doubly transgenic murine model of AD. These mice have cognitive features of an AD-like pathology and accumulate significant amounts of dense-core congophilic amyloid plaques starting from 4 to 6 month of age, regardless of the sex (Borchelt et al., 1997;Garcia-Alloza et al., 2006). The experimental groups, which included male and female animals, were subjected to retro-orbital injections of peptide pS8-Aβ 42 (10 µg in 125 µL of PS) starting from 2 months of age. After serial (at monthly intervals) inoculations with the peptide, the host mice were sacrificed at the age of 8 months. The brains were extracted, and sagittal brain sections (8-µm thick) were analyzed histochemically using Congo Red staining. The hippocampus was chosen as the target region for manual counting of the stained congophilic amyloid plaques using bright-field microscopy in the sections representing the brain layer located between 0.48 and 1.92 mm relative to the midline in lateral stereotaxic coordinates. The congophilic plaques found in the brains of all experimental animals were similar in terms of their location and size distribution in the brain parenchyma (Figure 2). Additionally, immunohistochemical characterization of the congophilic amyloid plaques revealed the presence of Aβ (Supplementary Figure 3). Quantitative analysis revealed a significantly lower number of congophilic amyloid plaques per section in the pS8-Aβ 42 -inoculated 8-month-old transgenic mice (−36.3%, P < 0.05) than that in the untreated littermates (Figure 2 and Table 1).
pS8-Aβ 42 Binds to Na + ,K + -ATPase Without Inhibiting Its Hydrolytic Activity For the measurements, we used purified Na + ,K + -ATPase from duck salt glands, which is a homolog of the α1β1 human isozyme. Earlier, we have demonstrated that the unmodified Aβ 42 inhibits the hydrolytic activity of the enzyme (Petrushanko et al., 2016). In contrast to Aβ 42 , which inhibited Na + ,K + -ATPase after 60min incubation with 40 µM peptide, pS8-Aβ 42 had no effect on enzyme activity ( Figure 3A). The interaction of Na + ,K + -ATPase with the Aβ 42 and pS8-Aβ 42 peptides was measured by ITC (Figures 3B,C). The stoichiometry of binding to Na + ,K + -ATPase was equal to 1 for both peptides, demonstrating that the peptides were predominantly in a monomeric state, as we have shown previously for Aβ 42 (Petrushanko et al., 2016). The binding constants of Aβ 42 and pS8-Aβ 42 with the enzyme were close to each other, and the energy profiles [enthalpy ( H) and entropy (T S)] for binding of both peptides were practically the same ( Table 2). This indicates that phosphorylation of Aβ does not affect the peptide interaction with Na + ,K + -ATPase.
Hydrophobic C-Terminal Domain of Aβ 42
Is Responsible for Its Binding to Na + ,K + -ATPase Structure analysis of the Aβ 42 :Na + ,K + -ATPase complex, which we performed earlier (Petrushanko et al., 2016), has shown that the Ser8 residue is located outside of the interaction site (Figure 4). Introduction of the phosphate group to Ser8 had no effect on the conformation of the Aβ 42 polypeptide chain and did not change the interaction interface (Figure 4).
Metal-Dependent Aggregation of Aβ 42 and pS8-Aβ 42 Correlates With Their
Ability to Inhibit Na + ,K + -ATPase Using DLS, we showed that Mg 2+ at concentrations of 3 mM and above induces the aggregation of Aβ 42 after 10 min of incubation (Supplementary Figure 4). By contrast, no oligomers of pS8-Aβ 42 were observed even after 20 min of incubation with 10 mM Mg 2+ . Since Na + ,K + -ATPase activity was measured in a buffer containing 3 mM Mg 2+ , we hypothesized that Mg 2+ -dependent oligomers of Aβ 42 are required to inhibit Na + ,K + -ATPase and the absence of such aggregates in pS8-Aβ 42 solution determines the absence of inhibition. To support this conclusion, we measured the Na + ,K + -ATPase inhibition by Aβ 42 at different time points and found that the degree of inhibition increases over time and plateaus after 30 min of incubation (Supplementary Figure 5). However, Aβ 42 incubated in Mg 2+ -containing solution for 30 min (in the absence of Na + ,K + -ATPase) did not inhibit Na + ,K + -ATPase if the activity was measured immediately after the addition of the oligomers to the enzyme-containing solution (Supplementary Figure 6). Since both Aβ 42 and pS8-Aβ 42 are able to bind to Na + ,K + -ATPase, we hypothesized that the initial binding of Aβ peptides to the enzyme acts as a seed for further aggregation of Aβ on the enzyme matrix and results in the inhibition of the enzyme. In the case of pS8-Aβ 42 , binding to the enzyme is not followed by oligomerization, which explains the absence of inhibition.
DISCUSSION
Since it was first proposed by Hardy in 1992, the amyloid hypothesis of AD has undergone a number of changes (Selkoe and Hardy, 2016). The focus of research has shifted FIGURE 3 | Interaction of Na + /K + -ATPase with beta-amyloid peptides. (A) Hydrolytic activity of purified Na +/ K + -ATPase was measured after 60 min incubation with 40 µM of Aβ 42 or pS8-Aβ 42 . The histogram represents the enzyme activity in the presence or absence of amyloid peptides, the enzyme activity without Aβ 42 is accepted as 100%. Data are mean values for three independent experiments ± SD. Statistical analysis was performed using one-way ANOVA (F = 1894.8, degree of freedom 2, P < 0.00001) with post hoc testing (using paired samples Student's t-test with Bonferroni correction); after a Bonferroni correction, a P-value < 0.016 was considered as statistically significant; * P < 0.001. ITC titration curves (upper panels) and binding isotherms (lower panels) for Aβ 42 (B), pS8-Aβ 42 (C), and Aβ 17-42 (D) interaction with Na,K-ATPase at 25 • C.
from senile plaques to soluble toxic oligomers of Aβ, and the inherent role of the tau-protein in the AD pathogenesis has been elucidated. Accumulation and aggregation of Aβ are still considered triggers of the AD pathological cascade (Musiek and Holtzman, 2015); however, recent failures of monoclonal antibodies against Aβ in clinical trials call for reassessment of the role of Aβ in AD pathogenesis (Abbott and Dolgin, 2016). It is possible that the pivotal factor is not the total amount of Aβ in the blood but the range and relative quantities of modified Aβ species. A number of different modified Aβ forms, such as isomerized (Roher et al., 1993;Shimizu et al., 2000), pyroglutamylated (Wirths et al., 2009;Nussbaum et al., 2012), truncated (Kummer and Heneka, 2014), and other peptides, have been identified in senile plaques. Previously identified chemical modifications of Aβ seem to increase the pathogenic properties of the peptide. However, based on the present data, Ser8 phosphorylation could be the first identified modification that reverses some disease-associated properties of Aβ.
Protein phosphorylation is a ubiquitous modification, which tightly and precisely regulates the structural and functional characteristics of proteins (Hunter, 1995;Cohen, 2001). Aberrant protein phosphorylation is a disease-modifying factor, one of the most prominent examples of which is the hyperphosphorylation of tau in AD (Cohen, 2001;Ballatore et al., 2007). Recently, pS8-Aβ was obtained in vitro by Aβ phosphorylation with protein kinase A and was subsequently identified in vivo (Kumar et al., 2011(Kumar et al., , 2013. Based on the oligomeric state of pS8-Aβ derived from brain tissue of AD model mice (Kumar et al., 2013) and FIGURE 4 | The model of Na + /K + -ATPase:pS8-Aβ 42 peptide complex on the basis of 2zxe PDB structure. Na + /K + -ATPase is shown as a translucent gray molecular surface, C-terminal part (17-42) of pS8-Aβ 42 peptide is shown in green and the N-terminal metal-binding domains (1-16) is shown in yellow. The phosphorylated Ser8 residue is located outside of the interaction region (highlighted in red). The zinc-binding site 11-14 and pSer8 residue are shown by the ball and stick representation. accelerated aggregation of pS8-Aβ 40 in vitro, Kumar et al. (2011) have proposed that pS8-Aβ represents a potentially pathogenic agent in AD. However, the aggregation of pS8-Aβ was previously studied in the absence of divalent cations, particularly zinc ions, which are involved in both physiological processes and AD pathogenesis (Takeda, 2000;Frederickson et al., 2005;Kumar et al., 2011;Jamasbi et al., 2017). In a complex with Aβ, zinc ions can form seeds of pathogenic aggregation (Frederickson and Bush, 2001;Miller et al., 2010;Bush, 2012). This is especially likely to occur in synapses, where concentrations of Zn 2+ may reach 100s of micromoles per liter (Paoletti et al., 2009).
We have previously shown that phosphorylation of Aβ at Ser8 leads to an increase in the zinc-dependent dimerization of Aβ (Kulikova et al., 2014). Zinc-induced dimerization of the unmodified Aβ occurs through residues 11EVHH14 , and the His6 residue is recruited for further oligomerization (Istrate et al., 2016). However, in pS8-Aβ dimers, His6 forms an additional intramolecular Zn 2+ -binding site with the pSer8 residue, which thereby excludes the His6 residue from further oligomerization (Kulikova et al., 2014). Thus, phosphorylation should lead to a decrease in the ability of Aβ to form zinc-induced aggregates. We investigated the effect of Ser8 phosphorylation on the pathogenic properties of the Aβ species Aβ 42 , which is more prone to aggregation than Aβ 40 and seems to trigger the disease in a number of models (Jarrett et al., 1993;Johnson-Wood et al., 1997;Gouras et al., 2000). As expected, Ser8 phosphorylation reduced the zinc-driven aggregation of Aβ 42 . The formation of aggregates in an equimolar mixture of pS8-Aβ 42 and Aβ 42 was also dramatically suppressed. This may be due to the fact that pS8-Aβ 42 forms zinc-induced heterodimers with the unmodified Aβ 42 , which are not capable of further aggregation. Formation of heterodimers between the metal-binding domains of Aβ and pS8-Aβ was previously observed in vitro (Mezentsev et al., 2016). We hypothesized that in vitro inhibition of zinc-dependent aggregation by pS8-Aβ 42 will result in an anti-amyloidogenic effect in vivo. To test the hypothesis, we studied the effect of retro-orbital injections of pS8-Aβ 42 on the progression of cerebral amyloidosis in a murine model of AD, B6C3-Tg(APPswe,PSEN1dE9)85Dbo/j. We have previously found that retro-orbital injection of isoD7-Aβ 42 , but not of Aβ 42 , promoted the amyloid plaque formation in mice of this line (Kozin et al., 2013). It was the first evidence of the facilitation of cerebral amyloidosis by a synthetic Aβ species injected into the bloodstream. The ability of blood-derived Aβ to induce AD-like pathology has recently been confirmed in a murine parabiosis model (Bu et al., 2017); however, the authors have not identified the Aβ species that triggered the pathogenesis. It is important to mention that isoD7-Aβ 42 exhibits increased zinc-dependent oligomerization in vitro (Istrate et al., 2016), and may serve as a seed of zinc-dependent aggregation of Aβ in brains of transgenic mice. In the case of pS8-Aβ 42 , which is unable to form zinc-dependent aggregates, we expected to observe an opposite effect. Indeed, the number of plaques in the hippocampus of the transgenic mice that received pS8-Aβ 42 injections was approximately two-third of that in the mice that received PS injections. Apparently, pS8-Aβ 42 is unable to serve as an aggregation seed in vivo and also partially prevents the amyloidogenic aggregation of the endogenous Aβ peptides. Based on zinc-dependent aggregation of the Aβ 42 /pS8-Aβ 42 mixture in vitro, it is likely that pS8-Aβ 42 forms heterodimers with Aβ 42 in the nerve tissue, which prevents the formation of senile plaques. This indicates that bloodstream-derived Aβ can serve not only as a trigger but also as an obstacle for cerebral amyloidosis progression, depending on the Aβ species composition. These findings further support the role of zinc-induced aggregation as an important event in the amyloidogenic process.
It is known that cognitive deficits in AD or in corresponding models do not always correlate with the appearance of amyloid plaques and often appear before Aβ aggregates can be detected (Musiek and Holtzman, 2015). Such effects are FIGURE 5 | The possible role of pS8-Aβ 42 in maintaining brain health. In healthy brain synapses (A), the release of Zn 2+ during synaptic transmission does not cause Aβ accumulation, as phosphorylated Aβ does not form zinc-induced aggregates and also prevents zinc-induced aggregation of the intact peptide through the formation of heterodimers. Phosphorylated Aβ does not inhibit Na + /K + -ATPase (NKA) and the heterodimers are easily cleared, therefore the synapse function is not impaired. If pS8-Aβ 42 is depleted (B), synaptic release of zinc ions promotes oligomerization of Aβ, inhibition of Na + ,K + -ATPase and, eventually, leads to the amyloid plaques formation, which impairs the synapse function. associated with soluble toxic oligomers of the Aβ peptide (Haass and Selkoe, 2007) or with receptor-mediated effects of Aβ (Dinamarca et al., 2012). Recently, we have shown that Aβ can bind to Na + ,K + -ATPase (Petrushanko et al., 2016), whose activity is critically important for maintaining electrogenic properties of neurons. Here, we found that binding of monomeric Aβ to the enzyme and subsequent oligomerization of the peptide on the Aβ:Na + ,K + -ATPase matrix leads to the inhibition of enzyme activity. This observation provides a possible explanation for the decrease in the activity of Na + ,K + -ATPase in brain tissues of AD patients (Zhang et al., 2013;Kairane et al., 2014;Petrushanko et al., 2016). Unlike the unmodified peptide, pS8-Aβ 42 does not show an inhibitory effect on Na + ,K + -ATPase. Surprisingly, the binding parameters of pS8-Aβ 42 to the enzyme were almost identical to those of the unmodified peptide; thus, it is not the initial binding that defines the inhibitory properties of Aβ 42 toward Na + ,K + -ATPase. We further showed that the N-terminal domain of Aβ 42 (including the pS8 residue) is not involved in the initial binding and is probably exposed to the solution. Since the N-terminal domain of the peptide governs its metal-dependent oligomerization, we suggest that the inhibition of Na + ,K + -ATPase is caused by the metaldependent formation of Aβ oligomers seeded by the solutionexposed N-terminal domain of the first Na,K-ATPase-bound peptide. Phosphorylation of the peptide at Ser8 seems to interfere with this process, which may correspond to its inability to form oligomers triggered by Mg 2+ (present in Na + ,K + -ATPase activity measurements buffer) or by Zn 2+ . Thus, a decrease in the level of phosphorylation of the peptide in the elderly may lead to the inhibition of Na + ,K + -ATPase, development of neurotoxic effects, and the disruption of nerve transduction long before the appearance of amyloid aggregates. The possible role of Aβ phosphorylation in the brain is presented at Figure 5.
CONCLUSION
In this study, we demonstrated that the phosphorylation of Aβ 42 at Ser8 changes its properties of zinc-driven aggregation, inhibition of Na + /K + -ATPase, and amyloidogenicity. The obtained data indicate that the phosphorylation of Aβ 42 neutralizes some of its AD-associated properties. Our findings provide the basis for discussion about the role of Ser8 phosphorylation in Aβ , which was previously considered only as a pathogenic modification. The anti-amyloidogenic properties of pS8-Aβ 42 in vivo support the intrinsic role of zinc-mediated aggregation in the formation of the senile plaques. Further studies addressing the role of pS8-Aβ in the human brain are needed for better clarification of both the significance of pS8-Aβ and the relevance of the obtained data to AD.
AUTHOR CONTRIBUTIONS
EB and IP performed most of the experiments with contributions from GT, AC, and SR. EB, IP, and VM drafted the paper. SK, OL, and AM contributed the conception and design of the study. All authors contributed to manuscript revision, read and approved the submitted version. | 8,043 | 2018-08-29T00:00:00.000 | [
"Biology",
"Chemistry"
] |
A Novel and Effective Recyclable BiOCl/BiOBr Photocatalysis for Lignin Removal from Pre-Hydrolysis Liquor
The presence of lignin hampers the utilization of hemicelluloses in the pre-hydrolysis liquor (PHL) from the kraft-based dissolving pulp production process. In this paper, a novel process for removing lignin from PHL was proposed by effectively recycling catalysts of BiOCl/BiOBr. During the whole process, BiOCl and BiOBr were not only adsorbents for removing lignin, but also photocatalysts for degrading lignin. The results showed that BiOCl and BiOBr treatments caused 36.3% and 33.9% lignin removal, respectively, at the optimized conditions, and the losses of hemicellulose-derived saccharides (HDS) were both 0.1%. The catalysts could be regenerated by simple photocatalytic treatment and obtain considerable CO and CO2. After 15 h of illumination, 49.9 μmol CO and 553.0 μmol CO2 were produced by BiOCl, and 38.7 μmol CO and 484.3 μmol CO2 were produced by BiOBr. Therefore, both BiOCl and BiOBr exhibit excellent adsorption and photocatalytic properties for lignin removal from pre-hydrolysis.
Introduction
The lignocellulosic biomass has a promising future and has been identified as a predictable, feasible and sustainable resource for value-added products [1]. The efficient separation and transformation of lignocellulosic components is an effective way to realize its comprehensive utilization, including the deep eutectic solvents (DESs) treatment for component isolation [2], lignin-based composite for photocatalytic degradation [3] and the production process of kraft-based dissolving pulp [4]. After the pretreating and cooking of raw materials, hemicellulose-based PHL, lignin-based black liquor and cellulose-based dissolving pulp were obtained independently [5]. The dissolving pulp is mainly used to prepare cellulose ether, acetate fiber, viscose fiber and other cellulose derivatives, which can be used in textiles and cigarette filters.
In the pulp mill, PHL is conventionally mixed with black liquor and burned for the recovery of chemicals and energy, which is unprofitable [6]. The hemicellulose-derived saccharides (HDS) in PHL can be utilized in the production of various value-added products, such as platform chemicals, biomaterials and biofuels [7,8]. In addition, xylooligosaccharide, which can improve the health of animal intestinal systems, is widely used in biochemical synthesis in the pharmaceutical and food industries [9]. The utilization of HDS is essential for the biorefinery concept, which can improve the comprehensive and efficient utilization of lignocellulosic biomass, increase revenue sources and reduce pollution load for enterprises. However, the presence of non-saccharide compounds in the PHL, especially lignin, hinders the separation and utilization of HDS. Therefore, the elimination of lignin is essential for the production of hemicellulose-based value-added products from PHL [10].
Many methods have been proposed for lignin removal from PHL which have been reported in the literature, including: (a) acidification [11][12][13]; (b) adsorption with lime [14,15], Nanomaterials 2021, 11, 2836 2 of 12 activated carbon (AC) [16,17] and ion exchange resin [18,19]; (c) flocculation with polymers [20,21]; (d) nanofiltration or microfiltration [22][23][24]; and (e) polymerization with laccase [25,26] and horseradish peroxidase [27,28]. All these processes had certain practical challenges, so they used to be combined for purifying the sugars of PHL [29,30]. Besides, most lignin removal methods were non-recyclable, which would result in increased costs and even pollution. Many studies reported that AC-treated PHL could be economically reused, such as in solvent extraction and thermal regeneration. However, solvent extraction has limited recovery efficiency, and thermal regeneration is similar to the production process of activated carbon; the effect depends on the type of AC, adsorption material and the choice of process [31,32]. Therefore, as there are some difficulties in recycling AC, a more recyclable process should be explored for removing lignin from PHL.
Photocatalysis is a green technology with important application prospects in energy and the environment [33]. Since the first report about the photocatalytic hydrogen production performance of TiO 2 by Fujishima and Honda in 1972 [34], numerous efforts have been concentrated on developing excellent semiconductor photocatalysts [35]. A new type of photocatalytic material, BiOX (X = F, Cl, Br, I), has attracted much attention from researchers all over the world with its excellent optical adsorption, electrical properties and high photocatalytic activity [36,37]. The essence of a semiconductor photocatalytic reaction is that the photocatalytic material absorbs the external radiant light energy, produces electron-hole pairs, and then conducts a series of chemical reactions with the adsorbed material on the catalyst surface. A photocatalytic reaction is one of the interactions between light and matter which is the fusion of photoreaction and catalytic reaction. The photo-excited catalyst generates electron-hole pairs and further acts through electron holes to generate a redox reaction [38,39]. Photocatalysis was highly considered as a desirable and green strategy which played an indispensable role in many fields [40]. Photocatalytic applications of semiconductors also received attention in the pulp and paper industry. For example, photocatalytic technology significantly increased the value of black liquor from papermaking [41], so it is destined to have a potential utility as PHL.
Compared with AC, the regeneration of catalysts-or, in other words, photocatalytic reaction-is a much simpler process. Photocatalytic degradation of lignin has many advantages: reaction conditions, low cost, full light energy utilization, low energy consumption, no secondary pollution, simple equipment operation, etc. The products are only CO, CO 2 and H 2 O, while CO can be used for fuels or mineralized into CO 2 by continued illumination.
The objective of the present study is to investigate a new recyclable method for removing lignin from PHL. The hypothesis is that BiOCl/BiOBr can remove lignin in PHL by electrostatic adsorption, and the catalyst can also be recycled after the complete degradation of lignin by photocatalysis. The process of removing lignin by BiOCl/BiOBr can also be combined with other processes, as the catalysts are stable and will not alter the chemistry of the PHL after treatment. In addition, the optimization of BiOCl/BiOBr treatment conditions for PHL are investigated, and the feasibility of repeated experiments of BiOCl/BiOBr is especially evaluated.
. PHL Preparation
Fast-growing poplar wood chips were obtained from Shan Dong Sun Paper Industry Joint Stock Co., Ltd., Jining, China. The washed wood chips were balanced in moisture by airing under sunlight, then qualified wood chips were picked and stored in a plastic receptacle at room temperature. Pre-hydrolysis of 500 g oven-dried wood chips was conducted in a horizontal rotary pulp digester (rotary type autoclave No.2611, Kumgai Riki Kogyo Co., Ltd. (Tokyo, Japan) with wood to water at a ratio of 1:6 (w/w), heated from room temperature to 170 • C at a speed of approximately 2.5 • C/min, and kept for 60 min. After pre-hydrolysis, the digester was rapidly taken off and cooled down with running water. Large particles and impurities in the PHL were removed by gauze in advance, and the liquid was filtered through a 0.22 µm nylon membrane, then placed in an encapsulated vial and stored at 4 • C before analysis.
Synthesis of BiOCl/BiOBr
At room temperature, 2 mmol Bi(NO 3 ) 3 ·5H 2 O and 3 mmol C 16 MIMCl were separately dissolved in 40 mL 2-methoxyethanol by stirring. Afterwards, the C 16 MIMCl solution was dropped into the Bi(NO 3 ) 3 ·5H 2 O solution slowly, then the mixture was continuously stirred for 30 min and decanted into a 100 mL Teflon-lined stainless steel autoclave. The autoclave was transferred into a 160 • C oven, heated and maintained for 1 h, and then cooled to room temperature naturally. The resulting BiOCl was washed several times with ethanol and water alternately, and dried at 60 • C. Meanwhile, the synthesis of BiOBr was under the same circumstances of the BiOCl synthesis, except replacing C 16 MIMCl with C 16 MIMBr.
BiOCl/BiOBr Treatment of PHL
A certain amount of BiOCl/BiOBr (1.0 wt% to 12.0 wt%) was added to 20 g PHL in a glass beaker, and the mixture was stirred at 350 rpm at room temperature. In addition, the effect of elapsed time (1 min to 60 min) on lignin removal was optimized. After the adsorption process, the solution was filtered through a 0.22 µm nylon membrane; subsequently, the filtrate was stored under the same condition as the PHL storage, and the precipitate was dried at 60 • C before photocatalytic regeneration.
Photocatalytic Degradation of Lignin
The regenerations of used BiOCl/BiOBr were conducted in a 100 mL airtight glass round-bottom flask. The used BiOCl/BiOBr after drying was suspended in 50 mL water before the experiment, the reaction mixture was stirred gently, and the reactor was kept at 30 • C by a cooling water circulation system. A 300 W xenon lamp (PLS-SXE300, Beijing Trusttech Co., Ltd., Beijing, China) was irradiated from the top of reactor. The gas in the reactor was extracted with a syringe every 3 h, measured by gas chromatograph (CEL-GC7900, Shimane, Japan), and assembled with a flame ionization detector (FID); ultrapure nitrogen was used as a carrier gas.
Cycle Experiments of BiOCl/BiOBr
The repeat experiments were carried out as previously mentioned. In detail, after 9 h photocatalysis in closed reactor, the mixture was exposed to air for illumination of 12 h to ensure the complete degradation of organic matter on the surface of the catalyst. After photocatalysis, solids and liquids were separated by filtration. Regenerated catalyst was recycled after drying, and the filtrate was preserved for subsequent analysis. Adsorption experiments were carried out with different cycle times under 6.0 wt% dosage and 10 min adsorption time at room temperature, and the regenerated catalysts were repeat-treated with 20 g fresh PHL.
Characterization of BiOCl/BiOBr
Characterizations of fresh BiOCl/BiOBr, used BiOCl/BiOBr and regenerated BiOCl/ BiOBr were all investigated. The crystal phases were performed on a Bruker AXS D8 X-ray diffraction (XRD) with Cu Kα radiation. The morphologies were observed by scanning electron microscopy (SEM, Hitachi Regulus8220). The UV-vis diffuse reflectance spectra (DRS) were measured on a Shimadzu UV-2550 recording spectrometer with BaSO 4 -coated integration sphere. The Fourier-transform infrared (FT-IR) spectroscopy was carried out on an ALPHA Fourier-transform infrared spectrometer.
Zeta Potentials and pH Values
A lignin suspension (1 g/L) and a solution with xylose, arabinose, galactose, glucose and mannose (all 1 g/L) were prepared in advance. The pH values of the former two liquids were measured along with PHL and water, then the lignin suspension, sugar solution and water were adjusted to the pH of PHL by acetic acid. The zeta potentials of these adjusted liquids were measured by using a Malvern Zetasizer Nano ZSP (Malvern, UK).
Analysis of PHL
The monosaccharide concentration in the PHL was determined by using an ion chromatography system (ICS-5000 + DC, Thermo Scientific, Waltham, MA, USA). The ion chromatography system was equipped with an ED40 electrochemical detector (Au working electrode, Ag/AgCl reference electrode), a CarboPA100 analytical column (3 mm × 150 mm) and a CarboPac PA100 protection column (3 mm × 30 mm). For measuring the concentration of total saccharides in the PHL, an additional acid hydrolysis step was carried out on the samples under the conditions of 4% (w/w) H 2 SO 4 and 121 • C for 1 h.
The lignin content in the PHL was determined by a UV/vis spectrophotometer (Agilent Technologies, Palo Alto, CA, USA) at 205 nm. The PHL was diluted a certain number of times with water as the blank group, and the absorbance value of the sample at 205 nm was measured by UV/vis spectrophotometer (the absorbance value was in the range of 0.2~0.7). Lignin content of the PHL was calculated by following Equation (1): where A, D and B represent the absorbance, dilution ratio and lignin content (g/L), respectively, and 110 is the absorption coefficient, L/(g·cm −1 ). The properties of the prepared PHL were as follows: HDS, 19.4 g/L; monosaccharides, 4.7 g/L; oligosaccharides, 14.7 g/L; dissolved lignin, 4.4 g/L. Varying concentrations of monosaccharide standard samples (arabinose, galactose, glucose, xylose and mannose) were prepared and detected separately by ion chromatography, and the standard curve was fitted according to the corresponding peak area. The PHL diluted to a certain concentration was detected by ion chromatography. The corresponding monosaccharide concentration was calculated through the peak area. The concentration of HDS was the sum of the five sugar contents.
A Novel Lignin Removal Method by Electrostatic Adsorption
The process of kraft-based dissolving pulp by pre-hydrolysis was considered as the most suitable production practice for biorefinery [5]. Figure 1 shows a proposed diagram for high-value utilization of lignocellulose by an environmental, recyclable and economical process. In the production process of dissolving pulp, hemicellulose and soluble smallmolecular lignin can be dissolved and removed by hot water pre-hydrolysis. As discussed before, the removal of lignin is essential prior to the value-added utilization of HDS, and the BiOCl/BiOBr was employed to remove lignin from PHL. After treatment, the PHL contained less lignin, and the HDS could be conveniently recovered/utilized for valueadded products by further purification. Lignin adsorbed on the surface of catalyst could be Nanomaterials 2021, 11, 2836 5 of 12 photocatalytically degraded into CO, CO 2 and H 2 O, and the catalysts would be regenerated to be used for treating the PHL again. economical process. In the production process of dissolving pulp, hemicellulose and soluble small-molecular lignin can be dissolved and removed by hot water pre-hydrolysis. As discussed before, the removal of lignin is essential prior to the value-added utilization of HDS, and the BiOCl/BiOBr was employed to remove lignin from PHL. After treatment, the PHL contained less lignin, and the HDS could be conveniently recovered/utilized for value-added products by further purification. Lignin adsorbed on the surface of catalyst could be photocatalytically degraded into CO, CO2 and H2O, and the catalysts would be regenerated to be used for treating the PHL again.
Optimization of BiOCl/BiOBr Treatment Conditions
Under the conditions of room temperature and adsorbent dosage 6.0 wt%, the effect of elapsed time on the removal of lignin in PHL by BiOCl/BiOBr is shown in Figure 2. Obviously, the adsorption experiment took place rapidly. The removal of lignin (relative to the lignin content in the pre-hydrolysis liquor) was 31.4% and 29.2% by BiOCl and BiOBr, respectively, at 1 min, then the adsorption equilibrium was achieved within 10 min, and 36.3% and 33.9% removals of lignin were achieved, respectively. This result indicates that the removal of lignin by BiOCl or BiOBr from PHL is a transitory process which is applicable for factories. Meanwhile, the processing time selected for the subsequent BiOCl/BiOBr treatments of the PHL was 10 min. As a comparison, the study investigated the effectiveness of wheat straw delignification, as well as the selectivity of the process, by using deep eutectic solvents. The highest lignin removal was observed for the ChCl and OX system (1:1, 57.9%), followed by the ChCl-lactic (1:10, 29.1%) and -malic (1:1, 21.6%) acid system [2]. The removed lignin could be synthesized for photoactive lignin/Bi4O5Br2/BiOBr bio-inorganic composites. The lignin-based composite decreased the dye concentration from 80 mg·L −1 to 12.3 mg·L −1 for RhB (85%) and from 80 mg·L −1 to 4.4 mg·L −1 for MB (95%). Therefore, the lignin as a main component of the composite was found to be an efficient and rapid biosorbent for nickel, lead, and cobalt ions [3].
Optimization of BiOCl/BiOBr Treatment Conditions
Under the conditions of room temperature and adsorbent dosage 6.0 wt%, the effect of elapsed time on the removal of lignin in PHL by BiOCl/BiOBr is shown in Figure 2. Obviously, the adsorption experiment took place rapidly. The removal of lignin (relative to the lignin content in the pre-hydrolysis liquor) was 31.4% and 29.2% by BiOCl and BiOBr, respectively, at 1 min, then the adsorption equilibrium was achieved within 10 min, and 36.3% and 33.9% removals of lignin were achieved, respectively. This result indicates that the removal of lignin by BiOCl or BiOBr from PHL is a transitory process which is applicable for factories. Meanwhile, the processing time selected for the subsequent BiOCl/BiOBr treatments of the PHL was 10 min. As a comparison, the study investigated the effectiveness of wheat straw delignification, as well as the selectivity of the process, by using deep eutectic solvents. The highest lignin removal was observed for the ChCl and OX system (1:1, 57.9%), followed by the ChCl-lactic (1:10, 29.1%) and -malic (1:1, 21.6%) acid system [2]. The removed lignin could be synthesized for photoactive lignin/Bi 4 O 5 Br 2 /BiOBr bioinorganic composites. The lignin-based composite decreased the dye concentration from 80 mg·L −1 to 12.3 mg·L −1 for RhB (85%) and from 80 mg·L −1 to 4.4 mg·L −1 for MB (95%). Therefore, the lignin as a main component of the composite was found to be an efficient and rapid biosorbent for nickel, lead, and cobalt ions [3].
In batch adsorption experiments, the removals of lignin and losses of HDS were measured at various adsorbent levels at 10 min and room temperature, and the results are shown in Figure 3. Figure 3a indicates that the removal of lignin was strongly affected by the dosage of BiOCl/BiOBr. Along with the increase in dosage from 1.0 wt% to 12.0 wt%, the removal of lignin increased from 15.5% to 50.1% by BiOCl, while the removal of lignin increased from 15.2% to 47.1% by BiOBr. The removals of lignin in the PHL by BiOCl and BiOBr were effective, and the loss of HDS should be minimized or even eliminated. Figure 3b shows the effect of BiOCl/BiOBr on the loss of HDS in the PHL. It can be seen that the losses of HDS were almost none at low dosage, and the losses of HDS in the PHL were both 0.1% at 6.0 wt% dosage of BiOCl and BiOBr, while the losses of HDS were 5.1% and 1.2% after 12.0 wt% dosage of BiOCl and BiOBr, respectively. The reason for this small portion of HDS lost was attributed to partial lignin adsorbed on BiOCl/BiOBr, which were covalently bonded to HDS [42]. The following experiments were conducted with 6.0 wt% dosage of BiOCl/BiOBr because the lignin removal was considerable and the HDS loss was negligible. In batch adsorption experiments, the removals of lignin and losses of HDS were measured at various adsorbent levels at 10 min and room temperature, and the results are shown in Figure 3. Figure 3a indicates that the removal of lignin was strongly affected by the dosage of BiOCl/BiOBr. Along with the increase in dosage from 1.0 wt% to 12.0 wt%, the removal of lignin increased from 15.5% to 50.1% by BiOCl, while the removal of lignin increased from 15.2% to 47.1% by BiOBr. The removals of lignin in the PHL by BiOCl and BiOBr were effective, and the loss of HDS should be minimized or even eliminated. Figure 3b shows the effect of BiOCl/BiOBr on the loss of HDS in the PHL. It can be seen that the losses of HDS were almost none at low dosage, and the losses of HDS in the PHL were both 0.1% at 6.0 wt% dosage of BiOCl and BiOBr, while the losses of HDS were 5.1% and 1.2% after 12.0 wt% dosage of BiOCl and BiOBr, respectively. The reason for this small portion of HDS lost was attributed to partial lignin adsorbed on BiOCl/BiOBr, which were covalently bonded to HDS [42]. The following experiments were conducted with 6.0 wt% dosage of BiOCl/BiOBr because the lignin removal was considerable and the HDS loss was negligible.
Affinity of Lignin to BiOCl/BiOBr
The strong adsorption of lignin was on account of its high molecular weight and In batch adsorption experiments, the removals of lignin and losses of HDS were measured at various adsorbent levels at 10 min and room temperature, and the results are shown in Figure 3. Figure 3a indicates that the removal of lignin was strongly affected by the dosage of BiOCl/BiOBr. Along with the increase in dosage from 1.0 wt% to 12.0 wt%, the removal of lignin increased from 15.5% to 50.1% by BiOCl, while the removal of lignin increased from 15.2% to 47.1% by BiOBr. The removals of lignin in the PHL by BiOCl and BiOBr were effective, and the loss of HDS should be minimized or even eliminated. Figure 3b shows the effect of BiOCl/BiOBr on the loss of HDS in the PHL. It can be seen that the losses of HDS were almost none at low dosage, and the losses of HDS in the PHL were both 0.1% at 6.0 wt% dosage of BiOCl and BiOBr, while the losses of HDS were 5.1% and 1.2% after 12.0 wt% dosage of BiOCl and BiOBr, respectively. The reason for this small portion of HDS lost was attributed to partial lignin adsorbed on BiOCl/BiOBr, which were covalently bonded to HDS [42]. The following experiments were conducted with 6.0 wt% dosage of BiOCl/BiOBr because the lignin removal was considerable and the HDS loss was negligible.
Affinity of Lignin to BiOCl/BiOBr
The strong adsorption of lignin was on account of its high molecular weight and being negatively charged on its surface, while the positive charges on the surface of BiOCl/BiOBr according to the determination of zeta potentials are shown in Table 1. A
Affinity of Lignin to BiOCl/BiOBr
The strong adsorption of lignin was on account of its high molecular weight and being negatively charged on its surface, while the positive charges on the surface of BiOCl/BiOBr according to the determination of zeta potentials are shown in Table 1. A new lignin removal process was explored based on the zeta potentials of lignin suspension, sugar solution, BiOCl suspension and BiOBr suspension, as the resulting values were −21.2, −2.06, +8.7 and +2.8, respectively. It reveals that the lignin suspension was electronegative, the BiOCl/BiOBr suspension was electropositive and the sugar solution was close to neutral. Thus, the opposite charges on the lignin and the BiOCl/BiOBr surface should result in a strong electrostatic attraction between the two, as shown in Figure 4. The electronegativity of lignin was much higher than that of HDS, which caused the strong affinity of lignin and weak affinity of HDS to BiOCl/BiOBr. In addition, as seen from the zeta potentials, the electropositivity of BiOCl was stronger than BiOBr, which corresponded to the finding that the removal of lignin by BiOCl was better than BiOBr. new lignin removal process was explored based on the zeta potentials of lignin suspension, sugar solution, BiOCl suspension and BiOBr suspension, as the resulting values were −21.2, −2.06, +8.7 and +2.8, respectively. It reveals that the lignin suspension was electronegative, the BiOCl/BiOBr suspension was electropositive and the sugar solution was close to neutral. Thus, the opposite charges on the lignin and the BiOCl/BiOBr surface should result in a strong electrostatic attraction between the two, as shown in Figure 4. The electronegativity of lignin was much higher than that of HDS, which caused the strong affinity of lignin and weak affinity of HDS to BiOCl/BiOBr. In addition, as seen from the zeta potentials, the electropositivity of BiOCl was stronger than BiOBr, which corresponded to the finding that the removal of lignin by BiOCl was better than BiOBr.
Characterization of BiOCl/BiOBr
The XRD spectra of fresh, used and regenerated BiOCl/BiOBr were displayed to investigate the crystalline structures, and the results are shown in Figure 5. As can be seen, all diffraction peaks of the products in Figure 5a could be matched to the tetragonal BiOCl (JCPDS no. 06-0249), and all diffraction peaks of the products in Figure 5b could be perfectly identified as tetragonal BiOBr (JCPDS no. 09-0393). Moreover, no additional diffraction peaks of impurities are detected, and this implies that synthesized catalysts had high purities and the structures of the catalyst remained after adsorption and photocatalysis. The used catalysts did not react with the organics in the PHL, and the generated catalysts certainly did not transform into Bi2O3 or Bi(OH)3. These are owing to the chemical stability of catalysts and having a tendency to complete the cycle of catalysts. In the photocatalytic experiment, reactive ⦁OH radicals could result in the degradation of lignin due to the scission β-O-4 bond. This process resulted in the generation of benzyl, alkoxy and alkyl free radicals, which took part in lignin depolymerisation reactions to form low molecular weight lignin fragments. The ⦁OH radicals could also directly attack
Characterization of BiOCl/BiOBr
The XRD spectra of fresh, used and regenerated BiOCl/BiOBr were displayed to investigate the crystalline structures, and the results are shown in Figure 5. As can be seen, all diffraction peaks of the products in Figure 5a could be matched to the tetragonal BiOCl (JCPDS no. 06-0249), and all diffraction peaks of the products in Figure 5b could be perfectly identified as tetragonal BiOBr (JCPDS no. 09-0393). Moreover, no additional diffraction peaks of impurities are detected, and this implies that synthesized catalysts had high purities and the structures of the catalyst remained after adsorption and photocatalysis. The used catalysts did not react with the organics in the PHL, and the generated catalysts certainly did not transform into Bi 2 O 3 or Bi(OH) 3 . These are owing to the chemical stability of catalysts and having a tendency to complete the cycle of catalysts. In the photocatalytic experiment, reactive •OH radicals could result in the degradation of lignin due to the scission β-O-4 bond. This process resulted in the generation of benzyl, alkoxy and alkyl free radicals, which took part in lignin depolymerisation reactions to form low molecular weight lignin fragments. The •OH radicals could also directly attack the phenyl rings of the lignin to form catechol, resorcinol and hydroquinone, and the lignin could be completely mineralized to CO 2 .
The SEM images of fresh, used and regenerated BiOCl/BiOBr are shown in Figure 6. It can be seen that both BiOCl and BiOBr were about 1-2 µm and exhibited hollow flowerlike morphologies, which presented clusters of some interlaced nanosheets. The ultrathin nanosheet structures contributed to preferable specific surface area, which may be beneficial for the adsorption of lignin. Additionally, the used and regenerated catalysts held the same morphologies, which confirmed that adsorption and photocatalysis did not change the structures of catalysts; this demonstrated the feasibility of the cycle of BiOCl/BiOBr by photocatalysis once more. the phenyl rings of the lignin to form catechol, resorcinol and hydroquinone, and the lignin could be completely mineralized to CO2. The SEM images of fresh, used and regenerated BiOCl/BiOBr are shown in Figure 6. It can be seen that both BiOCl and BiOBr were about 1-2 μm and exhibited hollow flowerlike morphologies, which presented clusters of some interlaced nanosheets. The ultrathin nanosheet structures contributed to preferable specific surface area, which may be beneficial for the adsorption of lignin. Additionally, the used and regenerated catalysts held the same morphologies, which confirmed that adsorption and photocatalysis did not change the structures of catalysts; this demonstrated the feasibility of the cycle of BiOCl/BiOBr by photocatalysis once more. As shown in Figure 7a,b, the major features of FT-IR spectra for these three samples of BiOCl were also similar, except that the four new peaks appeared after the lignin adsorption step was performed, with a similar result in BiOBr. The peaks at 1045 cm −1 and 1116 cm −1 could be ascribed to aromatic C-H in-plane deformation in the syringyl ring; the peaks at 1425 cm −1 corresponded to the C-H in-plane deformation with aromatic ring stretching; and the peaks at 1527 cm −1 could be assigned to aromatic skeletal vibration [43]. It was confirmed from FT-IR spectra that a portion of lignin was adsorbed onto BiOCl and BiOBr particles in the process of adsorption treatment. After the complete photocatalytic conversion of the adsorbed lignin, the characteristic FT-IR peaks of lignin disappeared. During the photocatalysis of lignin by BiOCl, some intermediate CO3 2− ion degradation of lignin was produced. The fundamental vibrations arose from the carbonate CO3 2− ion and were assigned to the asymmetric stretch (v3) at 1390 cm −1 and the out-of-plane bending (v2) vibration at 845 cm −1 . This phenomenon was considered to be among the prominent The SEM images of fresh, used and regenerated BiOCl/BiOBr are shown in Figure 6 It can be seen that both BiOCl and BiOBr were about 1-2 μm and exhibited hollow flower like morphologies, which presented clusters of some interlaced nanosheets. The ultrathi nanosheet structures contributed to preferable specific surface area, which may b beneficial for the adsorption of lignin. Additionally, the used and regenerated catalyst held the same morphologies, which confirmed that adsorption and photocatalysis did no change the structures of catalysts; this demonstrated the feasibility of the cycle o BiOCl/BiOBr by photocatalysis once more. As shown in Figure 7a,b, the major features of FT-IR spectra for these three sample of BiOCl were also similar, except that the four new peaks appeared after the ligni adsorption step was performed, with a similar result in BiOBr. The peaks at 1045 cm −1 an 1116 cm −1 could be ascribed to aromatic C-H in-plane deformation in the syringyl ring; th peaks at 1425 cm −1 corresponded to the C-H in-plane deformation with aromatic rin stretching; and the peaks at 1527 cm −1 could be assigned to aromatic skeletal vibration [43 It was confirmed from FT-IR spectra that a portion of lignin was adsorbed onto BiOCl an BiOBr particles in the process of adsorption treatment. After the complete photocatalyti conversion of the adsorbed lignin, the characteristic FT-IR peaks of lignin disappeared During the photocatalysis of lignin by BiOCl, some intermediate CO3 2− ion degradation o lignin was produced. The fundamental vibrations arose from the carbonate CO3 2− ion an were assigned to the asymmetric stretch (v3) at 1390 cm −1 and the out-of-plane bendin (v2) vibration at 845 cm −1 . This phenomenon was considered to be among the prominen As shown in Figure 7a,b, the major features of FT-IR spectra for these three samples of BiOCl were also similar, except that the four new peaks appeared after the lignin adsorption step was performed, with a similar result in BiOBr. The peaks at 1045 cm −1 and 1116 cm −1 could be ascribed to aromatic C-H in-plane deformation in the syringyl ring; the peaks at 1425 cm −1 corresponded to the C-H in-plane deformation with aromatic ring stretching; and the peaks at 1527 cm −1 could be assigned to aromatic skeletal vibration [43]. It was confirmed from FT-IR spectra that a portion of lignin was adsorbed onto BiOCl and BiOBr particles in the process of adsorption treatment. After the complete photocatalytic conversion of the adsorbed lignin, the characteristic FT-IR peaks of lignin disappeared. During the photocatalysis of lignin by BiOCl, some intermediate CO 3 2− ion degradation of lignin was produced. The fundamental vibrations arose from the carbonate CO 3 2− ion and were assigned to the asymmetric stretch (v 3 ) at 1390 cm −1 and the out-of-plane bending (v 2 ) vibration at 845 cm −1 . This phenomenon was considered to be among the prominent absorption features within carbonate spectra. Figure 7c,d each display the UV-vis DRS spectra of fresh, used and regenerated BiOCl and BiOBr. Used catalysts exhibited obvious red-shift compared to fresh catalysts, which corresponded to the changes of samples' colours-the fresh catalysts were both white, then changed into yellow after the adsorption of lignin. The visible light of absorption of both regenerated catalysts increased significantly, which may be attributed to the generation of oxygen vacancy after photocatalysis, as the regenerated catalysts both changed into dark colour [44]. absorption features within carbonate spectra. Figure 7c,d each display the UV-vis DRS spectra of fresh, used and regenerated BiOCl and BiOBr. Used catalysts exhibited obvious red-shift compared to fresh catalysts, which corresponded to the changes of samples' colours-the fresh catalysts were both white, then changed into yellow after the adsorption of lignin. The visible light of absorption of both regenerated catalysts increased significantly, which may be attributed to the generation of oxygen vacancy after photocatalysis, as the regenerated catalysts both changed into dark colour [44].
Photocatalytic Degradation of Lignin
Lignin on surface of BiOCl/BiOBr degraded into CO, CO2 and H2O, etc. by photocatalysis, and the yields of gases (CO and CO2) by BiOCl and BiOBr are shown in the Figure 8a,b, respectively. These illustrate the photocatalytic degradation mechanism of the lignin adsorbed on the surface of catalysts. CO and CO2 were significantly increased with the extension of illumination time, which was due to the further degradation of lignin. The reaction rates of lignin degradation by BiOCl and BiOBr were both significant within 9 h, and further increase in the illumination time had ordinary effects on the yields of CO/CO2, which may be caused by the consumption of lignin and limited O2 in the hermetic bottle. By 15 h photocatalysis, 50.1 μmoL CO and 556.0 μmoL CO2 were obtained by BiOCl, while 38.9 μmoL CO and 487.3 μmoL CO2 were obtained when using BiOBr.
Photocatalytic Degradation of Lignin
Lignin on surface of BiOCl/BiOBr degraded into CO, CO 2 and H 2 O, etc. by photocatalysis, and the yields of gases (CO and CO 2 ) by BiOCl and BiOBr are shown in the Figure 8a,b, respectively. These illustrate the photocatalytic degradation mechanism of the lignin adsorbed on the surface of catalysts. CO and CO 2 were significantly increased with the extension of illumination time, which was due to the further degradation of lignin. The reaction rates of lignin degradation by BiOCl and BiOBr were both significant within 9 h, and further increase in the illumination time had ordinary effects on the yields of CO/CO 2 , which may be caused by the consumption of lignin and limited O 2 in the hermetic bottle. By 15 h photocatalysis, 50.1 µmoL CO and 556.0 µmoL CO 2 were obtained by BiOCl, while 38.9 µmoL CO and 487.3 µmoL CO 2 were obtained when using BiOBr.
Cycle Performance of BiOCl/BiOBr
The used BiOCl and BiOBr were regenerated by photocatalysis, then the adsorbents were used to perform adsorb experiments with PHL under the same circumstances after degradation of the lignin adsorbed on their surface. The cycle experiment was carried out under 6.0 wt% dosage BiOCl and BiOBr for 10 min; as shown in Figure 9a,b, the removals of lignin by the regenerated adsorbent maintained well. The lignin removal rate dropped to 35.7% and 33.3% at the third time of BiOCl and BiOBr, respectively, which might be caused by the loss of adsorbent during filtration and drying in the experiments. The tiny decrease was unavoidable, so the cycle of BiOCl and BiOBr remained a remarkable adsorptive property.
The yields of CO and CO 2 in three cycles by BiOCl and BiOBr are shown in Figure 9c,d. As can be seen, the yields of CO and CO 2 decreased a little upon repeat experiments because of the loss of catalyst during experiments. The finding was in agreement with the reduction in the lignin removal. The decrease in lignin removal was also an important reason, as the lignin adsorbed on the catalysts tailed off. At the first experiment, 40.4 µmoL CO and 424.8 µmoL CO 2 were obtained by 9 h BiOCl photocatalysis, which then dropped to 39.1 µmoL CO and 410.7 µmoL CO 2 at the third cycle. Furthermore, 36.3 µmoL CO and 452.8 µmoL CO 2 were obtained by 9 h BiOBr photocatalysis the first time, which then dropped to 35.0 µmoL CO and 441.9 µmoL CO 2 after three cycles. This analysis shows that BiOCl and BiOBr regenerated by photocatalysis can be reused effectively, which is beneficial for large-scale implementation.
Cycle Performance of BiOCl/BiOBr
The used BiOCl and BiOBr were regenerated by photocatalysis, then the adsorbents were used to perform adsorb experiments with PHL under the same circumstances after degradation of the lignin adsorbed on their surface. The cycle experiment was carried out under 6.0 wt% dosage BiOCl and BiOBr for 10 min; as shown in Figure 9a,b, the removals of lignin by the regenerated adsorbent maintained well. The lignin removal rate dropped to 35.7% and 33.3% at the third time of BiOCl and BiOBr, respectively, which might be caused by the loss of adsorbent during filtration and drying in the experiments. The tiny decrease was unavoidable, so the cycle of BiOCl and BiOBr remained a remarkable adsorptive property. The yields of CO and CO2 in three cycles by BiOCl and BiOBr are shown in Figure 9c,d. As can be seen, the yields of CO and CO2 decreased a little upon repeat experiments
Cycle Performance of BiOCl/BiOBr
The used BiOCl and BiOBr were regenerated by photocatalysis, then the adsorbents were used to perform adsorb experiments with PHL under the same circumstances after degradation of the lignin adsorbed on their surface. The cycle experiment was carried out under 6.0 wt% dosage BiOCl and BiOBr for 10 min; as shown in Figure 9a,b, the removals of lignin by the regenerated adsorbent maintained well. The lignin removal rate dropped to 35.7% and 33.3% at the third time of BiOCl and BiOBr, respectively, which might be caused by the loss of adsorbent during filtration and drying in the experiments. The tiny decrease was unavoidable, so the cycle of BiOCl and BiOBr remained a remarkable adsorptive property. The yields of CO and CO2 in three cycles by BiOCl and BiOBr are shown in Figure 9c,d. As can be seen, the yields of CO and CO2 decreased a little upon repeat experiments
Conclusions
In summary, BiOCl/BiOBr treatment for lignin removal from PHL by electrostatic adsorption was very rapid, and equilibrium could be achieved within 10 min. Under the room temperature and 6.0 wt% dosage of BiOCl and BiOBr, 36.3% and 33.9% lignin were removed, respectively, and the losses of HDS were both 0.1%. The adsorbed lignin could transform into CO, CO 2 and H 2 O, etc. By 15 h illumination, 50.1 µmoL CO and 556.0 µmoL CO 2 were obtained by BiOCl, and 38.9 µmoL CO and 487.3 µmoL CO 2 were obtained by BiOBr. The catalysts were regenerated by the photocatalysis, and both BiOCl and BiOBr can be recycled and keep extreme adsorption and photocatalytic properties. Compared with traditional treatments of PHL, this novel process is more economical and environmentally friendly. Therefore, the catalysts have potential application for lignin removal from prehydrolysis liquor. Data Availability Statement: This manuscript comprises an original, unpublished material, which is not under consideration for publication elsewhere, and all authors have read and approved the text and consent to its publication. All experimental data are accurate and reliable. | 8,878.8 | 2021-10-25T00:00:00.000 | [
"Environmental Science",
"Chemistry",
"Materials Science"
] |
Evolution of Regional Economic Spatial Structure Based on IoT and GIS Service
Unbalanced regional development is an inevitable trend in the development of all countries in the world. The rapid development of the Internet of Things (IoT) technology has created tools for the study of regional development issues. IoT has many advantages and thus owns a very wide range of applications. This paper makes use of geographic information system (GIS) technology, which can be viewed as one of the IoT sensing information. Changes in spatial regional economic di ff erences and space and the evolution of the structure are particularly examined by processing spatial information such as maps, analyzing phenomena and events that exist on the earth, and exploiting Kriging and inverse distance weighting (IDW). The numerical results in this paper justify that the introduction of GIS technology to the study of economic diversity can upgrade regional economic research from a traditional qualitative and statistical level to a quantitative and spatial visualization level.
1. Introduction 1.1. Background and Significance. With the strong promotion of information technology, the Internet of Things (IoT) has gradually been applied in many aspects all over the world. The scale of the IoT industry has also continued to expand, becoming a new strategic industry. The IoT is not only a representative of a new generation of information technology but also an important development direction of a new generation of information technology. The IoT has many advantages, has a very strong permeability, and has a very wide range of applications. There is a close relationship between the IoT and the regional economy, which can promote the development of the regional economy, promote the transformation of the regional economy, and accelerate the growth rate of the regional economy. Therefore, research on regional economic differences based on the IoT is of great significance. Regional economic difference refers to the imbalance of the overall level of economic development between regions in a certain period of time [1]. Due to the imbalance of resources and development levels in each city (province, state), economic development cannot reach the same level in the same period, so there are regional economic differences [2]. Regional economic disputes are a global law in the process of regional development and a key issue in regional economic research. The study of regional economic spatial differences and their causes is helpful to understand the status quo of regional economic development, promote the economic development of underdeveloped areas, and consolidate the economic achievements of developed areas [3].
The intervention of Electronic System Design Automatic-Geographical Information System (ESDA-GIS) analysis method covers the shortcomings of traditional economic difference analysis, so it is possible to study the economic spatial relationship between regional units [4]. In addition, when selecting economic indicators for European Food Safety Authority analysis, this article did not use per capita GDP indicators, but selected 13 indicators representing different levels of economic disparity for the analysis of key factors, and narrowed the scope of the 13 economic disparity indicators. Through the weighted calculation obtained by processing, the final total evaluation value of the principal component analysis is used as the county-level unit economic evaluation score of the year, as a measure of ESDA-GIS analysis, and used to analyze the subsequent county-level economic spatial correlation [5,6].
Due to the importance of national economic research, more and more research teams have devoted themselves to the research of the national economy and have achieved very good results. For example, Farah bakhsh conducted a detailed study on the evolution of the regional economic spatial structure through the GIS method and from this infers the future economic development trend, but because it did not integrate the global environment, the conclusion is inaccurate [7]; Chhetri economy can be used to analyze specific economic conditions, but it only represents the direction of big data, and it is still not applicable to some scenarios [8]. The accuracy of economic research is very difficult [9,10].
In a bid to improve the accuracy of the regional economic spatial structure, this paper makes use of GIS-based IoT service to study the geographical environment of the region. In particular, we employ interpolation and local fitting approaches to ensure the accuracy of local economic data. As a result, a detailed division of the regional economic spatial structure was finally drawn out through controlled experiments.
The rest of the paper is organized as follows. Section 2 presents the data processing and analyzing approaches used in this paper, including interpolation-based overall/local fitting, correlation analysis, and spacial clustering. Section 3 demonstrates the general numerical results on the evolution of the regional economic spatial structure, whereas Section 4 focuses on those specifically related to the GIS service. Finally, Section 4 concludes the paper.
GIS-Based Regional Economic
Spatial Structure
Interpolation Method Based on Overall Fitting
Technology. The entire placement technique, the placement model, is determined by all the characteristic observations of all sampling points in the target area [11]. The characteristic of this interpolation technique is that it cannot provide the local characteristics of the interpolation area, so the model is mainly used for large-scale changes [12,13]. What we usually call the surface stress analysis method is to approximate the general trend of the sampled data by selecting a binary function [14]. The general form of the binary function is FðX, YÞ is the actual observed data value, B Rs is the fitted value of the trend surface, and P is the trend surface; when p = 0, it is the horizontal plane: When (X, Y) changes in space, when p = 1, it is an inclined plane, and B 0 , B 1X , ⋯, B XY are the coefficients of the polynomial: When p = 2, it is a quadric surface: Independent variable X, Y, dependent variable Z. The binary function must satisfy the least square sum of the difference between the observed value and the fitted value: Multiple regression techniques can be used to determine the aforementioned types of coefficients [15]. The use of a binary function to process surface interference voltage has the following characteristic: when p > 3, custom surfaces usually produce abnormally large or small values.
The placement residual is an independent error of the normal distribution and has a certain correlation. Before being used for local interpolation, the macro abnormal sampling value must be processed in advance. Total spatial interpolation was performed according to the empirical formula of one or more spatial parameters. This empirical equation is called the transformation function [16,17], and it is also a common method of total interpolation technique. Since this article does not use placement technology, it will not be described in detail here.
Interpolation Method Based on Local Fitting Technology.
The real surface of continuous space is difficult to express with mathematical polynomials. Therefore, the local placement technique is usually used to match the value interpolation through local sampling points [18]. The positioning method only uses neighboring values to estimate the value of unknown points. Generally, these are the following steps: (1) specify the adjacent search area or range, (2) search for points located in the adjacent area, (3) choose mathematical functions that express the spatial variation of these finite points, and (4) assign values to data points belonging to ordinary grid cells. Reset the operation of this step until all points on the grid are mapped [19].
Spline interpolation is a process of obtaining a set of curve functions by mathematically solving three bending moment equations through a series of smooth curves of shape values [20]. And, when all p − 1 degree derivatives and adjacent blocks at the limit of n degree polynomial are continuous, it is called a spline function [13]. The principle of the mobile placement method is to match the surrounding data points by defining appropriate local functions and solving the assembly function to find the interpolation of unspecified points. This method uses an unspecified point as the interference center [21]. The Kriging (lattice) interpolation method, cofounded by French geologist Georges Matheron and South African mining engineer DG Krige in 1997, is a geostatistical method and the best linear unbiased interpolation estimator (referred to as BLUE) [22,23]. Kriging put it this way: Suppose that Z is a regionalized variable carried by a point and is 2nd order stationary (or intrinsic), zðX I ÞðI = 1, 2, ⋯, NÞ, point bearer X I ðI = 1, 2, ⋯, NÞ. Now, 2 Wireless Communications and Mobile Computing we need to estimate the regionalization variables of the X 0 point bearing location, and the estimated amount used is Choose λ I to make the estimate of z′ðxÞ unbiased, and make the variance smaller than the variance of any linear combination of observations. That is, it satisfies the best: the deviation of the difference between the interpolation value and the true value is the smallest [24]; that is, the deviation of the difference between the interpolation value and the true value is the smallest, namely, Linear. The interpolation value is a linear combination of observations, namely, Unbiased Estimation. The expected value of the difference between the interpolated value and the observed value is zero, namely, In the MAPGIS software, we have provided us with three variable function models, namely, the power exponential model, the linear model, and the spherical (MATLON) model [25].
(1) Linear model: (2) c is the polynomial coefficient, power exponent model: (3) Spherical model: The range of influence or fragmentation effect is represented by the variable a, and the critical change value is rep-resented by the variable C. Meteorologists and geologists proposed the inverse distance weighting method, which was eventually called the Shepard method [26]. The concept is to place n points, the plane coordinate ðX j , Y j Þ is the vertical height, N = 1, 2, ⋯, I, and the reciprocal distance weighted interpolation function is Among them, is the horizontal distance from point ðX, YÞ to point ðX j , Y j Þ, J = 1, 2, ⋯, N. P is a constant greater than 0, called the weighted power exponent.
The advantage of this method lies in the fact that the formula is relatively simple, especially suitable for scattered nodes, not a problem of grid points. Its disadvantage is that it can only get the maximum and minimum values of the function at the node, as interpolation takes the weighted average of the values at each node.
Correlation Analysis Method.
There are some connections between many phenomena in nature. The relationship between the above two or more random variables is determined based on mathematical statistics and is called approximate relationship or correlation. The analysis and determination of this relationship are called correlation analysis [27].
The main task of correlation analysis is to study the closeness of the relationship between variables and to draw conclusions about whether the population is relevant based on the data sample. If we can get any information about another variable from a known variable, then these two related variables are called "independent variables." The correlation between two variables may be due to various complicated reasons, or one variable affects another variable, or there is an interaction between two variables, or there is no direct relationship between two variables; all variables are also affected by the third variable. In short, the relationship between the two involves certainty and random fluctuations. In the correlation model, both variables are random variables [28]. According to the closeness of the relationship between variables, correlation types can be divided into three types. That is, complete correlation, zero correlation, and statistical correlation.
A complete correlation (functional relationship) is between two variables x and y. If there is a correspondingly defined value y for any given value x, then the relationship between the two variables is complete correlation. Zero correlation (no relationship) is when there is no relationship between two variables or the change of one phenomenon (variable) does not affect the change of another phenomenon (variable). This relationship is called zero correlation or no relationship. If the relationship between two variables 3 Wireless Communications and Mobile Computing is between complete correlation and zero correlation, it is called correlation or statistical correlation. When only studying the correlation between two variables, it is called simple correlation; when studying the correlation between three or more variables, it is called multiple correlation. In mathematical statistics, the parameters that determine the degree of a close correlation between variables mainly include covariance and correlation coefficient.
Spatial Clustering Method.
In actual work, we often encounter the problem of sampling samples (or marking), and classification research is the basic method of scientific research. In statistics, cluster analysis is usually used to classify categories. The principle is to first treat a certain number of samples or indicators as one category and then divide the two categories by the highest affinity according to the degree of relevance of the sample (or indicator) and then consider the degree of the combined category and the other categories before the combination intimacy between. Repeat this process until all samples (or tags) are combined into one category. The spatial grouping method is different from the traditional statistical grouping analysis. First, spatial grouping is mainly based on the spatial location of geographic phenomena and reference-related feature information for grouping analysis; second, the purpose of spatial grouping is to analyze the spatial aggregation of spatial objects and their division into different subcomponents, groups (clas-ses), and different subgroups (classes) occupying different spatial areas. The formation of subgroups is a product of the geographic environment. Based on this, certain geographic mechanisms can be revealed, and they can also be used as the basis for other analyses. Third, spatial cluster analysis is different from traditional cluster analysis. It is based on the spatial correlation of geographic variables, while spatial cluster analysis can be based on spatial autocorrelation [29].
The spatial grouping analysis method used in this document belongs to spatial statistics, and the starting point of spatial statistics is to consider that the specific geographic phenomenon or specific feature value in the peripheral unit is related to the same phenomenon or feature value in the area and adjacent area units. Spatial location produces two types of spatial effects: spatial dependence and spatial heterogeneity. The former is usually also called spatial autocorrelation or spatial correlation. Similar values in variables tend to appear in nearby locations, leading to spatial grouping. For example, some high crime areas in cities are usually surrounded by other high crime areas.
3. Numerical Results on the Evolution of Regional Economic Spatial Structure the data, for better and in-depth analysis of the data, recent research and analysis of spatial data are often conducted to obtain the value of spatial attributes and the spatial distribution of data. For data association, it is very useful to understand the particularity of the data, which lays the foundation for the final GIS spatial analysis research. Here, I use the corresponding SPSS statistical tools to explore and analyze the data. SPSS is the English abbreviation for Statistics Product and Service Solutions, which is a statistical software package for social sciences. It is one of the most famous statistical analysis software in the world. First, we investigate and analyze the distribution and eigenvalues of spatial data. Here, we use the analysis and description function in SPSS software. The description of the data is shown in Table 1 and Figure 1.
In addition to the variance and standard deviation in the table, other indicators are relatively easy to understand.
Among them, variance and standard deviation are descriptions of deviation trends. Deviation trend refers to the characteristics of the data set, which deviates from the central value of the distribution, reflecting the degree to which the value of each variable deviates from its central value. Through the comparative analysis of variance and standard deviation, if the variance and standard deviation of a given data set are the smallest, it means that the difference of the data set is the smallest, so the data set is more representative than some study prospects.
Experiments on the Spatial Structure and Characteristics
of the Urban System. The urban system refers to a group of interconnected and evenly spaced towns in a relatively integrated region or country. Its characteristics are different types and clear division of labor. Due to regional differences and different natural environments, each city Industrial added value as a proportion of GDP The importance of large industrial enterprises in the economy 6 The proportion of tertiary industry output value in GDP Service-led economic transformation degree 7 The proportion of the added value of the primary industry in the area of commonly used cultivated land Industrial productivity in the primary industry 8 Total retail sales of consumer goods per capita Realization of the purchasing power of social goods 9 General budget revenue of local finance Reflect the level of economic scale 10 Per capita net income of rural residents Income of rural residents 11 Per capita net income growth rate Income of urban residents
Wireless Communications and Mobile Computing
will develop in a direction conducive to its development. As time goes by, urban agglomerations will form a clear division of labor and interconnections, for example, traffic cities, tourist cities, historical and cultural cities, coal cities, and complete cities. It is variable. The urban system is not static after its formation. The scale, structure, and form of the urban system will change over time and change in government planning. You must have integrity. The bourgeois system is not a closed social and economic system. It refers to a unified whole composed of a series of cities of different scales, different functions, and interconnected cities, with the designated central city as the core of the defined area. The cycle and exchange of energy, matter, and information continue with each other. The bourgeoisie and the outside world continue to exchange and cooperate in the fields of politics, economy, culture, science and technology, trade, etc., in order to strengthen the external relations of the bourgeois system and its own rapid and healthy growth.
The spatial structure of the urban system refers to the spatial interaction of towns in a region, which merges the spatially separated towns into an organic whole with a spe-cific structure and function. Studies have shown that the spatial distribution of urban systems has obvious scale-free characteristics and has a random fractal structure within a certain range. There are three basic fractal dimensions to describe the spatial structure characteristics of the peripheral urban system: one is the fractal set dimension, which starts from the point density and describes the same characteristics of the urban peripheral spatial distribution; the other is the dimension fractal correlation from multiple starting with point density; it describes the relative distribution of system components; the third is the fractal network dimension, which starts directly from the data distribution and describes the spatial structure of the system.
With the rise of transportation and information technology, the spatial structure of today's global system is based on the connections of channels, nodes, levels, flow, terrain, and networks. Nodes and channels are the material basis of the space structure. Domain and network are the functional elements that form the spatial characteristics. It can be considered that the spatial structure of the urban system is composed of different levels of cities, spatial flows, passages, regions, and networks.
Evolution of Regional Economic Spatial
Structure Based on GIS Location Services 4.1. Interaction Between Cities in the Study Area. The law of universal gravitation is the law of gravitation that explains the interaction between objects. It is the law of mutual attraction between objects due to their mass. Since the urban system is located in a relatively integrated region or country with the central city as the core, it is composed of a series of cities with different scales, different functions, and close connections. At the same time, the urban system is complete, hierarchical, and dynamic. The internal relationship of the urban system is like every planet has satellite orbits, and central cities also have satellite cities, county seats, and other systems, forming a complete urban system. And there is a relationship between attraction and repulsion according to the universal law of gravitation. When a city is far away from the central city, the attractiveness and aversion of the central city are relatively weak, and the space for independent development is larger, even affected by the surrounding cities; it is also close to another system and attracted by culture and its policy. When the distance is relatively close, the impact will be greater and the resources will be relatively greater, so it will be developed and promoted in collaboration with other parts of its own system.
Wireless Communications and Mobile Computing
There is a relationship between gravity and repulsion between towns. Under the influence of distance, economy, and culture, when the gravitational and repulsive forces between the two cities are balanced, the cities will develop together; otherwise, they will restrict and hinder each other and affect the city. The development of the city will lead to the unity or diversification of the city. In the urban system, due to distance, cities outside the system may be greatly affected by other systems, and the development of cities will also tend to other systems. This article introduces the types of gravity in physics to analyze the strength of the interaction between cities. Take the straight-line distance between cities as the radius, and calculate the attractiveness between the two as a feature to characterize the intensity of interaction between cities. The spatial interaction force M is used to measure the strength of the interaction between cities, the city's population size, economic development level, and other related indicators which are used as characteristic values to measure the quality of the city, and the interaction between cities is studied and calculated.
Taking into account the availability of data, the population data of 110 cities in the Yangtze River Economic Zone in 2014 and the GDP of the same year were selected, and the direct distance data between cities was used for calculation. According to the method of calculating the urban interaction force, the value of the interaction force among 110 cities in the Yangtze River Economic Zone was calculated. The results show that in the city system of the Yangtze River Economic Belt, Shanghai has the strongest interaction with other cities, the closest connection, and the highest sum of interaction possibilities, which is 36967794, which is significantly higher than other cities. He established Shanghai in the urban residential area of the Yangtze River Economic Zone. In order to facilitate the study of the overall urban spatial interaction of the Yangtze River Economic Belt, Chongqing in the upper reaches of the Yangtze River, Wuhan in the middle reaches of the Yangtze River, and Shanghai in the lower reaches of the Yangtze River were selected and their power to interact with cities in each region. Table 2 and Figure 2 show the top ten cities ranked from high to low in terms of interaction intensity with the three outer cities.
Land Use Changes on Both
Sides of the Region. Based on the interpretation results of land use in different periods of the above two typical plots, the interpretation results are then counted to obtain the percentages of different land use landscape types in the two study plots.
It can be seen from Table 3 and Figure 3 that the percentage of land use types in different periods is different. Generally, coniferous forests and deciduous forests account for a larger percentage, while other types of land use account for a relatively small percentage. At the same time, the 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 Rate (%) Value
Time (month)
Annual gross production Growth rate Table 4 and Figure 4, it can be seen that forest land is the main land use landscape type on both sides of the boundary between A and B, and forest land accounts for the largest proportion, followed by cultivated land. This is mainly due to the rehabilitation of human beings, which makes the cultivated area continue to increase. The analysis of the land use landscape types on both sides of the border between A and B in different periods shows that in all four periods, except for side B, the largest percentage of woodland in 1976, both sides accounted for a larger proportion. Both A and B sides showed a downward trend year by year; except for the larger proportion of B side in 1976, the proportion of cultivated land in other periods was higher than that of A side. The results show that the artificial arable land of side A has been restored to a large extent, and the two sides of the boundary have been increasing year by year, and the proportion of other lands on side B is relatively high, and the two sides of the boundary have a trend of increasing year by year. Table 5 and Figure 5 show the annual GDP of Province A from 2015 to 2019. The graph shows that the GDP of Province A has increased year by year, from 2,299 billion in 2015 to 40,154 billion yuan in 2019. In terms of the growth rate, the average annual growth rate of GDP has decreased since 2010, indi-cating an overall negative trend. The GDP growth rate is increasing from a quantitative standpoint. In 2010, the economic growth rate was the fastest, reaching 11.9%, followed by the annual GDP growth rate of A. Prices rise slowly. It can be seen that in the process of rapid economic growth, Area A has encountered some bottlenecks. These bottlenecks hinder the rapid growth and slow down A region's economy. Combined with the actual situation of A, part of the reason may be due to the environmental problems of the earth. Obstacles to economic growth, especially the land problem of A, make sustainable economic growth possible. Therefore, in the process of economic development, it is necessary to combine the actual environment of A to coordinate growth. Table 6 and Figure 6 show the total investment in fixed assets of society A from 2015 to 2019. The picture shows that the fixed asset investment of society A has increased from 990.6 billion yuan in 2015 to 2,355.5 billion yuan in 2015 and 2019, and the investment amount has almost doubled. Such a huge change is also an important reason for A's economic growth. Although the number of fixed assets has accumulated in a large amount, it is not difficult to see from the quantity that the annual growth rate of fixed asset investment has been declining very slowly since 2017, and the downward trend is very obvious. The growth rate has begun to decline from 22.9% per year. A's investment share will fall into a "recession" every year, as shown in Table 7 and Figure 7. 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 Rate (%) Value
Conclusions
This paper mainly studies the evolution of regional economic spatial structure based on GIS positioning-related IoT services. In particular, we analyze the economic growth and investigate the evolution process of the regional economic space structure. The data set used in this study is largely derived from a brief period of time on the Chinese market. Additionally, we will collect additional data over a longer period of time around the world in order to build a more universal theoretical framework in the near future.
Data Availability
All data used are given in the paper.
Conflicts of Interest
The author declares that they have no conflicts of interest. | 6,480.6 | 2021-09-27T00:00:00.000 | [
"Geography",
"Economics",
"Computer Science"
] |
Case study of residual stresses distribution in steel welded parts using ultrasound
Residual stresses occur in every welded manufactured structure. Different studies aimed to classify the methods of investigating the residual stresses and to highlight the advantages and disadvantages/limits of each of them. The conclusion is that ultrasonic investigation permits the evaluation of bulk stress state of welded components with acceptable accuracy. Our past research has addressed various aspects of ultrasonic investigation of residual stresses in welded steel parts: different cases of the analysis of residual stresses, edge effect on analyses of residual stresses. The purpose of this paper is to continue the previous research by introducing a new case: comparative study of equal size samples obtained by splitting an initial sample. The probe was conceived as a frame with a detachable side. After welding, the assembly is processed on a grinding machine to ensure parallel surfaces. The sample, as obtained, is ultrasonic investigated as part. After that, the detachable side is removed and the resulted part is split in six equal samples. Every so obtained sample was investigated and results are presented. New direction for future research is proposed.
Introduction
Residual stresses occur in every welded manufactured structure. Many investigations have been carried out to study this phenomenon. Different studies aimed to classify the methods of investigating the residual stresses and to highlight the advantages and disadvantages/limits of each of them [1][2][3][4][5][6][7][8][9][10].
Our past research has addressed various aspects of ultrasonic investigation of residual stresses in welded steel parts: different cases of the analysis of residual stresses, edge effect on analyses of residual stresses [11,12]. Observations made during experiments opened up new investigative directions. By studying, for example, edge effect we used samples of different thicknesses and we noticed an interesting distribution of stresses in the samples obtained by splitting an initial sample.
This paper aims to investigate the mode of distribution of residual stresses in identical samples obtained by splitting, perpendicular to the welding line, of an initially welded sample.
Theoretical backgrounds
There are several methods for the analysis of the phenomena and relationships arising from residual stresses and propagation of ultrasounds in metals. Our previous research [11,12], presented the theoretical principles used for experimentation. A synthesis is presented below. The approach is based on identifying a relationship between Young's modulus and velocity of a longitudinal wave.
Longitudinal wave velocity is given by Eq. 1 [1]: where: -Density of the material, [kg/m 3 ]; -Velocity of a longitudinal wave in an isotropic solid, [m/s]; , -Lamé moduli of an isotropic solid. According with [2] between Lamé moduli and Young's modulus there are the relationship presented by equation 2 and equation 3: where: -Young modulus; -Poisson ratio. (3) Using into equation 1 the values of Lamé moduli from equations 2 and 3 results: In Eq. 5 is explicit relation between Young's modulus, sound velocity, density of the material and Poisson ratio. (5) Relation between stress, strain and Young's modulus is assumed to be governed by Hooke's law.
Method
Experiments were conducted using a previews equipment and methodology [11,12] and a short description is presented in the following. The probe was conceived as a frame with a detachable side (see Figure 1). After welding, the assembly is processed on a grinding machine to ensure parallel surfaces. The sample, as obtained, is ultrasonic investigated as Part P (see Figure 7). After that, the detachable side is removed (see Figure 3) and the resulted part is splitting in six equal samples denominate as P1, P2, P3, P4, P5, P6 (see Figures 2 and 4). Every sample P1-P6 was processed on a grinding machine to ensure parallel surfaces (in same manner as the part P). Each of the P1-6 samples was marked with divisions to locate the measurement points (see # 5).
The sample material is USt.37-2 [ For each sample to be brought to the initial request position a screw extension was used (see Figures 6 and 8). The points of interest are marked as series of points A1, A2, A3, A4, A5, A6, A7; B1, B2, B3, B4, B5, B6, B7; C1, C2, C3, C4, C5, C6, C7 (see figure 9) and was established in our previous research [11,12] . These points were investigated for every sample (P, P1, P2, P3, P4, P5 and P6) ultrasonic in the state charged of the sample (with constriction), condition obtained by using the extension screw (as shown in the figure 6) and preserved by fastening on the base piece (as shown in the figure 8).
Results and discussion
As mentioned above the points of interest marked as series of points A1, A2, A3, A4, A5, A6, A7; B1, B2, B3, B4, B5, B6, B7; C1, C2, C3, C4, C5, C6, C7 were investigated for every samples P1-P6 in state charged (with constriction). For every point several measurements (minimum five) were done. To assessing whether one piece of experimental data from a set of observations, is likely to be spurious Chauvenet's criterion was applied. The results of the investigation are presented in tables 1-7, figures 10-12 and refer to points of interest A1, A2, A3, A4, A5, A6, A7 B1, B2, B3, B4, B5, B6, B7, C1, C2, C3, C4, C5, C6, C7 for each of the P, P1, P2, P3, P4, P5, P6 samples. On the tables below Δσ A , Δσ B , Δσ C represent stresses variation related to standard sample for the measuring points. The state of stresses in the P-piece is caused by the simultaneous action of forces F1 and F2 (see Figure 1). For the whole piece P the effects of the forces F1 and F2 determine the stress state shown in Table 1 and represented by line P on the graphs in Figures 10-12. It can be seen that by splitting the P-piece into six equal parts the stress distribution in the six P1-P6 pieces is different from the P-piece. For parts P1, P2, P3, P5 there is a decrease in the longitudinal wave speed, so a relaxation of the material by stretching.
It is also noted for the P4, P6 parts an increase in longitudinal wave speed, as compared to the P piece, which can be interpreted as a compression.
There are also local points P1 (C3, C4, C5), P2 (C3, C5), P5 (A5, B5, C4, C5) for which the speed is very different from the other points of the sample. These particular observed values may be linked with weld or fusion zone and heat-affected zone.
The distribution of most of the points for samples P1, P2, P3 and P4, P6 on either side of the sample P can be interpreted in relation to the direction in which the welding lines were made (see Figure 1).
It can be hypothesized that an influence on the distribution of the stresses in the measured points on each sample P1, P2, P3, P4, P5, P6 has the split order and the slot size.
Conclusions
By splitting the P-piece into six equal parts the stress distribution in the six P1-P6 pieces is different from the P-piece. For samples P1, P2, P3, P5 there is a decrease in the longitudinal wave speed, so a relaxation of the material by stretching. It is noted for the P4, P6 parts an increase in longitudinal wave speed, as compared to the P piece, which can be interpreted as a compression.
Weld or fusion zone and heat-affected zone have a particular and differentiated influence on the state of the stresses at the neighboring points. An example of this is the distribution of the stresses in the points P1 (C3, C4, C5), P2 (C3, C5), P5 (A5, B5, C4, C5).
The direction in which the welding layers are performed has an important influence. For the experiments the welding layers were performed in the direction of the arrow (see figure 1). It can be said that the sense of welding was from P6 to P1. It can be noticed that after splitting, the P1, P2, P3 pieces (located at the end of the welding cord) relaxed while the P4, P6 pieces (located towards the start of the welding cord) contracted.
The stress variation in the measured points, in absolute values, is between 0 and 132 N / mm2. Similar variations are identified in similar works in the literature [5]. Case studies are, however, particular, and the establishment of similarities is not always easy.
The behavior of the P5 piece rather similar to P1, P2, P3 and not similar to P4, P6 as expected would lead to the hypothesis that the order of splitting the original piece in the P1-P6 parts is important.
It can also be suspected as part of stress distribution in parts after splitting has the size of the cutting slit.
Taking into account the aforementioned, new research may be continued by taking into account the order of the splitting into parts, the parameters of the welding regime, and the characteristics of the heat-affected zone. | 2,084.6 | 2017-01-01T00:00:00.000 | [
"Engineering",
"Materials Science",
"Physics"
] |
Depth dose calculations
In this study, calorimetric method is used to derive the depth doses calculation. In the experiment at the KeV level, an Ir-192 source with three different gamma energies was used. The source activity is approximately 9.15 curie. The Ir-192 radioactive source is mainly used for industrial purposes. This study had been made to determine organ dose values in any industrial accident. The data are calculated as surface dose and depth dose. Values for different time intervals have been measured. Time and depth parameters for measurements and calculations were obtained for 30 different values. The results of the experiment and the calculations have been compared.
Introduction
Industrial radiography is a frequently used method of non-destructive testing for many industrial products.Industrial radiography is a routine operation requiring high radiation levels and relying primarily on human diligence in observing safety procedures to prevent accidents [1].
Ir-192 source is mostly used in industrial radiography.Ir-192 is normally produced by neutron activation of natural-abundance iridium metal, usually in nuclear reactors.The strength (or specific activity) of a resulting Ir-192 is related to the amount of neutron irradiation and length of time to which the naturalabundance iridium metal is exposed.During irradiation only the stable isotope Ir-191 is activated to produce Ir-192 by absorbing a neutron.Radioactive Ir-192 has a half-life of 73.83 days.Radioactive Ir-192 is used principally for non-destructive testing (NDT) and, to a lesser extent, as a radio-tracer in the oil industry.Industrial gamma radiography involves the testing and grading of welds on pressurized piping, pressure vessels, high-capacity storage containers, pipelines, and certain structural welds.Other tested materials include concrete (locating rebar or conduit within the concrete), machined parts, plate metal, and pipe wall.Gamma radiography is also used to identify flaws in metal castings and welded joints, as well as to indicate structural anomalies due to corrosion or mechanical damage [2].
Literature studies have shown that industrial radiography also has various accidents.Workers working with radioactive material in these incidents are exposed to radiation.Experiments are carried out using the calorimetric method to detect these radiation doses.In addition, organ doses taken by people who have been caught in the accident can be calculated by this method.For this purpose, depth dose calculations were made using water phantom and a farmer dosimeter.In this study, Ir-192 was used as a radioactive source.In this study, experiments were carried out in TAEK Cekmece Nuclear Research Centre Health Physics Department SSDL (Secondary Standard Dosimetry Laboratory).
Material and methods
In the field of ionizing radiations, the absorbed dose is one of the basic quantities commonly used in dosimetry.The main reason of that it is closely related to the biological effects of radiation.Water is of special interest as an absorbing material because it is very similar to human tissue.One of the duties of the national standards laboratories is to establish accurate standards for absorbed dose in water.To this end, much effort has been put into the development of various experimental approaches.One of the most frequently used ones is the calorimetric method because its basic principle is simple [3].
Therefore, the measurements made here were obtained in detail by this method.In these measurements, the values of pressure and temperature were recorded regularly and added to the calculations as correction parameters [1].
Depth dose calculations
The aim of the present work was to measure the depth dose rates by using a water phantom and 192Ir source to obtain experimental and theoretical results.The absorbed dose (D d ) at depths from 0 to 35 cm in tissue with the source in surface contact was calculated by; where, An equal the source activity in unit of curie (A is the exposure rate at 1 cm), Γ is the exposure rate constant, 0.96 is the conversion factor to rads/Roentgen, d is the tissue depth in unit of cm, 0.24 is the estimated encapsulation thickness in unit of cm, f(d) is the scatter and attenuation polynomial of Shalek and Stovall [4] for d ≤ 15 cm or exp(-0.035d)for d ≥ 15 cm.Since there is a likelihood that the source may have been at a distance of up to 2 cm from the surface, depth doses were also calculated for these circumstances using standard depth-dose improvement factors with greater distances from the surface.The results are shown in Fig. 3.
Experimental Arrangement
A schematic view of the experimental arrangement is given in Fig. 1.The Ir-192 source is 2 cm diameter and 0.56 cm in length.On the day of the experiment the source activity is 9.15 Ci.The half-life of the source is 74.2 days.Ir-192 is a meta-stable isotope of Iridium emitting gamma radiation with a low level of Linear Energy Transfer (LET) and penetration in human body.This isotope frequently has been used for industrial radiography [5,6].
The presence of impurities, especially those with long half-lives and high radiation energies, can significantly alter the absorbed doses.Its energy spectrum is extremely complex, because it has got 30 gammas and 40 X-rays [7].However, for an unshielded source, Table 1 shows the most important energies and frequencies of the emitted gamma and beta particles which have been taken from data compiled by Delacroix et al. [8].The average energies have been calculated and then added to Table 1.The distance of the source from the reference plane of measurement is 1 m.In this plane, the beam size is 10 cm×10 cm; the photon flux at the cross-section border is 50% of that on the beam axis.The water phantom consists of a cubic perspex tank, of side 30 cm, with walls 14 mm thick.The wall facing the beam has a thickness of 4.0 mm (0. 476 g.cm -2 ) over a section of 15 cm×15 cm to reduce the correction due to the non-equivalence of the perspex front face with water.The phantom, filled with demineralized water, is positioned in such a way that the reference plane is at a depth equivalent to 5g.cm-2 in water [9].
The ionization chamber, inserted into a perspex support (1.85 mm thick), can be moved inside the water along the beam axis by means of a translation table fixed on the phantom walls.This allows an accurate adjustment of the chamber position
Fig.2. Experimental Arrengement
To avoid long-term variation of the ionization current due to the possible influence of humidity on the ionization chamber, the phantom is filled up and drained every day.The temperature of the chamber is measured with a calibrated thermistor.To match the air conditions of the chamber cavity, the thermistor is placed in air inside a Perspex tube dipped into water.The temperature difference between the positions of the thermistor and of the chamber has been checked and found to be small (less than 0.01°C), so the error introduced is negligible [1].
A Baldwin-Farmer 0.6 cc ionization chamber with a ±%3 calibration accuracy (traceable to the National Bureau of standards) was exposed, in air, at a distance of 14 cm from the source.The positioning error was estimated to be less than ± 0.5 cm.This measurement indicated an average exposure rate on 15 June 2016, of 224.06 R/h at distances of 14 cm.The standard deviation of eight such measurements was less than %3.This corresponds to a source activity of 9.15 ± 0.65 Ci (0.33 ± 0.02 TBq), with the uncertainty reflecting possible positioning errors.This activity determination was based on an exposure rate constant of 4.8 R-cm 2 /mCi-h as reported in NCRP (National Council on Radiation Protection and Measurements) Pamplet 40 [7,10].Although estimates of the exposure-rate constant for 192 Ir range from 4.3 to 5.0 R-cm 2 /MCi-h, the value of 4.8 was selected to be consistent with the exposure and dose calculations of this paper which are based on the actual exposure measurements and are independent of the exact source activity [7,1].
Experimental and its calculation
Firstly, perspex tank has been filled with demineralised water. 192Ir source has been placed at a distance of 3 cm from the tank.Ionization chamber has been placed at a distance of 2.5 cm.Then, radiation count per minute has been carried out at this position.The exposure unit is generally expressed in unit of Ront/h.However, this expression can be transformed into rad/h by multiplying with a factor of 0,96 rad/Ront as given below: Using the proper conversion factors an expression for absorbed dose in the unit of rad/h is found as In this equation; calibration factor of ionization chamber (CP) and temperature-pressure factor (TP as also calculated from equation [11]
Results and discussion
The purpose of this study was to determine the depth dose.The distance between the radioactive source water phantom is 100 to 200 cm.The depth has been changed from 20 to 50 cm.When the results aren't obtained graphically, the depth-dependent dose change values change exponentially.However, at some points deviations have been observed due to back scattering.When the results are carefully analysed, the changes obtained when the gamma rays emitted from the radioactive source interact with the substance overlap with the classical approach.
Gamma rays interfere with the matter, resulting in photoelectric effect, Compton scattering and pair production (Fig. 4).Fig. 3 is similar to Fig. 5 when viewed.At low energies, the photoelectric effect and the Compton scattering dominate, while the high energies suppress pair production.(Fig. 5) The linear absorption coefficient for water varies between 0,167 and 0, 0706 for gamma rays between 100 KeV and 1000 KeV.
Conclusions
Gamma rays interaction with matter is important from the perspective of shielding against their effect on biological matter.They are considered as ionizing radiation whose scattering by electrons and nuclei leads to the creation of a radiation field containing negative electrons and positive ions.The main modes of interaction of gamma rays with matter are the photo effect both in its photoelectric and photonuclear forms, Compton scattering and electron positron pair production.To a minor extent, photo-fission, Rayleigh scattering and Thomson scattering also occur.Gamma rays have both material attenuation (effect of shielding) and geometric attenuation when interacting with matter.There is also the effect of scattering, which is found in a shielding material and called build up effect.The graph obtained for these reasons is similar to the graph of the interaction of gamma rays with matter.
Fig. 1 .
Fig.1. 192Ir Decay Scheme below) were 1.004 and 1laboratory temperature on the day of the experiment was taken as 23.5 o C and pressure as 1008.4mb.Furthermore, normal temperature and air pressure on sea level were 20 o C and 1013 mb, respectively.Then the water phantom was irradiated for 20 minutes with a change between SSDL = 100 and 200 cm and d = 20-50 cm.The results are shown in Fig. 3.
Fig. 4 .Fig. 5 .
Fig.4.The relative importance of various processes of gamma radiation interactions with matter
Table 1 .
[9]ma and beta energy distribution for an unshielded 192 I source[9] | 2,482.2 | 2017-09-01T00:00:00.000 | [
"Physics",
"Medicine"
] |
Effects of nano particles on antigen-related airway inflammation in mice
Background Particulate matter (PM) can exacerbate allergic airway diseases. Although health effects of PM with a diameter of less than 100 nm have been focused, few studies have elucidated the correlation between the sizes of particles and aggravation of allergic diseases. We investigated the effects of nano particles with a diameter of 14 nm or 56 nm on antigen-related airway inflammation. Methods ICR mice were divided into six experimental groups. Vehicle, two sizes of carbon nano particles, ovalbumin (OVA), and OVA + nano particles were administered intratracheally. Cellular profile of bronchoalveolar lavage (BAL) fluid, lung histology, expression of cytokines, chemokines, and 8-hydroxy-2'-deoxyguanosine (8-OHdG), and immunoglobulin production were studied. Results Nano particles with a diameter of 14 nm or 56 nm aggravated antigen-related airway inflammation characterized by infiltration of eosinophils, neutrophils, and mononuclear cells, and by an increase in the number of goblet cells in the bronchial epithelium. Nano particles with antigen increased protein levels of interleukin (IL)-5, IL-6, and IL-13, eotaxin, macrophage chemoattractant protein (MCP)-1, and regulated on activation and normal T cells expressed and secreted (RANTES) in the lung as compared with antigen alone. The formation of 8-OHdG, a proper marker of oxidative stress, was moderately induced by nano particles or antigen alone, and was markedly enhanced by antigen plus nano particles as compared with nano particles or antigen alone. The aggravation was more prominent with 14 nm of nano particles than with 56 nm of particles in overall trend. Particles with a diameter of 14 nm exhibited adjuvant activity for total IgE and antigen-specific IgG1 and IgE. Conclusion Nano particles can aggravate antigen-related airway inflammation and immunoglobulin production, which is more prominent with smaller particles. The enhancement may be mediated, at least partly, by the increased local expression of IL-5 and eotaxin, and also by the modulated expression of IL-13, RANTES, MCP-1, and IL-6.
Introduction
Previous epidemiological studies have indicated that long-term exposure to ambient particulate matter (PM) is linked to increases in mortality and morbidity related to respiratory diseases [1,2]. The concentration of PM of mass median aerodynamic diameter (a density-dependent unit of measure used to describe the diameter of the particle) < or 10 µm (PM10) is related to daily hospital admissions for asthma, acute and chronic bronchiolitis, and lower respiratory tract infections [3]. PM of mass median aerodynamic diameter < or 2.5 µm (PM2.5) are more closely associated with both acute and chronic respiratory effects and subsequent mortality than PM10 [4]. Our laboratory has researched health effects of diesel exhaust particles (DEP), main constituents of PM2.5 in urban areas, especially in vivo. We have reported that DEP exacerbate allergic asthma [5] and acute lung injury related to bacterial infection in murine models [6].
Recently, nano particles, particles less than 0.1 µm in mass median aerodynamic diameter, have been implicated to affect cardiopulmonary systems [4,7]. Indeed, two in vivo studies have demonstrated that nano particles induce prominent airway inflammation as compared with larger particles [8,9]. Nano particles which have a larger surface area than the particles with larger size are able to penetrate deeply into the respiratory tract and cause a greater inflammatory response [10,11].
Bronchial asthma has been recognized as chronic airway inflammation that is characterized by an increase in the number of activated lymphocytes and eosinophils. A number of studies have shown that various particles including carbon black (CB) can enhance allergic sensitization [12][13][14]. CB has been demonstrated to enhance proliferation of antibody forming cells and both IgE and IgG levels [15,16]. Ultrafine particles (PM and CB) reportedly exaggerate allergic airway inflammation in vivo [17,18]. However, all the studies have not described the size of particles they used. Therefore, no research has been addressed the size effects of particles or nano particles on allergic airway inflammation in vivo.
The aim of the present study was to elucidate the effects of two sizes of carbon nano particles (14 nm or 56 nm) on allergic airway inflammation, local expression of cytokines, chemokines, and 8-hydroxy-2'-deoxyguanosine , and production of total IgE and antigenspecific IgG 1 , IgG 2a , and IgE.
Animals
Male ICR mice 6 to 7 wk of age and weighing 29 to 33 g (Japan Clea Co., Tokyo, Japan) were used in all experiments. They were fed a commercial diet (Japan Clea Co.) and given water ad libitum. Mice were housed in an animal facility that was maintained at 24 to 26°C with 55 to 75% humidity and a 12-h light/dark cycle.
Study protocol
Mice were divided into six experimental groups (Fig. 1). The vehicle group received phosphate-buffered saline (PBS) at pH 7.4 (Nissui Pharmaceutical Co., Tokyo, Japan) containing 0.05% Tween 80 (Nakalai Tesque, Kyoto, Japan) once a week for 6 wk. The ovalbumin (OVA) group received 1 µg of OVA (Sigma Chemical, St. Louis, MO) dissolved in the same vehicle every 2 wk for 6 wk. The nano particle groups received 50 µg of nano particles (14 nm: PrinteX 90 or 56 nm: PrinteX 25, degussa, Dusseldorf, Germany) suspended in the same vehicle every week for 6 wk. The OVA + nano particle groups received the combined treatment in the same protocol as the OVA and the nano particle groups, respectively. The surface area of the 14 nm nano particles was 300 m 2 /g and that of 56 nm nano particles was 45 m 2 /g. The size of each particle was quantified by JEM-2010 transmission electron microscope (TEM; JEOL, Tokyo, Japan). Nano particles were autoclaved at 250°C for 2 h before use. The suspension was sonicated for 3 min using an Ultrasonic disrupter (UD-201; Tomy Seiko, Tokyo, Japan). In each group, vehicle, OVA, nano particles, or OVA + nano particles was dissolved in 0.1 ml aliquots, and inoculated by the intratracheal route through a polyethylene tube under anesthesia with 4% halothane (Hoechst, Japan, Tokyo, Japan). The animals were studied 24 h after the last intratracheal administration, with lung histology, bronchoalveolar lavage (BAL), protein levels of cytokines and chemokines in the lung tissue supernatants, immunohistochemistry for 8-OHdG, and with Igs. The studies adhered to the National Institutes of Health guidelines for the experimental use of animals. All animal studies were approved by the Institutional Review Board. Figure 1 Study Protocol.
Blood retrieval and analysis
Mice were anesthetized with diethyl ether. The chest and abdominal walls were opened, and blood was retrieved by cardiac puncture. Serum was prepared and frozen at -80°C until assayed for total IgE and antigen-specific IgG 1 , IgG 2a , and IgE.
Histologic evaluation
After exsanguinations, the lungs were fixed by intratracheal instillation with 10% neutral phosphate-buffered formalin at a pressure of 20 cm H 2 O for at least 72 h. Slices 2 to 3 mm thick of all pulmonary lobes were embedded in paraffin. Sections 3 µm thick were stained with Diff-Quik (International Reagents Co., Kobe, Japan) or periodic acid-Schiff (PAS) and examined by two of us (HT and KI) in a blind fashion.
Morphometric analysis for numbers of eosinophils, neutrophils, mononuclear cells, and goblet cells around the airways
Sections were stained with Diff-Quik to quantitate the numbers of infiltrated eosinophils, neutrophils, and mononuclear cells. The length of the basement membrane of the airways was measured by videomicrometer (Olympus, Tokyo, Japan) in each sample slide. The number of eosinophils, neutrophils, and mononuclear cells around the airways were counted with a micrometer under oil immersion. Results were expressed as the number of inflammatory cells per millimeter of basement membrane as described previously [5].
To quantitate goblet cells, sections were stained with PAS. The number of goblet cells in the bronchial epithelium was counted by micrometer. Results were expressed as the number of goblet cells per millimeter of basement membrane as described previously [5].
BAL
The trachea was cannulated after the collection of blood. The lungs were lavaged with 1.2 ml of sterile saline at 37°C, instilled bilaterally by syringe. The lavage fluid was harvested by gentle aspiration. This procedure was conducted two more times. The average volume retrieved was 90 % of the 3.6 ml that was instilled; the amounts did not differ by treatment. The fluid collections were combined and cooled to 4°C. The lavage fluid was centrifuged at 300 g for 10 min, and the total cell count was determined on a fresh fluid specimen using a hemocytometer. Differential cell counts were assessed on cytologic preparations. Slides were prepared using an Autosmear (Sakura Seiki Co., Tokyo, Japan) and were stained with Diff-Quik (International reagents Co.). A total of 500 cells were counted under oil immersion microscopy.
Immunohistochemistry
The production of 8-OHdG in the lung was detected by immunohistochemical analysis (n = 8 in each group) using anti-8-OHdG polyclonal antibody (Japan Institute for the Control of Aging, Shizuoka, Japan) as described previously [19,20]. Deparaffinized slides were blocked with 10% goat serum for 1 h. After blocking, anti-8-OHdG antibody (0.5 µg/ml) was incubated with the sections for 1 h at room temperature, followed by the incubation of a biotinylated secondary antibody and streptavidin-peroxidase conjugate. Then, the slides were incubated with 3-amino, 9-ethyl-carbazole chromogen, and counterstained with hematoxylin in AutoProbe III kit (Biomeda, Foster City, CA, USA). For each of the lung specimens, the extent and intensity of staining with anti-8-OHdG antibodies were graded on a scale of 0-4+ by two blinded observers on two separate occasions using coded slides as previously described [21]. A 4+ grade implies maximally intense staining, whereas 0 implies no staining.
Antigen-specific IgG determination
Antigen-specific IgG 1 or IgG 2a antibodies were measured by ELISA with solid-phase antigen [5,22]. In brief, microplate wells (Dynatech, Chantilly, VA) were coated with OVA overnight at 4°C and then incubated at room temperature for 1 h with PBS containing 1% bovine serum albumin (BSA; Sigma) containing 0.01% thimerosal (Nakalai Tesque). After washing, diluted samples were introduced to the microplate and incubated at room temperature for 1 h. After another washing, the wells were incubated at room temperature for 1 h with biotinylated rabbit anti-mouse IgG 1 or IgG 2a (Zymed Laboratories, San Francisco, CA). After yet another washing, the wells were incubated with horseradish-peroxidase-conjugated streptavidin (Sigma) at room temperature for 1 h. The wells were then washed and incubated with o-phenylenediamine and H 2 O 2 in dark at room temperature for 30 min. The enzyme reaction was stopped with 4 N H 2 SO 4 . Absorbance was read at 492 nm. Each plate incubated a previously screened standard plasma that contained a high titer of anti-OVA antibodies. The results were expressed in titers, calculated based on the titers of the standard plasma. Cut off values for antibody-positive plasma were set to hold as the mean value of absorbance of preimmune plasma.
Total IgE and antigen-specific IgE determination
Antigen-specific IgE antibody was measured by IgE-capture ELISA [5,22]. In brief, microplate wells were coated with a rat anti-mouse IgE monoclonal antibody (Yamasa Syoyu Co., Chiba, Japan) at 37°C for 3 h and then incubated at 37°C for 1 h with 1% BSA-PBS and 0.01% thimerosal. After washing with PBS containing 0.05% Tween 20 (PBST; Nacalai Tesque), diluted samples were introduced to the microplate and incubated overnight at 4°C. After washing with PBST, biotinylated OVA was added to each well and incubated for 1 h at room temperature with β-Dgalactosidase-conjugated streptavidin (Zymed). After the final washing, the wells were incubated with 4-methylumbelliferyl-β-galactoside (Sigma) as the enzyme substrate at 37°C for 2 h. The enzyme reaction was stopped with 0.1 M glycine-NaOH (pH, 10.3). The fluorescene intensity was read by a microplate fluorescene reader (Fluoroskan Flow Laboratories, Costa Mesa, CA). Each plate included a previously screened standard plasma that contained a high titer of anti-OVA antibodies. The results were expressed in titers, calculated based on the titers of the standard plasma. Cut off values for antibody-positive plasma were set two hold as mean fluorescene units of preimmune plasma. Total IgE was measured by capture ELISA in a manner similar to the detection of antigen-specific IgE. A biotinylated rat anti-mouse IgE (BD Biosciences Pharmingen, San Diego, CA) was used to detect captured IgE in place of biotinylated OVA. A 450 readings of the samples were converted to nanograms per milliliter using a standard curve generated with double dilutions of mouse IgE κ isotype standard (BD Biosciences Pharmingen).
Statistical analysis
Data were reported as mean ± SEM. Differences in the numbers of infiltrated inflammatory cells and goblet cells, cytokine protein levels, and immunogloblin concentrations and titers between groups were determined using analysis of variance (Stat view version 4.0; Abacus Concepts, Inc., Berkeley, CA) as described previously [5]. If differences between groups were significant (P < 0.05), Fisher's protected least significant difference test was used to distinguish between pairs of groups.
Effects of nano particles on antigen-related airway inflammation
To evaluate the effect of nano particles on antigen-related airway inflammation, we investigated the cellular profile of BAL fluid and lung histology.
The numbers of total cells and macrophages were significantly greater in the nano particle, OVA, and OVA + nano particle groups than in the vehicle group (P < 0.01: Table 1). Furthermore, the numbers were significantly greater in the OVA + 14 nm nano particle group than in the OVA group or the 14 nm particle group (P < 0.01: Table 1).
Although the numbers were greater in the OVA + 56 nm nano particle group than in the OVA group or the 56 nm nano particle group, the difference did not achieve significance. OVA challenge increased the number of eosinophils as compared with vehicle challenge without significance. The numbesr of eosinophils were greater in the OVA + nano particle groups than in the vehicle group (P < 0.05 for OVA + 14 nm nano particle, N. S. for OVA + 56 nm nano particle). The number was significantly greater in the OVA + 14 nm nano particle group than in the OVA group (P < 0.01) or 14 nm nano particle group (P < 0.05). The number was also greater in the OVA + 56 nm nano particle group than in the OVA group or 56 nm nano particle group, but the difference did not achieve significance. Challenge with nano particles significantly elevated the numbers of neutrophils as compared with vehicle challenge (P < 0.01 for 14 nm, P < 0.05 for 56 nm). OVA also elevated the number without significance as compared with vehicle challenge. The number was significantly greater in the OVA + 14 nm nano particle group than in the 14 nm nano particle group (P < 0.05) or the OVA group (P < 0.01). The number was also greater in the OVA + 56 nm nano particle group than in the nano particle group or the OVA group, the difference did not reach significance. Challenge with nano particles elevated the numbers of mononuclear cells as compared with vehicle challenge (P < 0.05 for 14 nm, N. S. for 56 nm). OVA also elevated the number of mononuclear cells without significance as compared with vehicle challenge. The number was significantly greater in the OVA + 14 nm nano particle group than in the vehicle (P < 0.01) or the OVA group (P < 0.05). The number was greater in the OVA + 56 nm nano particle group than in the OVA group, but difference did not achieve significance. There were no significant differ-ences between the nano particle groups and OVA + nano particle groups.
The magnitude and cellular profiles of airway inflammation were also evaluated in lung specimens stained with Diff-Quik. Intratracheal instillation of nano particles provided diffuse deposition of the particles into the bilateral lungs, including the bronchi and alveolar spaces. The particles were occasionally present within the subepithelial neutrophils and alveolar macrophages. The combined instillation of OVA + 14 nm nano particles for 6 wk led to a marked infiltration of eosinophils and mononuclear cells around the bronchi and bronchioles. OVA + 56 nm nano particles also induced severe airway inflammation, but the severity was less than that of OVA + 14 nm nano particles. Either OVA or nano particles alone resulted in slight recruitment of eosinophils and neutrophils. Vehicle administration caused little infiltration of inflammatory cells.
To quantitate the infiltration of inflammatory cells around the airways, we expressed the number of these cells per length of basement membrane of the airways ( Table 2). The number of eosinophils was greater in the OVA group than in the vehicle group without significance. The number of eosinophils was significantly greater in the OVA + 14 nm nano particle group than in the vehicle, the 14 nm nano particle, or the OVA group (P < 0.01). The number was greater also in the OVA + 56 nm nano particle group than in the OVA group or the 56 nm nano particle group, but the difference did not achieve significance.
OVA increased the number of neutrophils as compared with vehicle challenge without significance. In the presence of OVA, nano particles with a diameter of 14 nm significantly increased the number as compared with vehicle or OVA challenge (P < 0.01 for vehicle, P < 0.05 for OVA)). In the presence of OVA, nano particles with a diameter of 56 nm increased the number as compared with vehicle (P < 0.05) or OVA (N. S.). Challenge with nano particles increased the numbers as compared with vehicle challenge (P < 0.01 for 14 nm nano particle, N. S. for 56 nm nano particle). There were no significant differences between the OVA + nano particle groups and the nano particle groups. The number of mononuclear cells was significantly greater in the OVA group than in the vehicle group (P < 0.05). 14 nm nano particles significantly increased the number (P < 0.05 versus vehicle). The number was significantly greater in the OVA + 14 nm group than in the OVA group or the 14 nm nano particle group (P < 0.01). The number was also greater in the OVA + 56 nm nano particle group than in the 56 nm nano particle group (P < 0.05) or the OVA group, but the difference did not reach statistical significance.
Nano particles increase goblet cells after antigen challenge
To evaluate airway epithelial injury and hypersecretion of mucus, lung sections were stained with PAS (Table 2). OVA plus 56 nm nano particles increased the number of goblet cells as compared with vehicle without significance. The number was significantly greater in the OVA + 14 nm nano particle group than in the vehicle (P < 0.01), the OVA (P < 0.05), or the 14 nm nano particle group (P < 0.01). The number was greater also in the OVA + 56 nm nano particle group than in the vehicle (P < 0.05), the OVA (N. S.), or the 56 nm nano particle group (P < 0.05).
Effects of nano particles on local expression of Th2 cytokines in the presence of antigen
To explore the role of local expression of Th2 cytokines in the effects of nano particles on antigen-related airway inflammation, we quantitated protein levels of IL-5, IL-4, and IL-13 in the lung tissue supernatants (Table 3). OVA challenge increased the level of IL-5 as compared with vehicle challenge without significance. In the presence of OVA, nano particles significantly elevated levels of IL-5 as compared with vehicle (P < 0.01) or OVA (P < 0.05 for 56 nm, P < 0.01 for 14 nm). The levels were significantly greater in the OVA + nano particle groups than in the nano particle groups (P < 0.01). The levels of IL-13 were significantly greater in the OVA + 14 nm nano particle group than in the OVA group or 14 nm nano particle group (P < 0.01). The levels were greater also in the OVA + 56 nm nano particle group than in the OVA group (N. S.) or the 56 nm nano particle group (P < 0.05). The level of IL-4 was significantly lower in the OVA + 56 nm nano particle group than in the OVAgroup (P < 0.05). There were no other significant differences among the experimental groups.
Effects of nano particles on local expression of eotaxin, MCP-1, RANTES, and IL-6 in the presence of antigen
To investigate the local expression of eotaxin, MCP-1, RANTES, and IL-6, we measured protein levels of these cytokine and chemokines in the lung tissue supernatants (Table 4). OVA challenge increased the levels of eotaxin without significance as compared with vehicle challenge. The levels were significantly greater in the OVA + nano particle groups than in the vehicle (P < 0.01), the nano particle group (P < 0.05 for OVA + 56 nm nano particle, P < 0.01 for OVA + 14 nm nano particle), or the OVA (P < 0.05 for OVA + 56 nm nano particle, P < 0.01 for OVA + 14 nm nano particle) group. Nano particle challenge increased the levels of MCP-1 as compared to vehicle challenge (P < 0.01 for 14 nm, N. S. for 56 nm). OVA challenge slightly increased the levels without significance as compared with vehicle challenge. Nano particles combined with OVA enhanced the level as compared with nano particle alone (P < 0.01 for 14 nm nano particle, N. S. for 56 nm nano particle) or OVA alone (P < 0.01 for OVA + 14 nm nano particle group, P < 0.05 for OVA + 56 Six groups were intratracheally inoculated with vehicle, nano particles, OVA, or the combination of OVA and nano particles for 6 wk. Lungs were removed and frozen 24 h after the last intratracheal administration. Protein levels in the lung tissue supernatants were analyzed using ELISA. Results are shown as mean ± SEM. *P < 0.05 versus vehicle, **P < 0.01 versus vehicle, # P < 0.05 versus OVA, ## P < 0.01 versus OVA. $ P < 0.05 versus nano particles. $ P < 0.01 versus nano particles. nm nano particle group). The levels of RANTES were significantly greater in the OVA + nano particle groups than in the vehicle group (P < 0.01 for OVA + 14 nm nano particle, P < 0.05 for OVA + 56 nm nano particle), or the OVA group (P < 0.01 for OVA + 14 nm nano particle, P < 0.05 for OVA + 56 nm nano particle), or the nano particle groups (P < 0.01 for 14 nm, N. S. for 56 nm). The levels of IL-6 were significantly greater in the OVA + 14 nm and OVA + 56 nm nano particle groups than in the vehicle group (P < 0.01), the OVA group (P < 0.01 for OVA + 14 nm nano particle group, P < 0.05 for OVA + 56 nm nano particle group), or the nano particle groups (P < 0.01).
Effects of nano particles on 8-OHdG formations in the presence or absence of antigen
We next studied 8-OHdG formation generated from deoxyguanosine in DNA by oxidative stress in the lung. In the vehicle group, nuclear staining with 8-OHdG was barely detectable ( Fig. 2A). Nano particles or OVA challenge induced moderate staining with 8-OHdG (Fig. 2B, C, D). On the other hand, OVA plus nano particles resulted in intense immunoreactive 8-OHdG staining as compared to OVA or nano particles alone (Fig. 2E, F). The intensity and the extent of the immunoreactivity were more prominent in the OVA + 14 nm nano particle group (Fig. 1E) than in the OVA + 56 nm nano particle group (Fig. 2F). As typically shown in the OVA + nano particle groups, we found the expression of 8-OHdG in macrophages phagocyting nano particles as well as polymorphonuclear leukocytes (Fig. 2E, F).
Effects of nano particles on adjuvant activity for total IgE and antigen-specific production of IgG and IgE
To exanime whether nano particles have adjuvant activity for total IgE and antigen-specific Ig production, we measured total IgE and antigen-specific IgG 1 , IgG 2a , and IgE (Table 5). Total IgE levels were significantly greater in the OVA + nano particle groups than in the vehicle group (P < 0.01 for OVA + 14 nm nano particle group, P < 0.05 for OVA + 56 nm nano particle group). The levels were also significantly greater in the OVA + 14 nm nano particle group than in the OVA or the 14 nm nano particle group (P < 0.05). The antigen-specific IgG 1 was significantly greater in the OVA + 14 nm nano particle group than in the other groups (P < 0.01 versus each other group). The antigen-specific IgG 2a was not significantly different among the experimental groups. The combination of OVA plus 14 nm nano particles significantly increased antigenspecific production of IgE as compared with vehicle or OVA alone (P < 0.05).
Discussion
The present study demonstrated that nano particles administered by the intratracheal route enhanced airway inflammation associated with antigen challenge in mice. The inflammatory component was characterized by increased numbers of eosinophils, neutrophils, and mononuclear cells. Recruitment of these cells was accompanied by an increment in goblet cells in the bronchial epithelium. The airway inflammation induced by the combined administration of nano particles with antigen modulated local expression of IL-5, eotaxin, IL-13, Six groups were intratracheally inoculated with vehicle, nano particles, OVA, or the combination of OVA and nano particles for 6 wk. Lungs were removed and frozen 24 h after the last intratracheal administration. Protein levels in the lung tissue supernatants were analyzed using ELISA. Results are shown as mean ± SEM. *P < 0.05 versus vehicle, **P < 0.01 versus vehicle, # P < 0.05 versus OVA, ## P < 0.01 versus OVA. $ P < 0.05 versus nano particles. $$ P < 0.01 versus nano particles.
RANTES, MCP-1, and IL-6. The formation of 8-OHdG was moderately induced by nano particles or antigen alone, and was further enhanced by antigen plus nano particles as compared with nano particles or antigen alone. The enhancing effects were more prominent with 14 nm nano particles than with 56 nm nano particles. Furthermore, 14 nm nano particles enhanced total IgE and antigen-specific production of IgG 1 and IgE.
DEP exacerbate allergic diseases including allergic asthma [5]. Elementary carbon, which is mainly involved in the nuclei of DEP, can enhance allergic sensitization. We used CB in the present study, since CB is a useful prototypical particle for the research on the effects and their mechanisms of PM including DEP. Because CB is relatively inert, the effects of particle size can be elucidated without confounding factors [23]. Al-Humadi and coworkers have Immunohistological staining for 8-hydroxy-2'-deoxyguanosine in the lung obtained from (A) vehicle group, (B) 14 nm nano particle group, (C) 56 nm nano particle group, (D) OVA group, (E) OVA + 14 nm nano particle group, and (F) OVA + 56 nm nano particle group (n = 8 in each group) Figure 2 Immunohistological staining for 8-hydroxy-2'-deoxyguanosine in the lung obtained from (A) vehicle group, (B) 14 nm nano particle group, (C) 56 nm nano particle group, (D) OVA group, (E) OVA + 14 nm nano particle group, and (F) OVA + 56 nm nano particle group (n = 8 in each group). Lungs were removed twenty-four h after the last intratracheal instillation. Arrows denote positive staining. Original magnification × 300. demonstrated that CB exacerbates airway inflammation related to antigen in rats [18]. Last and colleagues have demonstrated that ambient particles with a diameter of less than 2.5 µm partially exacerbated lung inflammation related to antigen [17]. However, the comparative study focusing on the effects of particle size on antigen-related airway inflammation has never been conducted in vivo. In the present study, nano particles aggravated antigenrelated airway inflammation, which was confirmed by the counts of inflammatory leukocytes in BAL fluid and by the histological assesment. Furthermore, we showed that nano particles exaggerated goblet cell hyperplasia elicited by antigen. In overall trends, the enhancing effects were more prominent with 14 nm nano particles than with 56 nm nano particles. Furthermore, 14 nm nano particles had obvious adjuvant activity for the antigen-specific production of IgG 1 and IgE. These results clearly indicate that nano particles can aggravate antigen-related airway inflammation in vivo. Also, the effects are greater with smaller particles than with larger particles. We have previously examined the effects of DEP on allergic airway inflammation using 100 µg of DEP in vivo [5,24,25]. Based on the previous studies from our laboratory, we chose the dosage of 50 µg/body of nano particles, which can be considered to be involved in 100 µg of DEP as elementary carbon. Indeed, the enhancing effects of 14 nm nano particles on the airway inflammation and cytokine expressions are comparable to those of DEP in the previous study [5]. Another important point in this study is the surface area of the nano particles used. Surface area of particles exposed reportedly correlates magunitude of airway inflammation [26]. In our study, the surface area of the 14 nm nano particles was 6.7 fold larger than that of 56 nm nano particles (300 m 2 /g versus 45 m 2 /g). Nano particles with larger surface area are likely to attach more immunoregulative molecules than those with smaller surface area. As a result, smaller nano particles (14 nm) may lead to more prominent aggravation of antigen-related airway inflammation than larger nano particles (56 nm) in the present study. We did not examine the effects of the nano particles with the same particle number in the present study. However, the number of smaller nano particles is larger than that of larger nano particles when the particles make the same weight. Alternatively, our study has demonstrated not only the size effects of nano particles, but also the effects of their surface area and/or the effects of their number on the antigen-related airway inflammation. Future independent studies uniforming the surface area or particle number will provide better widestanding for the effects of the nano particles on antigen-related airway inflammation.
Allergic asthma is often associated with activation of IL-5 gene cluster, a pattern compatible with predominant activation of Th2-like T-lymphocyte population. IL-5 is essen-tial for maturation of eosinophils in the bone marrow and their release into the blood [27,28]. Also, these Th2 cytokines are implicated in the pathogenesis of allergic reactions via their roles in mediating IgG 1 and IgE production, and in differentiation, vascular adhesion, recruitment, activation, and survival of eosinophils. In our study, airway inflammation induced by the combined administration of nano particles and antigen were concomitant with the increased protein levels of IL-5. These results provide the first evidence that nano particles can accelerate antigen-related IL-5 expression and subsequent eosinophilic inflammation.
IL-13 is also recognized to regulate eosinophilic inflammation, and mucus secretion [29]. On the other hand, IL-6 is believed to participate in airway remodeling [30]. In the present study, nano particles enhanced the expression of the proteins in the presence of antigen. Therefore, nano particles may aggravate mucus hypersecretion and airway remodeling, at least partly, through the enhanced expression of IL-13 and IL-6. In fact, the OVA + nano particle groups showed enhancement in the mucus hypersecretion as compared with the OVA group or the nano particle groups.
Among chemokines, eotaxin is essential for eosinophil recruitment in antigen-related airway inflammation [31,32]. RANTES is a strong chemotactic and activating factor for eosinophils, and can modulate eosinophil adhesion [33,34]. In fact, our previous studies have confirmed that the exaggerated allergic airway inflammation induced by DEP paralleled the local elevation of the inflammatory proteins [5,22]. In the present study, nano particles enhanced the expression of these proteins in the presence of antigen as compared with antigen alone. The results suggest that nano particles aggravate allergic airway inflammation, at least in part, via the enhancement of the local expression of these proteins.
Interestingly, in our study, nano particles challenge increased the lung level of MCP-1 as compared to vehicle challenge. Also, the levels were significantly greater in the OVA + nano particle groups than in the vehicle or the OVA group. MCP-1 is a CC chemokine, and is chemoattractant for monocytes [34]. It also has a chemoattractant effect of CD4 + and CD8 + T lymphocytes [35]. MCP-1 also plays a role in recruitment of eosinophils to acute and chronic inflammatory sites [36]. Furthermore, some particles such as silica [37] and amosite asbestos [38] reportedly can induce MCP-1 in vitro and in vivo. Thus, our findings indicate that pulmonary exposure to carbon nano particles may induce MCP-1 expression in the airways. In addition, the aggravating effects of nano particles on antigen-related airway inflammation should be mediated, at least in part, via the enhanced expression of this chemokine.
In overall trends, the enhancing effects of nano particles on local expression of cytokines and chemokines related to antigen challenge were more prominent with 14 nm nano particles than with 56 nm nano particles. The differences in the enhanced expression of the proteins between the two sizes of nano particles may contribute, at least partly, to the differences in the magnitude of antigenrelated airway inflammation and goblet cell hyperplasia.
Redox imbalance is a critical factor for tissue injury in various pulmonary diseases including inflammation such as asthma [39,40]. Airway macrophages from individuals with asthma produce more reactive oxidative species (ROS) than those from control subjects [41]. On the other hand, nano particles have been implicated to induce and/ or enhance oxidative stress [42]. Furthermore, a recent study has demonstrated that the same type of nano particles as we used can induce ROS in the alveolar macrophages [43]. We, therefore, evaluated the contribution of oxidative stress to the deterious effects of nano particles on antigen-related airway inflammation. 8-OHdG is a proper marker of the oxidative stress. In our study, immunoreactivity of 8-OHdG in the lung was more intense in the OVA + nano particle groups than in the nano particle groups or the OVA group. Further, enhanced immunoreactivity for 8-OHdG was detected in alveolar macrophages as well as polymorphonuclear leukocytes. It is suggested that the enhanced oxidative stress is involved, at least partly, in the aggravation of antigen-related airway inflammation caused by nano particles.
Antigen-specific IgE is thought to contribute to inflammatory cell accumulation after antigen challenge via degranulation of mast cells [44]. On the other hand, IgG with antigen is a strong agonist for eosinophil degradation in vitro [45]. Furthermore, late asthmatic reactions are associated with IgG antibody [46]. Previous studies have reported that DEP with antigen demonstrate adjuvant activity for IgE production in vivo [47] and that those without antigen induce nonspecific IgE response in humans [48,49]. In addition, we have reported that DEP enhance antigen-specific production of IgE and IgG induced by intratracheal challenge with antigen in vivo [5,25]. In the present study, the combined intratracheal administration of 14 nm nano particles and antigen showed a significant greater increase in total IgE and antigen-specific IgG 1 and IgE than the other administration. The enhancement in the antigen-specific immunogloblin production by 14 nm nano particles can induce the enhanced release of a variety of inflammatory mediators such as histamine and leukotrienes, resulting in aggravated manifestations of allergic asthma.
Finally, in the real world, we inhale nano particles and antigen in ambient air, not particle suspension nor aliquot of antigen. The dose of nano particles injected in the present study can be estimated to be less than a hundred fold than that we inhale in daily life. Further, real PM including DEP are complex mixture of carbon, metals, and organics, which are different from CB used in the present study. Thus, it remains to be elucidated in future whether daily inhalation of nano particles with or without other compounds including organic chemicals than elementary carbon combined with occasional exposure of aerosol antigen lead to the same results as the present study.
Conclusion
The present study has shown evidence that nano particles can aggravate antigen-related airway inflammation. The effect may be mediated, at least partly, through the increased local expression of IL-5 and eotaxin and also by the modulated expression of IL-13, MCP-1, IL-6, and RANTES. Furthermore, 14 nm nano particles enhance total IgE and antigen-specific production of IgG 1 and IgE. These results suggest that nano particles can be a risk for exacerbation of allergic asthma. The aggravating effect may be larger with the smaller particles.
Publish with Bio Med Central and every scientist can read your work free of charge | 8,451.8 | 2005-09-16T00:00:00.000 | [
"Environmental Science",
"Materials Science",
"Medicine"
] |
Fuzzy radial basis function network for fuzzy regression with fuzzy input and fuzzy output
In this study, fuzzy regression (FR) models with fuzzy inputs and outputs are discussed. Some of the FR methods based on linear programming and fuzzy least squares in the literature are explained. Within this study, we propose a Fuzzy Radial Basis Function (FRBF) Network to obtain the estimations for FR model in the case that inputs and outputs are symmetric/nonsymmetric triangular fuzzy numbers. Proposed FRBF Network approach is a fuzzification of the inputs, outputs and weights of traditional RBF Network and it can be used as an alternative to FR methods. The FRBF Network approach is constructed on the basis of minimizing the square of the total difference between observed and estimated outputs. A simple training algorithm from the cost function of the FRBF Network through Backpropagation algorithm is developed in this study. The advantage of our proposed approach is its simplicity and easy computation as well as its performance. To compare the performance of the proposed method with those given in the literature, three numerical examples are presented.
Introduction
Regression analysis is one of the most widely used methods of estimation and it is applied to determine the functional relationship between independent and dependent variables. Fuzzy regression (FR) is a fuzzy type of classical regression in which some elements of the model are represented by any type of fuzzy numbers [35].
Fuzzy linear regression (FLR) first proposed by Tanaka et al. [46] is used to minimize the total spread of the fuzzy parameters subject to the support of the estimated values cover the support of the observed values for a certain α-level. In the light of Tanaka et al.'s [46] study, several methods have been developed for FR models. Another approach to FLR method is proposed by Diamond [16] to determine the fuzzy parameters in analog to conventional normal equations derived with a suitable metric. In general, there are two main approaches in FR analysis: linear programmingbased methods and FLS-based methods. The first one is based on minimizing fuzziness as an optimal criterion [4][5][6]8,20,33,[36][37][38][40][41][42]45,47], whereas the second one is based on least squares (LS) of errors as a fitting criterion [3,9,15,16,[25][26][27]31,48].
There are many studies in the literature related to FR since then proposed by Tanaka et al. [46]. Bardossy [5] developed a general form of regression equations for the fuzzy numbers and formulated the FR problem as a mathematical programming. Bardossy et al. [6] introduced a general methodology for FR and applied to an actual hydrological case study including the imprecise relationship between soil electrical resistivity and hydraulic permeability. Sakawa and Yano [40] developed LP-based methods for solving formulated three types of problems for obtaining the FLR models, where both input and output data are fuzzy numbers. Sakawa and Yano [41] introduced three types of multiobjective programming (MOP) problems for obtaining FLR models with fuzzy input and fuzzy output data. They developed an LP-based interactive decision making method to derive the satisfying solution of the decision maker for the MOP problems. Ming et al. [31] described a model for LS fitting of fuzzy input and fuzzy output data. Kao and Chyu [26] introduced the method of LS under fuzzy environment to handle fuzzy observations in regression analysis for three cases: crisp input-fuzzy output, fuzzy input-fuzzy output, and non-triangular fuzzy observations. Yang and Lin [48] proposed two estimation methods along with an FLS approach for considered FLR models with fuzzy inputs, fuzzy outputs and fuzzy parameters. Hojati et al. [20] proposed a simple goal programming-like approach for computation of FR for two cases: crisp inputs-fuzzy outputs and fuzzy inputs-fuzzy outputs. Chen and Dang [10] proposed a three-phase method to construct the FR model with variable spreads to resolve the problem of increasing spreads. Lu and Wang [30] proposed an enhanced fuzzy linear regression model (FLR FS ). Shakouri and Nadimi [43] introduced an approach to find the parameters of an FLR with crisp inputs and fuzzy outputs. Khan and Valeo [27] introduced a method, which is an extension of the Diamond's [16] FLS method, for FLR with fuzzy regressors, regressand and coefficients.
Many Neural Networks (NN) models are similar or identical to well-known statistical techniques such as linear regression, polynomial regression, nonparametric regression, discriminant analysis, principal components analysis and cluster analysis. Radial Basis Function Network (RBFN) is a special kind of NNs that consists of input layers, only one hidden layer and output layers. It has radial basis functions in hidden units and linear functions in output units, with adjustable weights. In recent years, various fuzzified versions of the NNs and the RBF Network have been developed for linear, nonlinear and nonparametric regression models.
NNs models have been applied in the FR analysis by various researchers. For example, Ishibuchi and Tanaka [23] introduced simple and powerful methods for FR analysis using NNs. Ishibuchi et al. [24] proposed an architecture of Fuzzy Neural Networks (FNN) that have crisp inputs, interval weights and interval outputs for FR analysis. Ishibuchi et al. [21] introduced an architecture of FNN with triangular fuzzy weights. Ishibuchi and Nii [22] proposed nonlinear fuzzy regression methods based on FNN with asymmetric fuzzy weights. Cheng and Lee [11] proposed FRBF Network that weights between input-hidden units and outputs considered as fuzzy numbers, but inputs and weights between hidden-output units considered as crisp numbers for FR analysis. Dunyak and Wunsch [17] described a method for nonlinear FR using NN models. Khashei et al. [28] proposed a hybrid method that yields more accurate results with incomplete data sets based on the basic concepts of NN and FR models to overcome the limitations in both methods. Mosleh et al. [35] presented a novel hybrid method based on FNN for approximate fuzzy parameters of fuzzy linear and nonlinear regression models with crisp inputs and fuzzy output. Cobaner et al. [14] proposed an adaptive neuro-fuzzy approach to estimate suspended sediment concentration on rivers. The potential of neuro-fuzzy technique is compared with Generalized Regression Neural Networks (GRNN), Radial Basis Function Neural Networks (RBFNN) and Multi-layer Perceptron (MLP) and also two different sediment rating curves (SRC). Haddadnia et al. [18] presented a fuzzy hybrid learning algorithm for the RBFNN. Roh et al. [39] presented a Fuzzy RBFNN based on the concept of information ambiguity. Hathaway et al. [19] presented a model that integrates three data types of numbers, intervals and linguistic assessment. Staiano et al. [44] described a novel approach to fuzzy clustering as a summation of a number of linear local regression models. Their approach is more effective in the training of RBFNN leading to improved performance with respect to other clustering algorithms. Alvisi and Franchini [2] proposed an approach under uncertainty using NN for water level (or discharge) forecasting. The parameters of the NN, i.e., the weights and biases, are represented by fuzzy numbers. Mitra and Basak [32] proposed a fuzzy version of the RBF Network.
To the best knowledge of the authors, there is no study on FRBF Network dealing with fuzzy regression with fuzzy input and fuzzy output. Therefore, we propose FRBF Network with fuzzy input, fuzzy output and also fuzzy weights, as an alternative to the existing FR methods in the literature. To show its appropriateness and effectiveness, our proposed method is applied to the three numerical examples and its performance is compared with existing FR methods. The results indicate that our proposed method is an effective method to estimate the output under fuzzy environment.
The remainder of the paper is organized as follows: in Sect. 2, fuzzy regression methods in the literature are reviewed. Our proposed Fuzzy Radial Basis Function Network approach is presented in Sect. 3. Three numerical examples are illustrated to compare the proposed approach with other FR methods given in Sect. 4. Finally, conclusions are drawn in Sect. 5.
Fuzzy regression methods
Fuzzy linear regression was first introduced by Tanaka et al. [46] and since then several different methods have been proposed for FR by various researchers. In general, fuzzy regression methods are divided into two categories: the first one is based on linear programming (LP) approach and the second one is based on the fuzzy least squares (FLS) approach. The first class which minimizes the total vagueness of the estimated values for the output includes Tanaka et al.'s [46] method and its extensions [20,33,40,45,46]. The sec-ond class includes FLS methods to minimize the total square of errors in the estimated values [15,16,31,48].
To determine the parameters of FR by minimizing the total square of errors in the estimated values, FLS and GFLS methods were proposed by Diamond [16] and Ming et al. [31], respectively. Fuzzy regression model for the methods of FLS and GFLS as considered as follows: where a 0 , a 1 ∈ are nonfuzzy parameters, X i , Y i ∈ E 1 are fuzzy numbers and E 1 is fuzzy number space.
are fuzzy outputs considered as triangular fuzzy numbers (TFNs). In fuzzy inputs, x i is the center, f i and f i are the left and right spread of X i , respectively. It is assumed that, The objective of the FLS and GFLS methods is defined as follows: In Eq. (2), two cases arise according to a 1 ≥ 0 or a 1 < 0. In case of a 1 ≥ 0, d(a 0 + a 1 X i , Y i ) 2 is given by; for FLS and GFLS, respectively. In Eqs. (3) and (4), the parameters a 0 and a 1 parameters are derived via ∂r ∂a 0 = 0 and ∂r ∂a 1 = 0 (for a 1 < 0; see [16,31]). Sakawa and Yano [40], and Hojati et al. [20] considered the following fuzzy regression model: where T and parameters A j = (a j , c j ) are considered as symmetric TFNs. Sakawa and Yano [40] formulated three types of problems for obtaining the FLR models with fuzzy input and fuzzy output using the three indices for equality between two fuzzy numbers as follows: Hojati et al. [20] proposed a goal programming-like approach which minimizes the total deviations of upper and lower points of α-certain predicted and associated observed intervals, for FLR model with fuzzy input and fuzzy output as follows: where ir L are deviation variables, "l" and "r " refer to the left (lower) and right (upper) points of the input intervals, "U " and "L" refer to the upper and lower points of the observed and predicted intervals, respectively (for details, see [20,40]).
Yang and Lin [48] proposed alternative FLS methods called as Approximate-distance fuzzy least squares (ADFLS) and Interval-distance fuzzy least squares (IDFLS), for FLR model with fuzzy input and fuzzy output as follows: where In the ADFLS method, the objective function is defined as follows: The objective function J (A 0 , A 1 , . . . , A k ) is minimized over A j subject toc j ≥ 0 and c j ≥ 0 for ADFLS method. In Eq. (9),m i ,l i ,r i , H 1 and H 2 are defined as follows: and In the IDFLS method, the objective function is defined as follows: The objective function ρ(A 0 , A 1 , . . . , A k ) is minimized over A j for IDFLS method (for details of ADFLS and IDFLS, see [48]).
Proposed approach
Radial Basis Function (RBF) Network is a special kind of NN which has input layers, a single hidden layer and output layers. The hidden layer contains hidden units, also called as radial basis function units, which have two parameters that describe the location of the function's center and its deviation (or width). Hidden units measure the distance between an input data and the functions's center. There are two sets of weights, one connecting the input layer to the hidden layer and the other connecting the hidden layer to the output layer. The weights between input and hidden layer which are also called as centers are determined by any clustering method, such as Fuzzy c-Means Clustering (FCM). The weights connecting the hidden layer to the output layer are used to form linear combinations of the hidden units for generating outputs of the RBF Network. RBF Network is trained by unsupervised learning or combining the supervised and unsupervised learning [12,13,50].
In this section, we propose a FRBF Network approach for FR model with fuzzy input and fuzzy output which are symmetric or nonsymmetric TFNs. Our proposed FRBF Network includes fuzzy input (X p ), fuzzy output (Y p ), fuzzy weights between input and hidden unit (W i j ) and also fuzzy weights between hidden and output unit (V j ). In this approach, the weights W i j and normalization factor σ 2 j are determined by unsupervised learning. W i j s are initialized by modified FCM algorithm given in Sect. 3.2 and V j s are randomly selected as TFNs. Then, W i j , V j and σ 2 j s are updated by BackPropagation (BP) algorithm which is supervised learning.
α-level sets of the fuzzy input X pi and the fuzzy out- , respectively. The weights between input and hidden units are symmetrical TFNs and denoted as where w L i j is the lower limit, w C i j is the center and w U i j is the upper limit of W i j . α-level sets of W i j are written as follows: The weights between hidden unit and output unit are TFNs and denoted as . α-level sets of V j can be written as same manner in W i j . Arithmetic operations on fuzzy numbers and intervals can be found in Alefeld and Mayer [1], Klir and Yuan [29] and Moore [34].
The hidden unit j is calculated as follows: Normalization factor of hidden unit j is determined as follows: Fuzzy estimated output for observation p of FRBF Network is calculated by; Let Y p be the fuzzy output corresponding to the fuzzy input X p . The cost function for the α-level sets of the fuzzy estimated outputŶ p and the corresponding fuzzy output Y p is introduced in Ishibuchi et al. [24] as follows: where, E L p,α and E U p,α indicate the squared errors for the lower limit and the upper limit of the α-level sets of E p , respectively. The total cost function E for the input-output pair (X p ,Y p ) is computed as follows:
Training algorithm of our proposed Fuzzy Radial Basis Function Network
Training algorithm of our proposed FRBF Network is constituted by Yapıcı Pehlivan [49]. In the algorithm, Choi Fig. 1.
The purpose of the proposed FRBF Network is to minimize total errors in estimations through the training algorithm. Let η be a learning constant, λ be a momentum constant and t indicates the number of iterations. The weights V j , W i j and normalization factor σ 2 j are updated by the training algorithm as follows: The fuzzy weights V j are updated by; If In Eqs. (17) and (18), v L j (t) and v U j (t) can be calculated using the cost function E p,α as follows: The derivatives in Eqs. (19) and (20) can be written as follows: The fuzzy weights W i j are updated by; If w L i j > w U i j then, In Eqs. (21) and (22), w L i j (t) and w U i j (t) can be computed using the cost function E p,α as follows: The derivatives in Eqs. (23) and (24) can be written as follows: can be computed in two ways as follows: The normalization factors σ 2 pj are updated by; where σ (t) pj can be calculated using the cost function E p,α as follows: The derivative where ζ L and ζ U can be computed in two ways as follows: From the above expressions, the training algorithm of the proposed FRBF Network can be summarized as follows: Step 1 Determine the fuzzy weights W i j using modified FCM algorithm given in Eqs. (27)- (29) Initialize the fuzzy weights V j as fuzzy numbers randomly Calculate the initial values of normalization factor by Eq. (13) Step 2 Repeat Step 3 for α 1 , α 2 , . . . , α s Step 3 Repeat the following procedures for p = 1, 2, . . . , n Step 3.1 h pj ,Ŷ p and E p,α are calculated by Eqs. (12)- (15) Step 3.2 Update the fuzzy weights V j by Eqs. (17)- (18) Step 3.3 Update the fuzzy weights W i j by Eqs. (21)- (22) Step 3.4 Update the normalization factors σ 2 pj by Eq. (25) Step 4 If the total number of iterations is satisfied, stop. Otherwise, go to Step 2.
Modified Fuzzy c-Means Clustering algorithm
The Fuzzy c-Means Clustering (FCM) algorithm is the most common cluster algorithm for RBF Network. It divides n data sets into c-fuzzy groups and estimates the cluster centers of each group [7,12].
In this study, we modified the FCM algorithm because of X i and W i j are fuzzy numbers. Modified FCM algorithm for our proposed FRBF Network is given as follows: Step 1 Set the number of clusters m and parameter b. Initialize cluster centers W i j and inputs X i for α = 0.
Step 2 Determine the membership values using W i j in two ways as; Step 3 Update the cluster centers W i j until the membership values are stabilized by;
Numerical examples
In this section, we considered three numerical examples to demonstrate the proposed FRBF Network approach that performs well while handling with FR model when input and outputs are triangular fuzzy numbers. Using these fuzzy data, we obtain an estimated fuzzy regression equationŶ = A 0 + A 1X with fuzzy parameters A 0 = (a 0 , c 0 ,c 0 ) and To compare the performance of the methods, we calculate the total errors in estimation using Eq. (2) for FLS and GFLS, Eq. (6) for SY, Eq. (7) for HBS, Eq. (9) for ADFLS and Eq. (10) for IDFLS methods.
Example 1 Sakawa and Yano [40] used an example to illustrate the regression model, in which input and outputs are symmetrical TFNs. The example has eight sets of the fuzzy observations (X i , Y i ) as shown in Table 1.
In the computations of the Example 1, we consider following specifications of our proposed FRBF Network approach for the training algorithm: To compare the performance of the seven FR methods in estimation given in Sect. 2, we applied to calculate the errors in estimating the observed outputs. Table 2 shows parameter estimations, predicted intervals of fuzzy outputs and sum of squares errors (SSE) in estimating the eight observations for these considered methods. In the methods of FLS, GFLS, SY, HBS, ADFLS, IDFLS and proposed FRBF Network approach, the results for α = 0 are used for comparison. In Table 2, SSE value of the FRBF Network approach is 9.9680, which is obviously better than FLS, GFLS, SY, HBS, ADFLS and IDFLS methods with 17.008, 22.162, 17.3682, 15.1991, 15.4723 and 10.3435 SSE values, respectively. Figure 2 illustrates the errors in estimations of FR methods and proposed FRBF Network approach. Example 2 Diamond [16] used an example to illustrate the regression model, in which inputs and outputs are nonsymmetrical TFNs. The example has eight sets of the fuzzy observations (X i , Y i ), see Table 3.
In the computations of the Example 2, we consider following specifications of our proposed FRBF Network approach for the training algorithm: Table 1 Fuzzy input-output data set from Sakawa and Yano [40] i Table 3 Fuzzy input-output data set from Diamond [16] i IDFLS and proposed FRBF Network approach, the results for α = 0 is used for comparison. In Table 4, SSE values of the IDFLS method is 1.4477 and FRBF Network approach is 1.5517, which are obviously better than FLS, GFLS and ADFLS methods with 2.4055, 3.0867 and 2.0843 SSE values, respectively. Figure 3 depicts the errors in estimations of FR methods and proposed FRBF Network approach.
Computational experience
The superiority of the proposed FRBF Network approach can be also observed through a test example from Diamond [16] and Ming et al. [31], in which inputs and outputs are symmetrical TFNs. This example has three sets of the fuzzy observations (X i , Y i ) as given in Table 5.
In the computations of the Example 3, we consider following specifications of our proposed FRBF Network approach for the training algorithm: 3.5904] which is calculated by the FCM method, and normalization factor as σ 2 1 = 1.614, σ 2 2 = 1.182 and fuzzy weights between hidden unit and output unit as To compare the performance of the seven FR methods in the estimation given in Sect. 2, we applied to calculate the errors in estimating the observed outputs. Table 6 shows parameter estimations, predicted intervals of fuzzy outputs and SSE values in estimating the eight observation for these considered methods. In the methods of FLS, GFLS, SY, HBS, ADFLS, IDFLS and proposed FRBF Network approach, Figure 4 shows the errors in estimations of FR methods and proposed FRBF Network approach. LINGO Software is used for solving the fuzzy regression methods. The training algorithm for the proposed FRBFN is coded in MATLAB Software and implemented on a Notebook (Intel Core 2 Duo) with CPU time of 2.0 GHz. The average relative performance of the proposed FRBF Network approach and other FR methods, measured by SSE values and CPU time, is shown in Table 7. Table 7 shows relative performance of the existing Fuzzy Regression methods and Fuzzy Radial Basis Function Network approach for Test Example from Diamond [16] and Ming et al. [31]. We compared the performance of considered methods with respect to the SSE values and CPU time. The SSE value of the proposed FRBF Network approach is 0.0770, whereas its CPU time is 233.626 s. As can be seen from Table 7, compared with FLS, GFLS, SY, HBS, ADFLS and IDFLS, the performance of FRBF Network approach improves substantially when the CPU time is increased. Although the CPU time of our proposed approach is more than the compared FR methods, SSE value of the estimations is obtained minimum than those. Because, it is expected to obtain the estimations with minimum SSE. It can be seen that our proposed approach gives better results than existing methods for FR models with fuzzy input and fuzzy output.
Conclusion
In this study, we have reviewed the relevant articles on Fuzzy Regression and provided an easily computation approach to estimate FR models with fuzzy input and fuzzy output. We presented a new estimation approach, Fuzzy Radial Basis Function Network, for Fuzzy Regression in the case that inputs and outputs are symmetric or nonsymmetric triangular fuzzy numbers. We derived a training algorithm of threelayer FRBF Network consisting of input, hidden and output layers. In the training algorithm, inputs, outputs and weights were defined by triangular fuzzy numbers. The construction of the algorithm is quite simple and the parameters of the FRBF Network, i.e., fuzzy weights and normalization factors, are systematically updated using this training algorithm given in Sect. 3.1. The effectiveness of the derived training algorithm is demonstrated by computation of three numerical examples performed for proposed FRBF Network approach using the Backpropagation algorithm. The examples show that our proposed approach performs better than the existing fuzzy regression methods based on Linear Programming and Fuzzy Least Squares.
This study is one of the approaches to derive training algorithm of FRBF Network approach which has fuzzy input, fuzzy output and fuzzy weights, as an alternative to FR methods in the literature. The advantage of this approach is its simplicity and easy computation as well as its performance, while its disadvantage is spending more time than the other FR methods. The proposed approach is more suitable than the existing FR methods: firstly, the proposed method is able to handle symmetric and nonsymmetric triangular fuzzy inputs and outputs. Secondly, Example 1 and Example 3 show that the FRBF Network approach is better than of the existing FR methods, in terms of the SSE values and predicted intervals in estimation.
As a conclusion, our proposed approach suggests an efficient alternative procedure to estimate predicted intervals for FR model with fuzzy input and output. As a limitation of our study, we only focused on fuzzy regression model in the case that input and output are assumed to be symmetric or nonsymmetric triangular fuzzy numbers. Therefore, we only considered FRBF Network when input, output and weights are triangular fuzzy numbers and we did not consider another types of fuzzy numbers in this study. Although the discussion of this study is confined to simple regression with one input and one output, it can be generalized to cope with cases of multiple inputs and outputs. For future studies, more general fuzzy inputs, outputs and weights such as trapezoidal fuzzy numbers could be handled with our FRBF Network approach and it could be applied to different FR models. | 5,889.8 | 2016-03-01T00:00:00.000 | [
"Computer Science",
"Mathematics",
"Engineering"
] |
Groundwater Level Complexity Analysis Based on Multifractal Characteristics: A Case Study in Baotu Spring Basin, China
Groundwater resources are important natural resources that must be appropriately managed. Because groundwater level �uctuation typically exhibits non-stationarity, revealing its complex characteristics is of scienti�c and practical signi�cance for understanding the response mechanism of the groundwater level to natural or human factors. Therefore, employing multifractal analysis to detect groundwater level variation irregularities is necessary. In this study, multifractal detrended �uctuation analysis (MF-DFA) was applied to study the multifractal characteristics of the groundwater level in the Baotu Spring Basin and further detect the complexity of groundwater level variation. The main results indicate that groundwater level variation in the Baotu Spring Basin exhibited multifractal characteristics, and multifractality originated from broad probability density function (PDF) and the long-range correlation of the hydrological series. The groundwater level �uctuations in wells 358 and 361 exhibited a high complexity, those in wells 287 and 268 were moderately complex, and the groundwater level �uctuations in wells 257 and 305 were characterized by a low complexity. The spatial variability of hydrogeological conditions resulted in spatial heterogeneity in the groundwater level complexity. This study could provide important reference value for the analysis of the nonlinear response mechanism of groundwater to its in�uencing factors and the development of hydrological models.
Introduction
Hydrological processes are nonlinear, complex, dynamic, and widely dispersed, making it necessary to assess the behaviour of hydrological processes at different scales (Li and Zhang 2007, Rakhshandehroo and Mehrab Amiri 2012, Shang and Kamae 2005).Due to the notable in uences of natural and human factors, hydrological time series exhibit highly complex characteristics (Ma et al. 2019).To reveal these complex characteristics and reasonably describe the complexity of groundwater level uctuation constitutes the basis for assessing the impact of natural factors and human activities on hydrological processes.This could provide a solid scienti c basis for hydrological prediction, construction of hydrological models and water resource management.
Hydrologists have long recognized the importance of studying the complexity and scale evolution of hydrological processes.Hydrological time series are typical fractal objects with nonlinear, chaotic and fractal characteristics (Liu et al. 2015).Therefore, fractal theory is widely used in the eld of hydrology (Bhuyan et al. 2009, Yuan et al. 2014).Fractal theory attempts to explain complex processes by determining simple underlying processes.The study of fractal theory began with Hurst's discovery of the long-range correlation in runoff records (Hurst 1951).Hurst rst used rescaled range (R/S) analysis to study the long-range correlation in time-series records of natural phenomena, and with the use of the calculated Hurst exponent, R/S analysis could provide a process for the quanti cation of the memory of time series (Mandelbrot and Wallis 1969).Subsequently, many other studies have reported similar longterm correlated uctuating behaviour in nature (Chakraborty and Chattopadhyay 2021, Koscielny-Bunde 1998, Koscielny-Bunde et al. 2006, Weron 2002).Currently, the concept of fractal theory has been extended and applied to many disciplines, such as medicine, nance, geophysics, hydrology, remote sensing and social sciences, in addition to other studies of nonlinear or stochastic processes (Ihlen 2012, Labat et al. 2011).In hydrology, fractal theory has been employed to explore the variability of rainfall processes (Sivakumar 2001), scaling behaviour of precipitation runoff and sediments (Wu et al. 2018), complexity of the rock distribution in rivers (Dwyer et al. 2021), long-range correlation of runoff (Zhao et al. 2017), and particle deposition during arti cial groundwater recharge (Wang et al. 2021).
Groundwater level uctuations comprise the dynamic response of a given groundwater system to recharge and discharge, which are in uenced by numerous factors.For example, natural factors, such as precipitation, evapotranspiration, seepage, soil moisture, and topography, and human factors, such as mining, irrigation, and construction of hydraulic projects (Li and Zhang 2007, Rakhshandehroo and Mehrab Amiri 2012), could vary on different spatial and temporal scales.Hydrogeological condition refers to conditions relating to groundwater formation, distribution and variation patterns, including groundwater recharge, burial, runoff, discharge, water quality and quantity.According to the concept of hydrogeological condition, groundwater recharge and discharge characteristics also belong to the category of hydrogeological condition.Therefore, it is because of the different hydrogeological conditions that there are different groundwater recharge-discharge laws, which lead to the varied characteristics of groundwater level uctuations.Moreover, due to the in uence of climatic conditions, the uctuations in groundwater level time series may exhibit periodicity and seasonal cycles.These uctuations are usually not consistent, and ordinary linear and deterministic models cannot be adopted to simulate these uctuations (Yu et al. 2016).
Non-stationary uctuation in the groundwater level could be described as fractional Brownian motion and could be largely long-range correlated rather than completely random (Rakhshandehroo and Mehrab Amiri 2012, Yu et al. 2016).This suggests that monofractals (homogeneous variable processes) cannot comprehensively describe uctuations in groundwater levels, and multifractal (complex nonlinear heterogeneous processes) analysis is required to detect irregularities in time series (Stanley 1999).
Multifractal detrended uctuation analysis (MF-DFA) is a common and effective tool to determine the multifractal behaviour and can be adopted to quantify the complexity of physical mechanisms (Ihlen 2012).MF-DFA has proven to be a valuable tool in the multifractal characteristics analysis of time series (Adarsh et Zhu et al. 2020).In previous decades, due to excessive groundwater exploitation, groundwater resources in the Baotu Spring Basin were seriously threatened, and even spring water was dry up.Reasonable protection of karst water resources in the Baotu Spring Basin is very important.At present, there is no research to analyze the multifractal characteristics and complexity of groundwater level in this area.The main objective of this study is to quantitatively calculate the complexity of the physical mechanisms that control the groundwater level through the multifractal method.The lower complexity is associated with the simpler physical mechanism, on the contrary, the groundwater level is more di cult to predict successfully.Therefore, the complexity of groundwater level can re ect the complexity of recharge and discharge in the groundwater system and the response sensitivity of the groundwater system to recharge and discharge.
Multifractal analysis is applied in this study to detect the multifractal behaviors and complexity in the time series of the groundwater level in the Baotu Spring Basin.Theoretically, this study is important for improved analysis of the nonlinear mechanism of the groundwater response to natural and anthropogenic factors.In addition, in practical engineering, this study can provide a scienti c basis for the development of hydrological models.In this study, rst, the detrended uctuation analysis (DFA) method is used to determine whether data processing is required before multifractal analysis.Multifractal analysis of the groundwater level in different aquifers in the study area is conducted via the MF-DFA method.The complexity of the physical mechanisms controlling the groundwater level can be assessed based on the complexity index (CI) (Lana et al. 2020), so the CI of the groundwater level is calculated.Finally, the source of the multifractality of groundwater level time series is examined.
Study Site And Data Collection
The Baotu Spring Basin is located in the midwestern part of Jinan city, Shandong Province (Fig. 1), and covers an area of 1730 km 2 .Karst water in the Baotu Spring Basin is widely distributed, and pore water mainly occurs in the northwestern part of the spring basin (Niu et al. 2021).
The northern part of the western boundary of the Baotu Spring Basin is the Mashan Fault, and the northern part of the eastern boundary is the Dongwu Fault.The northern boundary comprises Carboniferous-Permian igneous rocks, and the rest of the boundary comprises surface divides.
Controlled by the stratum dip and topography, the overall direction of karst groundwater runoff is from southeast to northwest (Qian et al. 2006).The hydrogeology in the Baotu Spring Basin is shown in Fig. 1, and the data is obtained from the hydrogeological investigation report of this area.According to the geological structure, the spring basin can be divided into magmatic rock, Cambrian and Ordovician karst carbonate rock strata, igneous rock, and Quaternary strata.The strata in the middle of the Baotu Spring Basin mainly include Cambrian and Ordovician karst carbonate strata, and the Cambrian strata are characterized by interbedded limestone and shale.The Ordovician strata comprise thick limestone, argillaceous limestone, and dolomitic limestone, with well-developed karst ssures and a high permeability, which facilitates groundwater recharge, runoff, and discharge.
Although there are many groundwater level observation wells in the Baotu Spring Basin, most of these wells are located in the north of the spring basin due to the mountainous area in the south of the spring basin.Moreover, most of these wells exhibit a short record time or suffer a large amount of missing data, and interpolation cannot be applied to obtain missing data.Considering the data availability and completeness, six groundwater level observation wells were selected in this study (Table 1), and the distribution of these wells in the study area is shown in Fig. 1.Wells 257, 358, and 361 are karst water level monitoring wells, and wells 305, 268, and 287 are pore phreatic water level observation wells.Wells 358 and 361 are located in the Cambrian-Ordovician limestone distribution area, and the limestone in this area is exposed on the surface.The area has well-developed karst pores, ssures and conduits, which provide huge space and channels for groundwater storage and transportation.Wells 257,268,287,305 are located in the limestone lie concealed region, overlying the Quaternary.According to the availability of data, the research time ranges from 1992 to 2012, and groundwater level data are every ten-day monitoring data, with a total of 756 measurements in 21 years.Step 1: In regard to time series x k (k = 1, …, N) to be analysed, the cumulative deviation of the time series is determined, also referred to as the 'pro le': Here, N is the length of the time series, and < x > denotes the average of {x k }.
Step 2: For each given scale s, is divided into N s nonoverlapping segments of equal length.
The setting of scale(s) in DFA and MF-DFA is the most noteworthy, the minimum scale should preferably be greater than 10 and the maximum scale should not be too large (Ihlen 2012).Therefore, in this study, when the calculation object is the groundwater level time series from 1992-2012, the minimum value of the scale is set to 16 and the maximum scale is 128 through debugging; when the calculation object is the groundwater level from 1992-2002, 2003-2012 or the precipitation time series from 2010-2019, because the time series length is reduced, the minimum value of the scale is set to 16 and the maximum scale is set to 64.
Step 3: Local trends are calculated for each segment via least squares tting.The variance can be determined as: Here, is the local variance for every one of the N s segments, and is the tting polynomial of any appropriate order within segment v (v = 1, 2, …, N s ).In this study, linear polynomials was employed Step 4: The qth-order uctuation function can be obtained as follows: where q denotes any real value other than zero.For q = 2, the standard DFA procedure is retrieved.In order to avoid large numerical errors, the selection of q-order should avoid large Step 5: The scaling behaviour of uctuation functions can be analysed by log-log plots of F q (s) versus s for each q.
Where H q is the slope of the curve of logF q (s) ~ log(s), and is referred to as the q-order Hurst exponent.
Before MF-DFA application, it is necessary to use DFA to calculate H 2 , and H 2 is called the generalized Hurst exponent.The time series persists when 0.5 < H 2 < 1 or H 2 > 1.5.Persistence represents long-range correlated process, that means if a time series increases (or decreases) over time, it is likely to continue to increase (or decrease).When 0 < H 2 < 0.5 or 1 < H 2 < 1.5, the time series is anti-persistent.Anti-persistence usually indicates that the trend of change of the time series is opposite to that of the previous period (Sun et al. 2019).Moreover, if 0 < H 2 < 1 then the signal is fGn (fractional Gaussian noise, which is a stationary process).If 1 < H 2 < 2 then it is fBm (fractional Brownian motion, which is nonstationary process) with some range of uncertainty in between (Eke 2002).When H 2 varies between 0.2 and 1.2, the time series resembles noise, at which point MF-DFA can be directly applied to the time series without any transformation, and conversely, the data should be processed before application (Ihlen 2012).Considering a monofractal time series, H q is independent of q.If there occurs multifractal behaviour, H q signi cantly depends on q.
Step 6: The level or strength of multifractals can be represented by the Hölder exponent spectrum (or singularity spectrum) f(α), and α is the Hölder exponent or singularity strength.α and f(α) can be obtained as: ∆α, α 0 and R are three important multifractal parameters.The spectrum width can be de ned as ∆α = α max -α min , ∆α is a measure of the α range, and the larger ∆α is, the richer the structure of the physical process and the stronger the multifractal.α 0 is the critical (central) Hölder exponent, which corresponds to the maximum value of f(α).The larger α 0 is, the more irregular the underlying process.The asymmetry of the multifractal spectrum can be quanti ed as R=(α max − α 0 )/(α 0 − α min ), with a right-skewed spectrum (R > 1) indicating that the physical process exhibits a ne structure and a left-skewed shape (R < 1) indicating a more regular or smoother structure.
The complexity index (CI) of groundwater level can be obtained by summing three normalized multifractal parameters, i.e., α 0 , ∆α, and R. Considering sequences with higher α 0 values, the width of the spectrum f(α) is larger, and right-skewed shapes (R > 1) are more complex than those with opposite features (Lana et al. 2020).In general, a higher CI value indicates that a more signi cant complexity on the physical mechanisms governing groundwater level uctuation and suggest a di cult success predictability, while a lower CI value indicates that is easier to predict.
Results And Discussion
Multifractal detrended uctuation analysis (MF-DFA)
To determine whether data conversion is required before MF-DFA application, DFA was applied to the groundwater time series.As shown in Fig. 3a, the generalized Hurst exponents were 1.2602, 1.7801, and 0.9291 for wells 257, 358, and 361, respectively.As shown in Fig. 3b and very close to 1, which is inconsistent with the non-stationary of groundwater level uctuations, but is also plausible to some extent due to the range of uncertainty between fGn and fBm.
Fluctuation functions F q (s) with different q values for the groundwater level time series were calculated via MF-DFA.For each q, a straight line was tted to the plots of log 2 (F q (s)) versus log 2 (s), and the slope of this line represents H q for that speci c q value.Figure 4 shows curves of the q-order Hurst exponents H q ~q.The curves of the q-order Hurst exponents H q versus q of the time series of the groundwater level in the observation wells show that H q nonlinearly decreases with increasing q, indicating that the time series exhibit multifractal characteristics.In Fig. 4, compared to the other karst water level monitoring wells, H q varied the most gently with q in well 257, and the intensity of the H q uctuation with q in well 358 varied between those in wells 257 and 361.Moreover, the drastic change in H q in well 361 indicate that the multifractal features of the groundwater level uctuation were highly complex.Among the pore water level observation wells, H q changed with q in well 268 to a similar extent as that in well 287, while in well 305, H q changed very slightly in well 305 for q < 1.
Figure 5 shows the multifractal spectra for the groundwater level records of the six observation wells.If a higher level of multifractality is observed in the signal, the f(α) spectrum widens, but this width converges to one point for a purely monofractal signal.Therefore, Fig. 5 shows that the groundwater level time series of these six observation wells were characterized by multifractality.As shown in Fig. 5, the structure and width of the multifractal spectra of the karst water level time series greatly differed, so the multifractal characteristics of the groundwater level time series in the study area exhibited very obvious spatial heterogeneity.
Multifractal parameters and complexity
To measure the multifractal strength and complexity of the groundwater level time series for the six observation wells, speci c parameters were calculated via MF-DFA, as listed in Table 2.The complexity of the physical mechanism controlling the time series could be quanti ed via the CI.The CI values of wells 358 and 361 were 3.50 and 3.48 respectively, which were the highest among all wells, and the difference between the two wells was small, so wells 358 and 361 were classi ed as high complexity.Similarly, wells 257 and 305 were classi ed as low complexity.The CI values of wells 287 and 268 were − 0.77 and − 1.46, respectively, signi cantly lower than CI with high complexity and higher than low complexity.Therefore, wells 257 and 305 were classi ed as moderate complexity.As shown in Fig. 5 and Table 2, the reason for the low complexity of wells 257 and 305 is that the shape of the multifractal spectrum of both wells is left-skewed (R < 1) and Δα of both wells is small.In addition, the high complexity of well 358 is due to the large width of the multifractal spectrum and the high complexity of well 361 is due to the large width of the multifractal spectrum and its right-skewed structure.Therefore, it is more di cult and uncertain to predict the groundwater level in wells 361 and 358, while the groundwater level of wells 305 and 257 exhibited a simple structure and could be more easily predicted.).In this study, the groundwater depths are all greater than 3m (Table 1) and groundwater evaporation is minimal and is not considered in this study.
(1) Functional areas of the Baotu Spring Basin From the perspective of the spring basin functional area where the observation well is located, according to its storage space, hydraulic characteristics and functional characteristics, the Baotu Spring Basin can be divided into a discharge area, direct recharge area and indirect recharge area, as shown in Fig. 6.
The direct recharge area of the Baotu Spring Basin is the Cambrian Chaomidian Formation-Middle Ordovician limestone area widely distributed in the hilly area in the south of the spring basin and the piedmont plain.Karst groundwater can directly accept both atmospheric precipitation and surface water leakage recharge in the direct recharge area.Because wells 361 and 358 are located in the direct recharge area, they respond acutely to precipitation and surface water leakage recharge, which is the reason for the more notable uctuation in the groundwater level in wells 361 and 358.Moreover, wells 358 and 361 are located in low mountainous areas where stepped terrain exists and the slope of the strata is high, providing favorable conditions for groundwater ow.
The discharge area of the Baotu Spring Basin contains a at terrain and a large thickness of Quaternary surface strata, which is conducive to receiving precipitation recharge.Most of the precipitation directly replenishes the pore water in the Quaternary aquifer.Due to the shallow groundwater depth in wells 268 and 287, and the shallow aquifer responds more dynamically to phenomena such as precipitation, evapotranspiration, vegetation, discharge, and soil capillarity, which may account for the higher complexity of pore water levels in wells 268 and 287 compared to karst water level in well 257.In addition, wells 257, 305, 268 and 287 are located in the plain area, with a small stratum gradient.
Compared with the mountain area, the groundwater ow rate is small.According to borehole information, there is a clay layer separates karst aquifer from Quaternary pore aquifer, and the precipitation can be recharged to the karst water only after passing through the Quaternary aquifer and clay layer, so the precipitation signal will be weakened by the ltering of the Quaternary aquifer and clay layer.Moreover, the karst water of Well 257 is not easily affected by other surface factors.Therefore, the groundwater level in well 257 remains stable. (
2) Precipitation
The four precipitation stations closest to the groundwater level observation wells are shown in Fig. 6.Since it is very di cult to obtain detailed precipitation data in the early stage, in this study, we only use the precipitation time series after 2010 to analyze the complexity of precipitation in different regions.This calculation result cannot fully represent the complexity of precipitation in the period of the groundwater level series used for calculation.However, it can re ect the complexity characteristics of precipitation in this region to a certain extent, and provide some reference for analyzing the complexity of groundwater time series.The CI was calculated for the cumulative daily precipitation time series of 1220 days during the 2010-2019 ood season (June, July, August, September), and the results are shown in Table 3.There is little difference in precipitation between the rainfall stations during the ood season, and the precipitation time series in descending order of complexity are Qiujiazhuang, Yanzishan, Donghongmiao, and Changqing.In addition, all six groundwater level observation wells are located in locations where the land use type is arable rather than impermeable, and therefore precipitation can recharge groundwater in the vertical direction.The Qiujiazhuang station, nearest to well 358, and the Yanzishan station, nearest to well 361, have higher precipitation time series complexity than the other stations, and the complexity of the time series of precipitation in ltration recharge obtained for wells 358 and 361 is therefore higher than for the other wells.2) shows that its uctuation pattern differs from that of the other wells in that the groundwater level in well 358 (Fig. 2b) declines sharply in 1998 and 2002 and then rises rapidly thereafter.It can therefore be speculated that there was irregular and very intense groundwater extraction as well as arti cial recharge activity in the vicinity of well 358, which has affected the multifractal results of groundwater levels.In summary, with approximately the same amount of precipitation recharge, the complexity of the precipitation time series near wells 358 and 361 is the highest, and karst water can receive direct recharge from precipitation through the limestone.In addition, wells 358 and 361 are located in a layer with a high slope and a high degree of karst development, which provides suitable conditions for groundwater transport.Although well 257 is also a karst well, the precipitation time series in the vicinity is less complex and the precipitation signal is weakened by the overlying strata, furthermore the groundwater ow rate at the location is low and therefore the water level is more stable.Wells 305, 368, 287 also have stable groundwater levels due to factors such as precipitation and the nature of the aquifer.
Multifractal source analysis
In general, the multifractality of time series can classi ed into two types (Tu et al. 2017, Ye et al. 2017).
The rst one is caused by the broad probability density function (PDF) of the time series, and the other one is due to the long-range correlation on different scales.Since randomly shu ed sequences can destroy the long-range correlation, the above two types can be distinguished by performing multifractal analysis of the randomly shu ed time series.If the multifractality is entirely attributable to the second type, it will disappear; otherwise, it will be retained.Moreover, the shu ed data could exhibit a lower multifractality than that of the original data when the multifractality is due to both types (Ihlen 2012, Yuan et al. 2014).
Figure 7 shows the multifractal spectra for the shu ed groundwater level data.As shown in Fig. 7, the width of the multifractal spectra was narrower than that of the original data, but this width did not converge to a point.As indicate in Table 5, Δα of the shu ed data was smaller than that of the original data.The larger Δα is, the higher the multifractality of the time series.As shown in Fig. 7 and Table 5, the multifractality diminished but did not disappear for all time series.This result shows that the multifractality of the groundwater level time-series could not be removed by shu ing, so the multifractality originated from both the long-range correlation and broad PDF of the groundwater level series.The high spatial variability of the groundwater level complexity is caused by highly uneven hydrogeological conditions.Therefore, by analysing the complexity of the physical mechanism controlling the groundwater level, we could further analyse the nonlinear mechanism of the groundwater response to natural and human factors, and the nonlinear response mechanism is the source of groundwater level complexity.In future research, complexity analysis of the groundwater level and nonlinear response mechanism analysis of groundwater to various factors could be combined by collecting more detailed hydrogeological data, which is an effective method to better understand the complexity of regional groundwater level change.
Declarations Figures
Hydrogeological
MethodMF-
DFA is a common mathematical tool for the detection of the scale characteristics and multifractality of time-series datasets.DFA can detect the non-stationarity of evolution, effectively eliminate the apparent long-range correlation attributed to external effects and identify uctuations only related to the characteristics of aquifers.In this study, MF-DFA was utilized to analyse the multifractal properties of non-stationary time series, and ultimately, the obtained information on the scale behaviour and parameters could be useful for multifractal modelling.The DFA and MF-DFA procedure comprises the following main steps (Adarsh et al. 2020, Gao et al. 2019, Kantelhardt 2002, Zhang et al. 2014):
negative and positive values (Little and Bloom eld, 2010).Therefore, referring to other studies (Labat et al. 2011, Little and Bloom eld 2010, Sun et al. 2019, Yu et al. 2016), MF-DFA was employed with the range of q values is [-5, 5], and the step of q used in MF-DFA is 0.1.
map of the Baotu Spring Basin Page 18/22
Figure 7 Multifractal
Figure 7 al. 2020, Ihlen 2012), image analysis applications and remote sensing (Aleksandrowicz et al. 2022, Krupiński et al. 2020).It is of theoretical and practical signi cance to better understand the statistical characteristics of groundwater table series according to multifractal parameters (Labat et al. 2011, Little and Bloom eld 2010, Sun et al. 2019, Yu et al. 2016), but there are few studies assessing the complexity of groundwater in different types of aquifers.The Baotu Spring Basin is one of the typical karst systems and the important water sources in North China.Moreover, the karst springs in the Baotu Spring Basin are of historical and tourism value (Gao et al. 2020, Kang et al. 2011, Mandelbrot and Wallis 1969, Xing et al. 2018,
Table 1 .
Basic information on the observation wells The presence of periodic uctuations in meteorological and hydrological time-series data could signi cantly affect the MF-DFA results(Li and Zhang 2007, Lu et al. 2021, Yuan et al. 2014, Zhang et al. 2019).Therefore, the spectral analysis is required to eliminate the periodic component of the groundwater level data (Lu et al. 2021).Groundwater level uctuations are shown in Figure 2.
(Yu et al. 2016)d Hurst exponents of wells 305, 268 and 287 were 1.2584, 1.2164 and 1.5166, respectively.The groundwater level time series of wells 358 and 361 behaves as persistence, and the groundwater level time series of other wells show anti-persistent correlation.The calculated generalized Hurst exponent ranged from 0.9291-1.7801.Because MF-DFA is best applied when the Hurst exponent varies between 0.2 and 1.2, the groundwater level time series should be converted via a one-order difference(Ihlen 2012).Although the generalized Hurst exponent of well 361 ranged from 0.2 to 1.2, the groundwater level time series of well 361 also requires one-order difference in order to ensure the consistency of the input data.The groundwater level time series of wells 358, 361, 287 behaves as persistence, and the groundwater level time series of other wells show anti-persistent correlation.Groundwater level uctuation is usually non-stationary and should be characterized by Brownian motion(Yu et al. 2016).In this study, the generalized Hurst exponents of wells 257,358,305,268,287 are all greater than 1.Thus, the uctuations of groundwater levels at these ve wells are fractional Brownian motion (fBm).The generalized Hurst exponent of well 361 is less than 1
Table 2
Multifractal parameters of the groundwater level time-series Yu et al. 20162011uctuations are the dynamic responses of aquifers to recharge and discharge.The high spatial variability of the complexity of groundwater level uctuations is due to the coupled effects of conditions affecting recharge and discharge, such as aquifer characteristics, precipitation, and anthropogenic disturbances, evaporation, etc(Labat et al. 2011, Rakhshandehroo and Mehrab Amiri 2012,Yu et al. 2016
Table 3
Since 2003, large-scale karst water exploitation has been prohibited in the Baotu Spring Basin.Therefore, the groundwater level time-series was divided into 1992-2002 and 2003-2012.As shown in Table4, after 2003, the CI of wells 257, 361, 305, 268 and 287 decreased slightly, indicating the complexity of groundwater level uctuations in these wells decreased, while the CI of well 358 increased.The groundwater level time series for well 358 (Fig.
Table 4
The CI of the groundwater level time-series
Table 5
Multifractal parameters of the shu ed groundwater level time series (Δα(shu ed) is calculated from the time series of groundwater level after random shu ing, Δα (original) is calculated from the original time series of groundwater level without random shu ing) This study applied MF-DFA to investigate the multifractal characteristics of the groundwater level in the Baotu Spring Basin, and the groundwater level complexity was analysed.The results indicate that the Baotu Spring Basin groundwater level time series exhibited multifractal characteristics, and each multifractal spectrum differed.The multifractality of the groundwater level time series could not be removed by shu ing, so the multifractality originated from both the long-range correlation of the groundwater level time series and broad PDF.The CI was calculated based on the multifractal parameters.The CI values of wells 358 361 were 3.50 and 3.48, respectively (high complexity), those of wells 287 and 268 were − 0.77 and − 1.46, respectively (moderate complexity), and those of wells 257 and 305 were the lowest, at -2.32 and − 2.43, respectively (low complexity).This result shows that the CI of the groundwater level indicated spatial heterogeneity. | 7,037.8 | 2023-11-28T00:00:00.000 | [
"Environmental Science",
"Geology"
] |
Model-based Analysis of ChIP-Seq (MACS)
MACS performs model-based analysis of ChIP-Seq data generated by short read sequencers.
Background
The determination of the 'cistrome', the genome-wide set of in vivo cis-elements bound by trans-factors [1], is necessary to determine the genes that are directly regulated by those trans-factors. Chromatin immunoprecipitation (ChIP) [2] coupled with genome tiling microarrays (ChIP-chip) [3,4] and sequencing (ChIP-Seq) [5][6][7][8] have become popular techniques to identify cistromes. Although early ChIP-Seq efforts were limited by sequencing throughput and cost [2,9], tremendous progress has been achieved in the past year in the development of next generation massively parallel sequencing. Tens of millions of short tags (25-50 bases) can now be simultaneously sequenced at less than 1% the cost of tradi-tional Sanger sequencing methods. Technologies such as Illumina's Solexa or Applied Biosystems' SOLiD™ have made ChIP-Seq a practical and potentially superior alternative to ChIP-chip [5,8].
While providing several advantages over ChIP-chip, such as less starting material, lower cost, and higher peak resolution, ChIP-Seq also poses challenges (or opportunities) in the analysis of data. First, ChIP-Seq tags represent only the ends of the ChIP fragments, instead of precise protein-DNA binding sites. Although tag strand information and the approximate distance to the precise binding site could help improve peak resolution, a good tag to site distance estimate is often unknown to the user. Second, ChIP-Seq data exhibit regional biases along the genome due to sequencing and mapping biases, chromatin structure and genome copy number variations [10]. These biases could be modeled if matching control samples are sequenced deeply enough. However, among the four recently published ChIP-Seq studies [5][6][7][8], one did not have a control sample [5] and only one of the three with control samples systematically used them to guide peak finding [8]. That method requires peaks to contain significantly enriched tags in the ChIP sample relative to the control, although a small ChIP peak region often contains too few control tags to robustly estimate the background biases.
Here, we present Model-based Analysis of ChIP-Seq data, MACS, which addresses these issues and gives robust and high resolution ChIP-Seq peak predictions. We conducted ChIP-Seq of FoxA1 (hepatocyte nuclear factor 3α) in MCF7 cells for comparison with FoxA1 ChIP-chip [1] and identification of features unique to each platform. When applied to three human ChIP-Seq datasets to identify binding sites of FoxA1 in MCF7 cells, NRSF (neuron-restrictive silencer factor) in Jurkat T cells [8], and CTCF (CCCTC-binding factor) in CD4 + T cells [5] (summarized in Table S1 in Additional data file 1), MACS gives results superior to those produced by other published ChIP-Seq peak finding algorithms [8,11,12].
Modeling the shift size of ChIP-Seq tags
ChIP-Seq tags represent the ends of fragments in a ChIP-DNA library and are often shifted towards the 3' direction to better represent the precise protein-DNA interaction site. The size of the shift is, however, often unknown to the experimenter. Since ChIP-DNA fragments are equally likely to be sequenced from both ends, the tag density around a true binding site should show a bimodal enrichment pattern, with Watson strand tags enriched upstream of binding and Crick strand tags enriched downstream. MACS takes advantage of this bimodal pattern to empirically model the shifting size to better locate the precise binding sites.
Given a sonication size (bandwidth) and a high-confidence fold-enrichment (mfold), MACS slides 2bandwidth windows across the genome to find regions with tags more than mfold enriched relative to a random tag genome distribution. MACS randomly samples 1,000 of these high-quality peaks, separates their Watson and Crick tags, and aligns them by the midpoint between their Watson and Crick tag centers ( Figure 1a) if the Watson tag center is to the left of the Crick tag center. The distance between the modes of the Watson and Crick peaks in the alignment is defined as 'd', and MACS shifts all the tags by d/2 toward the 3' ends to the most likely protein-DNA interaction sites.
When applied to FoxA1 ChIP-Seq, which was sequenced with 3.9 million uniquely mapped tags, MACS estimates the d to be only 126 bp (Figure 1a; suggesting a tag shift size of 63 bp), despite a sonication size (bandwidth) of around 500 bp and Solexa size-selection of around 200 bp. Since the FKHR motif sequence dictates the precise FoxA1 binding location, the true distribution of d could be estimated by aligning the tags by the FKHR motif (122 bp; Figure 1b)
Peak detection
For experiments with a control, MACS linearly scales the total control tag count to be the same as the total ChIP tag count. Sometimes the same tag can be sequenced repeatedly, more times than expected from a random genome-wide tag distribution. Such tags might arise from biases during ChIP-DNA amplification and sequencing library preparation, and are likely to add noise to the final peak calls. Therefore, MACS removes duplicate tags in excess of what is warranted by the sequencing depth (binomial distribution p-value <10 -5 ). For example, for the 3.9 million FoxA1 ChIP-Seq tags, MACS allows each genomic position to contain no more than one tag and removes all the redundancies.
With the current genome coverage of most ChIP-Seq experiments, tag distribution along the genome could be modeled by a Poisson distribution [7]. The advantage of this model is that one parameter, λ BG , can capture both the mean and the variance of the distribution. After MACS shifts every tag by d/ 2, it slides 2d windows across the genome to find candidate peaks with a significant tag enrichment (Poisson distribution p-value based on λ BG , default 10 -5 ). Overlapping enriched peaks are merged, and each tag position is extended d bases from its center. The location with the highest fragment pileup, hereafter referred to as the summit, is predicted as the precise binding location.
In the control samples, we often observe tag distributions with local fluctuations and biases. For example, at the FoxA1 candidate peak locations, tag counts are well correlated between ChIP and control samples (Figure 1c,d). Many possible sources for these biases include local chromatin structure, DNA amplification and sequencing bias, and genome copy number variation. Therefore, instead of using a uniform λ BG estimated from the whole genome, MACS uses a dynamic parameter, λ local , defined for each candidate peak as: where λ 1k , λ 5k and λ 10k are λ estimated from the 1 kb, 5 kb or 10 kb window centered at the peak location in the control sample, or the ChIP-Seq sample when a control sample is not available (in which case λ 1k is not used). λ local captures the (e,f) MACS improves the motif occurrence in the identified peak centers (e) and the spatial resolution (f) for FoxA1 ChIP-Seq through tag shifting and λ local . Peaks are ranked by p-value. The motif occurrence is calculated as the percentage of peaks with the FKHR motif within 50 bp of the peak summit. The spatial resolution is calculated as the average distance from the summit to the nearest FKHR motif. Peaks with no FKHR motif within 150 bp of the peak summit are removed from the spatial resolution calculation. influence of local biases, and is robust against occasional low tag counts at small local regions. MACS uses λ local to calculate the p-value of each candidate peak and removes potential false positives due to local biases (that is, peaks significantly under λ BG , but not under λ local ). Candidate peaks with p-values below a user-defined threshold p-value (default 10 -5 ) are called, and the ratio between the ChIP-Seq tag count and λ local is reported as the fold_enrichment.
For a ChIP-Seq experiment with controls, MACS empirically estimates the false discovery rate (FDR) for each detected peak using the same procedure employed in the previous ChIP-chip peak finders MAT [13] and MA2C [14]. At each pvalue, MACS uses the same parameters to find ChIP peaks over control and control peaks over ChIP (that is, a sample swap). The empirical FDR is defined as Number of control peaks / Number of ChIP peaks. MACS can also be applied to differential binding between two conditions by treating one of the samples as the control. Since peaks from either sample are likely to be biologically meaningful in this case, we cannot use a sample swap to calculate FDR, and the data quality of each sample needs to be evaluated against a real control.
Model evaluation
The two key features of MACS are: empirical modeling of 'd' and tag shifting by d/2 to putative protein-DNA interaction site; and the use of a dynamic λ local to capture local biases in the genome. To evaluate the effectiveness of tag shifting based on the MACS model d, we compared the performance of MACS to a similar procedure that uses the original tag positions instead of the shifted tag locations. The effectiveness of the dynamic λ local is assessed by comparing MACS to a procedure that uses a uniform λ BG from the genome background. Figure 1e,f show that both the detection specificity, measured by the percentage of predicted peaks with a FKHR motif within 50 bp of the peak summit, and the spatial resolution, defined as the average distance from the peak summit to the nearest FKHR motif, are greatly improved by using tag shifting and the dynamic λ local . In addition, FoxA1 is known to cooperatively interact with estrogen receptor in breast cancer cells [1,15]. As evidence for this, we also observed enrichment for estrogen receptor elements (3.1-fold enriched relative to genome motif occurrence) and its half-site (2.7-fold) [15] within the center 300 bp regions of MACS-detected FoxA1 ChIP-Seq peaks.
λ local is also effective in capturing the local genomic bias from a ChIP sample alone when a control is not available. To demonstrate this, we applied MACS to FoxA1 ChIP-Seq and control data separately. Using the same parameters, all the control peaks are, in theory, false positives, so the FDR can be empirically estimated as Number of control peaks / Number of ChIP peaks. To identify 7,000 peaks, the FDR for MACS is only 0.4% when a control is available and λ local is used. To get 7,000 peaks when a control is not available, the FDR could still remain low at 3.8% if MACS estimates λ local from the ChIP sample, whereas it would reach 41.2% if MACS uses a global λ BG . This implies that the λ local is critical for ChIP-Seq studies when matching control samples are not available [5,9].
Method comparisons
We compared MACS with three other publicly available ChIP-Seq peak finding methods, ChIPSeq Peak Finder [8], Find-Peaks [11] and QuEST [12]. To compare their prediction specificity, we swapped the ChIP and control samples, and calculated the FDR of each algorithm as Number of control peaks / Number of ChIP peaks using the same parameters for ChIP and control. For FoxA1 and NRSF ChIP-Seq (an FDR for CTCF is not available due to the lack of control), MACS consistently gave fewer false positives than the other three methods (Figure 2a,b).
Determining the percentage of predicted peaks associated with a motif within 50 bp of the peak center for FoxA1 and NRSF ChIP-Seq, we found MACS to give consistently higher motif occurrences (Figure 2c,d). Evaluating the average distance from peak center to motif, excluding peaks that have no motif within 150 bp of the peak center, we found that MACS predicts peaks with better spatial resolution in most cases (Figure 2e,f). For CTCF, since QuEST does not run on samples without controls, we only compared MACS to ChIPSeq Peak Finder and FindPeaks. Again, MACS gave both higher motif occurrences within 50 bp of the peak center and better spatial resolutions than other methods ( Figure S1 in Additional data file 1). In general, MACS not only found more peaks with fewer false positives, but also provided better binding resolution to facilitate downstream motif discovery.
Comparison of ChIP-Seq to ChIP-chip
A comparison of FoxA1 ChIP-Seq and ChIP-chip revealed the peak locations to be fairly consistent with each other ( Figure 3a). Not surprisingly, the majority of ChIP-Seq peaks under a FDR of 1% (65.4%) were also detected by ChIP-chip (MAT [13] cutoff at FDR <1% and fold-enrichment >2). Among the remaining 34.6% ChIP-Seq unique peaks, 1,045 (13.3%) were not tiled or only partially tiled on the arrays due to the array design. Therefore, only 21.4% of ChIP-Seq peaks are indeed specific to the sequencing platform. Furthermore, ChIP-chip targets with higher fold-enrichments are more likely to be reproducibly detected by ChIP-Seq with a higher tag count ( Figure 3b). Meanwhile, although the signals of array probes at the ChIP-Seq specific peak regions are below the peak-calling cutoff, they show moderate signal enrichments that are significantly higher than the genomic background (Wilcoxon p-value <10 -320 ; Figure 3c). Indeed, 835 out of 1,684 ChIP-Seq specific peaks could also be detected in ChIP-chip, when the less stringent FDR cutoff of 5% is used. Another reason why peaks detected by ChIP-Seq may be undetected by ChIPchip is that ChIP-Seq specific peaks are usually slightly shorter than similar fold-enrichment peaks found by both ChIP-Seq and ChIP-chip ( Figure 3d) and may not be detectable on the array due to insufficient probe coverage. On the Zhang et al. R137.5 Genome Biology 2008, 9:R137 other hand, ChIP-chip specific peak regions also have significantly more sequencing tags than the genomic background (Wilcoxon p-value <10 -320 ; Figure S2 in Additional data file 1), although with current sequencing depth, those regions cannot be called as peaks.
Comparison of MACS with ChIPSeq Peak Finder, FindPeaks and QuEST
Comparing the difference between ChIP-chip and ChIP-Seq peaks, we find that the average peak width from ChIP-chip is twice as large as that from ChIP-Seq. The average distance from peak summit to motif is significantly smaller in ChIP-Seq than ChIP-chip (Figure 3e), demonstrating the superior resolution of ChIP-Seq. Under the same 1% FDR cutoff, the FKHR motif occurrence within the central 200 bp from ChIPchip or ChIP-Seq specific peaks is comparable with that from the overlapping peaks (Figure 3f). This suggests that most of the platform-specific peaks are genuine binding sites. A comparison between NRSF ChIP-Seq and ChIP-chip ( Figure S3 in Additional data file 1) yields similar results, although the overlapping peaks for NRSF are of much better quality than the platform-specific peaks.
Discussion
ChIP-Seq users are often curious as to whether they have sequenced deep enough to saturate all the binding sites. In principle, sequencing saturation should be dependent on the fold-enrichment, since higher-fold peaks are saturated earlier than lower-fold ones. In addition, due to different cost and throughput considerations, different users might be interested in recovering sites at different fold-enrichment cutoffs. Therefore, MACS produces a saturation table to report, at different fold-enrichments, the proportion of sites that could still be detected when using 90% to 20% of the tags. Such tables produced for FoxA1 (3.9 million tags) and NRSF (2.2 million tags) ChIP-Seq data sets ( Figure S4 in Additional data file 1; CTCF does not have a control to robustly estimate foldenrichment) show that while peaks with over 60-fold enrichment have been saturated, deeper sequencing could still recover more sites less than 40-fold enriched relative to the chromatin input DNA. As sequencing technologies improve their throughput, researchers are gradually increasing their sequencing depth, so this question could be revisited in the future. For now, we leave it up to individual users to make an informed decision on whether to sequence more based on the saturation at different fold-enrichment levels.
The d modeled by MACS suggests that some short read sequencers such as Solexa may preferentially sequence shorter fragments in a ChIP-DNA pool. This may contribute to the superior resolution observed in ChIP-Seq data, especially for activating transcription and epigenetic factors in open chromatin. However, for repressive factors targeting relatively compact chromatin, the target regions might be harder to sonicate into the soluble extract. Furthermore, in the resulting ChIP-DNA, the true targets may tend to be longer than the background DNA in open chromatin, making them unfavorable for size-selection and sequencing. This implies that epigenetic markers of closed chromatin may be harder to ChIP, and even harder to ChIP-Seq. To assess this potential bias, examining the histone mark ChIP-Seq results from Mikkelsen et al. [7], we find that while the ChIP-Seq efficiency of the active mark H3K4me3 remains high as pluripotent cells differentiate, that of repressive marks H3K27me3 and H3K9me3 becomes lower with differentiation (Table S2 in Additional data file 1), even though it is likely that there are more targets for these repressive marks as cells differentiate. We caution ChIP-Seq users to adopt measures to compensate for this bias when ChIPing repressive marks, such as more vigorous sonication, size-selecting slightly bigger fragments for library preparation, or sonicating the ChIP-DNA further between decrosslinking and library preparation.
MACS calculates the FDR based on the number of peaks from control over ChIP that are called at the same p-value cutoff. This FDR estimate is more robust than calculating the FDR from randomizing tags along the genome. However, we notice that when tag counts from ChIP and controls are not balanced, the sample with more tags often gives more peaks even though MACS normalizes the total tag counts between the two samples ( Figure S5 in Additional data file 1). While we await more available ChIP-Seq data with deeper coverage to understand and overcome this bias, we suggest to ChIP-Seq users that if they sequence more ChIP tags than controls, the FDR estimate of their ChIP peaks might be overly optimistic.
Conclusion
As developments in sequencing technology popularize ChIP-Seq, we propose a novel algorithm, MACS, for its data analysis. MACS offers four important utilities for predicting protein-DNA interaction sites from ChIP-Seq. First, MACS improves the spatial resolution of the predicted sites by empirically modeling the distance d and shifting tags by d/2. Second, MACS uses a dynamic λ local parameter to capture local biases in the genome and improves the robustness and specificity of the prediction. It is worth noting that in addition to ChIP-Seq, λ local can potentially be applied to other high throughput sequencing applications, such as copy number variation and digital gene expression, to capture regional biases and estimate robust fold-enrichment. Third, MACS can be applied to ChIP-Seq experiments without controls, and to those with controls with improved performance. Last but not least, MACS is easy to use and provides detailed information for each peak, such as genome coordinates, p-value, FDR, fold_enrichment, and summit (peak center).
Dataset
ChIP-Seq data for three factors, NRSF, CTCF, and FoxA1, were used in this study. ChIP-chip and ChIP-Seq (2.2 million ChIP and 2.8 million control uniquely mapped reads, simplified as 'tags') data for NRSF in Jurkat T cells were obtained from Gene Expression Omnibus (GSM210637) and Johnson et al. [8], respectively. ChIP-Seq (2.9 million ChIP tags) data for CTCF in CD4 + T cells were derived from Barski et al. [5].
ChIP-chip data for FoxA1 and controls in MCF7 cells were previously published [1], and their corresponding ChIP-Seq data were generated specifically for this study. Around 3 ng FoxA1 ChIP DNA and 3 ng control DNA were used for library preparation, each consisting of an equimolar mixture of DNA from three independent experiments. Libraries were prepared as described in [8] using a PCR preamplification step and size selection for DNA fragments between 150 and 400 bp. FoxA1 ChIP and control DNA were each sequenced with two lanes by the Illumina/Solexa 1G Genome Analyzer, and yielded 3.9 million and 5.2 million uniquely mapped tags, respectively.
Software implementation
MACS is implemented in Python and freely available with an open source Artistic License at [16]. It runs from the command line and takes the following parameters: -t for treatment file (ChIP tags, this is the ONLY required parameter for MACS) and -c for control file containing mapped tags; -format for input file format in BED or ELAND (output) format (default BED); --name for name of the run (for example, FoxA1, default NA); --gsize for mappable genome size to calculate λ BG from tag count (default 2.7G bp, approximately the mappable human genome size); --tsize for tag size (default 25); --bw for bandwidth, which is half of the esti-mated sonication size (default 300); --pvalue for p-value cutoff to call peaks (default 1e-5); --mfold for high-confidence fold-enrichment to find model peaks for MACS modeling (default 32); --diag for generating the table to evaluate sequence saturation (default off).
In addition, the user has the option to shift tags by an arbitrary number (--shiftsize) without the MACS model (-nomodel), to use a global lambda (--nolambda) to call peaks, and to show debugging and warning messages (-verbose). If a user has replicate files for ChIP or control, it is recommended to concatenate all replicates into one input file. The output includes one BED file containing the peak chromosome coordinates, and one xls file containing the genome coordinates, summit, p-value, fold_enrichment and FDR (if control is available) of each peak. For FoxA1 ChIP-Seq in MCF7 cells with 3.9 million and 5.2 million ChIP and control tags, respectively, it takes MACS 15 seconds to model the ChIP-DNA size distribution and less than 3 minutes to detect peaks on a 2 GHz CPU Linux computer with 2 GB of RAM. Figure S6 in Additional data file 1 illustrates the whole process with a flow chart.
Authors' contributions
XSL, WL and YZ conceived the project and wrote the paper. YZ, TL and CAM designed the algorithm, performed the research and implemented the software. JE, DSJ, BEB, CN, RMM and MB performed FoxA1 ChIP-Seq experiments and contributed to ideas. All authors read and approved the final manuscript.
Additional data files
The following additional data are available. Additional data file 1 contains supporting Figures S1-S6, and supporting Tables S1 and S2.
Additional data file 1 Figures S1-S6, and Tables S1 and S2 Figures S1-S6, and Tables S1 and S2. Click here for file | 5,117 | 2008-09-17T00:00:00.000 | [
"Biology",
"Computer Science"
] |
Information flows from hippocampus to auditory cortex during replay of verbal working memory items
The maintenance of items in working memory (WM) relies on a widespread network of cortical areas and hippocampus where synchronization between electrophysiological recordings reflects functional coupling. We investigated the direction of information flow between auditory cortex and hippocampus while participants heard and then mentally replayed strings of letters in WM by activating their phonological loop. We recorded local field potentials from the hippocampus, reconstructed beamforming sources of scalp EEG, and – additionally in four participants – recorded from subdural cortical electrodes. When analyzing Granger causality, the information flow was from auditory cortex to hippocampus with a peak in the [4 8] Hz range while participants heard the letters. This flow was subsequently reversed during maintenance while participants maintained the letters in memory. The functional interaction between hippocampus and the cortex and the reversal of information flow provide a physiological basis for the encoding of memory items and their active replay during maintenance.
At the network level, synchronized oscillations have been proposed as a mechanism for functional interactions between brain regions (Fries, 2015;Pesaran et al., 2018). It is thought that these oscillations show temporal coupling of the low-frequency phase for long-range communication between cortical areas (Sarnthein et al., 1998;Polanía et al., 2012;Maris et al., 2011;Johnson et al., 2018a;Johnson et al., 2018b;Boran et al., 2019;Solomon et al., 2017). This synchronization suggests an active maintenance process through reverberating signals between brain regions.
We here extend previous studies with the same task (Michels et al., 2008;Boran et al., 2019) by recording from four participants with hippocampal LFP and direct cortical recordings (ECoG) from electrodes over primary auditory, parietal, and occipital cortical areas. Given the low incidence of the epileptogenic zone in parietal cortex, parietal ECoG recordings are rare. To benefit from the wide spatial coverage of scalp EEG, we analyzed the directed functional coupling between hippocampal LFP and the beamforming sources of scalp EEG in all 15 participants. We found that the information flow was from auditory cortex to hippocampus during the encoding of WM items, and the flow was from hippocampus to auditory cortex for the replay of the items during the maintenance period. eLife digest Every day, the brain's ability to temporarily store and recall information -called working memory -enables us to reason, solve complex problems or to speak. Holding pieces of information in working memory for short periods of times is a skill that relies on communication between neural circuits that span several areas of the brain. The hippocampus, a seahorse-shaped area at the centre of the brain, is well-known for its role in learning and memory. Less clear, however, is how brain regions that process sensory inputs, including visual stimuli and sounds, contribute to working memory.
To investigate, Dimakopoulos et al. studied the flow of information between the hippocampus and the auditory cortex, which processes sound. To do so, various types of electrodes were placed on the scalp or surgically implanted in the brains of people with drug-resistant epilepsy. These electrodes measured the brain activity of participants as they read, heard and then mentally replayed strings of up to 8 letters. The electrical signals analysed reflected the flow of information between brain areas.
When participants read and heard the sequence of letters, brain signals flowed from the auditory cortex to the hippocampus. The flow of electrical activity was reversed while participants recalled the letters. This pattern was found only in the left side of the brain, as expected for a language related task, and only if participants recalled the letters correctly.
This work by Dimakopoulos et al. provides the first evidence of bidirectional communication between brain areas that are active when people memorise and recall information from their working memory. In doing so, it provides a physiological basis for how the brain encodes and replays information stored in working memory, which evidently relies on the interplay between the hippocampus and sensory cortex.
Task and behavior
Fifteen participants (median age 29 years, range , 7 male, Table 1) performed a modified Sternberg WM task (71 sessions in total, 50 trials each). In the task, items were presented all at once rather than sequentially, thus separating the encoding period from the maintenance period. In each trial, the participant was instructed to memorize a set of 4, 6, or 8 letters presented for 2 s (encoding). The number of letters was thus specific for the memory workload. The participants read the letters themselves and heard them spoken at the same time. Since participants had difficulties reading eight letters within the 2 s encoding period, also hearing the letters assured their good performance. After a delay (maintenance) period of 3 s, a probe letter prompted the participant to retrieve their memory (retrieval) and to indicate by button press ('IN' or 'OUT') whether or not the probe letter was a member of the letter set held in memory (Figure 1a). During the maintenance period, participants rehearsed the verbal representation of the letter strings subvocally, i.e., mentally replayed the memory items. Participants had been instructed to employ this strategy, and they confirmed after the sessions that they had indeed employed this strategy. This activation of the phonological loop (Baddeley, 2003) is a component of verbal WM as it serves to produce an appropriate behavioral response (Christophel et al., 2017).
The mean correct response rate was 91% (both for IN and OUT trials). The rate of correct responses decreased with set size from a set size of 4 (97% correct responses) to set sizes of 6 (89%) and 8 (83%) (Figure 1b). Across the participants, the memory capacity averaged 6.1 (Cowan's K, [correct IN rate +correct OUT rate -1]×set size), which indicates that the participants were able to maintain at least six letters in memory. The mean response time (RT) for correct trials (3045 trials) was 1.1±0.5 s and increased with workload from set size 4 (1.1±0.5 s) to 6 (1.2±0.5 s) and 8 (1.3±0.6 s), 53 ms/item ( Figure 1c). Correct IN/OUT decisions were made more rapidly than incorrect decisions (1.1±0.5 s vs 1.3±0.6 s). These data show that the participants performed well in the task and that the difficulty of the trials increased with the number of letters in the set. In further analysis, we focused on correct trials with set size 6 and 8 letters to assure hippocampal activation and hippocampo-cortical interaction as shown earlier (Boran et al., 2019).
Power spectral density in cortical and hippocampal recordings
To investigate how cortical and hippocampal activity subserves WM processing, we analyzed the LFP recorded in the hippocampus (Figure 1d, Figure 1-figure supplement 1, Supplementary file 1) together with ECoG from cortical strip electrodes (Figure 2a, Figure 3a and f). In the following, we present power spectral density (PSD) time-frequency maps from representative electrode contacts. In an occipital recording of Participant 1 (grid contact H3), strong gamma activity (>40 Hz) in the relative PSD occurred while the participant viewed the letters during encoding (increase >100% with respect to fixation, Figure 2b). Similarly, encoding elicited gamma activity in a temporal recording over auditory cortex (increase >100%, grid contact C2, Figure 2c), similar as in Kumar et al., 2021. Gamma increased significantly only in temporal and occipital-parietal contacts (permutation test with z-score >1.96, Figure 2a).
After the letters disappeared from the screen, activity occurred in the [11 14] Hz range (high alpha/ low beta, Figure 2b) toward the end of the maintenance period in temporal and occipital contacts (permutation test p<0.05, Figure 2d). Similarly, the temporal scalp EEG of Participant 2 (black rimmed disk denotes electrode site T3 in Figure 3a) showed activity during encoding and maintenance, albeit at lower frequencies ( Figure 3b); this pattern was found only in scalp EEG and not in ECoG, probably because the strip electrode was not located over auditory cortex. In Participant 3, a similar pattern occurred in the PSD of a temporo-parietal recording (most posterior strip electrode contact, Figure 3f), where the appearance of the letters prompted gamma activity and the maintenance period showed alpha activity ([8 11] Hz, Figure 3g). Similarly, in the electrode contacts on right parietal cortex of Participant 4 (Figure 3k), the letter stimulus elicited gamma activity and the maintenance period showed alpha activity (8-11 Hz, Figure 3l).
The site of the participants' maintenance activity coincides with the generator of scalp EEG that was found in the parietal cortex for the same task (Michels et al., 2008). The PSD thereby confirmed the findings of local synchronization of cortical activity during WM maintenance (Michels et al., 2008;Bidelman et al., 2021;Pavlov and Kotchoubey, 2022). For each participant, we report the atlas parcels that contained EEG sources with the maximal t-value and the t-value of sources in auditory cortex (Heschl gyrus) during encoding and maintenance (non-parametric cluster-based permutation test p<0.05). In each participant, the vast majority of the significant LCMV sources were in the left hemisphere, both during encoding (≥87%) and during maintenance (≥81%). We also report the net information flow (ΔGranger) for correct and incorrect trials in the direction auditory cortex → hippocampus during encoding and in the direction hippocampus → auditory cortex during maintenance. Figure 1. Task and recording sites. (a) In the task, sets of consonants are presented and have to be memorized. The set size (4, 6, or 8 letters) determines working memory workload. In each trial, presentation of a letter string (encoding period, 2 s) is followed by a delay (maintenance period, 3 s). After the delay, a probe letter is presented. Participants indicate whether the probe was in the letter string or not. (e) Hippocampal PSD shows sustained beta activity toward the end of maintenance. (f) Phase-locking value (PLV) between hippocampus and auditory cortex (contact C3) during fixation (black), encoding (blue), and maintenance (red). The PLV spectra show a broad frequency distribution. The PLV during maintenance is higher than during fixation. Red bars: frequency ranges of significant PLV difference (p<0.05, cluster-based non-parametric permutation test against a null distribution with scrambled trials during fixation and maintenance). In the hippocampus of all four participants, we found elevated activity in the beta range ([12 24] Hz) toward the end of the maintenance period (increase >100%, Figure 2e, Figure 3c, h and m), confirming the hippocampal contribution to processing of this task (Boran et al., 2019).
Functional coupling between hippocampus and cortex
To investigate the functional coupling between cortex and hippocampus, we first calculated the phaselocking value (PLV). In Participant 1, we found high PLV over a broad frequency range in contacts over auditory cortex throughout the trial. Compared to encoding, maintenance showed enhanced PLV in the theta range between hippocampal LFP and cortical ECoG (PLV=0.4 in contact C3, permutation test p<0.05, Figure 2f). PLV in the [4 8] Hz theta range increased significantly with several contacts over auditory cortex (permutation test p<0.05, Figure 2g). This speaks for a functional coupling between auditory cortex and hippocampus mediated by synchronized oscillations (Rezayat et al., 2021).
Directed functional coupling between hippocampus and ECoG
What was the directionality of the information flow during encoding and maintenance in a trial? We used spectral Granger causality (GC) as a measure of directed functional connectivity to determine the direction of the information flow between auditory cortex and hippocampus in Participant 1 during the trials. During encoding, the information flow was from auditory cortex to hippocampus with a maximum in the theta frequency range (dark blue curve in Figure 2h). The net information flow ΔGranger (GC hipp→cortex -GC cortex→hipp) during encoding was significant in the [6 8] Hz range (blue bar in Figure 2h, p<0.05 permutation test against a null distribution). During maintenance, the information flow in the theta frequency range was reversed (dark red curve), i.e., from hippocampus to auditory cortex (dark red curve in Figure 2h). The net information flow ΔGranger during maintenance was significant in the [5 8] Hz range (red bar in Figure 2h, p<0.05 permutation test against a null distribution). Concerning the spatial spread of the theta GC, the maximal net information flow ΔGranger (GC hipp→cortex -GC cortex→hipp) during encoding occurred from auditory cortex to hippocampus (p<0.05, permutation test, Figure 2i). During maintenance, the theta ΔGranger was significant from hippocampus to both auditory cortex and occipital cortex (permutation test p<0.05, Figure 2j). Interestingly, in Participant 1, the distribution of high ΔGranger coincides with the distribution of high PLV: both show a spatial maximum to grid contacts over auditory cortex and both appear in the theta frequency range.
We next tested the statistical significance of the spatial spread of contacts with high ΔGranger ([4 8] Hz) during maintenance ([−2 0] s). To provide a sound statistical basis, we tested the spatial distribution of GC on the grid contacts against a null distribution. The activation on grid contacts was reshaped into a grid vector. The spatial collinearity of two grid vectors was captured by their scalar product. We next performed 200 iterations of random trial permutations. For each iteration, we selected two subsets of trials, and we calculated the scalar product between the two vectors corresponding to these subsets. We then tested the statistical significance of the scalar product ( Figure 2k). The true distribution (red) is clearly distinct from the null distribution (gray, blue bar marks the 95th percentile). The analogous procedure was performed for PSD ( Figure 2a and d), PLV (Figure 2g), and GC during encoding (Figure 2i), which gave equally significant results in all cases. 8] Hz) during encoding ([−5 −3] s). ECoG over auditory cortex predicts hippocampal local field potentials. (j) Net information flow ΔGranger ( [4 8] Hz) during maintenance ([−2 0] s). Hippocampus is maximal in predicting auditory cortex (contact C2 and surrounding contacts). (k) Statistical significance of the spatial spread of contacts with high ΔGranger ([4 8] Hz) during maintenance ([−2 0] s). We calculated the scalar product between two spread vectors. We then tested the statistical significance of the scalar product. The true distribution (red) is clearly distinct from the null distribution (gray, blue bar marks 95th percentile). (l) The Granger time-frequency map illustrates the time course of the spectra of panel (h). During encoding, net information (ΔGranger) flows from auditory cortex to hippocampus (blue). During maintenance, the information flow is reversed from hippocampus to auditory cortex (red) indicating the replay of letters in memory. Grid contacts with significant increase are marked with a yellow rim (p<0.05, cluster-based nonparametric permutation test against a null distribution with scrambled trials). The time course in time-frequency maps is shown relative to the fixation period (b, c, e). Colors of Granger spectra indicate information flow: dark blue, cortex to hippocampus during encoding; light blue, hippocampus to cortex during encoding; dark red, hippocampus to cortex during maintenance; light red, cortex to hippocampus during maintenance. ΔGranger is the difference between spectra, where ΔGranger <0 denotes information flow cortex→hippocampus and ΔGranger >0 denotes information flow hippocampus→cortex. Grid contacts are identified by column (anterior A to posterior H) and row (inferior 1 to superior 8).
Figure 2 continued
As a further illustration of the ΔGranger time course, the time-frequency plot (Figure 2l) shows the difference between GC spectra (GC hipp→cortex -GC cortex→hipp) at each time point, where blue indicates net flow from auditory cortex to hippocampus and red indicates net flow from hippocampus to auditory cortex. (b, c, g, h, l,m). Colors of Granger spectra indicate information flow: dark blue, cortex to hippocampus during encoding; light blue, hippocampus to cortex during encoding; dark red, hippocampus to cortex during maintenance; light red, cortex to hippocampus during maintenance. ΔGranger is the difference between spectra where ΔGranger <0 denotes information flow cortex→hippocampus and ΔGranger >0 denotes information flow hippocampus→cortex. Bars: frequency range of significant ΔGranger (p<0.05), clusterbased non-parametric permutation test against a null distribution with scrambled trials during encoding and maintenance, respectively.
Similarly in Participant 2, the time course of GC followed the same pattern between auditory cortex (anterior strip electrode contact in Figure 3a) and hippocampus (Figure 3d and e). Among the participants that had both LFP and temporo-parietal ECoG recordings, Participant 3 had an electrode contact over left visual cortex; the sensory localization was indexed by the strong gamma activity in the most posterior contact of the strip electrode (Figure 3g). The time course of information flow between visual cortex and hippocampus (Figure 3i and j) followed the same pattern as described for the auditory cortex above. Interestingly, the pattern appeared with LFP recorded from right hippocampus in Participant 3 (Supplementary file 1). However, in Participant 4, the recordings from the right cortical hemisphere (Figure 3k) did not show significant GC between LFP and ECoG during task performance (Figure 3n and o).
Thus, we showed in recordings from the left cortical hemisphere that letters were encoded with information flow from sensory cortex to hippocampus; conversely, the information flow from hippocampus to sensory cortex indicated the replay of letters during maintenance.
Source reconstruction of the scalp EEG
We used beamforming (Oostenveld et al., 2011) to reconstruct the EEG sources during encoding and maintenance for each of the 15 participants ( Table 1). We tested whether the sources during fixation differed from sources during encoding and during maintenance (non-parametric cluster-based permutation t-test Maris and Popov et al., 2018). In each participant, the proportion of significant sources in the left hemisphere exceeded 80% of all significant sources. Across all participants, the spatial activity pattern during both encoding and maintenance showed the highest significance in frontal and temporal areas of the left hemisphere ( Figure 4-figure supplement 1).
Directed functional coupling between hippocampus and averaged EEG sources
The synchronization between hippocampal LFP and EEG sources (N=15 participants) confirmed the directed functional coupling found in the three participants with ECoG. We first calculated the GC between hippocampus and the EEG beamforming sources in the auditory cortex. We found that the mean GC spectra resembled the GC spectrum for ECoG in the theta frequency range ([4 8] Hz, Figure 4a). During encoding, the net information flow was from auditory cortex to hippocampus (light blue curve -dark blue curve, blue bar, p<0.05, group cluster-based permutation test). During maintenance, the net information flow was reversed (dark red curve -light red curve, red bar, p<0.05, group cluster-based permutation test), i.e., from hippocampus to auditory cortex. Interestingly, the pattern appeared with LFP recorded from the right hippocampus in several participants (Supplementary file 1). A similar GC pattern emerged when using the signals from left temporal scalp electrodes but was eliminated when using a Laplacian derivation. Thus, both for ECoG and EEG beamforming sources, GC showed the same bidirectional effect in theta between auditory cortex and hippocampus.
To explore the spatial distribution, we computed GC also for other areas of cortex. We averaged the net information flow (ΔGranger) in the theta range across the participants and projected it onto the inflated brain surface (Figure 4b and c). During encoding, the mean information flow was strongest from auditory cortex to hippocampus (ΔGranger=−0.049, p=0.0009, Kruskal-Wallis test, Figure 4b). For all other areas, the mean ΔGranger was also from cortex to hippocampus but the effect was weaker (mean ΔGranger = [-0.03 0], Dunn's test, Bonferroni corrected). During maintenance ( Figure 4c) the information flow was reversed. While all areas had information flow from hippocampus to cortex (ΔGranger = [0.02], Dunn's test, Bonferroni corrected), the strongest flow appeared from hippocampus to auditory cortex (ΔGranger=0.034, p=0.001, Kruskal-Wallis test).
Directed functional coupling and the participants' performance
The reversal of ΔGranger appeared in all 15 participants individually (Figure 4d). We averaged ΔGranger for each participant in the [4 8] Hz theta frequency range. The ΔGranger between hippocampus and auditory cortex, was negative during encoding and was positive during maintenance in the theta frequency range (p=4.1e-10, paired permutation test). The directionality and its reversal was missing for all other areas, e.g., lateral prefrontal cortex (p=0.16, paired permutation test, Figure 4e). Of note, all analyses up to here were performed on correct trials only.
M ai nt en an ce
En co di ng g Figure 4. Granger causality (GC) between hippocampal local field potentials (LFP) and EEG sources. (a) Spectral GC between hippocampal LFP and auditory EEG sources, averaged over all N=15 participants. The shaded area indicates the variability across the population. During encoding, the net Granger (ΔGranger) indicates information flow from auditory cortex to hippocampus ([6 10] Hz, blue bar). During maintenance, ΔGranger indicates information flow from hippocampal LFP to auditory cortex (red bars, [6 9] Hz, [13 15] Hz). Bars: frequency range of significant ΔGranger (p<0.05), group cluster-based non-parametric permutation t-test against a null distribution with scrambled trials during encoding and maintenance. Colors of Granger spectra indicate information flow: dark blue, cortex to hippocampus during encoding; light blue, hippocampus to cortex during encoding; dark red, hippocampus to cortex during maintenance; light red, cortex to hippocampus during maintenance. (b) The median net information flow (ΔGranger) in the [4 8] Hz range during encoding is projected onto an inflated brain surface. The maximal ΔGranger appeared from temporal superior gyrus (median ΔGranger=-0.049) indicating information flow from auditory cortex to hippocampus. Negative values of median ΔGranger appeared also in other areas, albeit less intense (ΔGranger>-0.03). (c) The median net information flow (ΔGranger) in the [4 8] Hz range during maintenance is projected onto an inflated brain surface. The maximal ΔGranger appeared from temporal superior gyrus (median ΔGranger=0.034) indicating an information flow from hippocampus to auditory cortex. Positive values of median ΔGranger appeared also in other areas, albeit less intense (ΔGranger <0.02). (d) The maximal ΔGranger in the [4 8] Hz range was negative during encoding (blue, auditory cortex → hippocampus, median ΔGranger=-0.049) and positive during maintenance (red, hippocampus → auditory cortex, median ΔGranger=0.034) for each participant (red and blue connected marker, paired permutation
Figure 4 continued on next page
Finally, we established a link between the participants' performance and ΔGranger. For incorrect trials, the net information flow ΔGranger from auditory cortex to hippocampus did not show the same directionality in all participants and did not reverse in direction (p=0.37, paired permutation test, Figure 4f). Since participants performed well (median performance 91%), we balanced the numbers of correct and incorrect trials. We calculated the GC in a subset of correct trials (median of 200 permutations of a number of correct trials that equals the mean percentage of incorrect trials=10%); the effect was equally present for the subset of correct trials (p<0.0005). This suggests that timely information flow, as indexed by GC, is relevant for producing a correct response.
Discussion
WM describes our capacity to represent sensory input for prospective use. Our findings suggest that this cognitive function is subserved by bidirectional oscillatory interactions between memory neurons in the hippocampus and sensory neurons in the auditory cortex as indicated by phase synchrony and GC. In our verbal WM task, the encoding of letter items is isolated from the maintenance period in which the active rehearsal of memory items is central to achieve correct performance. First, analysis of task-induced power showed sustained oscillatory activity in cortical and hippocampal sites during the maintenance period. Second, analysis of the inter-electrode phase synchrony and the directional information flow showed task-induced interactions in the theta band between cortical and hippocampal sites. Third, the directional information flow was from auditory cortex to hippocampus during encoding, and during maintenance, the reverse flow occurred from hippocampus to auditory cortex. This pattern was found only to the left cortical hemisphere, as expected for a language-related task. Fourth, the comparison between correct and incorrect trials suggests that the participants relied on timely information flow to produce a correct response. Our data suggests a surprisingly simple model of information flow within a network that involves sensory cortices and hippocampus (Figure 4g): during encoding, letter strings are verbalized as subvocal speech. The incoming information flows from sensory cortex to hippocampus (bottom-up). During maintenance, participants actively recall and rehearse the subvocal speech in their phonological loop (Baddeley, 2003;Christophel et al., 2017). The GC indicates the information flow from hippocampus to cortex (top-down) as the physiological basis for the replay of the memory items, which finally guides action.
The current study is embedded in previous studies using the same or similar tasks. Persistent firing of hippocampal neurons indicated hippocampal involvement in the maintenance of memory items (Boran et al., 2019;Kamiński et al., 2017;Kornblith et al., 2017). An fMRI study reports salient activity in the auditory cortex during maintenance in an auditory WM task (Kumar et al., 2016), which indicates that sensory cortical areas are involved in the maintenance of WM items. During encoding, the activity of local assemblies was associated with gamma frequencies and local processing (Figure 2a b c, Figure 3g l) while GC inter-areal interactions took place in theta frequencies, in line with previous reports (Solomon et al., 2017;von Stein and Sarnthein, 2000). Parietal generators of theta-alpha EEG indicated involvement of parietal cortex in WM maintenance (Michels et al., 2008;Tuladhar et al., 2007;Näpflin et al., 2008;Boran et al., 2019;Boran et al., 2020). The hippocampo-cortical phase synchrony (PLV) was high during maintenance of the high workload trials (Boran et al., 2019). Building on these previous studies, the current study focused on high workload trials and extended them by the analysis of directional information flow. test, correct trials only). The mean values and statistical significance derive only from 10% of the correct trials in order to balance the number of incorrect trials. (e) The net information flow between hippocampal LFP and lateral prefrontal cortex in the [4 8] Hz range has a lower median than to auditory cortex and higher variability (correct trials only, p=0.16, paired permutation test, not significant). (f) For incorrect trials, the maximal ΔGranger in the [4 8] Hz range is highly variable (p=0.37, paired permutation test, not significant). (g) Bidirectional information flow between cortical sites and hippocampus in the working memory network. The GC analysis suggests a surprisingly simple model of information flow during the task. During encoding, letter strings are verbalized as subvocal speech; the incoming information flows from auditory cortex to hippocampus. During maintenance, participants actively recall and rehearse the subvocal speech in the phonological loop; GC indicates an information flow from hippocampus to cortex as the physiological basis for the replay of the memory items.
The online version of this article includes the following figure supplement(s) for figure 4:
Figure 4 continued
In the design of the task, we aimed to separate in time the encoding of memory items from their maintenance. In the choice of the 2 s duration for the encoding period were guided by the magic number 7±2, which may correspond to 'how many items we can utter in 2 s' (Baddeley, 2003;Christophel et al., 2017). The median Cowan's K=6.1 shows that high workload trials were indeed demanding for the participants, where both encoding and maintenance may limit performance. We therefore presented the letters both as a visual and an auditory stimulus. Certainly, maintenance processes are likely to appear already during the encoding period as maintenance neurons ramp up their activity already during encoding (Boran et al., 2019). Furthermore, encoding may extend past the visual stimulus (t=-3 s). We therefore focused our analysis on the last 2 s of maintenance [-2 0] s. With this task design, we found patterns of GC that were clearly distinct between encoding and maintenance.
Our study capitalizes on a unique dataset. We first benefitted from direct cortical recordings that assured the neuronal origin of the signals. We then confirmed the GC results by using the wide spatial coverage of scalp EEG, where we used beamforming to localize the cortical sources that generate the scalp EEG. The interaction between recordings from different brain regions has to be discussed with respect to volume conduction (Trongnetrpunya et al., 2015). On the recording level, the choice of two separate references for LFP and ECoG has been shown to avoid spurious effects in GC (Bastos and Schoffelen, 2015). On the level of scalp EEG analysis, we used beamforming as a source reconstruction technique (Popov et al., 2018) to characterize the primary neuronal generators that were localized specifically in left auditory cortex. A similar GC pattern emerged when using the signals from left temporal scalp electrodes but it was eliminated when using a Laplacian derivation. When looking at the GC frequency spectra, there was a strong frequency dependence of GC from hippocampus to ECoG (Figure 2h, Figure 3d i). Likewise, GC to EEG sources showed a strong frequency dependence (Figure 4a). This speaks against volume conduction because the transfer of signal through tissue by volume conduction is independent of frequency in the range of interest here (Miceli et al., 2017). Finally, there was a strong task dependence of GC (Figure 2h, Figure 3d i, Figure 4a d), again speaking against a strong contribution of volume conduction.
In the literature, there are several studies investigating the WM network. However, only few report directional interactions. One of these (Johnson et al., 2018a) reports cross-spectral directionality between intracranial recordings in frontal cortex and the medial temporal lobe in theta frequencies.
One study on episodic memory suggests directional information flow to and from hippocampus (Griffiths et al., 2019). Within hippocampus, directional information flow from posterior to anterior hippocampus indicated successful WM maintenance (Li et al., 2022). The frequencies of GC found in the current study were in the ([4 8] Hz) theta band, in line with scalp EEG findings during WM tasks (Sarnthein et al., 1998;Polanía et al., 2012) and other tasks (Solomon et al., 2017) that activate oscillations in long-range recurrent connections (Fries, 2015;Pesaran et al., 2018).
Taken together, our results corroborated earlier findings on the WM network and extended them by providing a physiological mechanism for the active replay of memory items.
Materials and methods Task
We used a modified Sternberg task in which the encoding of memory items and their maintenance was temporally separated (Figure 1a). Each trial started with a fixation period ([−6, -5] s), followed by the stimulus ([−5, -3] s). The stimulus consisted of a set of eight consonants at the center of the screen. The middle four, six, or eight letters were the memory items, which determined the set size for the trial (4, 6, or 8 respectively). The outer positions were filled with 'X', which was never a memory item. The participants read the letters and heard them spoken at the same time. After the stimulus, the letters disappeared from the screen, and the maintenance interval started ([−3, 0] s). Since the auditory encoding may have extended beyond the 2 s period, we restrict our analysis to the last 2 s of the maintenance period ([−2, 0] s). A fixation square was shown throughout fixation, encoding, and maintenance. After maintenance, a probe was presented. The participants responded with a button press to indicate whether the probe was part of the stimulus. The participants were instructed to respond as rapidly as possible without making errors. After the response, the probe was turned off, and the participants received acoustic feedback regarding whether the response was correct or incorrect. The participants performed sessions of 50 trials in total, which lasted approximately 10 min each. Trials with different set sizes were presented in a random order, with the single exception that a trial with an incorrect response was always followed by a trial with a set size of 4. The task can be downloaded at http://www.neurobs.com/ex_files/expt_view?id=266.
Participants
The participants in the study were patients with drug resistant focal epilepsy. To investigate a potential surgical treatment of epilepsy, the patients were implanted with intracranial electrodes. Electrodes were placed according to the findings of the non-invasive presurgical evaluation, where the epileptologists hypothesized the epileptic foci to be localized (Zijlmans et al., 2019). Since the presumed epileptic foci included the hippocampus in all patients, electrodes were placed in the hippocampus. In four patients, additional electrodes were placed on the cortex because an epileptic focus in the cerebral cortex was considered. The participants provided written informed consent for the study, which was approved by the institutional ethics review board (PB 2016-02055). The participants were right handed and had normal or corrected-to-normal vision. For nine participants (5-14), the PSD and PLV have been reported in an earlier study (Boran et al., 2019).
Electrodes for LFP, ECoG, and EEG
The depth electrodes (1.3 mm diameter, eight contacts of 1.6 mm length, spacing between contact centers 5 mm, Ad-Tech, adtechmedical.com) were stereotactically implanted into the hippocampus for LFP recording. Subdural grid and strip electrodes (platinum electrode contacts with 4 mm 2 contact surface and 1 cm inter-contact distance, Ad-Tech) were placed directly on the cortex for ECoG recordings. For scalp EEG recording, cup electrodes (Ag/AgCl) were placed on the scalp and filled with electrolyte gel (Signagel, Parker Laboratories) to obtain an impedance <5 kΩ.
Electrode localization
The stereotactic depth electrodes were localized using post-implantation CT and post-implantation structural T1-weighted MRI scans. The CT scan was registered to the post-implantation scan as implemented in FieldTrip (Stolk et al., 2018). A fused image of CT and MRI scans was produced and the electrode contacts were marked visually. The position of the most distal hippocampal contact was projected in a hippocampal surface (Figure 1d, Figure S1).
To localize the ECoG grids and strips, we used the participants' postoperative MRI, aligned to CT and produced a 3D reconstruction of the participants' pial brain surface. Grid and strip electrode coordinates were projected on the pial surface as described in Groppe et al., 2017; Figure 2a and Figure 3a and f.
The scalp EEG electrodes were placed at the sites of the 10-20 system by experienced technicians and no further localization was performed. While the 10-20 standard is 21 scalp electrodes, in some patients some electrode sites stayed vacant to assure the sterility of the leads to the intracranial electrodes, resulting in a median of 17 scalp sites per patient.
Some of the intracranial electrode contacts were found in tissue that was deemed to be epileptogenic and that was later resected. Still, neurons in this tissue have been found to participate in task performance in an earlier study (Boran et al., 2019).
Recording setup, re-referencing, and preprocessing All recordings (LFP, ECoG, and scalp EEG) were performed with the Neuralynx ATLAS system (sampling rate 4000 Hz, 0.5 1000 Hz passband, Neuralynx, neuralynx.com). ECoG and LFP were recorded against a common intracranial reference. Signals were analyzed in MATLAB (Mathworks, Natick MA, USA). We re-referenced the hippocampal LFP against the signal of a depth electrode contact in white matter. We re-referenced the cortical ECoG against a different depth electrode contact. The choice of two separate references for LFP and ECoG has been shown to avoid spurious functional connectivity estimates (Bastos and Schoffelen, 2015). The scalp EEG was recorded against an electrode near the vertex and was then re-referenced to the averaged mastoid channels. All signals were downsampled to 500 Hz. All recordings were done at least 6 hr from a seizure. Trials with large unitary artifacts in the scalp EEG were rejected. We focused on the trials with high workload (set sizes 6 and 8) for further analysis. We used the FieldTrip toolbox for data processing and analysis (Oostenveld et al., 2011).
Power spectral density
We first calculated the relative PSD in the time-frequency domain (Figure 2b). Time-frequency maps for all trials were averaged. We used 3 multitapers with a window width of 10 cycles per frequency point, smoothed with 0.2×frequency. We computed power in the frequency range [4 100] Hz with a time resolution of 0.1 s. The PSD during fixation ([−6 -5]
Phase-locking value
To evaluate the functional connectivity of hippocampus and cortex, we calculated the PLV between hippocampal LFP channels and ECoG grid (multitaper frequency transformation with two tapers based on Fourier transform, frequency range [4 100] Hz with frequency resolution of 1 Hz).
where PLVi,j is the PLV between channels i and j, N is the number of trials, X(f) is the Fourier transform of x(t), and (•)* represents the complex conjugate.
Using the spectra of the 2-s epochs, phase differences were calculated for each electrode pair (i,j) to quantify the inter-electrode phase coupling. The phase difference between the two signals indexes the coherence between each electrode pair and is expressed as the PLV. The PLV ranges between 0 and 1, with values approaching 1 if the two signals show a constant phase relationship over all trials. In
Source reconstruction of the EEG sources
We reconstructed the scalp EEG sources using linearly constrained minimum variance (LCMV) beamformers in the time domain. To solve the forward problem, we used a precomputed head model template and aligned the EEG electrodes of each participant to the scalp compartment of the model via interactive scaling, translation, and rotation (ft_electrode_realign.m). We then computed the source grid model and the leadfield matrix, wherein we determined the grid locations according to the brain parcels of the automated anatomical atlas (AAL) (Tzourio-Mazoyer et al., 2002). We solved the inverse problem by scanning the grid locations using the LCMV filters separately for encoding and maintenance. The EEG sources were baselined with respect to the fixation period and presented as a percent of change from the pre-stimulus baseline. We defined cortical areas from multiple parcels since AAL is a parcellation based on sulci and gyri. We performed all the steps of the source reconstruction with FieldTrip (Oostenveld et al., 2011) and projected the sources onto an inflated brain surface.
Spectral Granger causality
In order to evaluate the direction of information flow between the hippocampus and the cortex, we calculated spectral non-parametric GC as a measure of directed functional connectivity analysis (Oostenveld et al., 2011). We evaluated the direction of information flow in the (Sarnthein et al., 1998;Li et al., 2022;[4 20]) Hz frequency range. To compute the GC, we first downsampled the signals to the Nyquist frequency=40 Hz. We then computed the GC between hippocampal contacts and ECoG grid contacts. We also computed GC between the same hippocampal contacts and EEG sources located over the regions of interest. GC examines if the activity on one channel can forecast activity in the target channel. In the spectral domain, GC measures the fraction of the total power that is contributed by the source to the target. We transformed signals to the frequency domain using the multitaper frequency transformation method (two Hann tapers, frequency range [4 20] Hz with 20 s padding) to reduce spectral leakage and control the frequency smoothing.
We used a non-parametric spectral approach to measure the interaction in the channel pairs at a given interval time (Bastos and Schoffelen, 2015). In this approach, the spectral transfer matrix is obtained from the Fourier transform of the data. We used the FieldTrip toolbox to factorize the transfer function H(f) and the noise covariance matrix Σ. The transfer function and the noise covariance matrix were then employed to calculate the total and the intrinsic power, S(f)=H(f)ΣH×(f), through which we calculated the Granger interaction in terms of power fractions contributed from the source to the target.
where S xx (f) is the total power and S xx(f) is the instantaneous power. To average over the group of participants, we calculated the Granger spectra for the selected channel pairs and averaged these spectra over participants (Figure 4a).
To illustrate the time course of GC over time, we calculated time-frequency maps with the multitaper convolution method of Fieldtrip (Oostenveld et al., 2011).
Statistics
To analyze statistical significance, we used cluster-based non-parametric permutation tests. To assess the significance of the difference of the Granger between different directions, we compared the difference of the true values to a null distribution of differences. We recomputed GC after switching directions randomly across trials, while keeping the trial numbers for both channels constant. Then we computed the difference of GC for the two conditions. We repeated this n=200 times to create a null distribution of differences. The null distribution was exploited to calculate the percentile threshold p=0.05. In this way, we compare the difference of the dark and light spectra against a null distribution of differences. We mark the frequency range of significant GC with a blue bar for encoding (dark blue spectrum exceeds light blue spectrum, information flow from cortex to hippocampus) and with a red bar for maintenance (dark red spectrum exceeds light red spectrum, information flow from hippocampus to cortex).
To test the statistical significance of the spatial spread of contacts with high PSD, PLV, or ΔGranger, we calculated the spatial collinearity on the grid contacts against a null distribution. First, we transform the activation on grid contacts into a grid vector. We then performed 200 iterations of random trial permutations. For each iteration, we selected two subsets (50%) of trials and we calculated the scalar product between the vectors corresponding to the two subsets. The null distribution was created by randomly mixing trials from the two task periods fixation and encoding. We finally tested the statistical significance of the scalar product. The true distribution was established to be statistically distinct from the null distribution if it exceeded the 95th percentile of the null distribution.
We assess if the reconstructed EEG sources during encoding and maintenance are significantly different from the pre-stimulus baseline (fixation). We use the FieldTrip's method ft_sourcestatistics (Oostenveld et al., 2011), wherein we apply a non-parametric permutation approach to quantify the spatial activation pattern during the encoding of the memory items and their active replay.
Due to high average performance of the participants (91%) the number of correct and incorrect trials is imbalanced. To balance the number of correct trials with the number of incorrect trials, we randomly selected 10% of the correct trials and recomputed the GC spectra and then the net information flow (ΔGranger). We repeated this n=200 times and presented the mean ΔGranger for each participant.
For comparisons between two groups, we used the non-parametric paired cluster-based permutation test. We created a null distribution by performing N=200 random permutations.
To test the directionality of the information flow in the group of the participants, we used the group cluster-based permutation t-test from the FieldTrip toolbox (Oostenveld et al., 2011) with multiple comparison correction using the false discovery rate approach. Statistical significance was established at p<0.05. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
Data availability
All data needed to evaluate the conclusions in the paper are present in the paper. The codes used to produce the results in the paper are freely available at the repository https://github.com/vdimakopoulos/verbal_working_memory. The task can be downloaded at http://www.neurobs.com/ex_files/ expt_view?id=266. Part of the data has been published earlier [36]. Additional data and code are indexed in http://www.hfozuri.ch/.
The following previously published dataset was used: | 9,916.2 | 2022-08-12T00:00:00.000 | [
"Biology",
"Psychology"
] |
A new species of Pseudochalcura ( Hymenoptera , Eucharitidae ) , with a review of antennal morphology from a phylogenetic perspective
Pseudochalcura alba Heraty & Heraty, sp. n. is described from Santiago del Estero and Catamarca provinces in northwestern Argentina. Th e male and female have long dorsal rami on all of the fl agellomeres basal to the terminal segment, which is a unique feature within the genus and shared only with some species of Rhipipalloidea. Antennal modifi cations are compared across the Stilbula clade, of which all are parasitoids of Camponotini (Formicinae). A phylogenetic hypothesis for the group is proposed based on an analysis of 28S and 18S sequence data for 28 species. Ramose antennae are derived independently in both males and females across the clade, but with fully ramose female antennae restricted to the New World prolata group of Pseudochalcura and to some species of the Old World genus Rhipipalloidea. A sister group relationship between these genera is proposed based on both morphological and molecular data. Female antennae in other species of these genera, and other genera in the clade are at most dorsally lobate or serrate, but more commonly cylindrical. Monophyly of species of Obeza and Lophyrocera is supported and linked to a behavioral trait of oviposition into fruits as opposed to fl ower heads or leaf buds. Within the Stilbula clade, a dichotomy between New and Old World taxa suggest relatively recent post-Miocene exchanges across the Northern Hemisphere.
Introduction
Eucharitidae (Hymenoptera: Chalcidoidea) are parasitoids of ant pupae (Clausen 1940, Heraty 2002).A variety of methods are employed to gain access to the ant brood, but all involve oviposition outside of the nest and the active transport of the fi rst-instar planidial larva.Th e Stilbula clade is a distinct group within the Eucharitini (Eucharitidae: Eucharitinae), with all known host records from Camponotini (Formicinae) (Clausen 1940, Heraty 2002).Oviposition within this group is either in large egg masses that resemble fruit, or into the skin of fruits themselves, thus introducing larvae into the foraging proximity of their hosts (Clausen 1940;Heraty andBarber 1990, Torréns et al. 2008).Based on morphology, the clade is either monophyletic or paraphyletic (Heraty 2002), but monophyletic for either molecular data alone (28S and 18S) or in combination with morphological data (Heraty et al. 2004).Pseudochalcura exemplify the behavior typical of the clade and are the focus of this paper, along with the description of a species with unusual antennal features for both the genus and family.
Pseudochalcura are comprised of 10 recognized species that range in the New World from Chile and Argentina to the Yukon and Alaska, although several new species have been discovered since their fi rst revision (Heraty 1986, JMH unpublished).Pseudochalcura gibbosa (Provancher) oviposit masses of 1000-2000 eggs into leaf buds of various Ericaceae and Malvaceae (Cook 1905;Pierce and Morrill 1914;Clausen 1940;Heraty & Barber 1990).Eggs of this species overwinter in the leaf bud and likely fall to the ground as the buds expand and the bud scales drop in the spring.Simultaneous hatching of the egg mass upon stimulation by the host ant is seen as a means of both attracting ants (recruitment) and for gaining transport back to the ant brood as a food resource (Heraty and Barber 1990).First-instar planidial larvae of P. gibbosa initially attack larvae of Camponotus novaeboracensis (Fitch), with development completed by one to three individuals on a single host pupa (Wheeler 1910;Heraty and Barber 1990).Th e only other confi rmed host record is for P. nigrocyanea Ashmead from Brazil, for which adults were observed exiting from a nest of Camponotus and ovipositing into fl ower buds of an unidentifi ed Rosaceae (Heraty unpublished).Deposition of egg masses into fl ower buds is shared with Stilbula (Clausen 1940) and Substilbula (Heraty, unpublished), and is likely a plesiomorphic behavior for the Stilbula clade with oviposition into fruit regarded as a derived behavior (Heraty and Barber 1990;Torréns et al. 2008).
Th e form of the antennal fl agellum in Eucharitidae is highly variable, ranging from cylindrical or lobate, serrate or ramose in both males and females (Heraty 2002).Antennae may be simple to serrate, with single long rami on the fl agellomeres, or with paired elongate rami as found in some species of Saccharissa and Chalcura (Heraty 2002).Antennal rami can be either cylindrical, fl attened, or even antler-like as in Tricoryna alcicornis Bouček (Bouček 1988;Heraty 2002).Some genera even have an increased number of fl agellomeres, with Eucharissa erugata Heraty having as many as 18.Morphological diversity in the antennal fl agellum is restricted to the tribe Eucharitini, with all other Eucharitidae (Oraseminae, Gollumiellinae, and Psilocharitini [Eucharitinae]) having simple, cylindrical fl agellomeres.Ramose antennae are independently developed in various clades within Eucharitini, and, in males, often fi xed within larger groups such as in virtually all of the poneromorph parasitoids.Similar morphological changes are rare within females, which often have at most dorsally serrate antenna.Th e Stilbula clade is unique in its extreme level of antennal variation which is paralleled in both males and females.Also notable is a similar amount of antennal variation within Pseudochalcura.Only within the Stilbula clade do the female antennae have the same diversity of form as the males.Th e variation has not been examined previously from a phylogenetic perspective in a closely related group to observe whether changes are homoplastic or synapomorphic.Herein we describe the extreme variation possible in a new species and an undescribed species of Rhipipalloidea, as well as resolution of these changes on a phylogeny proposed for members of the Stilbula clade.We also discuss changes in their behavior and correlated biogeographic considerations.
Materials and methods
Terms for descriptive purposes follow Heraty (2002).Images were obtained using GT-Vision® Ento-Vision software operating on a Leica M16 zoom lens linked to a JVC KY-F75U 3-CCD digital video camera.Images were enhanced with Adobe Photoshop CS2.Museums for deposition include the Entomology Research Museum, University of California Riverside (UCRC) and the Instituto Fundación Miguel Lillo, Tucumán, Argentina (IFML).Images for antenna of other species of are based on a variety of material, with information on locality and specimen deposition available from the senior author upon request.Th e undescribed species of Rhipipalloidea is from Tenompok (Sabah, Malaysia) and is deposited in the Bishop Museum.Illustrations of antennae in Fig. 23 were redrawn from Heraty (1985Heraty ( , 2002)).
Molecular Analyses -Th irty-four individuals were sequenced representing 28 species of Eucharitinae (Eucharitidae), with three species of Pseudometagea selected as the outgroup (Table 1).Pseudometagea are parasitoids of Lasiini (Formicinae) and restricted to the Nearctic Region and Central America (Heraty 1985).Th e remaining 7 genera are included in the Stilbula clade of parasitoids of Camponotini (Heraty 2002).Conspecifi c individuals with identical sequence were eliminated from the analysis but indicated on Fig. 23.Th e complete matrix included 30 terminal taxa.Voucher specimens, except where indicated as "not available", are deposited at UCRC (Table 1).
Analysis -Parsimony analyses were implemented using PAUP*4.0b10 (Swoff ord 2002) using 500 random-addition heuristic searches and TBR branch swapping.Uninformative sites were included in all analyses.Gaps were treated as missing.Trees were condensed for branches with a minimum length of zero.Successive approximation character weighting was applied to the resulting trees using the maximum value of the rescaled consistency index and a base weight of 1000, with the resulting tree rescaled to unity character weights and compared in length to the most parsimonious trees (Carpenter 1988;Heraty et al. 2004).Bootstrap values for parsimony analyses were evaluated with 1000 random replicates and 2 random addition searches per replicate.In addition, a second partitioned likelihood analysis was performed using RaxML with three partitions and 200 rapid bootstrap replicates (Stamatakis et al. 2008) as implemented on the CIPRES server (http://www.phylo.org/sub_sections/portal/).Each partition was analyzed under a separate GTR+I+Γ model as applied in Heraty et al. (2004).
Description
Pseudochalcura alba Heraty & Heraty, sp.n. urn:lsid:zoobank.org:act:97405532-D55F-449C-9303-2F8BDFE1137EFigs 1-9 Diagnosis.Th is is the only species of Pseudochalcura in which the male has fl agellomeres 1-9 each with a long branch, the female has a long branch on fl agellomeres 1-7, and both sexes have a 10-segmented fl agellum.Th is species is placed in the condyla group of species based on the bare wing veins, sculptured (strigate) gastral sternite 1 in the female, absence of a metatibial spur, and ramose female antenna (to couplet 11 in Heraty 1986).No other member of this group has a 12-segmented female antenna or completely ramose male antenna.Th e closest species is considered to be P. prolata Heraty (Argentina: Chaco Province), which has the basal 5-7 fl agellomeres of the female antenna each with a stout branch, and a maximum of 8-11 antennal segments (male unknown).Th e basal branch of the female of P. prolata is stouter and 5.4 times as long as the basal length, as compared to 6.2 times in P. alba.Male (Holotype).Length 2.35 mm.Black with faint metallic luster; scape, pedicel, mandible, petiole, and basal quarter of femora light brown; antennal fl agellum and rest of legs white.Wings hyaline, stigma clear.
Head 1.79 times as broad as high (Fig. 4).Posterior ocellar line (POL) 2.8 times lateral ocellar line (LOL); POL 3.3 times ocellar ocular line (OOL).Frons and face irregularly costate (Fig. 4); clypeus and supraclypeal area smooth; genal bridge emarginate behind the mouthparts.Eyes separated by 2.2 times their height.Malar space 1.1 times height of eye, malar depression absent.Apical tooth of mandible long and overlapping opposing genal area.Labrum not observed.Antenna 12-segmented (Fig. 4); scape short and cylindrical, not reaching to median ocellus; all fl agellomeres but the last with long branches ranging from 7.4 to 12.3 times as long as basal length, fl agellomere branches alternating in origin from the base and slightly fl attened; apical fl agellomere unbranched and 3.6 times longer than broad; fl agellomeres with dense short setae, and multiporous plate sensilla small and recessed into depressions of the antennal wall.Mesosoma mostly with areolate sculpture, interstices broad with verrucose sculpture (Figs 2, 3, 5).Mesoscutum 2.0 times as broad as long; scutoscutellar sulcus transverse and deeply foveate; scutellar-axillar complex 1.4 times as long as maximum width of scutellum.Propodeal disc slightly rounded with broad alveolate sculpture medially and a few scattered setae dorsally, disc laterally relatively smooth; callus and metapleuron alveolate.Femoral groove deeply impressed.Proepisternum rugulose, becoming smooth apically.Mesocoxa weakly sculptured and lacking a lateral carina; metacoxa mostly smooth, laterally with small scattered pits.Legs stout (Figs 1, 2, 3), metafemur expanded medially, 3.5 times as long as broad; with short sparse appressed setae later- Figures 2-9.Pseudochalcura alba sp.n.: 2-5 male holotype: 2 lateral habitus 3 mesosoma, lateral 4 head, frontal 5 mesosoma, dorsal 6-9 female paratype: 6 antenna, lateral 7 head, frontal 8 mesosoma, dorsal 9 habitus, lateral.Inset is magnifi cation of mesoscutal sidelobe.ally; metatibia with sparse semi-erect setae dorsally and dense appressed setae ventrally; metatibial spur absent.Forewing 1.9 times as long as broad; costal cell 0.31 times as long as wing, without setae; submarginal vein and basal area of wing bare; wing veins clear and diffi cult to discern beyond submarginal vein, stigma elongate oval and about 4 times as long as broad; disc of wing with microsetae ventrally.Hind wing broad and apically rounded, 3.4 times as long as broad.
Petiole 2.7 times as long as broad in lateral view, 1.6 times as long as metacoxa (Fig. 3); very slightly curved in profi le; bare with fi ne granulate sculpture.Gaster globose, fi rst tergite (Gt 1 ) 1.7 times as long as broad, smooth with few short setae dorsally; fi rst gastral sternite smooth.
Female (Paratype; diff erences from male).Length 3.72 mm.Dark brown to black with faint metallic luster; antenna, mandible, petiole and basal two thirds of femora brown; rest of legs light brown.
Mesosoma with areolate sculpture, interstices narrow and smooth (Figs 8, 9, and inset).Mesoscutum 2.1 times as broad as long; scutellar-axillar complex 1.1 times as long as maximum width of scutellum.Propodeal disc fl at with broad alveolate sculpture medially, disc laterally areolate (Fig. 9).Metafemur 3.4 times as long as broad.Forewing 2.1 times as long as broad; costal cell 0.31 times as long as wing.Hind wing 3.4 times as long as broad.
Phylogenetic Results
Only two genera of the Stilbula clade (sensu Heraty 2002) were not included in these analyses.Neostilbula (Madagascar) is now considered as a member of the Eucharis clade in the Eucharitini (Heraty unpublished).Specimens of Striostilbula (Australia) were not available for molecular analysis, although its placement in the Stilbula clade is based on weak morphological support and its inclusion is suspect (Heraty 2002).Parsimony analysis of the 7 genera in the Stilbula clade resulted in a single tree (247 steps, r.i.0.88; Fig. 23).Successive approximations weighting generated the same tree, suggesting the data provide stable results (Carpenter 1988).Bootstrap support was high for most nodes across the tree.Th e RAxML results provided nearly the same tree topology, but with Stilbuloida sister group to the rest of the Stilbula clade; however the RAxML bootstrap results supported a monophyletic Stilbuloida + Stilbula (100%; Fig. 23).Th e RAxML results also placed P. gibbosa as sister group to the P. americana clade, but with almost no bootstrap support (52% as compared to 51% for the placement on the parsimony tree (Fig. 23).For Pseudochalcura, the results presented in Fig. 23 are more concordant with the morphology-based phylogeny presented in Heraty (1986).Relationships within and among the other genera were the same in both results (Fig. 23).
Two groups occur within the Stilbula clade (Fig. 23): (A) Leurocharis (Australian), Substilbula (Australian), Rhipipalloidea (Indoaustralian) and Pseudochalcura (New World); weak parsimony bootstrap support (52%), but strong RAxML bootstrap support (99%); and (B) Stilbuloida (Australian), Stilbula (Old World), Obeza and Lophyrocera (both New World); strong parsimony bootstrap support (100%) but not monophyletic in RAxML.Genera in group B all share a bare callar region on the mesosoma, and usually have strong transverse carinae on the lower face.Sister group relationships between Rhipipalloidea + Pseudochalcura and Lophyrocera + Obeza have been proposed based on both morphological and molecular data (Heraty 2002;Heraty et al. 2004).Rhipipalloidea and Pseudochalura have the postgenae or hypostomae forming a complete bridge posterior to the mandibles in association with a reduction of the maxillary complex (Heraty 2002), and monophyly of Obeza and Lophyrocera share strong projections on the propodeum and callus (Heraty 2002).Based on morphology, Leurocharis was treated as either a sister group to all of these genera when monophyletic, or excluded when paraphyletic (Heraty 2002).Subsequent analyses of an even larger molecular data set consistently put all of these Stilbula clade genera together in a monophyletic group (Heraty unpublished).
Th e new species, P. alba, is placed together with P. prolata Heraty (Fig. 23, grey box).Pseudochalcura prolata was placed by Heraty (1986) with two other species, P. condylus Heraty and P. sculpturata Heraty, in the prolata species group based on six synapomorphies, including males having 7 basal rami on the antennal fl agellomeres, the fi rst gastral sternite striate, and lack of a metatibial spur, which are all features shared with P. alba.Pseudochalcura alba is the only male with rami on all fl agellar segments, including an elongation of the terminal segment (Fig. 4).Females of P. alba, P. sculpturata, and P. prolata have at least the basal fl agellomeres ramose (Figs 6,14,23g).Females of all other species of Pseudochalcura have either cylindrical or at most dorsally serrate antennae (Figs 10,12,16,23e).Based on the lack of costal cell setae, P. pauca was treated as a potential sister group of the prolata group, but it has serrate, not ramose, antennae (Fig. 23f ); the male is unknown (Heraty 1986).Pseudochalcura gibbosa, along with P. septuosa Heraty and P. atra Heraty, was proposed as the sister group of P. pauca + the prolata group (Heraty 1986), which is consistent with the current results (Fig. 23).Males of the P. americana clade herein and P. gibbosa have only the basal 4-6 fl agellomeres ramose (Figs 11,13).Based on the phylogenetic hypothesis presented in Fig. 23, in males, a fl agellum bearing 4-6 basal rami originating along the dorsal midline is considered plesiomorphic for the genus (Figs 11,13), whereas a fl agellum with 7-9 rami, with an alternating origin along the midline (Figs 4,15,17,18) is apomorphic within Pseudochalcura.
Females of Rhipipalloidea from Australia have either simple cylindrical fl agellomeres as in R. gruberi (Girault) (Fig. 19), dorsally serrate fl agellomeres as in R. mira (Girault) (Fig. 20), or completely ramose fl agellomeres as in several undescribed species from Sarawak (Fig. 21), Vietnam, and the Philippines.One undescribed species from Mudigere (western India) has dorsally serrate fl agellomeres.Males of all species of Rhipipalloidea have long dorsal rami on fl agellomeres 1-7 (Figs.22, 23d).Female rami always originate along the midline, whereas male rami have alternating sites of origin.Th us the range of antennal morphology in these Australasian species parallels the diversity found in their New World sister group.
Discussion
Th e two groups proposed on the basis of morphological and molecular evidence not only revise our understanding of the evolutionary history of the group, but also suggest evolutionary character transitions that are of particular interest in the Eucharitidae.Four unusual evolutionary changes across the group are of particular interest within the family: coloration of the head and mesosoma, fusion of the postgenal area behind the mouthparts, a shift in oviposition habits from leaf buds to fruit, and morphological changes in the male and female antennae.
A yellow-patterned coloration of the mesosoma with contrasting black to dark green or blue patches is rare within Eucharitidae and occurs only within the Stilbula clade and some species of the eucharitine genus Eucharis (Heraty 2002).Among related Chalcidoidea, a similar color pattern occurs only within the distantly related Philomidinae (tentatively placed within Perilampidae).Within the Stilbula group A, only Pseudochalcura sculpturata Heraty (prolata group, gray box, Fig. 23; P. prolata not included on tree) has a yellow and black patterning of the mesosoma, with other species either brown, black, or black with faint metallic tints.Each of the genera in group B (Stilbuloida, Stilbula, Obeza and Lophyrocera) have some or all species with a black and yellow pattern on the mesosoma (Heraty 2002).Among these species, the base color of the mesosoma is almost always yellow, with a general orange color found only in Obeza fl oridana (Ashmead) (Heraty 2002).Coloration can be highly variable, with individuals of a species at a single locality ranging from almost completely black or metallic blue to mostly yellow (e.g.Stilbula septentrionalis (Brues), Heraty 1985;Lophyrocera variabilis, Torréns et al. 2008).Based on outgroup comparison across Eucharitidae, a uniformly brown, black or metallic mesosoma is plesiomorphic.Th erefore a yellow In Pseudochalcura and Rhipipalloidea, the postgenae are completely fused as a transverse bridge posterior to their highly reduced maxillary complex.In Substilbula, the hypostomae are fused dorsally just below the foramen, but with the postgenae widely separated and the maxillary complex slightly reduced (fi g. 338, Heraty 2002).In Obeza, Lophyrocera, Leurocharis and Substilbula, the postgenae are strongly produced medially and narrowly separated, but never fused and the maxillary complex is not reduced (Heraty 2002).Only in Stilbuloida, Stilbula and Striostilbula are the hypostomal and genal lobes broadly separated and maxillary complex fully developed as in other Eucharitidae.Mouthparts are reduced in a few other eucharitids (e.g.Indosema indica Husain & Agarwal and Pseudometagea nefrens Heraty), but without a correlated extension of the postgenae.Outside of the Stilbula clade, only Orasema simulatrix Gahan (Eucharitidae: Oraseminae) has the postgenae strongly produced, but without the corresponding reduction of the maxillary complex.None of the hypotheses for fusion of the postgenae in the Stilbula clade, either herein or based entirely on morphology (Heraty 2002) support a linear transformation series for the fusion of the genae from broadly separated, to closely associated, to completely fused.Instead, multiple independent origins of each of the two derived character states is supported.
Obeza and Lophyrocera are the only eucharitids known to oviposit small batches of eggs (60-100) under the epidermal layer of small fruits; other members of the Stilbula clade oviposit into existing cavities within leaf or fl ower buds (Fig. 23; Heraty Heraty 1986Heraty , 2002)).Red line is development of dorsal rami on female antenna; blue line is development of dorsal rami on male antenna.Clades A and B marked.Species names are followed by country codes and DNA voucher number (see Table 1).1990;Heraty 2002;Torréns et al. 2008).Th e outgroup for our analysis, Pseudometagea, oviposit into cavities in the seed heads of Poaceae (Heraty and Darling 1984), and most Eucharitini oviposit into fl ower buds or scatter their eggs on the leaf surface (Heraty 2002).Fruit oviposition and the direct association with a frugivorous ant host is unique (Heraty and Barber 1990).Across the Stilbula clade behaviors range from oviposition by Pseudochalcura of large masses of eggs (>1000) into leaf buds (Heraty and Barber 1990;Heraty 2002), oviposition by Stilbuloida into the base of nectary-fi lled trumpet-shaped fl owers of Loranthus (Heraty 2002, unpublished), to a variety of behaviors in Stilbula such as oviposition of large masses (>10,000 eggs) into leaf buds or small masses on the side of wind-dispersed achenes (Clausen 1940).In all cases, there is either a direct association with fruit (apomorphic), an association of the eggs with ant-attractive fl ower nectaries, or with large egg-masses that resemble fruit and potentially attract the host ants (Heraty and Barber 1990).Th ese are considered as apomorphic behaviors for the entire Stilbula clade, but plesiomorphic to fruit oviposition, and all involve an association with recruitment behavior and the frugivorous habits of camponotine ants (Heraty and Barber 1990).
Antennal morphology across the Stilbula clade is highly diverse.Th e plesiomorphic condition, as based on comparison with the other subfamilies of Eucharitidae (Gollumiellinae and Oraseminae) as well as within Eucharitinae (Psilocharitini), is a simple cylindrical fl agellum (Heraty 2002).Pseudometagea all have a simple 8-10 segmented fl agellum and have been proposed as the sister group of the remaining Eucharitini (Heraty et al. 2004;Fig. 23).Within the Stilbula clade, Obeza, Stilbula, Stilbuloida and Substilbula have simple or at most lobate fl agellomeres in both sexes (Fig. 23c,i,j).Stilbuloida doddi is the only species with intraspecifi c variation in the antennae of females, ranging from simple to lobate (Fig. 23h).Basal fl agellomeres of females that are either slightly or strongly lobate (Figs 10,12,16) are considered plesiomorphic, whereas a basally or completely ramose fl agellum (Figs 6, 14, 23g) is apomorphic.Th e current hypothesis (Fig. 23) would suggest that long rami on the fl agellomeres of females (red line) is derived independently in the Asian species of Rhipipalloidea and in the prolata group of Pseudochalcura.Ramose male antennae have developed independently at least three times in the Stilbula clade (blue lines, Fig. 23).A dorsal ramus on at least the basal fl agellomeres is found in males of Rhipipalloidea, Pseudochalcura, Stilbuloida and Lophyrocera.In Leurocharis, the female has a simple fl agellum and the male has a dorsally strongly lobate fl agellum (Fig. 23b).Striostilbula females have simple fl agellomeres, whereas males have relatively short basal fl agellomeres similar to Stilbuloida (Heraty 2002).Females always have the projections oriented medially along the midline, as do males of Leurocharis, Stilbuloidea, Striostilbula and some Pseudochalcura.Th e dorsal origin of rami in males of Lophyrocera, Rhipipalloidea and some Pseudochalcura alternate along the dorsolateral and dorsomedial surfaces of each fl agellomere.Dorsal rami with linear or alternating dorsal origins have arisen independently in males of the Eucharis (uncommonly), Chalcura (commonly), and Kapala (all species) clades, but always with the rami along all segments, and never in females.
Only Pseudochalcura and Rhipipalloidea have developed ramose antennae in females.While ramose antennae in males are common, they have developed independ-ently in diff erent groups of Eucharitini, and often in diff erent forms.Across Eucharitini, there are similar diff erences in the origin of the rami, either medial or off set from the midline in males.Th e latter may be interpreted as a way of increasing the surface area for detecting pheromones.However it is not so easy to postulate why females would develop elongate rami.Th e question which remains is whether this is a pleiotropic eff ect linked to male antennal development, or a unique origin related to fi nding the accurate host plant for oviposition?Th e biology of both Rhipipalloidea and the prolata group of Pseudochalcura is unknown, and only further natural history insights from the fi eld will likely help to resolve these issues.
Th ere is strong support from both morphological and molecular data for a parallel split in both group A and B between the New World and Old World genera.In group A, there is a grade between the strictly Australian genera (Leurocharis and Substilbula), the Australasian Rhipipalloidea, and the New World Pseudochalcura.Rhipipalloidea extends north to the Philippines, whereas Pseudochalcura occurs as far north as Alaska and the Yukon (Heraty 1986(Heraty , 2002)).Th e chances for a recent exchange across a Beringian land connection appear likely.In group B, there is a split between the Australian Stilbuloida and the widespread Old World genus Stilbula, with the latter having an extensive northern distribution in Japan and western Russia (Heraty 2002).Lophyrocera and Obeza are restricted to the New World, with Obeza found only in subtropical areas of North America, but with Lophyrocera occurring as far north as the state of Washington in western North America.Again a recent faunal exchange seems likely.Based on molecular data, Carmichael (2003) predicted an age of divergence for the Stilbula clade ranging between 20-25 million years ago.Th is would fi t well with a late Miocene or later dispersal event of these diff erent groups within the Stilbula clade across the northern hemisphere (Davis et al. 2002;Renner et al. 2004).
Table 1
List of taxa, with DNA voucher codes (p = primary; s = secondary), collection localities, UCR museum specimen identifi ers (na = no voucher), and Genbank accession numbers. | 5,909.4 | 2009-09-14T00:00:00.000 | [
"Biology"
] |
Connecting METTL3 and intratumoural CD33+ MDSCs in predicting clinical outcome in cervical cancer
Methyltransferase-like 3 (METTL3) is a member of the m6A methyltransferase family and acts as an oncogene in cancers. Recent studies suggest that host innate immunity is regulated by the enzymes controlling m6A epitranscriptomic changes. Here, we aim to explore the associations between the levels of METTL3 and CD33+ myeloid-derived suppressor cells (MDSCs) in tumour tissues and the survival of patients with cervical cancer (CC). Specimens of paraffin embedded tumour from 197 CC patients were collected. The expression levels of METTL3 and CD33 were measured by immunohistochemical (IHC) staining. The clinical associations of the IHC variants were analysed by Pearson’s or Spearman’s chi-square tests. Overall survival (OS) and disease-free survival (DFS) were estimated by the Kaplan–Meier method and log-rank test. Hazard ratios (HRs) and independent significance were obtained via Cox proportional hazards models for multivariate analyses. METTL3 in CD33+ cells or CC-derived cells was knocked down by METTL3-specific siRNA, and MDSC induction in vitro was performed in a co-culture system in the presence of METTL3-siRNA and METTL3-knockdown-CC-derived cells compared with that of the corresponding controls. We found that tumour tissues displayed increased levels of METTL3 and CD33+ MDSCs compared with tumour-adjacent tissues from the same CC patients. Importantly, METTL3 expression was positively related to the density of CD33+ cells in tumour tissues (P = 0.011). We further found that the direct CD33+CD11b+HLA-DR− MDSC induction and tumour-derived MDSC induction in vitro were decreased in the absence of METTL3. The level of METTL3 in tumour microenvironments was significantly related to advanced tumour stage. The levels of METTL3 and CD33+ MDSCs in tumour tissues were notably associated with reduced DFS or OS. Cox model analysis revealed that the level of METTL3 in tumour cells was an independent factor for patient survival, specifically for DFS (HR = 3.157, P = 0.022) and OS (HR = 3.271, P = 0.012), while the CD33+ MDSC number was an independent predictor for DFS (HR: 3.958, P = 0.031). Interestingly, in patients with advanced-disease stages (II–IV), METTL3 in tumour cells was an independent factor for DFS (HR = 6.725, P = 0.010) and OS (HR = 5.140, P = 0.021), while CD33+ MDSC density was an independent factor for OS (HR = 8.802, P = 0.037). Our findings suggest that CD33+ MDSC expansion is linked to high levels of METTL3 and that METTL3 and CD33+ MDSCs are independent prognostic factors in CC.
Background
Cervical cancer (CC) is one of the most common tumours, ranking fourth for both incidence and mortality in women worldwide [1][2][3]. CC is the result of continuous infection with some strains of human papillomavirus (HPV), such as HPV16 and HPV18 [4,5]. Though there are abundant measures of prevention and cure, cervical cancer continues to exhibit high invasion and poor prognosis [6]. In the past decade, researchers worldwide have found that the expression levels of molecular markers in the tumour microenvironment could be an essential factor for cervical cancer (CC) growth and metastasis [7,8]. In addition to traditional prognostic factors, including age, WHO grade, TNM stage and clinical status, some of the molecular markers could be new predictors of CC prognosis [6,9,10]. However, there are no confirmed molecular markers for tumour progression or prognosis in CC patients. The related viral proteins E6 and E7 have been the focal points of research for the past several years [11,12]. In other words, easily detectable and meaningful molecular markers need to be confirmed.
Methyltransferase-like 3 (METTL3) is associated with N 6 -methyladenosine (m 6 A) RNA methylation, which is the most abundant modification ubiquitously occurring in eukaryotic mRNAs [13,14]. This modification regulates mRNA stability or translation and can affect many functions, such as immune cell differentiation, cell development, circadian periods and tumour growth [15,16]. In previous studies, METTL3 was found to have an adverse influence on acute myeloblastic leukaemia (AML), breast cancer (BC), ovarian carcinoma, bladder cancer (BC) and gastric cancer (GC) [17][18][19][20][21][22]. Additionally, m 6 A modifications are carried out by a combination of m 6 A methyltransferases (also named writers: METTL3, METTL14 and WTAP), m 6 A demethylases (also named erasers: FTO and ALKBH5) and specific RNA-binding proteins (also named readers: YTHDF1/2/3, HNRNPA2B1, IGF2BP1/2/3, eIF3 [22][23][24]. CD33-positive cells are usually defined as myeloidderived suppressor cells (MDSCs) with suppressive influence on human tumour tissues [25,26]. MDSCs in the tumour environment were confirmed to be an independent indicator of poor prognosis in patients with many solid tumours [25,27,28]. In our previous studies, the MDSC proportion was expanded in the tumour microenvironment and showed extensive negative regulatory function for antitumour immunity in malignancies [29][30][31]. Recent studies have indicated that the differentiation of myeloid cells is regulated by m 6 A methyltransferases, including METTL3 [22,32,33]. We hypothesized that MDSC expansion may be linked to the level of METTL3 in the microenvironment of CC.
In the present study, we detected the levels of METTL3 and CD33 + MDSCs in tumour specimens from 197 CC patients by immunohistochemical (IHC) staining. We observed increased levels of METTL3 and CD33 + MDSCs in tumour tissues and positive associations between the levels of METTL3 and CD33 + MDSCs. The high levels of METTL3 and CD33 + MDSCs in the CC tumour microenvironment were significantly associated with poor disease-free survival (DFS) and overall survival (OS) in CC patients. Importantly, METTL3 and CD33 + MDSCs were independent prognostic predictors for CC patients. These findings suggest that METTL3 and MDSCs contribute to the development of disease and that METTL3 may respond to MDSC expansion in tumour microenvironments in CC.
Patients and tissue samples
A total of one hundred ninety-seven CC patients who received therapy at Sun Yat-Sen University Cancer Center in Guangzhou, China, and who accepted medical follow-up that continued until 2019 were included. Paraffin tumour specimens from 197 CC patients were collected at Sun Yat-Sen University Cancer Center between 2008 and 2010. In this retrospective study, none of the patients received antitumour treatment before tumour tissue was obtained, and all 197 patients were histologically confirmed as having primary CC. As shown in Table 1
Immunohistochemistry (IHC) and immunofluorescence staining
Paraffin-embedded tissues were continuously sectioned at a thickness of 4 μm, and the immunohistochemistry kit was used according to the manufacturer's instructions. In brief, tissue sections were deparaffinized by xylene, rehydrated in graded alcohols and immersed in EDTA (PH 8.0). Microwave (95 °C 12 min) was applied for antigen retrieval, and samples were cooled to room temperature. The endogenous enzyme block reagent was used to block the activities of endogenous peroxidase. The goat serum was applied to block nonspecific binding sites at room temperature for 30 min. Primary antibodies, including rabbit anti-METTL3 antibody (1:400), rabbit anti-CD33 antibody (1:200), and rabbit mAb IgG control (1:200) were incubated at 37 °C for 1 h and developed with peroxidase. After staining by haematoxylin, images were taken under a microscope (NIKON ECLIPSE 80i). The expression of METTL3 on CD33 + cells was measured by immunofluorescence staining; DAPI was used to stain the nuclei. The images were taken with a fluorescence microscope (NIKON ECLIPSE C1).
The METTL3 expression level was scored in tumour cells in five to ten separate × 400 high-power fields (HPFs). We scored METTL3 expression in the tumour cells of each specimen using a semiquantitative immunoreactivity scoring system, which ranged from 0 to 12 18:393 and was equal to multiplication of the intensity of immunohistochemical staining (zero: no staining; one: weak staining; two: moderate staining; and three: strong staining) and the percentage of positive tumour cells (one: less than 25%; two: 25-50%; three: 50-75%; and four: more than 75%). The expression of CD33 was determined by counting CD33-positive cells from five to ten separate × 400 HPFs from the same patient. METTL3 expression level in tumour-infiltrating cells (TILs) was evaluated based on the mean percentage from five to ten separate × 400 HPFs from the same patient. These METTL3-and CD33-positive scores were determined separately by two pathologists. An isotype control IgG antibody was included.
Knockdown of METTL3 by siRNA
To knock down METTL3 in HeLa cells or CD33 + cells, we generated METTL3-specific siRNA (siMETTL3) with the help of RiboBio; a control-siRNA vector was also generated. The siRNAs were transiently transfected into HeLa or CD33 + cells by Lipofectamine ™ LTX Reagent with PLUS ™ Reagent according to the manufacturer's instructions. After 48 h, the cells were harvested for immunoblotting and MDSC induction. The sequences of siMETTL3 siRNAs were as follows: siMETTL3_001 5′-CAA GTA TGT TCA CTA TGA A-3′; siMETTL3_002 5′-GAC TGC TCT TTC CTT AAT A -3′; and.
MDSCs induction in vitro and fluorescence-activated cell sort (FACS) staining
Peripheral blood mononuclear cells (PBMCs) were derived from the peripheral blood of healthy donors by gradient centrifugation separation. The CD33 + cells were sorted by human CD33-antibody-linked magnetic beads. After isolation, the 1 × 10^6 CD33 + cells were seeded in a 24-well plate (outer well) and co-cultured with HeLa cells (inner well, at 1:2 ratio) with or without METTL3 knockdown in a Transwell system (3421, Corning, New York, NY, USA) for 48 h. The cells were harvested for FACS staining and detected by cytometry CytExpert software (Beckman Coulter, San Jose, USA) or immunoblotting. Cells were pipetted into single-cell suspensions and incubated with corresponding fluorescence-labelled antibodies according to the manufacturer's instructions. The flow cytometer used was a cytoFLEX (Beckman), and all data were analysed by the original analysis software provided with the flow cytometer (CytExpert). The CD33 + CD11b + HLA-DR − cells were defined as peripheral MDSCs in this study.
Immunoblotting
The harvested cells were lysed with pre-cooled RIPA buffer, and the proteins were quantified by a BCA protein assay kit (23227, Invitrogen, Carlsbad, USA) and then separated using a 10% SDS-PAGE. Proteins were transferred onto polyvinylidene difluoride membrane (IPVH00010; Millipore, Massachusetts, USA). The membrane was blocked with 5% milk and incubated with the corresponding primary antibodies at 4 °C overnight. Next, the membrane was incubated with HRP-coupled secondary antibodies at room temperature and detected using a West dura extended duration substrate.
Statistics
SPSS 19.0 software (SPSS Inc., Chicago, USA) was used to analyse all the data, and GraphPad Prism 7 software (La Jolla, USA) was used to obtain the curves. The median values were used as cut-off values to divide the patients into two groups (high level and low level). We used Pearson's chi-square test or Spearman's chi-square test to analyse the relationships between immunohistochemical variants in different cell populations and patients' clinical parameters. The relationships among the expression of METTL3 in tumour cells, METTL3 in tumour-infiltrating immune cells and CD33 in tumourinfiltrating immune cells were determined using Pearson's or Spearman's correlation coefficient and linear regression analyses. Cut-off selection was based on X-tile (Version 3.6.1, New Haven, USA). Then, we evaluated prognostic factors in univariate and multivariate analyses using the Cox proportional hazards model. In our research, *P < 0.05 was considered significant. Raw data of this article have been deposited in the Research Data Deposit (RDD) (www.resea rchda ta.org.cn) with accession number RDDB2020000943.
The level of METTL3 is positively linked to the number of CD33 + MDSCs and contributes to tumour development
In the present study, the levels of METTL3 and CD33 + MDSCs were examined in tumour specimens from 197 patients with CC by IHC. METTL3 was located in the nuclei of tumour cells and tumour-infiltrating immune cells, while CD33 + cells were scattered mainly in the tumour stroma ( Fig. 1a, b); isotype IgG was used as a control (Fig. 1c). Importantly, we found that CD33 and METTL3 co-localized in some tumour-infiltrated immune cells (Fig. 1d). We further demonstrated that the percentage of CD33 + CD11b + HLA-DR − peripheral MDSCs was increased in CC patients compared with healthy donors, as was the percentage of tumourderived CD33 + CD11b + HLA-DR − MDSCs compared with that of tumour-adjacent tissues (Fig. 1e, f, n = 3). Consistent with the increase in the MDSC population in CC patients, the level of METTL3 was increased in the peripheral and tumour-infiltrating immune cells compared with the corresponding controls ( Fig. 1g, h). Among the 197 patients with CC, the median survival time was 96 months (range: 0-120 months), and the 10-year DFS and 10-year OS rates were 88.83 and 86.80%, respectively (Additional file 1: Figure S1A and S1B). Table 2 shows the results of the relationships between clinicopathological features and immunohistochemical variants in different cell types in the tumour microenvironment. High METTL3 expression in the tumour and in tumour-infiltrating immune cells was linked to tumour stage (P = 0.040 and 0.020, respectively). In addition, we analysed the relationship between METTL3 expression in tumour cells and in tumourinfiltrating immune cells and the number of CD33 + MDSCs via Spearman's correlation coefficient and linear regression. The expression of METTL3 in tumour cells was positively correlated with that in tumour-infiltrating immune cells (R = 0.264, P < 0.001) (Fig. 2a). The number of CD33 + cells was positively correlated with the expression of METTL3 in tumour cells (R = 0.145, P = 0.041) and tumour-infiltrating immune cells (R = 0.182, P = 0.011) (Fig. 2b, c).
Moreover, we found that the METTL3 level in tumour cells was positively correlated with TILs in the early (R = 0.049, P = 0.012) and advanced stage (R = 0.129, P = 0.002) (Additional file 1: Figure S1E and S1F), while we found that in the advanced stage, the number of CD33 + cells was positively correlated with the METTL3 level in TILs (R = 0.088, P = 0.013) (Additional file 1: Figure S1J).
To further investigate the role of METTL3 in the regulation of MDSC expansion, we knocked down METL3 expression in CD33 + cells or HeLa cells. We found that CD33 + CD11b + HLA-DR − MDSCs and tumour-derived MDSCs were decreased when METTL3 was knocked down in CD33 + cells or HeLa cells ( Fig. 2d-g).
METTL3 and CD33 + MDSCs are independent factors for patient prognosis
As shown in Table 3, univariate analysis showed that in addition to lymph node involvement and clinical stage, high levels of METTL3 in tumour cells (HR: 4.244, P = 0.002) and in tumour-infiltrating immune The immunohistochemical staining for METL3 and CD33 CC specimens (× 400). c The isotype antibody IgG was included (× 400). d Immunofluorescence staining for METTL3 (red) and CD33 + (green) in CC specimens; the white arrows point to the METTL3 + and CD33 + cells. The images were taken by fluorescence microscope. HLA-DR − CD33 + CD11b + cells were gated by a FACS gating strategy and were defined as MDSCs in this study. e, f Representative density plots showed the MDSC population in the peripheral blood of healthy donors (HD) or CC patients, as well as in the immune cells from tumour tissues (TIL) or tumour-adjacent tissues (NIL). A statistical graph is included for the comparison between the indicated groups. (G-H) Representative immunoblotting shows the expression of METTL3 in the peripheral blood, TILs and NILs. A statistical graph is included for the comparison between the indicated groups. The experiments in e, f were performed at least three times, and the data were plotted as the mean ± SEM. Statistics were conducted with an unpaired Student's t test, *P < 0.05, and ***P < 0.001 vs. the corresponding control (Table 3). When we performed multivariate Cox proportional hazards regression analysis in Table 4 (Table 4).
METTL3 and CD33 + MDSCs have predictive value for patients with early and advanced disease stages
We further divided the 197 patients into two subgroups based on the clinicopathological stage: 127 of the total patients were in early disease stage (stage I), while 70 of the total patients were in advanced disease stage (stage II-IV). Through the Kaplan-Meier method, we found that the high expression of METTL3 in tumour-infiltrating immune cells was significantly correlated with poor DFS (P = 0.033) and OS (P = 0.019) (Additional file 2: Figure S2C and S2D) in patients with early disease stage, while there was no significant association between the high expression of METTL3 in tumour cells (P = 0.400 vs P = 0.183) and the number of CD33 + MDSCs (P = 0.393 vs P = 0.227) with the DFS and OS of patients with earlystage disease (Additional file 2: Figure S2A, S2B, S2E and S2F). For patients with advanced-stage disease (n = 70), a high level of METTL3 in tumour cells was dramatically correlated with decreased DFS (P < 0.001, Fig. 4a) and OS (P < 0.001, Fig. 4b), and a high level of METTL3 in tumour-infiltrating immune cells was negatively correlated with DFS (P = 0.004, Fig. 4c) and OS (P < 0.001, Fig. 4d); the increased number of CD33 + MDSCs was dramatically correlated with poor DFS (P < 0.001, Fig. 4e) and OS (P < 0.001, Fig. 4f ). Using multivariate Cox The association between METTL3 expression in tumour cells and intratumoural CD33 + MDSC number (R = 0.145, P = 0.041). c The association between METTL3 expression in TILs and intratumoural CD33 + MDSC number (R = 0.182, P = 0.011). CD33 + cells were isolated from PBMCs of healthy donors with human anti-CD33 beads, and the METTL3 levels in CD33 + cells or HeLa cells were knocked down by siMETTL3. d Immunoblotting showed the METTL3 expression in CD33 + cells with or without METTL3 knockdown. e HLA-DR − CD33 + CD11b + MDSC induction from CD33 + cells in the presence of siMETTL3 or siControl (SiNC). A statistical graph is included for the comparison between the indicated groups. f Immunoblotting showed METTL3 expression in HeLa cells with or without METTL3 knockdown. g Tumour-associated HLA-DR − CD33 + CD11b + MDSC induction from CD33 + cells in coculture with Hela-siMETTL3 or Hela-siControl cells in a Transwell System for 48 h. A statistical graphs is included for the comparison between the indicated groups. Representative flow cytometry density plots (left) and statistical bar chart (right). The statistical analysis was performed using Spearman's correlation and linear regression. R, Spearman's correlation, is the correlation coefficient. The experiments in e, g were performed at least three times, and the data were plotted as the mean ± SEM. Statistics were conducted with an unpaired Student's t test, **P < 0.01, and ***P < 0.001 vs. the corresponding control Table 4).
The combination of METTL3 levels and CD33 + MDSCs was associated with the survival of patients with CC
Finally, considering that METTL3 levels were positively correlated with high CD33 + MDSC infiltration, we calculated the significance of the combination of these two biomarkers for the survival of CC patients. All 197 patients were divided into three groups. Patients with low levels of both METTL3 in tumour-infiltrating immune cells and CD33 + MDSCs were included in the combined low expression group, those with high levels of only one of the two biomarkers were included in the combined medium expression group, and those with high levels of both were included in the combined high expression group. The high combination of METTL3 and intratumoural CD33 + MDSCs was associated with reduced DFS (P < 0.001, Fig. 5a) and OS (P < 0.001, Fig. 5b). In the patients (127) with early-stage disease, the high combination of METTL3 and CD33 + MDSCs was not related to DFS (P = 0.063, Fig. 5c) but was clearly negatively related to OS (P = 0.037, Fig. 5d). In the patients (70) with advanced-stage disease, the combination of high METTL3 levels and CD33 + MDSCs was clearly related to unfavourable DFS (P < 0.001, Fig. 5e) and OS (P < 0.001, Fig. 5f (Table 3), suggesting that the combination of high METTL3 levels and CD33 + MDSCs improved patient prognostic stratification in those with advanced disease.
Discussion
The development of tumour cells depends on the tumour microenvironment, which includes tumour cells, various other cells and extracellular components [7]. The immunosuppressive cells in the tumour microenvironment, such as Tregs and MDSCs, not only affect each other, but their changes in number and types will affect tumour development [34,35]. METTL3 is one of the 'writers' , and its role is to catalyse the m 6 A methylation of mRNA (and other nuclear RNAs); after the methylation of m 6 A, RNAs will nucleate and transport to the cytoplasm faster and then produce more proteins for function and proliferation. Some studies have shown that METTL3 expression can promote tumour cell proliferation, leading to poor patient prognosis. The tumour-infiltrated MDSC population usually induces antitumour immunity tolerance by inhibiting the proliferation and function of T cells, such as hindering antigen presentation by antigen-presenting cells [36]. Increased METTL3 levels and CD33 + MDSCs have been found in tumour microenvironments and lead to a poor prognosis [37][38][39][40]. In this study, we focused on the distribution of METTL3 and CD33 + MDSCs in the tumour microenvironment of 197 patients with CC. The positive association between METTL3 levels and CD33 + MDSCs and the prognostic value of these two variants in CC patients were demonstrated. Importantly, we demonstrated that knockdown of METTL3 in CD33 + cells or HeLa cells could attenuate MDSC or tumour-associated MDSC differentiation in vitro. M 6 A methyltransferases, especially METTL3, can affect many physiological and pathological diseases through p53 and other genes [14]. At the nucleic acid level, silencing m 6 A methyltransferase significantly affects gene expression and mRNA splicing patterns, leading to changes in normal cell signalling pathways and apoptosis [33]. In bladder cancer cells, m 6 A-modified direct targets IKBKB and RELA (two key regulators of the NF-κB pathway) mediated by METTL3 become factors that promote tumour development [13]; in glioblastoma stem cells (GSCs), knocking down METTL3 can induce changes in m 6 A-enriched mRNA and alter the mRNA expression of genes with key biological functions Table 4 The multivariate cox regression analysis in cervical cancer patients The significant different factors in univariate analysis were analyzed by multivariate analysis, and the factors which were not significant in univariate analysis were not included in GSCs (such as ADAM1937) [41]. In recent studies, high METTL3 levels were related to tumour invasion and poor outcomes in breast cancer and acute myeloid leukaemia (AML) [21,42]. Our results are consistent with the results of these studies, showing that high METTL3 expression results in poor prognosis in CC patients. METTL3 regulates haematopoietic stem cell differentiation and induces the development of leukaemic cells by upregulating MYC expression [42,43]. Therefore, we wondered whether METTL3 expression may be linked to the density of tumour-infiltrated MDSCs. Our data identified a positive association between METTL3 expression in tumour cells and in tumour-infiltrating immune cells and intratumoural CD33 + MDSC density.
The results indicate that METTL3 could directly induce CD33 + CD11b + HLA-DR − MDSC differentiation or tumour-associated MDSC differentiation in vitro. Moreover, both METTL3 and CD33 + MDSCs were independent factors for the prognosis of CC patients, and the combination of METTL3 levels and CD33 + MDSC density displayed prognostic value for CC patients, including patients at early or late disease stages. The function, distribution and clinical relevance of the proportion of tumour-derived CD33 + MDSCs have been explored in recent years. MDSCs are generally elevated in tumour tissues and in the peripheral blood of cancer patients and are linked to antitumour immunity suppression, resulting in tumour growth and metastasis [25,34,44]. In our study, CC patients with a high infiltration of MDSCs in the cervical cancer microenvironment showed a poor prognosis, which is consistent with observations in other solid cancers. The tumour microenvironment is a main battleground between tumour cells and the host immune system. Tumour cells usually 'educate' infiltrated immune cells through many factors, such as cytokines or tumour-derived exosomes, to affect the proliferation, differentiation and function of tumourinfiltrating immune cells, resulting in the expansion of suppressive immune cells, including M2 macrophages, MDSCs and Tregs, and limiting the antitumour effect of cytotoxic T cells. Epigenetic modifications such as RNA modification, DNA methylation and histone modifications can rapidly regulate infiltrated immune cell differentiation and activities in tumour microenvironments [45]. Here, our data suggest that METTL3-mediated m 6 A RNA modification is positively associated with the increase in MDSC expansion and affects tumour development and prognosis in CC and induces CD33 + cells to differentiate into MDSCs in the tumour microenvironment. We further demonstrate the prognostic value of the combination of the METTL3 level in tumour-infiltrating immune cells and CD33 + MDSC density in CC patients, especially for those in advanced disease stages. A mechanistic study to support the role of METTL3 in the regulation of tumour-derived MDSC differentiation is currently underway, and the underlying mechanisms will be clarified in the near future.
Conclusions
Our study demonstrated a comprehensive result of the relationship between METTL3 and CD33 in CC and revealed that METTL3 could induce direct MDSC and tumour-associated MDSC differentiation in vitro. The results showed that both biomarkers were adverse indicators for prognosis and may have significant relationships in the microenvironment of CC. Our research may offer clues for further research into the mechanism behind METTL3 in the regulation of MDSC-mediated immune suppression in the CC microenvironment. | 5,829.6 | 2020-10-15T00:00:00.000 | [
"Biology",
"Medicine"
] |
Beam Trajectory Analysis of Vertically Aligned Carbon Nanotube Emitters with a Microchannel Plate
Vertically aligned carbon nanotubes (CNTs) are essential to studying high current density, low dispersion, and high brightness. Vertically aligned 14 × 14 CNT emitters are fabricated as an island by sputter coating, photolithography, and the plasma-enhanced chemical vapor deposition process. Scanning electron microscopy is used to analyze the morphology structures with an average height of 40 µm. The field emission microscopy image is captured on the microchannel plate (MCP). The role of the microchannel plate is to determine how the high-density electron beam spot is measured under the variation of voltage and exposure time. The MCP enhances the field emission current near the threshold voltage and protects the CNT from irreversible damage during the vacuum arc. The high-density electron beam spot is measured with an FWHM of 2.71 mm under the variation of the applied voltage and the exposure time, respectively, which corresponds to the real beam spot. This configuration produces the beam trajectory with low dispersion under the proper field emission, which could be applicable to high-resolution multi-beam electron microscopy and high-resolution X-ray imaging technology.
Introduction
The electron source was one of the crucial components for the development of electron beam microscopy, constructed by Knoll and Ruska in 1932, as its point-to-point resolution was mostly limited by the quality, especially spherical aberration, of the magnetic lenses [1]. In the last 91 years, there have been only four different electron sources used practically in the electron microscope: (i) thermionic emission with tungsten filament, (ii) thermionic emission with lanthanum hexaboride, (iii) Schottky type thermionic emission, and (iv) electric field-induced emission of the tungsten filament. The significant improvement was important for high current density, high resolution, high brightness, and low energy spread [1][2][3]. Considerable attention was paid to the development of the electron source for electron beam microscopy [4][5][6][7][8][9]. In recent years, carbon nanotubes (CNTs) have been considered field emission (FE) materials because they have remarkable properties such as high aspect ratio [10], high electrical conductivity [9,11], high thermal conductivity [12,13], and high mechanical strength [14], which make them extremely attractive as nanoscale reinforcements in high-performance composites. These properties have made CNTs a promising tip material for electron microscopy, e.g., scanning electron microscopy (SEM) [9] and atomic force microscopy [15]. There is great interest in CNT arrays for the successful demonstration of CNT-based field emission displays (FEDs) and field emission lamps. In the fabrication of CNT emitters, oxidation [16], doping [17], and laser irradiation [18] processes have been proposed to improve the field emission of electron beams [19]. The high aspect ratio of CNT emitters play an important role in the field emission electron beam for high stability [11], high brightness [4,20], and low angle of dispersion [21]. The emitter contains 14 × 14 emitters as one beam source. The height and dot size of CNT growth were controlled by the photolithography process and the growth conditions on the Si wafer substrate. The height of 40 µm and the spot size of 3 µm of each CNT morphology was observed using the SEM, and the tip size is measured to be 50~100 nm. Figure 1 shows the schematic diagram of the experimental setup of the negative voltage supply and electrode system of the field emission electron beam of CNT with 14 × 14 emitters for the image-capturing process using the MCP (Hamamatsu MCP F6959, Japan). The negative high voltage was in the range of 0 to −2 kV for the CNT emitters of the island. The stainless-steel gate electrode (SUS 304) was fixed at a distance of 250 µm from the cathode, which is shown in Figure 1a. The field emission electron beam started to emit at the peak point of the CNT emitter. The aging process of the CNT was maintained to achieve stable field emission of the electron beam, and the electron emission characteristics and uniformity were significantly improved [19]. The field emission electron beam was measured at the anode using the Keithley instrument 6517A (electrometer/high resistance meter, Tektronix). The MCP was replaced by the anode to measure the current. Figure 1b shows the CNT holder and the SEM image of one island CNT of 14 × 14 emitters with an average height of 40 µm and a peak dot diameter of 50~100 nm. In Figure 1c, the left top, right top, left bottom, and the right bottom show the top view of the module in which the CNT substrate is attached, the high voltage electrode, gate guide, and the SEM image of the mesh electrode (SUS 304) with a diameter of 300 µm attached to the gate guide. Figure 1d shows the FEM image on the phosphor screen of the MCP at 900 V. In this experiment, the vacuum chamber was evacuated by the mechanical and turbomolecular pumps with a base pressure of 1.0 × 10 −8 Torr.
Microchannel Plate Design
The MCP plate is essentially a plate (disc) made of an electrically insulating material (usually glass) containing a hexagonal array of tiny holes, as shown in Figure 2a. The conventional phosphor screen provides insufficient information about the trajectory of the electron beam and is affected by the effect of signal noise due to the applied high voltage [24]. In addition, the CNT emitter can be damaged by the high electric field due to the joule heat effect [24,30]. MCP has more advantages compared to conventional phosphor screens. The incident electron beam is amplified near the threshold voltage by the generation of secondary electrons and protects the CNTs from irreversible damage during the vacuum arc. Figure 2b shows the schematic diagram of the MCP, which contains three electrodes, namely the MCP-in electrode, the MCP-out electrode, and the phosphor electrode. Furthermore, this MCP was used to study the trajectories of the field emission electron beam with the vertically aligned one-island CNT with 14 × 14 emitters near threshold voltage. After applying negative voltage to the CNT emitter, the field emission electron beam started to emit from the peak point of each CNT under the high electric field and accelerated toward the MCP fixed at a distance of 25 mm from the mesh electrode. The effective diameter of this MCP was 28 mm, and the inner electrode was a mesh electrode and was grounded. The MCP-out electrode was placed at a distance of 1 mm from the MCP-in electrode and was biased with a positive voltage of up to 1.5 kV. Moreover, the phosphor was located at a distance of 1 mm from the MCP-out electrode and was biased with a positive voltage of up to 3 kV to convert the electron beam into a light signal. The phosphor electrode converted the electron signal into a light signal. A Nikon D700 digital single-lens reflex (DSLR) camera was used to capture the emitted light signal throughout the field emission process. measure the current. Figure 1b shows the CNT holder and the SEM image of one island CNT of 14 × 14 emitters with an average height of 40 µ m and a peak dot diameter of 50~100 nm. In Figure 1c, the left top, right top, left bottom, and the right bottom show the top view of the module in which the CNT substrate is attached, the high voltage electrode, gate guide, and the SEM image of the mesh electrode (SUS 304) with a diameter of 300 µ m attached to the gate guide. Figure 1d shows the FEM image on the phosphor screen of the MCP at 900 V. In this experiment, the vacuum chamber was evacuated by the mechanical and turbomolecular pumps with a base pressure of 1.0 × 10 −8 Torr.
Microchannel Plate Design
The MCP plate is essentially a plate (disc) made of an electrically insulating material (usually glass) containing a hexagonal array of tiny holes, as shown in Figure 2a. The conventional phosphor screen provides insufficient information about the trajectory of the electron beam and is affected by the effect of signal noise due to the applied high voltage [24]. In addition, the CNT emitter can be damaged by the high electric field due to the joule heat effect [24,30]. MCP has more advantages compared to conventional phosphor screens. The incident electron beam is amplified near the threshold voltage by the generation of secondary electrons and protects the CNTs from irreversible damage during the vacuum arc. Figure 2b shows the schematic diagram of the MCP, which contains three electrodes, namely the MCP-in electrode, the MCP-out electrode, and the phosphor electrode. Furthermore, this MCP was used to study the trajectories of the field emission electron beam with the vertically aligned one-island CNT with 14 × 14 emitters near threshold voltage. After applying negative voltage to the CNT emitter, the field emission electron beam started to emit from the peak point of each CNT under the high electric field and accelerated toward the MCP fixed at a distance of 25 mm from the mesh electrode. The effective diameter of this MCP was 28 mm, and the inner electrode was a mesh electrode and was grounded. The MCP-out electrode was placed at a distance of 1 mm from the MCP-in electrode and was biased with a positive voltage of up to 1.5 kV. Moreover, the phosphor was located at a distance of 1 mm from the MCP-out electrode and was biased with a positive voltage of up to 3 kV to convert the electron beam into a light signal. The phosphor electrode converted the electron signal into a light signal. A Nikon D700 digital single-lens reflex (DSLR) camera was used to capture the emitted light signal throughout the field emission process.
Current-Voltage Characteristics of One-Island Carbon Nanotubes
Many researchers have performed experiments [9,21,22,[39][40][41][42] and reported the currentvoltage characteristics of the field emission electron beam of the different structures of CNT emitters. The I-V diagram describes the field emission performance, which is influenced by the type of material, arrangement, and surface morphology of the emitters [30]. Carbon nanotubes are considered an ideal material for the fabrication of field emitters due to their high aspect ratio, mechanical strength, and chemical stability [39]. In our previous experiment [9], the emission current and brightness of the CNT emitter were investigated with different tip diameters, geometric field enhancement factors (β geo ), and the number of samples. Due to the high brightness, low threshold voltage, and high stability of the electron beam, we optimized the group 1 sample from our previous experiment for the different purposes of this study [9]. To understand the characteristics of the field emission electron beam profile from an island beam source, a flat anode was used to measure the I-V curve in the high vacuum chamber of this experiment. Figure 3 shows the I-V curve of the field emission electron beam from vertically aligned single-island CNT emitters as a function of the applied voltage. The threshold voltage of the single-island CNT emitter was measured at 810 V with an emission current of 10 nA. After the threshold voltage, the emission current increased dramatically with the applied voltage. The inset in Figure 3a,b shows the Fowler-Nordheim (F-N) plot and the cone-shaped structure SEM image of the vertically aligned CNT with 14 × 14 emitters, where the dot size, pitch size, and height of each CNT emitter are 3 µm, 15 µm, and 40 µm, respectively, that occupy the total area of 39.2 × 10 3 µm 2 . The F-N plot of ln(I/V 2 ) versus 1/V explains the electron beam's field emission behavior [9,30]. The relationship between the CNT's morphology and correspondence field emission can be established [9,21].
Current-Voltage Characteristics of One-Island Carbon Nanotubes
Many researchers have performed experiments [9,21,22,[39][40][41][42] and reported the current-voltage characteristics of the field emission electron beam of the different structures of CNT emitters. The I-V diagram describes the field emission performance, which is influenced by the type of material, arrangement, and surface morphology of the emitters [30]. Carbon nanotubes are considered an ideal material for the fabrication of field emitters due to their high aspect ratio, mechanical strength, and chemical stability [39]. In our previous experiment [9], the emission current and brightness of the CNT emitter were investigated with different tip diameters, geometric field enhancement factors (βgeo), and the number of samples. Due to the high brightness, low threshold voltage, and high stability of the electron beam, we optimized the group 1 sample from our previous experiment for the different purposes of this study [9]. To understand the characteristics of the field emission electron beam profile from an island beam source, a flat anode was used to measure the I-V curve in the high vacuum chamber of this experiment. Figure 3 shows the I-V curve of the field emission electron beam from vertically aligned single-island CNT emitters as a function of the applied voltage. The threshold voltage of the single-island CNT emitter was measured at 810 V with an emission current of 10 nA. After the threshold voltage, the emission current increased dramatically with the applied voltage. The inset in Figure 3a,b shows the Fowler-Nordheim (F-N) plot and the cone-shaped structure SEM image of the vertically aligned CNT with 14 × 14 emitters, where the dot size, pitch size, and height of each CNT emitter are 3 µ m, 15 µ m, and 40 µ m, respectively, that occupy the total area of 39.2 × 10 3 µm 2 . The F-N plot of ln(I/V 2 ) versus 1/V explains the electron beam's field emission behavior [9,30]. The relationship between the CNT's morphology and correspondence field emission can be established [9,21].
Measurement of Field Emission Microscopy Image of CNT Emitters
To obtain quantitative information about the microscopic properties of the CNT emitter, the FEM image was examined using the MCP. The single island containing the 14 × 14 emitter worked as a one-beam source in which the emission current drastically increased under the slowly increasing applied voltage as shown in Figure 3. The center exposed area in the FEM pattern reflects the major electron emission from the protruding point at the end of each emitter tip [24]. From the tip to the trunk of the CNT emitters, the surface was smooth, which contributed to enhancing the electric field at the peak point. The trajectory of the electron beam could not be distinguished from each emitter because of the overlap of the electron beams [43][44][45]. Figure 4 shows the FEM image of the one-island CNT with 14 × 14 emitters without a focusing electrode. When the variable negative voltage was applied to the CNT, the emitted electron beam was focused directly on the MCP plate due to the very small gap distance (250 µm) between the cathode and the gate electrode. The applied voltage increased continuously from 800 V to 940 V in the voltage difference range of 10 V so that the electron beam trajectory increased continuously with increased field emission current, as shown in Figure 4. In Figure 4a, there is no image of the electron beam trajectory because there is no emission current flowing. In Figure 4b-o, the field emission electron beam spot is continuously captured. In Figure 4b, the bright spot of the electron beam appears on the center of the phosphor screen of the MCP at 810 V. The FEM image with a high-density bright spot is visible in the center of the MPC. From Figure 4f, the spot size increases continuously until Figure 4k. From Figure 4k-o, its size remains constant with its full width half maximum (FWHM) of 2.71 mm. The uniform size of the high-density bright spot remains constant from Figure 4k-o, but the size of the green spot increases continuously because the divergence of the electron beam was limited by the electron emission from the vertically aligned CNT emitters [46,47]. The vertically aligned CNT shows a small beam divergence of electron beam trajectory at a distance of 25 mm of the MCP with a high-dense bright spot opening angle of 2.9 • [9, [48][49][50]. Figure 4a-o confirms that the field emission electron beam from 14 × 14 CNT emitters works as a one-beam source and is focused on a point with a high-density bright spot without a focusing electrode near the threshold voltage. Many experiments are repeated to confirm the size of the high-density bright spot. Figure 5 shows the confirmation of the size of the focused electron beam of the 14 × 14 CNT emitters under the same conditions as in Figure 4. In Figure 5a-f, the applied voltage is increased from 850 V to 900 V with an increasing range of 10 V to capture the high-dense bright spot of the FEM on the MCP. In Figure 5f, the size of the high-dense bright spot is calculated to be 2.71 mm with its FWHM. Hence, the symmetrically distributed FEM image is captured under the variation of the applied voltage on the phosphor screen of the MCP.
Spot Size Trajectory Analysis of Field Emission Electron Beam
The pattern of the high-dense bright spot has a circular spot, and the intensity profile of the beam spot can be well described by a Gaussian distribution, , where A, x 0 , and σ represent the peak intensity, the maximum point of peak intensity and the standard deviation, respectively. The fullwidth half maximum of the electron beam can be expressed as FWHM = 2 √ 2ln2σ, where σ is the standard deviation [21,51]. Figure 6 explains the analysis of the intensity profile of the high-dense electron beam spot of Figure 4. In Figure 6a, the intensity increases continuously up to 900 V, after which it becomes constant and reaches its maximum intensity at 255 atomic units. As the emission current increases, the intensity profile of the dense bright spot expands. The FWHM is expected to become more compact at higher beam voltage [52]. The FWHM measures the actual size of high-density bright spot of the electron beam. The size of the high-density bright spot increases with the increase in the applied voltage from 800 V to 900 V and then remains constant with a size of FWHM of 2.71 mm, as shown in Figure 6b. In our previous experiment [9], the simulated result is explained in detail and compared with the experimental result, the change in the spot size of the beam path was simulated by varying the distance between the gate electrode and the phosphor screen. Figure 6c shows the variation of the beam divergence when varying the distance between the gate electrode and the phosphor screen according to our previous experiment. The electron beam trajectory follows the fitting parameter W z = w 0 + aZ b where W z is the beam trajectory, and w 0 , a, and b are fitting parameters with values −0.06270, 0.11089, and 0.75970, respectively. The electron emission and the beam trajectory depend on the structure of the emitter of the nano-tip emitter, the gap distance from the cathode to the mesh electrode, and the gap distance from the mesh electrode to the phosphor anode, respectively. Figure 6d is the schematic diagram for the decrease in the size of the electron beam spot from (i) to (v) with the decrease in the distance between the mesh electrode and the phosphor electrode of MCP, which is clearly explained in Figure 6c. In our experiment, the emitter density is high although the beam is highly focused at a point on the MCP without a focusing electrode. Nonetheless, the proposed MCP approach can generate uniform beam trajectories with the high-density beam spot of the CNT emitters with the best morphology structures under the proper field emission.
Nanomaterials 2022, 12, x FOR PEER REVIEW 7 of 14 explained in detail and compared with the experimental result, the change in the spot size of the beam path was simulated by varying the distance between the gate electrode and the phosphor screen. Figure 6c shows the variation of the beam divergence when varying the distance between the gate electrode and the phosphor screen according to our previous experiment. The electron beam trajectory follows the fitting parameter Wz = w0 + aZ b where Wz is the beam trajectory, and w0, a, and b are fitting parameters with values −0.06270, 0.11089, and 0.75970, respectively. The electron emission and the beam trajectory depend on the structure of the emitter of the nano-tip emitter, the gap distance from the cathode to the mesh electrode, and the gap distance from the mesh electrode to the phosphor anode, respectively. Figure 6d is the schematic diagram for the decrease in the size of the electron beam spot from (i) to (v) with the decrease in the distance between the mesh electrode and the phosphor electrode of MCP, which is clearly explained in Figure 6c. In our experiment, the emitter density is high although the beam is highly focused at a point on the MCP without a focusing electrode. Nonetheless, the proposed MCP approach can generate uniform beam trajectories with the high-density beam spot of the CNT emitters with the best morphology structures under the proper field emission.
Noise Effect of Focus Spot Size
Several processes deteriorate the signal-to-noise ratio (SNR) of electron microscopy images, such as the noise of the primary electron beam, the secondary electron beam in MCP, and the noise of the final detection system [53,54]. In the MCP, the input electron energy increases the generation of secondary electrons compared to the noise [54]. For the measurement of the electron beam spot, exposure time plays an important role in the SNR effect [55,56]. The electron beam spot size is captured in a short exposure time to analyze the spot size of the electron beam from the CNT emitter on the MPC's phosphor screen. The exposure time is the time in which the light can reach the sensor. The focused electron beam is kept at a fixed position and the beam spot size of FEM is controlled by the exposure time [57][58][59]. Figure 7 shows the reduction of the beam spot by reducing the exposure time at a fixed applied voltage of 900 V. In Figure 7a, the FWHM of the high-density bright spot is calculated to be 2.71 mm for an exposure time of 1 s. Moreover, in Figure 7c-e, the FWHM is calculated to be 2.71 mm at an exposure time of 1/3, 1/4, and 1/5 s, respectively, showing the real electron beam trajectory at the center of the MCP. As the exposure time of the electron beam increased, the noise effect in the beam spot size also increased. Figure 8 shows the line profile of the electron beam spot from Figure 7a,b, in which the noise effect is explained by the variation of the exposure time. The FWHM of the real beam spot size is 2.71 mm at 1/5 s. This experiment explains the electron beam trajectory with its spot size by optimizing the applied voltage and exposure time. Hence, the SNR is increasing with the variation of exposure time at the optimized voltage. The FWHM of the high-density beam spot is calculated as 2.71 mm at 900 V under the variation of the exposure time. The real beam spot size is very important to obtain the actual electron beam trajectory.
Noise Effect of Focus Spot Size
Several processes deteriorate the signal-to-noise ratio (SNR) of electron microscopy images, such as the noise of the primary electron beam, the secondary electron beam in MCP, and the noise of the final detection system [53,54]. In the MCP, the input electron energy increases the generation of secondary electrons compared to the noise [54]. For the measurement of the electron beam spot, exposure time plays an important role in the SNR effect [55,56]. The electron beam spot size is captured in a short exposure time to analyze the spot size of the electron beam from the CNT emitter on the MPC's phosphor screen. The exposure time is the time in which the light can reach the sensor. The focused electron beam is kept at a fixed position and the beam spot size of FEM is controlled by the exposure time [57][58][59]. Figure 7 shows the reduction of the beam spot by reducing the exposure time at a fixed applied voltage of 900 V. In Figure 7a, the FWHM of the highdensity bright spot is calculated to be 2.71 mm for an exposure time of 1s. Moreover, in Figure 7c-e, the FWHM is calculated to be 2.71 mm at an exposure time of 1/3, 1/4, and 1/5 s, respectively, showing the real electron beam trajectory at the center of the MCP. As the exposure time of the electron beam increased, the noise effect in the beam spot size also increased. Figure 8 shows the line profile of the electron beam spot from Figure 7a,b, in which the noise effect is explained by the variation of the exposure time. The FWHM of the real beam spot size is 2.71 mm at 1/5 s. This experiment explains the electron beam trajectory with its spot size by optimizing the applied voltage and exposure time. Hence, the SNR is increasing with the variation of exposure time at the optimized voltage. The FWHM of the high-density beam spot is calculated as 2.71 mm at 900 V under the variation of the exposure time. The real beam spot size is very important to obtain the actual electron beam trajectory. Figure 9a shows a schematic representation of the electron beam trajectory f Figure 7 where the exposure time is reduced by 1/2, 1/3, 1/4, and 1/5 s respectively. Fig 9b shows a simulation of electron beam trajectory in which opera simulation 3D [6 used to analyze the high-density bright spot with its current density. The simulatio modelled based on an emitter spot size of 3 µ m with a height of 40 µ m. In addition diameter of the mesh hole is set to 160 µ m with a thickness of 100 µ m. The dist between the cathode and gate, and gate and anode is set to 150 µ m, and 1 m respectively. Figure 9c shows the simulation results of the electron beam spot from Fig 9b, where the beam is strongly focused on the center of the phosphor screen (with color). The red, yellow, and green colors in Figure 9c represent the scattering of electron beam from Figure 9b, which can act as a noise effect on the phosphor screen Figure 7 where the exposure time is reduced by 1/2, 1/3, 1/4, and 1/5 s respectively. Figure 9b shows a simulation of electron beam trajectory in which opera simulation 3D [60] is used to analyze the high-density bright spot with its current density. The simulation is modelled based on an emitter spot size of 3 µm with a height of 40 µm. In addition, the diameter of the mesh hole is set to 160 µm with a thickness of 100 µm. The distance between the cathode and gate, and gate and anode is set to 150 µm, and 1 mm, respectively. Figure 9c shows the simulation results of the electron beam spot from Figure 9b, where the beam is strongly focused on the center of the phosphor screen (with blue color). The red, yellow, and green colors in Figure 9c represent the scattering of the electron beam from Figure 9b, which can act as a noise effect on the phosphor screen. Nanomaterials 2022, 12, x FOR PEER REVIEW 11 of 14
Conclusions
Vertically aligned cone-shaped CNTs with 14 × 14 emitters were prepared as an island in the Si wafer substrate by sputter coating, photolithography, and PE-CVD for the high resolution and low dispersion of the electron beam. This island of CNT emitters was called a single beam source. This beam source was perfectly aligned with the center of the gate electrode (SUS 304) to ensure uniform field emission of the highly focused electron beam. The threshold voltage of the single island CNT emitter was measured at 810 V with an emission current of 10 nA. After the threshold voltage, the emission current increased dramatically with the applied voltage. The uniform FEM image was captured by varying the applied voltage and exposure time to study the symmetrically distributed electron beam trajectory and the beam spot size. The high-density beam spot increased continuously under the variation of the applied voltage and remained constant after 900 V with its FWHM of 2.71 mm on the phosphor screen of the MCP. In addition, the FEM image was captured with variation of the exposure time to investigate the noise effect.
Conclusions
Vertically aligned cone-shaped CNTs with 14 × 14 emitters were prepared as an island in the Si wafer substrate by sputter coating, photolithography, and PE-CVD for the high resolution and low dispersion of the electron beam. This island of CNT emitters was called a single beam source. This beam source was perfectly aligned with the center of the gate electrode (SUS 304) to ensure uniform field emission of the highly focused electron beam. The threshold voltage of the single island CNT emitter was measured at 810 V with an emission current of 10 nA. After the threshold voltage, the emission current increased dramatically with the applied voltage. The uniform FEM image was captured by varying the applied voltage and exposure time to study the symmetrically distributed electron beam trajectory and the beam spot size. The high-density beam spot increased continuously under the variation of the applied voltage and remained constant after 900 V with its FWHM of 2.71 mm on the phosphor screen of the MCP. In addition, the FEM image was captured with variation of the exposure time to investigate the noise effect. | 6,910.4 | 2022-12-01T00:00:00.000 | [
"Engineering",
"Physics",
"Materials Science"
] |
Research Article Spatial, Temporal, and Interchannel Image Data Fusion for Long-Distance Terrestrial Observation Systems
This paper presents methods for intrachannel and interchannel fusion of thermal and visual sensors used in long-distance terrestrial observation systems. Intrachannel spatial and temporal fusion mechanisms used for image stabilization, super-resolution, denoising, and deblurring are supplemented by interchannel data fusion of visual- and thermal-range channels for generating fused videos intended for visual analysis by a human operator. Tests on synthetic, as well as on real-life, video sequences have confirmed the potential of the suggested methods.
INTRODUCTION
Long-distance terrestrial observation systems have traditionally been high-cost systems used in military and surveillance applications.Recent advances in sensor technologies (such as in infrared cameras, millimeter wave radars, and low-light television cameras) have made it feasible to build low-cost observation systems.Such systems are increasingly used in the civilian market for industrial and scientific applications.
In long-distance terrestrial observation systems, infrared sensors are commonly integrated with visual-range chargecoupled device (CCD) sensors.Such systems exhibit unique characteristics-thanks to the simultaneous use of both visible and infrared wavelength ranges.Most of them are designed to give the viewer the ability to reliably detect objects in highly detailed scenes.The thermal-range and visualrange channels have different behaviors and feature different image distortions.Visual-range long-distance observations are usually affected by atmospheric turbulence, which causes spatial and temporal fluctuations to the index of refraction of the atmosphere [1], resulting in chaotic geometrical distortions.On the other hand, thermal channels are less vulnerable to the turbulent effects [2][3][4][5][6][7] but usually suffer from substantial sensor noise and reduced resolution as compared to their visual-range counterparts [8].One way to overcome those problems is to apply data fusion techniques.
In recent years, a great deal of effort has been put into multisensor fusion and analysis.Available fusion techniques may be classified into three abstraction levels: pixel, feature, and semantic levels.At the pixel level, images acquired in different channels are combined by considering individual pixel values or small arbitrary regions of pixels in order to make the fusion decision [9][10][11][12].At the feature-level fusion, images are initially subjected to feature-driven segmentation in order to produce a set of regions with various properties that are used to determine which features from which image are to be included in the fused image [13][14][15][16].Semanticdriven methods transform all types of input data into a common variable space, where the data is fused [17].In developing new data fusion methods, it is possible to extract different types of features from different channels before fusing them-a concept that the above-mentioned methods fail to do since they apply the same fusion criteria to all input channels.Unlike previous methods, the method in this paper applies sensor-specific criteria to each channel before fusing the data.
The development of fusion algorithms using various kinds of pyramid/wavelet transforms has led to numerous pixel-and feature-based fusion methods [18][19][20][21][22][23][24].The motivation for the pyramid/wavelet-based methods emerges from observations that the human visual system is primarily sensitive to local contrast changes, for example, edges and corners.However, observation systems are characterized by diversity in both location and size of the target of interest; therefore, rigid decomposition, which is characteristic for multiresolution fusion methods, turns out to be less suitable for longrange observation tasks [13,16,25].This paper describes a video processing technology designed for fusion of thermal-and visual-range input channels of long-range observation systems into a unified video stream intended for visual analysis by a professional human operator.The suggested technology was verified using synthetic, as well as real-life, thermal and visual sequences from a dedicated database [26,27].The database contained video sequences acquired in the near vicinity of the camera as well as sequences of sites located as far as 25 km away.
SYSTEM DESCRIPTION
The proposed video processing technology is outlined in a schematic block diagram in Figure 1.The outlined method is based on a two-stage process.The first stage consists of intrachannel-interframe processing methods used to perfect each input channel independently.Each channel's processing method is designed according to the sensor's specific limitations and degradations.For the visual-range channel processing, spatial-temporal fusion is implemented for compensating turbulence-induced image geometrical distortions, as well as super-resolution above the visual sensor's sampling rate.For the thermal channel, spatial-temporal fusion is implemented for sensor noise filtering and resolution enhancement by means of 3D (spatial-temporal) local adaptive filtering.These visual-and thermal-range intrachannel fusion schemes are thoroughly described in Sections 3 and 4, respectively.The second stage is interframeintrachannel fusion.At this stage, thermal-and visual-range channel image frames, corrected and enhanced, are fused frame by frame using a multiple-criteria weighted average scheme with locally adapted weights.The second stage is detailed in Section 5.
Channel characterization and processing principles
In remote sensing applications, light passing long distances through the troposphere is refracted by atmospheric turbulence, causing distortions throughout the image in the form of chaotic time-varying local displacements.The effects of turbulence phenomena on imaging systems were widely recognized and described in the literature, and numerous methods were proposed to mitigate these effects.
One method for turbulence compensation is adaptive optics [28,29].Classical adaptive optics, which uses a single deformable mirror, provides correction for a limited field of view (FOV).Larger FOV corrections can be achieved by several deformable mirrors optically conjugated at various heights [30][31][32][33].In modeling images with distortion caused by atmospheric turbulence, light from each point in the acquired scene is assumed to possess a slightly different tilt and low-order aberration, and it can be modeled by convolving a raw image with a space-variant pseudorandom point spread function [34].Therefore, multiconjugate adaptive optics techniques require complex structure and reconstruction processes, making them unsuitable for operational systems.
Other turbulence compensation methods use an estimation of modulation transfer function (MTF) of the turbulence distortions [35][36][37][38].The drawback of those methods is that they require some prior knowledge about the observed scene, which is often unavailable.
Methods that require no prior knowledge are suggested in [3-7, 13, 39-42].The principal idea is to use, for reconstructing distortion-compensated image frames, an adaptively controlled image resampling method based on the estimate of image local displacement vectors.Using those concepts, turbulence compensation algorithms which preserve genuine motion in the scene are suggested in [43][44][45][46][47].
In this paper, these techniques are further elaborated and improved upon in order to obtain super-resolution in addition to turbulence compensation.The new techniques are used as an interframe-interchannel fusion mechanism for the visual-range input channel.As shown in the flow diagram, presented in Figure 2, visual-range video processing consists of three processing stages: (i) estimation of the reference frames, (ii) determination of the motion vectors for all pixels in image frames and motion vector analysis for realmotion extraction, and (iii) generation of stabilized frames with super-resolution and preservation of the real motion.Those stages are thoroughly described, respectively, in Sections 3.2, 3.3, and 3.5.
Estimation of the reference frames
The reference images, which are the estimation of the stable scene, are obtained from the input sequence.The reference images are needed for measuring the motion vectors for each current video frame.One way to measure the motion vectors of each image frame is by means of elastic registration with the previous frame.However, this method does not allow reliable discrimination of real movements in the scene from those caused by the atmospheric turbulence.For this task, estimation of the stable scene is required.We adopt the approach of [48] and suggest using a pixelwise rank gray-level filtering of video frames in a temporal sliding window for generating such an estimation intended to serve as the reference frame.The use of rank smoothing filters such as median or alpha-trimmed-mean filters is substantiated in two ways.First, distribution of a light beam propagating through a turbulent atmosphere has a mean of zero.This means that the center of the deflection is located at the same point that the light beam would have hit if there was no turbulence present.Therefore, the statistical expectation of the gray-level values is relatively close to the mean of the trace of the same pixel's values over a long period of time.The reason for using a rank filter instead of a mean filter is the fact that for moving objects that accommodate a pixel for a short period of time, the gray-level distribution for this pixel is found to be tail-heavy.When applying rank filters, the distribution tails will be eliminated from evaluation of estimated values.Rank filtering might result in resolution degradation.This will be dealt with in subsequent processing stage, which suggests resolution enhancement (see Section 3.4).It was found experimentally that the use of a temporal median filter provides an acceptable solution in terms of both stable scene evaluation quality and computational efficiency [49,50].
The length in time of the filter temporal window, N, is determined by the correlation interval of turbulence effect over time; that is, the longer the time correlation of the turbulence effect is, the larger the size of the temporal sliding window becomes.Our experiments have shown that for correlation intervals of atmospheric turbulence of order of seconds, temporal window size should be of the order of 100 frames for frame rate of 25 frames per second.
Temporal pixelwise median filtering for estimating the stable scene as a reference image is illustrated in Figure 3, where part (a) presents a sample frame taken from a turbulent distorted sequence acquired with a camera acquiring images in size of 4 times common intermediate format (4CIF-704 × 576 pixels) in a frame rate of 25 frames per second (the sequence can be found at [26]). Figure 3(b) depicts the estimation of the stable scene calculated by temporal median over 117 frames.One can notice that the geometrical distortions in Figure 3(a), in particular around the dune's rim on the left-hand side of the image, are removed from the stabilized estimation in Figure 3 In principle, the median filtering in a moving time window presents high computational complexity.Utilizing a fast recursive method for median filtering [48,49] enables a realtime implementation at common video rates.
Motion vector analysis for real-motion discrimination
In order to avoid distortion of real motion due to the turbulence compensation process, real motion should be detected in the observed scene.To this end, a real-time two-stage decision mechanism is suggested in [44,45,49].This method forms, for each pixel in each incoming frame, a real-motion separation mask (RMSM ( p) , where p is the space-time coordinate vector, p = [x, y, t]).At the first step, a straightforward fast algorithm is utilized for extracting areas, such as background, that are most easily classified as stable.In most cases, the majority of the image pixels are extracted at this stage.Those parts are not further processed.Only the pixels, which were not tagged as stable at the first phase, are dealt with at the second phase.The second stage uses a more sophisticated though more time-consuming algorithm.
Stage I
At the first stage, the gray-level difference between the current value of each pixel of the incoming frame and its temporal median is calculated as "real-motion measure."This is referred to as distance-from-median (DFM) measure: where t is an index of the current processed frame, and I p is its median over the temporal window (Ω) centered at p: If the distance, DFM ( p ) , is below a given predefined threshold, the pixel is considered to be of a stationary object.The threshold is determined by exploiting the observer's limitation of distinguishing between close gray-levels.In realtime applications, the threshold is an adjustable parameter of the algorithm that can be adjusted by the observer in course of the scene visual analysis.Background areas, which do not belong to a moving object nor are located near edges, will be resolved in this way.All other pixels that are not resolved at this stage are processed at the next one.
Figure 4(a) presents a frame extracted from a real-life turbulent degraded video sequence with moving objects (see [26]). Figure 4(b) is the reference frame computed by applying elementwise temporal median filtering over 117 frames, as described in Section 3.2.Figure 4(c) represents darker pixels, which were tagged as real-motion at the first stage.As one can see, while this first stage detects most of the background pixels as such, it produces some "false alarms" (marked with arrows).Figure 4(d) represents, in darker tones, pixels that contain real motion.As one can see, the real-motion detection errors are eliminated at the second processing stage, which is described in the following section.
In its simplest form, the optical flow method assumes that it is sufficient to find only two parameters of the translation vector for every pixel.The motion vector {Δ x, Δ y} = {x − x, y − y}, for every pixel, is the vector difference between the pixel's location in the original image I (x,y) and its location in the reference image I ( x, y) .For the subsequent processing stages, the translation vector is presented in polar coordinates as {M ( p ) , θ ( p ) } through its magnitude {M ( p ) } and angle {θ ( p ) }, which are subjected to cluster analysis for discriminating real movement against that caused by atmospheric turbulence.
Real-motion discrimination through motion field magnitude distribution
For cluster analysis of the motion vector magnitude distribution function for all pixels (x, y) in a particular frame, each pixel in the frame is assigned with a certainty grade, the magnitude-driven mask (MDM ( p ) ).The MDM ( p ) measure ranges between 0 and 1 and characterizes the magnitudebased likelihood that particular pixel belongs to objects in a real motion.Figure 5 presents the certainty as a function of the motion vector's magnitudes.It is natural to assume that minor movements are caused by turbulence, and larger movements correspond to real motion.The intermediate levels comprise motion vectors' magnitudes upon which concise decision cannot be made.The magnitudes' thresholds T L and T H are application-dependent parameters and can be set by the user.Based on the analysis of our visual database, in our experiments with real-life videos, T L and T H were set to 2 and 4 pixels, respectively.
Real-motion discrimination through motion field's angle distribution
A pixel's motion discrimination through angle distribution is achieved by means of statistical analysis of the angle component of the motion field.For the neighborhood of each pixel, the variance of angles is computed.As turbulent motion has fine-scale chaotic structure, motion field vectors in a small spatial neighborhood distorted by turbulence have considerably large angular variance.Real motion, on the other hand, has strong regularity in its direction and therefore the variance of its angles over a local neighborhood will be relatively small.The neighborhood size, in which the pixel's angular standard deviation is computed, should be large enough to secure a good statistical estimation of angle variances, and as small as possible to reliably localize small moving objects.In our experiments with the dedicated real-life database [26, 27], it was found that neighborhood's sizes of 11 × 11 and 15 × 15 present a reasonable compromise.
As a result of variance analysis, each pixel is assigned with an angle-driven mask (ADM ( p ) ), which presents an angle distribution-based likelihood that this pixel belongs to an object in a real motion.This is illustrated in Figure 6.Real moving objects have bounded angular variances, T L and T H .Both turbulent and background areas should be regarded as stable.This means that pixels with angular variance smaller than T L or higher than T H are regarded as stationary.Those values are set by the observer.In our experiments with real-life video, they were set to (π/6) 2 and (π/3) 2 , respectively.
Real-motion separation mask
Having both MDM ( p ) and ADM ( p ) , a combined real-motion separation mask (RMSM ( p ) ) is formed as follows: Equation ( 3) implies that the ADM measure is more accurate than the MDM when the term |ADM ( p ) − 1/2| has a higher value than |MDM ( p ) − 1/2|.In this case, the ADM measure will be used; otherwise the MDM value will be applied.Figure 4(d) presents the RMSM ( p ) , where real moving objects are represented in darker pixels.
Generation of super-resolved stabilized output frames
In turbulence-corrupted videos, consequent frames of a stable scene differ only due to small atmospheric-turbulenceinduced movements between images.As a result, the image sampling grid defined by the video camera sensor may be considered to be chaotically moving over a stationary image scene.This phenomenon allows for the generation of images with larger number of samples than those provided by the camera if image frames are combined with appropriate resampling [2,[60][61][62][63].
Generally, such a super-resolution process consists of two main stages [2,[64][65][66][67][68].The first is determination, with sub-pixel accuracy, of pixel movements.The second is combination of data observed in several frames in order to generate a single combined image with higher spatial resolution.A flow diagram of this stage of processing is shown in Figure 7.
For each current frame of the turbulent video, inputs of the process are a corresponding reference frame, obtained as a temporal median over a time window centered on the current frame, and the current frame displacement map.The latter serves for placing pixels of the current frame, according to their positions determined by the displacement map, into the reference frame, which is correspondingly upsampled to match the subpixel accuracy of the displacement map.For upsampling, different image interpolation methods can be used.Among them, discrete sinc-interpolation is the most appropriate as the one with the least interpolation error and may also be computed efficiently [69].As a result, output stabilized and enhanced in its resolution frame is accumulated.In this accumulation process, it may happen that several pixels of different frames are to be placed in the same location in the output-enhanced frame.In order to make the best use of all of them, these pixels must be averaged.For this averaging, the median of those pixels is computed in order to avoid the influence of outliers that may appear due to possible errors in the displacement map.
After all available input frames are used in this way, the enhanced and upsampled output frame contains, in positions where there were substitutions from input frames, accumulated pixels of the input frames and, in positions where there were no substitutions, interpolated pixels of the reference frame.Substituted pixels introduce to the output frame high frequencies outside the baseband defined by the original sampling rate of the input frames.Those frequencies were lost in the input frames due to the sampling aliasing effects.Interpolated pixels that were not substituted do not contain frequencies outside the baseband.In order to finalize the processing and take full advantage of the super-resolution provided by the substituted pixels, the following iterative reinterpolation algorithm was used.This algorithm assumes that all substituted pixels accumulated, as described above, are stored in an auxiliary replacement map containing pixel values and coordinates.At each iteration of the algorithm, the discrete Fourier transform (DFT) spectrum of the image obtained at the previous iteration is computed and then zeroed in all of its components outside the selected enhanced bandwidth, say, double of the original one.After this, inverse DFT is performed on the modified spectrum, and corresponding pixels in the resulting image are replaced with pixels from the replacement map, thus producing an image for the next iteration.In this process, the energy of the zeroed outside spectrum components can be used as an indicator when the iterations can be stopped.
Once iterations are stopped, the output-stabilized and resolution-enhanced image obtained in the previous step is subsampled to the sampling rate determined by selected enhanced bandwidth and then subjected to additional processing aimed at camera aperture correction and, if necessary, denoising.
Figure 8 illustrates the feasibility of the method.Figure 8(a) is a frame extracted from turbulent degraded real-life sequence, while Figure 8 Atmospheric turbulence also affects thermal-range videos.Figure 9 demonstrates application of the method to intermediate infrared wavelengths (3-8 μm), turbulent video sequence.Figure 9(a) shows an example of a super-resolved frame generated from the thermal sequence (whose stable reference corresponding frame is presented in Figure 4(b)).The marked fragments of Figure 9 In the evaluation of the results obtained for real-life video, one should take into account that substantial resolution enhancement can be expected only if the camera fillfactor is small enough.The camera fill-factor determines the degree of lowpass filtering introduced by the optics of the camera.Due to this low-pass filtering, image high frequencies in the baseband and aliasing high-frequency components that come into the baseband due to image sampling are suppressed.Those aliasing components can be recovered and returned back to their true frequencies outside the baseband in the described super-resolution process, but only if they have not been lost due to the camera low-pass filtering.The larger the fill-factor is, the heavier the unrecoverable resolution losses will be.For quantitative evaluation of the image resolution enhancement achieved by the proposed super-resolution technique, we use a degradation measure method described in [70].The method compares the variations between neighboring pixels of the image before and after lowpass filtering.High variation between the original and blurred images means that the original image was sharp, whereas a slight Table 1: Quantitative evaluation of the super-resolved images.The degradation grade is ranging from 0 to 1, which are, respectively, the lowest and the highest degradations.
Original Super-resolved Visual-range video (see Figure 8) Entire original image (see Figure 8 variation between the original and blurred images means that the original image was already blurred.The comparison result presented in a certain normalized scale as in the image degradation measure ranged from 0 to 1 is shown in [70] to very well correlate with subjective evaluation of image sharpness degradation with 0 corresponding to the lowest sharpness degradation and 1 to the highest degradation.
The described method might be biased at the presence of substantial noise.To eliminate this, in this example, both visual-and thermal-range sequences were acquired in lighting and sensor conditions to minimize the noise level.Table 1 shows the results of the comparison, using this measure, between images presented in Figures 8 and 9 and their individual fragments before and after applying the described superresolution process.It is clearly seen from the table that the super-resolved images present better quality in terms of this quantitative quality measure.
Generation of output frames with preservation of real motion
The algorithm of generating the stabilized output frame F ( p ) is defined by where "•" denotes elementwise matrix multiplication, I ( p ) is the estimation of the stable scene as described in Section 3.2 or the super-resolved stable scene as described in Section 3.4, I ( p ) is the current processed frame (t), DFM is the "distancefrom-median" mask described in Section 3.3.1,and RMSM is the real-motion separation mask detailed in Section 3.3.2. Figure 10 illustrates results of the described turbulence compensation process.Figure 10( frame (marked with a white arrow) is retained, while the turbulence-induced distortion of the still rim situated on the frame's left-hand side (marked with striped arrows) is removed.
THERMAL-RANGE IMAGE FUSION FOR DENOISING AND RESOLUTION ENHANCEMENT
As detailed in Section 2, the first stage of the fusion algorithm consists of intrachannel-interframe processing.The visualrange channel processing was described in Section 3. The thermal channel processing for sensor noise filtering and resolution enhancement by means of 3D (spatial-temporal) local adaptive filtering is depicted in this section.
Channel characterization and filtering principle
Thermal sensors suffer from substantial additive noise and low image resolution.The thermal sensor noise can be described in terms of the spatial (x, y) and temporal (t) axes using 3D noise models [71,72].Resolution degradation is associated with the finite aperture of the sensor sensitive cells.Video frames usually exhibit high spatial and temporal redundancy that can be exploited for substantial noise suppression and resolution enhancement.In [48,73], a sliding window transform domain two-dimensional (2D) filtering for still image restoration is described.In this paper, an extension of this method to three-dimensional (3D) spatial/temporal denoising is suggested for thermal image sequence processing [13].A block diagram of the filtering is shown in Figure 11.For each position of the window, the DFT or the discrete cosine transform (DCT) of the signal volume within the spatial/temporal window is recursively computed from that of the previous position of the window.Recursive computation substantially reduces the filter's computational complexity [73,74].The signal's spectral coefficients are then subjected to soft or hard thresholding according to where β and β represent input and modified transform coefficients, correspondingly, and λ represents the set of coefficients of the frequency response of the camera (spatial and temporal indices are omitted for the sake of brevity).The division of image spectra by frequency response of the camera is the implementation of camera aperture correction by means of pseudoinverse filtering [48].
The window spectra which are modified in this way are then used to generate the current image sample of the output, by means of the inverse transform of the modified spectrum.Note that, in this process, the inverse transform need not be computed for all pixels within the window, but only for the central sample, since only the central sample has to be determined in order to form the output signal.
Tests and results
For the purpose of testing, two sets of artificial movies were generated, having various levels of additive Gaussian noise.The first artificial test movie contains bars with different spatial frequencies and contrasts, and the second is of a fragment of a text.Figure 12 shows results of applying a 3D filtering for image denoising.The images in Figures 12(a) and 12(b) correspond to the original frames.Figures 12(c) and 12(d) show the corresponding frames originating from a sequence possessing temporal and spatial random additive noise.Figures 12(e) and 12(f) show corresponding frames obtained using 3D filtering.Numerical results on noise suppression capability of the filtering obtained for the test images, in terms of residual filtering error, are provided in Table 2.These images and the table data clearly demonstrate the high noise suppression capability of the filtering stage.Full videos can be found in [27].
The results of 3D filtering of real-life video sequences are illustrated in Figure 13.Figures 13(a) and 13(c) are frames taken from real-life thermal sequences; Figures 13(b) and 13(d) are the corresponding frames from the filtered sequences.As one can see, while noise is substantially suppressed, object edges in the scene are not only well preserved but even sharpened-thanks to aperture correction implemented in the filtering in addition to noise suppression.
Fusion principles
In accordance with the linear theory of data fusion for image restoration [75], the interchannel fusion process is implemented as a linear combination of thermal-and visual-range channel frames: where I Thermal Several methods for assigning weight coefficients for data acquired from dissimilar sensors' modalities are known [16][17][18][19]25].Those methods suggest applying a single metric for each channel.This means that the weights are extracted using only one feature of the acquired images.As the aim of the fusion process in the visual observation systems is presenting a superior output (in human observation terms), typically the visual output quality of observation systems is defined by several criteria, such as edge preservation, noise presence, and how active are different areas of the scene.This implies that a composite assignment of weight coefficients, based on those criteria, has to be formulated.To this end, we compose both w Thermal p and w Visual p of three sets of weights as The first set of weights w VI is associated with user-defined "visual importance" ("VI") in the thermal and visual channels.The second set of weights w Noise suggests using noise estimation techniques in the fusion process for noise reduction in the fused output.Many observation system applications are intended to evaluate activity of a scene, for example, a car entering a driveway or people in motion.Therefore, the third set of weights w Motion is designed to represent the activity level of the scene.Methods for computation of w VI p , w Noise
Visual importance weights (1) Visual channel
Weighing fused images with local weights determined by visual importance of sequences was suggested in [13,25].The local spatial/time variances were suggested as the visualrange weights.However, local-variance weighing has some limitations associated with it.First, neighborhoods with only moderate changes in the visual images are assigned with zero weights and are omitted from the fused output even if they may be important visually.Other limitations are due to frequent changes of the same sample's neighborhood variance in sequential frames.This may cause flickering in the output fused sequence and make the observation task more difficult.This is most common in background areas and in areas which are highly affected by noise.As the presence of noise manifests itself in higher local variances, using this criterion will boost noise presence in the output fused image.
The flickering effect can be significantly reduced by using temporal smoothing of the weights.The noise boost presented by the visual-channel VI-weights is dealt with in Section 5.2.2.In order to cope with the omission of visual data, we propose to compute visual VI-weights as follows: where w Visual p are the computed weights in location, ( p ) and σ V p are local intensity standard deviations computed in a spatial running window centered in p, and g Visual 8) and (10) for the visual and thermal channels.
Based on the test videos used, g Visual 1 and g Visual 2 were selected to be 1 and 10, respectively. (
2) Thermal channel
The thermal channel VI-weights are specified under the assumption that importance of pixels in the thermal image is determined by their contrast with respect to their background and they are defined as [13,25] where I IR p is the input frame from the thermal channel and I IR p is its local average estimates.
As images are usually highly inhomogeneous, the weight for each pixel should be controlled by its spatial neighborhood.The selection of the size of the neighborhood is application-driven.In our implementation, it is user-selected and is defined as twice the size of the details of objects of interest.Different techniques can be used for estimating the average over the pixel neighborhood, such as local-mean and median [76].Both methods have shown good results in experiments without undesired artifacts.
As for background or smooth areas, a similarity can be drawn between the visual and thermal weights.In both weighing mechanisms, those areas are assigned to have weights equal to zero and are omitted from the output image.Therefore, it is suggested to use the user-defined scalars, g IR 1 and g IR 2 , in the same manner.This brings (9) into the following format: The considerations for setting the values of g Visual We illustrate the described VI-controlled interchannel image fusion in Figure 14. Figure 14(c The brick wall in the image is built from bricks with poles of cement holding them together.The location of the poles might become crucial for military and civil-engineering applications.While it is quite difficult to see the poles in Figure 14(c), they are clearly noticeable in Figure 14(d).This is also true for the hot spots that appear in the field in the lower-left part of the image.Those spots are seen in more detail in Figure 14(d).
Noise-defined weights
We assume that sensor noise acting in each channel can be modeled as additive white signal-independent Gaussian noise [8,77,78].It follows from the linear theory of data fusion for image restoration [79] that noise-defined weights assigned to each sample of the input channels should be proportional to the signal-to-noise ratio (SNR): where σ V p and σ IR p are the image local standard deviations in visual-and thermal-range channels, and N V p and N IR p are the corresponding channel noise standard deviations for the sample neighborhood centered at position p.
Two methods for evaluating the noise level of every pixel over its neighborhood may be considered: noise variance through evaluation of noise floor in image local spectra in a running window [76,79].
The estimation of the noise level yields a quantity measure for each sample.The lower the pixel's noise level estimate is, the heavier the weight assigned to it will be: Figure 15 illustrates weighing fused images according to their local SNR estimates.Figure 15(c) presents the output when fusing Figures 15(a (3) Details are better presented.The target of interest might not be the power plant itself, but its surrounding.Observing Figure 15(d) reveals more details and allows the observer to make better decisions (dotted arrows ).Additionally, more details can be extracted from the buildings themselves.The chessboard arrows ( ) point to the building floors which are spotted in Figure 15(d) and not in Figure 15(c).
Quantitative assessment of the noise levels in Figures 15(c) and 15(d) is presented in Figure 16 that shows the rowwise average power spectra of Figures 15(c) and 15(d) which were fused with (solid) and without (dotted) noise-defined weights.One can see from this figure that noise floor in the fused image generated with noise-defined weights is substantially lower.
Motion-defined weights
Observation system applications frequently require evaluation of activity of a scene in time.This section suggests a fusion mechanism, which assigns moving objects in the scene with heavier weights.To accomplish that, a quantitative real-motion certainty-level measurement denoting the confidence level of whether this sample is a part of a real moving object, as described in Section 3.3, is used to assign input samples with a weight proportional to their motion level.
Figure 17 presents a typical road scene where a car (marked with striped arrows) is followed by a bus or a truck (marked with blank arrows).The car happens to be very hot, and therefore it exhibits itself as a bright spot in the thermal channel (see Figure 17(a)).The truck is bigger and cooler than the car, and it manifests itself in the visual channel.Both the car and the truck are assigned with higher motion weights in the corresponding channels.The motion-vectordefined weight matrices of the thermal and visual images are shown in Figures 17(a Figure 18(a) shows an image that was fused using noisedefined and VI-weights, as described in Sections 5.2.1 and 5.2.2, with no motion taken into consideration.It might be difficult to track the vehicles in these images.Modification of the fusion scheme to include motion-defined weights resulted in the output fused image presented in Figure 18(b) in which both car and truck can be spotted much easier than in Figure 18(a)) (see marked arrow).
CONCLUSIONS
A new multichannel video fusion algorithm, for longdistance terrestrial observation systems, has been proposed.It utilizes spatial and temporal intrachannel-interframe and intrachannel fusion.In intrachannel-interframe fusion, new methods are suggested for importance, local SNR level, and local motion activity.While each of the described methods can stand on its own and has shown good results, the full visual-and thermal-range image fusion system presented here makes use of them all simultaneously to yield a better system in terms of visual quality.Experiments with synthetic test sequences, as well as with real-life image sequences, have shown that the output of this system is a substantial improvement over the sensor inputs.
Figure 1 :
Figure 1: Fusion algorithm flow diagram: visual-range spatial and temporal image fusion for image stabilization and super-resolution (upper branch) and thermal-range spatial and temporal image fusion for image denoising and resolution enhancement (bottom branch).
Figure 2 :
Figure 2: Flow diagram of video processing in visual-range channel. (b).
Figure 3 :
Figure 3: Temporal median rank filter as an estimation of the stable scene: (a) is a frame extracted from a turbulent degraded reallife video sequence, while (b) is the stable scene estimation using pixelwise temporal median.
Figure 4 :
Figure 4: Motion extraction and discrimination: (a) is a frame that has been extracted from a real-life turbulent degraded thermal sequence; (b) depicts the stable scene estimation computed over 117 frames; (c) is the result of real-motion extraction of phase I, while (d) is the result of real motion extracted after phase II.
Figure 5 :
Figure 5: Magnitude-driven mask (MDM) certainty level as a function of the motion vector's magnitude.
Figure 6 :Figure 7 :
Figure 6: Angle-driven mask (ADM) certainty level as a function of the motion vector's local spatial standard deviation.
Figure 8 :
Figure 8: Super-resolution through turbulent motion-visual-range sequence.(a) shows a raw video frame; (b) shows a super-resolved frame generated from a visual-range turbulent degraded real-life video; (c)-(d) are the magnified fragments marked on (b)-the left-hand side shows the fragment with simple interpolation of the initial resolution and the right-hand side shows the fragment with super-resolution.
(b) is its super-resolved stable one.Figures 8(c) and 8(d) are magnified fragments from Figure 8(b).The fragments are marked with black boxes on Figure 8(a).In both Figures 8(c) and 8(d), the original fragments are shown on the left-hand side, while the superresolved fragments are shown on the right-hand side.
(a) are presented in Figures 9(b) and 9(c), in which fragments with initial resolution are given on the left-hand side, while the super-resolved fragments, extracted from Figure 9(a), are given on the righthand side.
Figure 9 :
Figure 9: Super-resolution through turbulent motion.(a) presents a super-resolved frame generated from a thermal-range turbulent degraded real-life video; (b)-(c) are the magnified fragments marked on (a)-the left-hand side shows the fragment with simple interpolation of the initial resolution and the right-hand side shows the fragment with super-resolution.
a) is a frame extracted from a real-life turbulent degraded image (see [26]), and Figure 10(b) shows the stabilized frame.As one can notice, the motion of the flying bird located near the upperleft corner of the plaque on the right-hand side of the
Figure 10 :
Figure 10: Turbulence compensation: (a) is a frame extracted from a turbulent degraded sequence, while (b) is the corresponding turbulent compensated frame.
p and I
Visual p are pixel intensities in thermal and visual channels, correspondingly, and w Thermal p and w Visual p are the related channel weights.
Motion p are described in Sections 5.2.1, 5.2.2, and 5.2.3, respectively.
Table 2 : 3 Figure 12 :Figure 13 :
Figure 12: Local adaptive 3D sliding window DCT domain filtering for denoising video sequences.Figures (a) and (b) show noise-free test image frames.Figures (c) and (d) are corresponding images with additive white Gaussian noise.Figures (e) and (f) are corresponding filtered frames.
scalars that secure nonzero contribution of the channel in uniform areas, where the local standard deviation is small.Scalars g Visual 1 and g Visual 2 are set by the user and are application-dependent.For instance, if the user would like to emphasize edges and changes in higher frequencies, he would choose large g Visual 2 with relation to g Visual 1 .However, this might result in flickering output and omission of visual information of uniform areas from the composite output.
Figure 14 :
Figure 14: Fusing visual-and thermal-range channel images using two described methods for computing the VI-weights.Figure (c) is the fused image using variance and distance from the local average as weights for the visual-and thermal-range channels, respectively.Figure (d) presents the same input images (a) and (b) fused using VI-weights as defined by (8) and(10) for the visual and thermal channels.
the ones, described under Section 5.2.1(1), used to set g Visual 1 and g Visual 2 .
Figure 15 :
Figure 15: Fusion applying noise-defined weights.Figure (c) is the fused output of Figures (a) and (b) using VI-weights.Figure (d) represents the same input images fused using VI-weights and noise-defined weights.
Figure 14 (
Figure 14(d) shows the same input frames fused applying g 1 and g 2 on each channel.The brick wall in the image is built from bricks with poles of cement holding them together.The location of the poles might become crucial for military and civil-engineering applications.While it is quite difficult to see the poles in Figure14(c), they are clearly noticeable in Figure14(d).This is also true for the hot spots that appear in the field in the lower-left part of the image.Those spots are seen in more detail in Figure14(d).
(i) estimation of the additive noise variance through local autocorrelation function in a running window; (ii) estimation of the additive Advances images according to their local signal-to-noise ratio
Figure 16 :
Figure 16: Rowwise mean power spectra of image fused with (solid) and without (dotted) SNR weighing.
( 1 )
Figure 15 illustrates weighing fused images according to their local SNR estimates.Figure 15(c) presents the output when fusing Figures 15(a) and 15(b), applying only VIweights.Figure 15(d) shows the same two input frames fused while applying VI-weights along with noise-defined weights.The evaluation of the additive noise variance was performed through analysis of image local correlation function.Local SNRs were evaluated in a moving window of 11 × 11 pixels.In evaluating images of Figure 15, observation professionals have pointed out what follows.(1) Background noise reduction (see areas pointed by blank arrows ): on video sequences, this type of noise tends to flicker and annoy the user observing the video for several hours.(2) Edges preservation (see areas indicated by striped arrows ): one can easily notice how the building edges are better presented in Figure 15(d).(3) Details are better presented.The target of interest might not be the power plant itself, but its surrounding.Observing Figure 15(d) reveals more details and allows the observer to make better decisions (dotted arrows ).Additionally, more details can be extracted from the buildings themselves.The chessboard arrows ( ) point to the building floors which are spotted in Figure 15(d) and not in Figure 15(c).
) and 17(b), respectively, where heavier weights are shown in darker pixels.
( 1 )
compensation for visual-range atmospheric turbulence distortions, (2) achieving super-resolution in turbulence-compensated videos, (3) image denoising and resolution enhancement in thermal videos.The former two methods are based on local (elastic) image registration and resampling.The third method implements real-time 3D spatial-temporal sliding window filtering in the DCT domain.The final interchannel fusion is achieved through a technique based on the local weighted average method with weights controlled by the pixel's local visual
Figure 17 :Figure 18 :
Figure 17: Motion weights extracted from real-life sequences.(a) is a sample frame from the thermal channel; (b) is the corresponding frame from the visual-range one.(c)-(d) are the matching motion-defined weights. | 9,703.4 | 2008-02-12T00:00:00.000 | [
"Engineering"
] |
BMEFIQA: Blind Quality Assessment of Multi-Exposure Fused Images Based on Several Characteristics
A multi-exposure fused (MEF) image is generated by multiple images with different exposure levels, but the transformation process will inevitably introduce various distortions. Therefore, it is worth discussing how to evaluate the visual quality of MEF images. This paper proposes a new blind quality assessment method for MEF images by considering their characteristics, and it is dubbed as BMEFIQA. More specifically, multiple features that represent different image attributes are extracted to perceive the various distortions of MEF images. Among them, structural, naturalness, and colorfulness features are utilized to describe the phenomena of structure destruction, unnatural presentation, and color distortion, respectively. All the captured features constitute a final feature vector for quality regression via random forest. Experimental results on a publicly available database show the superiority of the proposed BMEFIQA method to several blind quality assessment methods.
Introduction
In view of the increasing development of image processing technologies, it has become more feasible to help humans perceive the real world in high-quality images. For example, multi-exposure fusion (MEF) and high dynamic range (HDR) imaging both belong to image enhancement technology, which can provide excellent detail information and an ideal, natural appearance [1]. HDR imaging technology requires additional processing, such as generation, image conversion to a low dynamic range image, and visualization on common displays. Unfortunately, all of these procedures can produce artifacts that affect the quality of HDR images [2]. Multi-exposure fusion technology is relatively simple, it does not need intermediate operations, and it has also been applied in some practical fields [3]. However, the unique process of integration from multiple different exposure images into one ultimate image will result in distortion due to the fusion weighting assignment. Obviously, the distortion will influence human perception and result in different human opinion scores. Hence, to measure the distortion degree objectively, it is necessary to develop a quality assessment method for MEF images.
In recent years, many researchers have developed some image quality assessment (IQA) methods [4][5][6][7][8][9][10][11][12][13][14]. Those methods can be categorized into full-reference (FR) methods, reduced-reference (RR) methods, and no-reference (NR) methods. The FR and RR methods both need original image information to make a comparison; however, it is opposite to the truth that there is never a pre-defined reference in the real application. Therefore, it is more urgent to design NR methods for IQA tasks.
Based on the considerations of the inherent characteristics of MEF images, a new NR-IQA method is proposed in this paper to predict the quality of MEF images more accurately, and it is named BMEFIQA.
The main contributions are detailed as follows: (1) Inspired by the various characteristics of MEF images, structural, naturalness, and colorfulness features are extracted from different standpoints to perceive their distortion. (2) On account of structure loss produced by abnormal exposure, the exposure map is weighted to the gradient similarity to detect structural distortion. (3) The experimental results demonstrate that the proposed method is competent for MEF images and superior to several NR-IQA methods.
The remainder of this paper is constructed as follows: In Section 2, related works are presented. In Section 3, the proposed BMEFIQA method is described in detail. The experimental results and analysis based on a public MEF image database are presented in Section 4. Finally, the conclusion is drawn in Section 5.
Related Works
Currently, there are many efficient NR-IQA methods for ordinary images. For instance, Moorthy et al. [5] proposed an NR-IQA method named the distortion identification-based image verity and integrity evaluation (DIIVINE), which is based on natural scene statistics (NSSs). In [6], Saad et al. utilized an NSS model of discrete cosine transform coefficients to design the IQA method, which is referred to as BLINDS-II. In [7], Mittal et al. extracted features from the empirical distribution of locally normalized luminance and the products of locally normalized luminance in the spatial domain; the method is named BRISQUE. Liu et al. [8] constructed the CurveletQA method by utilizing curvelet transform to extract a set of statistical features. Xue et al. [9] combined the gradient magnitude map with the Laplacian of Gaussian response to perceive the structural information of images and showed highly competitive performance, dubbed as GradLog. Fang et al. [10] employed the degree of deviation from NSS models to represent the unnatural character of contrastdistorted images, which is termed ContrastQA. Li et al. [11] proposed an NR-IQA method based on structural degradation, which is described by the gradient-weighted histogram of local binary pattern (LBP) calculation on the gradient map (GWH-GLBP). Liu et al. [12] developed the oriented gradient (OG) IQA method by studying the quality relevance of the relative gradient orientation. Gu et al. [13] combined local and global considerations to design an NR-IQA method called NIQMC. Oszust [14] captured the information carried by image derivatives of different orders by local features and used it for image quality prediction. Zhang et al. [15] integrated the features of natural image statistics and created a multivariate Gaussian model of image patches for quality assessment. Xu et al. [16] proposed an NR-IQA method based on high-order statistics aggregation, which needs a small codebook.
Except for the above IQA methods for ordinary images, some studies have also contributed to predicting the quality of tone-mapped images. For example, Gu et al. [17] devised an effective blind tone-mapped quality index (BTMQI) via the analysis of information, naturalness, and structure. Kundu et al. [18] derived the HDR image gradient-based evaluator (HIGRADE) based on standard measurements of the bandpass and newly conceived differential NSS. Although the above IQA methods for ordinary and tone-mapped images showed good performance, they are not appropriate for the prediction of the quality of MEF images, as their distortion type is distinguished from the above two types of images.
For MEF images, there also exist some IQA methods, which belong to FR methods. For instance, Zheng et al. [19] proposed a quantitative metric called the ratio of spatial frequency error; this is derived from the definition of spatial frequency, which reflects local intensity variation. Ma et al. [20] proposed an objective method based on the principle of structural similarity and a novel measure of patch structural consistency. Xing et al. [21] combined contrast information with structural similarity and saturation similarity to predict the quality of an MEF image. Fang et al. [22] utilized a pyramid subband contrast preservation scheme and an information theory-adaptive pooling strategy to establish a quality assessment method for MEF images of both static and dynamic scenes. Deng et al. [23] designed a method by extracting color, texture, and structural features. Martinez et al. [24] utilized the multi-scale scheme to compute structural similarities. Considering the limitation of such FR methods, which need reference information, an NR method was proposed in our previous work [25]. However, there is room for improvement. Therefore, to predict the quality of an MEF image more accurately, its characteristics should be taken into consideration more comprehensively.
The Proposed BMEFIQA Method
Since MEF images are fused by multiple images with different exposure levels [26], the weight assignment process will introduce some artifacts, such as detail loss, structural degradation, unnaturalness, and color distortion. Figure 1 gives a vivid presentation of MEF images generated by three different MEF algorithms, which show the different visual effects. Figure 1a, generated by Merten's algorithm [27], has the highest mean opinion score (MOS) and is the best quality. The detailed information is well preserved, except for the over-exposed part (i.e., outside the entrance of the cave). Moreover, it also has much more abundant color information than the others, and it seems more natural. At a first glance of Figure 1b, which is generated by Raman's algorithm [28], the blackened scene is the initial perception of humans. In fact, the brightness decline also leads to detail and color information loss. The MEF image generated by local energy weighting [29] shows proper brightness, but it loses the original naturalness, and it also introduces some artifacts around the wall and stones. Therefore, structural, naturalness, and color information are the main factors that influence the visual quality of MEF images. To fill the gap of quality prediction for MEF images under the condition with a lack of reference information, a novel method named BMEFIQA is proposed in this paper, which considers three factors (i.e., structure, naturalness, and colorfulness). Figure 2 presents the pipeline of the proposed BMEFIQA method. Specifically, three types of quality-sensitive features corresponding to the above three factors are extracted to generate overall feature vectors. Then, random forest is utilized to train a quality regression model, thus aggregating all excavated features. More details are given in the following subsections.
Structural Features
Structural information usually carries the basic visual content of a scene, and the human visual system (HVS) has strong adaptability to extract the structure for visual perception [30]. For an MEF image, distortion introduction will always destroy its structural information, such as abnormal exposure (i.e., over-exposure or under-exposure). Therefore, the visual quality of an MEF image can be determined by measuring whether the structural information is damaged or not, especially for over-exposed and under-exposed regions. First, the exposure of the MEF image needs to be measured to distinguish the regions of over-exposure and under-exposure. Given MEF image I and converting it into a gray-scale one, its corresponding exposure map E I can be calculated by measuring the distance between its normalized pixel intensity and the constant 0.5. Specifically, when the normalized pixel intensity is close to 0 or 1, the corresponding pixel is regarded as under-exposed or over-exposed. Therefore, E I can be defined as where I y is the normalized pixel intensity of the gray-scaled MEF image I, and τ is the standard variance of the Gaussian function, which is set to 0.2 according to previous experience [31]. Figure 3 shows three exposure maps of the MEF images in Figure 1. By comparing with Figure 1, it can be observed that the brighter regions are the under-exposed or over-exposed regions, and the fainter regions represent the normal exposed regions in Figure 3. As an effective structural information feature, the gradient is utilized to describe the structure loss phenomenon. To further measure the influence of abnormal exposure on the acquisition of gradient information, some fake MEF images are obtained by darkening and brightening the real MEF images, which are denoted as I f . They can be produced as follows: where C = c, c∈{1/3.5, 1/5, 1/6.5, 3.5, 5, 6.5}, and I r represents the gray-scaled MEF image I. After that, six fake MEF images can be generated. Then, gradient maps G of such fake MEF images are calculated as where p x and p y are the Prewitt filters, along with the horizontal and vertical directions, respectively. ⊗ indicates the convolution operator. Additionally, the gradient map of the real MEF image I is also obtained by (3), which is denoted as G I . Then, the gradient similarities between I and the generated six fake MEF images are calculated. Let G s be the similarities of G and G I , which are calculated to quantify the influence of abnormal exposure. They are defined as where C 1 is a constant to avoid zero denominators, which is set to 10 −8 .
Since the distortions of the over-exposed and the under-exposed regions are easier for the human eye to perceive, combining the gradient similarities G s with the exposure map E I can make the distortion perception more accurate. Therefore, gradient combined with exposure weighting is defined as where k and f are the indexes of the horizontal and vertical pixels.
As the same number of fake MEF images, after combining six gradient similarity maps with E I , the obtained G e is a 6-dimensional feature that describes the structure loss of the MEF image.
From the statistical perspective, the NSS model is utilized in the gradient domain to represent the structural variation in the MEF images. Specifically, the gradient map G I is processed by the local mean subtraction and divisive normalization to obtain the mean subtracted contrast normalized (MSCN) coefficients [7], which are expressed bŷ whereĜ I (i, j) are the MSCN coefficients of G I at the position of (i, j). µ(i, j) and δ(i, j) are the local mean and standard deviation of G I (i, j), respectively. Different degrees of structural distortion will inevitably affect the distribution of the MSCN coefficients of G I . As the distribution has a Gaussian-like appearance, a generalized Gaussian distribution (GGD) is utilized to match the MSCN coefficients, and the mathematical expression is given by where β = σ Γ(1/α)Γ(3/α), Γ(·) represents the gamma function. The parameters α and σ 2 represent the shape and variance of the Gaussian distribution, respectively. Therefore, the two parameters α and σ 2 are taken as the 2-dimensional global structural features. In addition, the pairwise products of neighboring MSCN coefficients are also calculated to capture the relationship of the neighboring pixels. The MSCN coefficients are processed in four directions, and this process is expressed as where H(i, j), V(i, j), D 1 (i, j), and D 2 (i, j) are the results processed along with the horizontal, vertical, main diagonal, and sub-diagonal directions, respectively.
Then, an asymmetric generalized Gaussian distribution (AGGD) is utilized to fit each pairwise product, which is defined as where The above parameters v, ϕ, ω 2 l , and ω 2 r constitute the compensation features that perceive the global structural distortion. As a result, the compensation features calculated in the four directions form a 16-dimensional feature vector.
However, the phenomena of structure loss also contain the loss of detailed information, which can be measured via entropy. Block entropy is calculated to perceive the local detail information variation, which is denoted as e b . After calculating the local entropy in 8 × 8 blocks of each MEF image, the entropy calculation is used to measure the distribution of the obtained local entropy, which can be expressed as where m is the block number of each MEF image, and p(·) is the probability density of the m-th block entropy value. Furthermore, the mean value and standard deviation of the block entropy are also calculated to measure the overall detail loss, which are denoted as e m and e s , respectively. Finally, e, e m , and e s are integrated to form another set of 3-dimensional structural features.
Naturalness Features
Generally, a high-quality MEF image has a natural-like appearance. MEF algorithms may disrupt the natural statistical regularities in the spatial domain. From a global perspective, the naturalness can be quantified via the NSS-based model. An MEF image, I, is converted to a gray-scale image, I r , and then the MSCN coefficients of I r are calculated according to Equation (6). Moreover, the GGD model defined in Equation (7) is utilized to capture the statistical property, and the shape and variance parameters are taken as the first group of global naturalness features.
From another perspective, naturalness is also affected by the overall brightness and contrast, which is seriously influenced by under-exposed and over-exposed conditions. It has been demonstrated that the mean and standard deviation values of the image intensity can represent the brightness and contrast of an image [32]; the entropy can also describe the distortion of brightness. Therefore, these moment features (mean and standard deviation) and the entropy feature are applied to build NSS models to capture the naturalness variation undergoing the distortions. Among them, Gaussian probability density functions are used to fit the mean and standard deviation. The specific definitions are as follows: where a and s are the mean and standard deviation values, respectively. d a and d s are the possibilities of the MEF image being natural when a and s are given. The parameters in Equations (12) and (13) are set to ξ a = 26.063, τ a = 118.559, ξ s = 12.858, and τ s = 57.274, respectively. In addition, the entropy feature is fitted by the extreme value probability density function, which is defined as where o is the entropy value of I r .
Colorfulness Features
Abundant color information is important for an outstanding MEF image, as it illustrates that the image has proper color saturation and realistic scene chroma. The weight assignment processes in the different MEF algorithms may emphasize different parts of the image content, as different humans do not always focus on the same things. As Figure 1 shows, different emphasis brings different distortion. Nevertheless, the color distortion measure is necessary and right. In previous work, it was proved that the color perception of human vision is mainly processed in the opponent color space [33]. Therefore, the opponent color space is obtained by the red-green-blue (RGB) color channels in the RGB color space. The transformation processes are expressed as T 1 = R − G, T 2 = (R + G)/2 − B; T 1 is the obtained red-green channel, and T 2 is the obtained yellow-blue channel.
A combination of the two transformed channels is utilized to represent the colorfulness of the MEF image. The specific definition is as follows: where γ 2 T 1 and γ 2 T 2 denote the variances in T 1 and T 2 , respectively. a T 1 and a T 2 denote the mean value of T 1 and T 2 , respectively.
In Figure 1, it can be seen that over-exposure and under-exposure will both affect the representation of excellent color with a decrease in contrast. Therefore, the contrast energies are calculated in the opponent color space to perceive the color distortion. Let C e be the calculated contrast energy; it is obtained by where g ∈ {T 1 , T 2 }; I T 1 and I T 2 denote the red-green and yellow-blue channels of I, respectively. ρ I g = I g ⊗ p x 2 + I g ⊗ p y 2 ; χ is the contrast gain to correct and normalize all filter responses, ε is the maximum value of ρ I g , and ϑ g is the noise threshold. χ and ϑ g are fixed at the same settings as in [34]. Finally, the mean value of C e is calculated as another feature except for C t . In fact, according to the opponent color space, two channels can be used to perform the contrast energy calculation process, and 2-dimensional features are obtained.
Moreover, to obtain a more complete sense of color distortion, the NSS model is also executed in the two opponent color channels. Specifically, T 1 and T 2 are processed with the MSCN coefficient calculation defined in Equation (6) and the GGD model defined in Equation (7), respectively. The obtained shape and variance parameters constitute 4-dimensional complementary color features.
Feature Aggregation and Quality Regression
So far, 39-dimensional features are extracted via structure, naturalness, and colorfulness analyses, denoted as S F , N F , and C F , respectively. According to the obtained quality-sensitive features, the mapping relationship between the features and human subjective scores should be learned to achieve the purpose of quality prediction. As an effective regression manner, random forest (RF) is used to pool the high-dimensional features to indicate the quality of the MEF image. The specific expression can be represented as follows: where Q is the final quality score of the MEF image; η(·) denotes the mapping function by RF; and F denotes the final all-feature vectors, F = {S F , N F , C F }.
Experimental Protocol
In the experiment, an open MEF subjective assessment database [35] is selected to evaluate the performance of the proposed BMEFIQA method. The database consists of 136 MEF images, which portray different scenarios. The corresponding perceptual quality score is subjectively tested by numerous observers, that is, human subjective scores, and the MOS values of each MEF image are obtained from them. These MEF images are generated by distinct types of MEF algorithms, including local energy weighted linear combination, global energy weighted linear combination, Li12 [36], Raman09 [28], ShutaoLi12 [37], Mertens07 [27], Gu12 [29], and ShutaoLi13 [38]. Specifically, 136 MEF images are derived from 17 source multi-exposure images. The detailed contents are reported in Table 1 and Figure 4. Three standard performance evaluation criteria are utilized to evaluate the performance of the proposed method, and they include the Person linear correlation coefficient (PLCC), Spearman's rank-order correlation coefficient (SROCC), and root-mean-square error (RMSE). Specifically, PLCC, SROCC, and RMSE are used to evaluate the prediction accuracy, prediction monotonicity, and prediction error, respectively. Moreover, before the calculation of the PLCC value, a five-parameter logistic regression is used to perform nonlinear mapping between the predicted score and the subjective quality scores; the definition is given by where Q f is the fitted score, and λ 1 , λ 2 , λ 3 , λ 4 , and λ 5 are the fitting parameters, which can be obtained via the nlinfit function in MATLAB software. The initial values of these five fitting parameters are guided by the video quality expert group (VQEG) [39]. The final values are determined by nonlinear least-squares optimization between the subjective quality scores and Q f . When calculating the PLCC value, the input (i.e., predicted score Q) should be replaced by Q f . An excellent quality assessment method should always have higher PLCC and SROCC values and a lower RMSE value. To obtain a robust criterion, we employ 17-fold crossvalidation to test the proposed BMEFIQA method. Specifically, the MEF image database is divided into 17 subsets, 16 of which are used for RF model training, and the remaining are used for testing. Each subset contains MEF images belonging to the same scene. Finally, we report the average criteria values (i.e., PLCC, SROCC, and RMSE) across all fold trails when the train/test cycles are over for all the scenes.
Performance Comparison
To verify the superiority of the proposed method, two types of the NR-based method, which has demonstrated effectiveness for other images, are used to make the performance comparison. The first type contains ten methods, which are designed for ordinary images, dubbed as 2D NRIQA, namely, DIIVINE [5], BLIND-II [6], BRISQUE [7], CurveletQA [8], GradLog [9], ConstrastQA [10], GLBP [11], OG [12], NIQMC [13], and SCORER [14]. The second type contains three methods designed for tone-mapped images, dubbed as TM-NRIQA, namely, BTMQI [17], HIGRADE-1 [18], and HIGRADE-2 [18]. All the comparison methods are learning-based ones following the 17-fold cross-validation to obtain their performance for MEF images, which is consistent with the proposed method. To ensure that the results are not biased, the performances of the comparison methods are all obtained by running the original released codes from the authors.
The overall performance comparison results are presented in Table 2. To highlight the best performance, the PLCC, SROCC, and RMSE values are shown in bold. Obviously, the proposed BMEFIQA method reveals its better performance compared to the other comparison methods, which verifies its validity. We can draw some observations from Table 2. On the one hand, most 2D NRIQA methods exhibit poor performance, except for GradLog; the worst performance indicators for PLCC and SROCC are only 0.163 and 0.113, respectively. The reasons for this phenomenon can be roughly summarized as follows: First, the distortion of an MEF image presents differently from an ordinary 2D image, such as difficult semantic understanding and color deterioration when suffering from over-exposure and under-exposure. Second, the GradLog method combines gradient amplitude maps with the Laplace of Gaussian responses to sense structural information, which can describe the structural distortion of MEF images well. Hence, its performance outperforms other comparison 2D NRIQA methods. On the other hand, compared with 2D NRIQA methods, TM-NRIQA methods achieve better performance. This can be attributed to the fact that the distortion of tone-mapped images seems similar to that of MEF images. However, such results should also be improved to predict the quality of MEF images more accurately. The proposed BMEFIQA method considers the structure, naturalness, and colorfulness of an MEF image from three aspects. Its comprehensive consideration advocates that it obtains superior performance to the competing methods. In order to compare the performance of the different methods more intuitively, Figure 5 shows the scatter plots for the objective predicted scores and MOS values of fourteen NR-IQA methods. They are conducted using the whole MEF database, and the scatter points are fitted by the logistic function. The closer these scatter points are to the fitting line, the better the performance. In Figure 5, it can be seen that the predicted scores of the proposed method show a higher correlation with the MOS values than the other NR-IQA methods.
Impacts of Different Features and Block Sizes
Based on the above performance comparison, the proposed BMEFIQA method is demonstrated to have a good-quality prediction capability. However, for the individual features, their role in the overall method remains ambiguous. To realize their purpose, the individual features S F , N F , and C F and their different combinations are used to train the corresponding regression models to predict the quality of the MEF images separately. As shown in Table 3, the obtained regression models are written as Model-t, t ∈ {1, 2, . . . , 7}, which indicates the number of individual features and their possible combinations. Their respective performance results are also given in Table 3, and the best performance results are shown in bold. The following can be observed: First, among the individual features, compared with the naturalness features N F and color features C F , the structural features S F show a relatively strong performance. For the overall performance, S F contributes the most. Second, among the different combinations, the combination of S F and C F produces the best performance. Although the performance of C F itself is relatively weak, when it is combined with N F , its performance is improved, indicating that it provides a good auxiliary effect. Third, when all S F , N F , and C F features are incorporated together, the best performance can be obtained. Therefore, we believe that S F , N F , and C F features are complementary to each other. [6]; (c) BRISQUE [7]; (d) CurveletQA [8]; (e) GradLog [9]; (f) ContrastQA [10]; (g) GWH-GLBP [11]; (h) OG [12]; (i) NIQMC [13]; (j) SCORER [14]; (k) BT-MQI [17]; (l) HIGRADE-1 [18]; (m) HIGRADE-2 [18]; (n) Proposed. In the block entropy calculation process, as different block sizes may affect the quality prediction performance, the block size should be determined to obtain the optimal performance. In the experiment, performances are tested under different block sizes, as shown in Table 4. Specifically, the block size varies from 8 to 64, and the corresponding PLCC, SROCC, and RMSE values are given. Additionally, the run time under different block sizes is also developed. From the results, we can observe that the PLCC and SROCC values and the run time decrease with an increase in block size. Under comprehensive consideration, the block size is set to 8.
Run Time
In a practical application, in addition to performance, efficiency is also important for the methods. Therefore, the mean implementation times for all MEF images in the MEF database of the proposed method and the other competing methods are reported in Table 5. All the original codes are implemented on a Windows 10 1.80 GHz Intel Core i7-8565U CPU with 8GB RAM and NVIDIA MX 150, using MATLAB 2017b. In Table 5, it can be seen that the proposed BMEFIQA method has a moderate execution time.
Discussion
In this study, a novel NR-IQA method that can handle the quality prediction task for MEF images is proposed. Specifically, the proposed BMEFIQA method considers the structure, naturalness, and colorfulness of MEF images, and it extracts the corresponding features of these three aspects. Through the performance comparison in the previous section, the proposed method shows its potential in predicting the quality of MEF im-ages. It can be used in the application terminal to monitor quality with no reference information provided.
Although the proposed BMEFIQA method considers some image attributes, it is limited for kaleidoscopic images in real cases. They may be influenced by some special situations, which cannot be foreseen. As a result, regarding the performance of the proposed method, there is still room for improvement. For instance, the human attention mechanism will affect the perception of MEF image quality, and humans will have different artistic biases toward MEF images in different scenes. Deep learning is well known for its ability to automatically learn the features of some images. In future work, deep learning-based methods or more handcrafted features with the attention mechanism can be incorporated to improve the performance of the BMEFIQA method. Moreover, the MEF image database should be expanded for data training and testing to increase the robustness and practicability of the method.
Conclusions
This paper proposes a blind quality assessment method for multi-exposure fusion (MEF) images based on structure, naturalness, and colorfulness analyses of MEF images, named BMEFIQA. For the structure analysis, exposure map calculation is implemented to weight the gradient similarities, statistical modeling is used in the gradient domain, and entropy calculation is performed to characterize the loss of structural information. For the naturalness analysis, statistical modeling in the spatial domain combines the statistics of moment features and entropy, which are built to characterize scene naturalness. For colorfulness analysis, opponent color space is used to calculate contrast energy and build the statistical model. Finally, random forest is used to fuse the extracted features into prediction quality. Experiments on the public MEF image database demonstrate the superiority of the proposed BMEFIQA method. | 6,921.6 | 2022-02-01T00:00:00.000 | [
"Computer Science"
] |
A new methodology for detecting adhesion location in aluminum tube expansion
During the expansion forming of aluminum tube, the efficiency of heat exchanger diminishes due to the adhesion of groove and expansion ball inside the tube. Despite its importance, a limited number of researches on the adhesion problems in aluminum tube expansion have been published. This study aims to analyze the adhesion occurring during the expansion forming of aluminum tube for heat exchanger and to identify its location. For this, the method of using the statistical analysis of geometry by image processing and the slope of force measured from expansion forming was suggested. The new method discovers the adhesion location from the standard deviation of groove height measured before and after expansion of tube and the differentiation of force measured from expansion forming. To prove this method, the area with deviation of groove height above average was discovered, and it was confirmed that from the actual expansion of tube cross-sectional images, the height of some grooves was abnormally shorter due to adhesion. Also, from this method, it was confirmed that the changes in differentiation of force occurring from the expansion of the tube also include the information on adhesion location.
Introduction
The fin-tube heat exchanger is a heat transfer device used as condenser or evaporator in the refrigeration and air-conditioning sector and exchanges heat between the fluid flowing inside the tube and external environment through the fin connected with the tube. As the efficiency of heat exchanger is directly affected by the forms of the fin and the tube, the effect of various forms of fins and tubes has previously been studied by calculating thermal contact conductance between fins and tubes. 1 Furthermore, Tang et al. 2 have proposed a new methodology for improving the thermal contact conductance performance at the expansion forming process. Recently, the effect of tube geometry on heat exchanger efficiency has been studied, 3 and heat exchanger coefficients and pressure drop for various tube geometries have been computed. 4 Products with inner grooved tubes have shown increases in the heat exchange efficiency by 100% 5 and 111%-207% 6 higher than those with smooth tubes. Regarding tube materials, although most use copper as the material of tube, aluminum is considered to be used for some applications such as air conditioners because of its cheaper material and production cost as well as its lighter weight, compared to copper. Aluminum tubes are one-fourth times cheaper and 35% lighter than copper tubes. 7 Furthermore, aluminum tubes have shown higher heat transfer efficiency by 136% with inner grooved geometry than without it. 7 It has also been investigated that Al-Cu composed tubes have reduced the material cost of the heat exchangers, but guaranteed heat exchanging performance. 8 In the manufacturing process of fin-tube heat exchanger, the expansion forming that expands the outer diameter of tube in order to make sure that the tube and fin are securely joined together is required to increase heat exchanging efficiency. However, in this expansion forming, adhesion occurs between the fin and tube because the upper portion of inner grooves sticks to the surface of expansion ball. This adhesion occurs more in aluminum tubes than copper tubes and can be observed by measuring the height of inner groove. 8 Adhesion phenomenon can diminish the performance of heat exchanger as adhesion causes irregularity of tube expansion and loss of inner grooves.
Most researches done on expansion forming process and heat exchanging efficiency have been focused on the driving force during expansion and/or the level of expansion dependent on the groove geometry as well as expansion ball geometry, 9 through simple expansion experiments and finite element analysis. For the level of expansion, Lee and Park 10 have compared nonuniform grooved tube with uniform grooved tube. Almeida et al. 11 have studied the driving force during expansion dependent on expansion ball geometry, while Seibi 12 has found the effect of expansion ratio and expansion ball angle on the expansion force through simple expansion experiments. Numerical analyses have also been conducted to improve the expansion process 13 and to measure changes in the driving force between diverse expansion balls of various sizes and a smooth tube. 14 Some investigators have evaluated heat exchanger performance by measuring the expansion ratio that represents the fin-tube contact quality, 15 while others have optimized tube expanding process to maximize heat exchanging efficiency. 16 Moreover, the heat exchanging efficiency has been improved by optimizing the expanding velocity for tube expanding process, 17 and structural integrity of expansion balls and tubes has been numerically validated by calculating the stress fields produced on tubes during expansion process. 18 However, the tube material in these studies has been limited to copper, without using aluminum tubes.
Therefore, this study aims to investigate the adhesion phenomenon in aluminum tubes with inner grooves during its expansion forming process as well as the location of adhesion in aluminum tubes and to propose a new method of predicting the occurrence and the location of adhesion during expansion forming. For this, experiments simulating the expansion forming process were performed on aluminum tubes, and the images of tube cross sections before and after expansion were analyzed. Through the image processing method on the cross-sectional images captured, the adhesion phenomenon in aluminum tube expansion was investigated, and based on these results, a new method of predicting the occurrence and the location of adhesion was suggested from the driving force of aluminum tube expansion.
Adhesion in aluminum tube expansion forming
The manufacturing process of heat exchangers includes the expansion forming that joins the fin and tube. Expansion forming is a metal forming process that increases the outer diameter of tube by passing the expansion ball with diameter greater than its inner diameter through the tube ( Figure 1). Expansion balls with increased hardness, obtained by coating or heat processing, are often used for repeated expansion work. The surface of expansion balls with increased hardness is stained with aluminum after expansion forming; this is due to the adhesion phenomenon. Adhesion phenomenon is a form of abrasion and refers to the phenomenon in which, during the friction of two surfaces, the surface with lower hardness breaks due to the friction heat and sticks to the surface with higher hardness. 19 In fact, this adhesion phenomenon frequently occurs during expansion forming at factories. The adhesion of expansion ball results in the loss of product manufacturing time and cost due to replacement of expansion balls and diminished quality of heat exchangers.
Expanding experiments
During the processing procedure, it is required to identify the frequency of occurrence and occurrence time of adhesion phenomenon to prevent performance degradation of heat exchangers. To determine the adhesion phenomenon occurring from expansion forming, expansion simulation experiment was performed ( Figure 2). Expansion simulation experiment can be largely divided into expansion experiment and data collection. We create expansion using the alternating current (AC) singleaxis controller (RCS-6000G; Robostar, Korea) to push the expansion ball into the grooved aluminum tube. Here, the driving force between expansion ball and tube is measured using the load cell (CDFS-100; Bongshin, Korea; Figure 3).
Image processing for the cross section of tube
Looking at the force graph generated during expansion, over a certain level of force is required for expansion. However, it is hard to determine only from the force graph where the adhesion started. Therefore, there is a need to precisely analyze the internal information of plastic deformation by taking the cross-sectional images of the expanded tube, and for this, image analysis method was used.
The cross section of aluminum tube was taken after completion of expansion forming, and to obtain the boundary data, 330 mm of the expanded part was cut into 10 mm intervals and the cross section was taken. When taking the image of each cross section, the cross section was polished using 1 mm diamond suspension, and then, the image was taken using an optical microscope (Nikon LV100D; Japan). From the obtained cross-sectional images, the boundary data of aluminum tube were obtained using image binarization. The boundary data were divided into the outer boundary on the outside and the inner boundary on the groove side.
To calculate the radius inside the aluminum tube, each cross-sectional image was rotated and the distance from the center was calculated. The center of the tube was located in the image, the image was then rotated counterclockwise with intervals of 0.1°based on the center, and then the continuous inner and outer boundaries of the tube was reconstructed by the linear interpolation of the discrete inner and outer boundary image data ( Figure 4). The distance from the center of the tube to the continuous inner and outer boundaries was calculated according to the relation of 1 pixel = 0.0048 mm and then was aligned by the angle generated when the cross-sectional image of the tube was rotated counterclockwise. Then, the spread-out state of the cross section of the tube was plotted by the distance between the center and the inner and outer boundaries of the tube versus the rotation angle ( Figure 5). Using the obtained boundary data, the variables for the radius inside the aluminum tube and groove were defined (Table 1, Figure 6). R ga and R gb refer to the radius to the apex of the groove and to the bottom in the inner boundary, respectively, and R out refers to the radius to the outer boundary. H g is the groove height and H w is the tube thickness or the distance from the center of two points on the bottom of groove to the outer boundary, which can be calculated in the following formula To examine the relationship between the change in each variable to cross section and the adhesion phenomenon, the data obtained from each cross section were statistically processed ( Figure 7). As the total number of grooves for one cross section is 54, averages and standard deviations (SDs) of the groove height (H g ), tube thickness (H w ), and outer boundary radius (R out ) values for each cross-sectional image were calculated. In particular, when we use SD values, these values are small. Therefore, we used normalized values of SD (i.e. ratio of SD) that was calculated by the ratio of SD to the average size of the groove height before expansion and expressed as percentage for comparison. This ratio of SD (i.e. s i (%)) is formulated by equation (3), and here, m à i is the average value of the groove height before expansion and s i is the SD of the groove height data after expansion
Results and discussion
The geometric data were obtained from image processing, and then, the cross section of tube before and after expansion was compared ( Figure 8). Compared to before expansion, outer boundary radius (R out ) increased by 0.1908 mm (5.45%) and groove height (H g ) decreased by 0.0246 mm (9.49%). The tube thickness (H w ) did not show great difference compared to before expansion ( Table 1).
The H g was uneven in the area with high SD of groove height (H g ) and was low in some grooves (Figure 9(a)). The fact that the height of some grooves was greatly lower than others reflects the loss of material forming the grooves. The adhesion phenomenon occurred during expansion forming is the estimated cause of groove loss, and from the cross section of an area with large SD of H g , grooves of abnormally low height were observed (Figure 9(b)). Also, in the cross-sectional image of the same area after actual expansion, abnormally low height compared to other grooves was confirmed ( Figure 10). Therefore, it is estimated that the areas at 30, 160, 190, 230, and 250 mm with higher than average SD are the areas of adhesion.
The expansion process of aluminum tube is a process which involves expanding the ball to expand the outer wall with grooves by inducing the tube inside to undergo plastic deformation. This process is a form of plastic deformation where the force expanding the diameter of tube and the friction force between materials act as the reaction force to the expansion ball. In this process, adhesion occurs when the aluminum sticks to the expansion ball before some grooves experience sufficient plastic deformation due to increase in local friction heat. Assuming that the force required for expansion of tube is constant, the change in force during expansion includes the change in friction due to adhesion. Therefore, in this study, in order to verify the occurrence of adhesion, we examined the increase and decrease in force by differentiating the force data obtained from expansion experiment over time.
In order to examine the change in force from adhesion, more detailed understanding in adhesion phenomenon is required. Heat is continuously generated from the friction of expansion ball and groove from the expansion forming, and locally, adhesion occurs where the groove and expansion ball stick to each other. Although friction mainly occurs between the aluminum and the steel expansion ball before adhesion, local friction partly occurs between aluminum and other aluminum attached in the expansion after adhesion. Generally, the friction factor between aluminum and steel is approximately 0.45-0.61, but in contrast, the friction factor between aluminum and steel is high at 1.05-1.35. 20 Therefore, in case of adhesion, momentary change in force occurs. As this is a local phenomenon, it is hard to observe from the overall data, but the adhesion phenomenon can be observed when the force is differentiated over time and viewed by the formation of its increase and decrease ( Figure 11).
There are largely two reasons that the confirmed adhesion location does not precisely match. First, it is because the actual aluminum tube contracts in the axial direction in the expansion experiment. The aluminum tube of 400 mm before expansion contracts by 20 mm of length after expansion. The force generated during expansion was expressed relative to the location of tube before expansion, while geometric data summarize the location of tube after expansion, leading to difference in location. Second, as the geometry was obtained from the cross section of cutting the tube with an interval of 10 mm, there is less continuity of SD of groove height compared to the continuous force data. Therefore, there is a slight difference for the SD of groove height to follow the change in the derivative of the continuous force data over the tube length. When taking into consideration these aspects, it can be regarded that the locations of adhesion estimated by two methods nearly match. However, although the change in differentiation data is related to the adhesion of the groove height as explained above, the fact that the actual aluminum tube contracts in the axial direction in the expansion experiment results in the overall increase in the driving force as a function of expansion ball location ( Figure 3). Moreover, in the driving force, initially there is a sudden increase in the expansion ball location of 0 to ;5 mm ( Figure 3) that presumably occurs while the expansion ball with a length of 5 mm is being inserted.
There are several researches related to expansion forming using the driving force of expansion because it is the easiest to measure the force from related experimental data. However, as mentioned earlier, the existence of adhesion during expansion cannot be determined only from the force data, and even if the change in force is examined, it cannot be the sole basis to predict the occurrence of adhesion. Therefore, this study determined the occurrence of adhesion using the statistical data of groove height obtained from image processing of cross section after expansion and the cross-sectional images. Using these results, it was confirmed that the change in the slope of force (i.e. the differentiation of force) during expansion is closely related to adhesion.
Conclusion
This study suggests a new method to correlate differentiation of expansion force measured through expansion forming processes with a statistical analysis of the geometry derived from image processing in locating the adhesion sites occurring at the expansion forming process of aluminum tubes for heat exchangers. Here, geometric data obtained by image processing of tube cross sections are outer boundary radius, groove height, and tube thickness before and after the expansion, while force at the expansion forming process was measured through experiments. The location of adhesion was estimated using not only the SD of the groove heighta geometric data that have much to do with the adhesion-but also the differentiation of force occurring at the expansion. In this study, the adhesion was evidenced through investigating actual cross-sectional images, and it was confirmed that the adhesion location is related to a sudden increase or decrease in the expansion force. It was also confirmed that using the variation in the slope estimated using the derivative of the expansion force versus time, it can be determined whether or not adhesion occurred. | 3,950.8 | 2017-11-01T00:00:00.000 | [
"Engineering",
"Materials Science"
] |
raxmlGUI 2.0 beta: a graphical interface and toolkit for phylogenetic analyses using RAxML
RaxmlGUI is a graphical user interface to RAxML, one of the most popular and widely used software for phylogenetic inference using maximum likelihood. Here we present raxmlGUI 2.0-beta, a complete rewrite of the GUI, which replaces raxmlGUI and seamlessly integrates RAxML binaries for all major operating systems providing an intuitive graphical front-end to set up and run phylogenetic analyses. Our program offers automated pipelines for analyses that require multiple successive calls of RAxML and built-in functions to concatenate alignment files while automatically specifying the appropriate partition settings. While the program presented here is a beta version, the most important functions and analyses are already implemented and functional and we encourage users to send us any feedback they may have. RaxmlGUI facilitates phylogenetic analyses by coupling an intuitive interface with the unmatched performance of RAxML.
Introduction
Phylogenetic inference is a keystone in evolutionary biology research. It provides the foundations for tackling a wide range of questions, from population dynamics to taxonomy of higher taxa. RAxML (Stamatakis, 2014) is one of the most widely used programs in phylogenetic analysis, implementing extremely fast algorithms to analyze large datasets using maximum likelihood. Despite the undisputed efficiency of RAxML, the program is only available through a command-line interface. This requires users to be familiar with the shell environment and to navigate through the ever-growing number of commands implemented in the program, which may exclude many potential users without such experience. RaxmlGUI (Silvestro and Michalak, 2012) is a graphical interface intended to facilitate phylogenetic analyses using RAxML by providing a graphical front-end to help users set up their analysis. Although this interface has been widely used, there are many areas of improvement in terms of accessibility, usage and performance.
Here, we present the first public beta version of raxmlGUI 2.0, a complete rewrite of the raxmlGUI program. This version brings a new cross-platform design, novel functionalities and a seamless integration with RAxML 8.2. Similarly to its predecessor, raxmlGUI 2.0 is designed to be easy to use, providing the user with an intuitive interface with access to the model settings required to setup and run a phylogenetic analysis. Here we describe the available options of raxmlGUI 2.0-beta and outline the upcoming features that will be supported in the official release of raxmlGUI 2.0.
Main features
The program comes with pre-compiled integrated versions of RAxML for the major operating systems (MacOS, Windows, Linux), including the PTHREADS and SSE3 versions (Stamatakis, 2014) allowing the user to run faster analyses using parallel computing, when multiple CPUs are available. RaxmlGUI 2.0 is structured in two parts The input panel provides options to load new alignments and create a concatenated file and to specify partition-specific substitution matrix. The analysis panel provides options to specify the type of analysis, evolutionary models and outgroup selection. The output panel gives easy access to the folder with the input files and a list of output files that appears upon completing the analysis. On the right side of the window the user can select the version of RAxML, start the analysis, and visualize the RAxML output.
( Fig. 1), providing on the left all the commands and options to load input files, set up the analysis, define substitution models and partitions, among other features. On the right panel, it provides options to choose the RAxML version and start the analysis. A RAxML console is integrated in the GUI showing the progress of the analysis, the commands used to launch the analysis and all the screen output produced by RAxML.
Basic setup
RaxmlGUI 2.0 supports alignment files in two formats: extended PHYLIP and FASTA (example files are available in the program's repository). Upon loading an alignment, the program parses the names attributed to each sequence (e.g. the species name) and creates a list of taxa in the Outgroup menu button, which can be used to root the tree based on a user-defined outgroup (note that maximum likelihood trees can always be re-rooted after the analysis using tree-viewing software such as FigTree (Rambaut, 2012)).
Phylogenetic analyses can be run based on different types of data: nucleotide sequences (DNA, RNA), amino acid sequences, discrete binary and multi-state characters (e.g. used for descriptions of morphological data). Since each data type requires a specific class of substitution models, raxmlGUI automatically recognizes the data type from the loaded input file and provides the user with a drop-down menu showing all the substitution models compatible with the alignment.
Analytical pipelines
The default analysis includes a maximum likelihood search of the best tree, The most important output of this analysis is named "RAxML_bipartitions.input.tre" (where input is by default the file name of the alignment) and includes the maximum likelihood tree topology and branch lengths with labels reporting the bootstrap scores for each node (bipartition) in the tree. All output files are by default saved in the same directory of the input file.
Other types of analysis are available in raxmlGUI 2.0. Some analyses integrate multiple calls to RAxML to simplify the user experience in a single pipeline. For instance, the ML + thorough bootstrap option launches a sequence of three RAxML calls 4 to 1) infer the maximum likelihood tree through a user-defined number of independent searches; 2) run a user-defined number of thorough non-parametric bootstrap replicates; and 3) draw the bootstrap support values onto the maximum likelihood tree.
Automatic concatenation of alignments
An important feature of raxmlGUI 2.0 is the automated concatenation and partitioning of alignments, which simplifies the analysis of multiple genes or combination of different data types, e.g. amino acids sequences and morphological data.
After loading the first alignment, the user can add new ones to concatenate them into a single analysis. Upon loading additional alignments, raxmlGUI 2.0 performs the following tasks: • Parse the data to determine the data type (nucleotides, amino acids, multistate) • Parse the taxa names to make sure the concatenation of sequences occurs across matching taxa even if they are listed in different order among input files • Create a combined dataset file in the same directory as the input files • Create a file defining the boundaries of each partition in the concatenated alignment (each alignment file is assigned a new partition) and the respective substitution matrix (for amino acid data only) • For any mismatch between taxa of different partitions, give option to automatically create sequences of missing data in the concatenated alignment or drop taxa with missing sequences in any partition.
These features facilitate the concatenation of different alignment files and the creation of the partition files. They also reduce the probability of errors stemming from manually merging sequences by matching taxa names. Finally, raxmlGUI 2.0 also facilitate the generation of sparse matrices resulting from the combination of alignments with different and only partly overlapping taxonomic coverage.
Upcoming features
Upcoming updates of raxmlGUI will include full support of the latest version of RAxML-ng (Kozlov et al., 2019), which provides improved performance for very large datasets.
Additional features will allow users to enforce topological tree constraints, to compute Robinson-Foulds distances (Pattengale et al., 2007), and to enable more flexibility setting substitution models and dataset partitions.
Implementation
RaxmlGUI 2.0 is built with Electron, a framework for creating cross-platform desktop applications using web technologies like JavaScript, HTML, and CSS. The user interface is built with Material-UI, a React UI framework with components that implement Google's Material Design.
The Electron base improves the portability and compatibility across platforms and operating systems compared to the previous version of raxmlGUI that uses an obsolete Python 2.x codebase. The installation is extremely simple and does not require any additional external libraries or dependencies, nor does it require admin rights on the machine.
On machines featuring multiple CPUs (i.e. most desktop and laptop computers) the GUI allows users to easily use RAxML's powerful parallel computing, which can drastically speed up the analyses. RaxmlGUI 2.0 includes pre-compiled versions of the PTHREAD version of RAxML and a dropdown menu button to specify the desired number of CPUs allocated for the analysis.
Availability and users' feedback
A public beta version of raxmlGUI 2.0 is available at antonellilab.github.io/raxmlGUI/ The program is open source and licensed under a 6 GNU Affero General Public License v3 (AGPL-3.0). We encourage users to report any issues, feature requests, and general feedback either as issues on GitHub github.com/AntonelliLab/raxmlGUI/issues (this requires a GitHub account) or by email at raxmlgui.help[at]gmail.com. | 1,960 | 2019-10-10T00:00:00.000 | [
"Computer Science",
"Biology"
] |
Nanozyme-based sensing of dopamine using cobalt-doped hydroxyapatite nanocomposite from waste bones
Dopamine is one of the most important neurotransmitters and plays a crucial role in various neurological, renal, and cardiovascular systems. However, the abnormal levels of dopamine mainly point to Parkinson’s, Alzheimer’s, cardiovascular diseases, etc. Hydroxyapatite (HAp), owing to its catalytic nature, nanoporous structure, easy synthesis, and biocompatibility, is a promising matrix material. These characteristics make HAp a material of choice for doping metals such as cobalt. The synthesized cobalt-doped hydroxyapatite (Co-HAp) was used as a colorimetric sensing platform for dopamine. The successful synthesis of the platform was confirmed by characterization with FTIR, SEM, EDX, XRD, TGA, etc. The platform demonstrated intrinsic peroxidase-like activity in the presence of H2O2, resulting in the oxidation of 3,3′,5,5′-tetramethylbenzidine (TMB). The proposed sensor detected dopamine in a linear range of 0.9–35 μM, a limit of detection of 0.51 µM, limit of quantification of 1.7 µM, and an R2 of 0.993. The optimization of the proposed sensor was done with different parameters, such as the amount of mimic enzyme, H2O2, pH, TMB concentration, and time. The proposed sensor showed the best response at 5 mg of the mimic enzyme, pH 5, 12 mM TMB, and 8 mM H2O2, with a short response time of only 2 min. The fabricated platform was successfully applied to detect dopamine in physiological solutions.
Introduction
Dopamine is an important neurotransmitter in the human body, and it is implicated in various debilitating diseases.It plays a central role in the renal, neurological, cardiovascular, hormonal, and metabolic systems (Dhanasekaran et al., 2018).Dopamine plays an important role in the brain's ability to receive signals, and low levels of it can cause a variety of neurological diseases, including epilepsy, schizophrenia, and Parkinson's disease (Howes et al., 2017).Its normal level ranges from 1.3 to 2.6 μM.Beyond the given limits, it can point to heart diseases, Parkinson's disease, cardiovascular problems (Nishan et al., 2020), abnormal thyroid hormone levels, neuromuscular problems, and schizophrenia (Kienast and Heinz, 2006).The correlation between dopamine and several devastating and fatal illnesses underscores the need to sense it accurately, selectively, and cost-effectively.
In recent years, various sophisticated methods have been reported for the sensing of dopamine, such as fluorescence (Wang et al., 2015), electrochemistry (Xu et al., 2016), chemiluminescence (Xu et al., 2011), high-performance liquid chromatography (Carrera et al., 2007), electrochemiluminescence (Wang et al., 2017), etc.Despite their merits, the majority of these techniques require complex sample pretreatment.These techniques are not only expensive to acquire and sustain but also laborious and time-consuming.The need for highly skilled operators is another major limitation of the aforementioned techniques.Background interference and low reproducibility have also been reported for the mentioned methods.These shortcomings associated with the reported techniques put a question mark on their viability as techniques of choice for dopamine sensing and biosensing.Therefore, much work needs to be done to provide straightforward, quick, and effective techniques for the highsensitivity detection of dopamine.Colorimetric assays offer a viable alternative in comparison to the above-stated techniques.They have attracted considerable attention owing to their naked-eye observation capability, easy operation, and lower cost (Nishan et al., 2021).Moreover, the naked eye observation can further be confirmed through a UV-Vis spectrophotometer for accurate quantification (Shi et al., 2016).Natural enzyme-based sensing platforms have been used for the detection of various biomarkers, but they are expensive, have a low shelf life, are difficult to handle, are temperature-sensitive, and lose their activity under harsh conditions (Fang et al., 2006).
The emergence of mimetic enzymes (nanozymes) is a new viable alternative to overcome the limitations of natural enzymes (Zhu et al., 2017).Nanozymes are artificial enzymes based on nanomaterials that mimic the role of natural enzymes.Their superior catalytic properties and ability to endure harsh conditions have impressed researchers (Jiang et al., 2019).They are also used in bioassays (Zhu et al., 2017), as industrial catalysts (Maurya et al., 2015), in food processing (Oueslati et al., 2018), agriculture (Anjum et al., 2016), and environmental monitoring (Lu et al., 2013).Mimic enzymes offer the advantages of lower cost, high stability, a large surface area, and easy operation (Han et al., 2015).Transition metal-based enzyme mimics are the most effective class of artificial enzymes owing to their high conductivity and different oxidation states.It has been reported that Co−Fe 3 O 4 /graphene (Zhu et al., 2018), Co 3 O 4 @NiO (Hosseini et al., 2017), CoFe 2 O 4 /CoS (Yang et al., 2018), and CoS nanospheres (Luong et al., 1988) function as good peroxidase mimetic catalysts with favorable catalytic activity and high stability.Cobalt nanoparticles have been extensively used as nanozymes for sensing and catalytic activities due to their high stability and low cost.However, the problem of agglomeration limits their use as reliable players in the fabrication of sensing platforms for bioanalysis (Nana et al., 2021).Researchers have used various strategies to overcome the problem of agglomeration in metal nanoparticles.
For this purpose, in most cases, metal nanoparticles are mixed with other functional elements to provide a synergistic effect.Different matrix materials have been used to achieve the deagglomeration of the metal nanoparticles.Hydroxyapatite (HAp) has fantastic osteointegrative and osseoconductive properties.It is a nanoporous substance that is biocompatible, catalytic (owing to the presence of basic and acidic moieties), bioactive, degradable, and ubiquitous (Roopalakshmi et al., 2017).These qualities make it a popular candidate for its use as a matrix material for the fabrication of metal-based mimic enzymes (Irfan and Irfan, 2020).It can easily be synthesized from waste materials such as bovine bone (Barua et al., 2019), fish bone (Muhammad et al., 2016), coral (Pountos and Giannoudis, 2016), egg shells, etc (Gergely et al., 2010).HAp is a promising material for the fabrication of such platforms due to its exceptional ability to withstand a wide range of cationic and anionic substituents (Fihri et al., 2017).Doping with different kinds of metal ions improves the physical, chemical, and biological properties of HAp (Ratha et al., 2021).
In this work, we have used cobalt-doped HAp nanocomposite as a peroxidase mimic for the colorimetric sensing and biosensing of dopamine for the first time.Co-doped HAp, as a mimic enzyme, can catalyze the oxidation of TMB with the assistance of H 2 O 2 .The colorless solution changes to a blue-green product in the presence of H 2 O 2 , acting as an oxidizing agent.This transformation can be observed with the naked eye and confirmed with a UV-Vis spectrophotometer.As dopamine was added to the reaction mixture, it reduced the oxidized TMB to TMBred, with the consequent disappearance of the blue-green color to colorless.The fabricated sensing platform was successfully applied in a physiological solution for the sensing of dopamine.
Instrumentation
Using Fourier transform infrared spectroscopy on the Cary 630 FTIR spectrometer (Agilent Technologies, Danbury, Connecticut, USA), the distinctive peaks of HAp and Co-HAp were characterized.The materials' FTIR spectra were obtained within the given range of 4,000-400 cm −1 .The parameters set for FTIR experiments were 256 scans per sample and 4 cm −1 resolution.Scanning electron microscopy linked to energy dispersive X-ray spectroscopy (SEM-EDX) using a TESCAN VEGA (LMU) SEM with an INCAx-act (Oxford Instruments) EDX attachment working at 20 kV was used to analyze the morphology of the produced materials.The produced materials' phase was investigated using X-ray diffraction (Shimadzu, LabX XRD-6100 with Cu-Kα radiation) with a scan range of 10 °-80 °.The voltage for acceleration was set at 30 kV, while the current was set at 20 mA.Cu Kα radiations were utilized with a monochromatic wavelength of (λ = 1.5405Å).Using JCPDS, file No. 04-0783, the phase of the produced HAp and Co-HAp nanocomposite was identified.Under nitrogen gas flow, thermal gravimetric analysis (TGA) was performed at temperatures between 50 °C and 800 °C at a heating rate of 10 °C/min.Using a UV-Vis spectrophotometer (Shimadzu, UV, 1,800, Japan), absorption spectra were captured.
Preparation of HAp from waste bones
The waste bones of goats were used to prepare HAp.The bones were boiled for two to 3 hours to remove the remains of organic components.To ensure further cleaning, the bones were treated with an acetone solution (70% v/v solution in water) for 3 hours.Water was used to rinse the bones several times.The treated bones were dried for 24 h at 100 °C in an oven.The dried bones were pulverized with the help of a pestle and mortar.To obtain the desired HAp, calcination was carried out at 800 °C for 3 hours in a furnace (Khawar et al., 2019).The work received ethical approval from the concerned forum of Kohat University of Science and Technology, KUST, Kohat via No. KUST/Ethical Committee/1023.
Preparation of Co-HAp nanocomposite
Cobalt (II) chloride hexahydrate (CoCl 2 .6H 2 O) was mixed with the synthesized HAp powder for doping at a weight ratio of 1:9, respectively.The ingredients were mixed with a pestle and mortar to form a homogenous mixture.The mixture was calcined at 800 °C for 3 hours in a furnace.The overall process for the preparation HAp and Co-HAp is shown in the Scheme 1.
Dopamine sensing through Co-HAp nanocomposite
The synthesized Co-HAp nanocomposite (5 mg) was suspended in a 500 μL PBS solution.At the same time, 100 μL of TMB solution (12 mM in DMSO) and 100 μL of H 2 O 2 (8 mM) were added to the solution mixture.The catalytic ability was measured by recording the changes in absorbance over time and acquiring the absorption spectra.TMB was used as a substrate for the colorimetric detection of dopamine, and the peroxidase-like activity was measured.The expected colorimetric change was confirmed using both the naked eye and UV-Vis spectra.
In order to validate the catalytic role of Co-HAp in the reaction system, the sensing of dopamine was performed under different conditions.These include TMB + Co-HAp, TMB + H 2 O 2 , and TMB + H 2 O 2 +Co-HAp.
Sensing of dopamine in physiological solution
The physiological solution was utilized to detect dopamine.In this solution, three different concentrations of dopamine were spiked.In the prepared samples, colorimetric detection of dopamine was performed and confirmed by a UV-Vis spectrophotometer (Zheng et al., 2011).The peak around 568 cm −1 represents the bending mode of phosphate functionality, while the band at 1030 cm −1 indicates the stretching mode of the phosphate moiety.The peak in the range of 1400-1600 cm −1 shows the presence of a carbonate group in the HAp.The peak at 3,627 cm −1 represents the hydroxyl group that comes from moisture.Co-HAp nanocomposite indicated peaks at 565 cm −1 and 1025 cm −1 .These peaks represent the phosphate functionality in the synthesized nanocomposite (Sahana et al., 2013).There is no more peak of carbonate functionality due to the removal of carbonyl functionality in the form of carbon dioxide.However, there is no significant difference between the pure and Codoped HAp (Yazdani et al., 2019).(Palierse et al., 2022), the distinct diffraction peaks were compared based on the anatase HAp and Co-HAp phases.According to the standard data sheet the XRD pattern of the doped samples 1 (B) was found to be comparable to that of crystalline HAp (Ca 10 (PO 4 ) 6 (OH) 2 ).With the exception of a minor widening of the peaks at 32 °, 33 °, and 34 °2θ, the XRD profiles of the doped HAp did not change significantly from those of pure HAp and did not show any new peaks belonging to cobalt.The probable anionic replacement of OH ions in the HAp by chloride ions of cobalt chloride, in addition to the replacement of calcium by cobalt, might be the cause of this drop in crystallinity.Crystal lattice distortion, crystallite size, and percentage crystallinity are the factors that influence peak broadening (Khaliq et al., 2023).Using the Scherrer equation for the anatse phase of HAp and Co-HAp and the most intense peak, the average particle size was determined (Nachit et al., 2016).The average crystal sizes of the anatase phases of HAp and Co-HAp were estimated to be about 38 and 32 nm, respectively.Shows that at (A) 5 mg Co-HAp nanocomposite, 100 µL TMB (12 mM), 500 µL PBS, 100 µL H 2 O 2 (9 mM), and (B) shows that the addition of 100 µL dopamine (35 µM) results in the reduction of the oxidized TMB.
EDX analysis of prepared HAp and Co-HAp
Figure 1v, vi; Table 1 show the elemental composition and Co-HAp.The EDX spectrum of pure HAp shows presence of calcium, oxygen, and phosphorus in the weight percentage of 37.78, 49.97, and 12.25, respectively, as shown in Table 1.Whereas Co-HAp, along with the already present elements calcium (32.59%), oxygen (45.32%), and phosphorus (16.0%), shows the presence of chlorine and cobalt in the weight percentages of 2.19 and 3.90, respectively.This confirms the successful doping of Co on the surface of the synthesized HAp.
TGA analysis of prepared HAp and Co-HAp
In order to ascertain the performance of the fabricated platform in harsh thermal conditions, it is essential to determine its thermal stability.Figure 1vii shows the thermal gravimetric analysis study of the synthesized HAp and Co-HAp nanocomposite.The TGA of pure HAp from 200 to 400 °C is 0.601% weight loss, and from 500 to 800 °C, the weight loss is 1.612%.The overall weight loss in pure HAp is 2.213%.The TGA of Co-HAp weight loss is 1.172%; it is from 500 to 800 °C.There is no considerable weight loss in Co-HAp because HAp is already thermally stable even at higher temperatures.
Proof of catalytic activity of the mimic enzyme
The Co-HAp nanocomposite demonstrated improved peroxidaselike catalytic activity.Using the peroxidase substrate TMB and H 2 O 2 , the peroxidase-like catalytic activity of the Co-HAp nanocomposite was investigated.The shift in absorbance of the oxidized TMB (TMB oxi ) at 652 nm was monitored to observe the progress of the reaction.It is clear from Figure 2 that there is no oxidation when the TMB and Co-HAp are present only in the system.There is a very little oxidation of TMB when H 2 O 2 is present.However, the combination of H 2 O 2 and Co-HAp, significant TMB oxidation of takes place (Ivanova et al., 2019).It is clearly evident from this that H 2 O 2 and Co-HAp are both necessary for the oxidation of TMB.This suggests that the catalyst accelerates the oxidation of TMB in the presence of H 2 O 2 by exhibiting peroxidaselike activity.
Colorimetric sensing of dopamine
The fabricated sensing platform, i. e., Co-HAp nanocomposite (5 mg), was used for dopamine colorimetric sensing, with TMB serving as a chromogenic substrate.In a typical experiment, 100 µL of hydrogen peroxide (9 mM) was combined with 100 µL of TMB (12 mM) in 500 µL of PBS solution.The color changed from transparent to blue-green, with a visible colorimetric change (A).After the addition of 100 µL of dopamine (35 µM), the blue-green color changed to transparent (B) due to the reduction of oxidized TMB, as confirmed by the UV-Vis spectrophotometer.Inset Figure 3 shows both the colorimetric change and the UV-Vis spectra.
Sensing mechanism of dopamine via Co-HAp nanocomposite
The mimic enzyme (Co-HAp nanocomposite) can work as an excellent platform for the sensing of dopamine.In this process, the oxidizing potential of hydrogen peroxide assisted the mimic enzyme in the oxidation of TMB.A visible colorimetric change to a blue-green color occurs, indicating the oxidation of TMB.Upon the addition of dopamine, the reaction complex changes from blue-green to transparent.This colorimetric change was confirmed through a UV-visible spectrophotometer.The catalytic action was based on a remarkable color change from colorless.The oxidation of TMB to a blue-green color is mediated by a hydroxyl radical that is generated as a result of hydrogen peroxide breakdown assisted by the mimic enzyme.The generated hydroxyl free radical removes the electron from the amino group of the TMB and oxidizes it.Here, the OH free radical acts as an oxidizing agent, and the resulting oxidized TMB gives a bluegreen color to the reaction mixture.In reverse, the addition of dopamine to the reaction mixture reduces the oxidized TMB to its original colorless form and itself oxidizes to dopamine quinone (Zheng et al., 2021).A detailed mechanism is shown in Scheme 2. (i): Showing the calibration plot of absorbance versus dopamine concentration.(ii) UV-Vis spectra and the alterations that correlate to variations in dopamine concentrations.
Effect of Co-HAp nanocomposite amount
In order to assess the effect of Co-HAp, different amounts in the range of 2-8 mg of the proposed platform were As the amount of Co-HAp increased, color of the solution mixture started to fade.The solution color completely vanished 5 mg of the Co-HAp.With the increase in the concentration of Co-Hap, the color of the reaction mixture starts to intensify to blue-green.For these experiments, 100 µL of TMB solution (12 mM), 500 µL of phosphate-buffer saline, 100 µL of H 2 O 2 , and 100 µL of a 35 µM dopamine solution were taken.Under the prevailing conditions, the reaction took only 2 min.Ivanova et al. reported 7 mg of mimic enzyme as optimal in their work (Ivanova et al., 2019).Figure 4i illustrates the correlation between different Co-HAp nanocomposite concentrations, at given amounts of TMB, H 2 O 2 , and dopamine.
Impact of pH on the fabricated sensor
The fabricated sensing platform was tested in the pH range 3-10, as shown in Figure 4ii.As the pH increases from three to five, the color of the chromogenic substrate gradually vanishes until it completely disappears at pH 5.As the pH increased further, the color of the reaction mixture appeared again and intensified untill it reached pH 10.In these experiments, 5 mg doped Co-HAp nanocomposite, 100 μL TMB solution (12 mM), incubation time 2 min, 100 μL H 2 O 2 , and 100 μL dopamine solution (35 µM) were The Figure demonstrates the selectivity of the fabricated sensor for dopamine in the presence of other analytes, including glucose, K + , Mg 2+ , HSA, uric acid, nitrite, and ascorbic acid.used.As a result, a pH of was determined to be the ideal pH for the suggested probe.pH four was proposed as the optimal pH for the work reported in an earlier published work (Ivanova et al., 2019).
Effect of TMB concentration
The optimization results for TMB concentration for the sensing of dopamine shown in Figure 4iii.The TMB concentration was in the range of 3-21 mM in the optimization experiments.It is clear from the inset Figure at 12 mM of TMB concentration we get the best response.A TMB concentration, lower or higher than the mentioned amount does not produce any desirable results.In these experiments, 100 µL of TMB (3-21 mM), 100 µL of H 2 O 2 (9 mM), and 5 mg of doped Co-HAp nanocomposite were used.Wu et al. reported 0.5 mM TMB as optimal in their work (Wu et al., 2018).In the subsequent experiments, 12 mM of TMB concentration was used.
Effect of H 2 O 2 concentration
In order to obtain the optimal H 2 O 2 concentration, the sensing experiments were performed at different concentrations of H 2 O 2, ranging from 2 to 14 mM, as shown in Figure 4iv.Results indicate that the best response was achieved at 8 mM of H 2 O 2 .These experiments were performed under the following conditions: 5 mg of Co-HAp nanocomposite; 100 µL of TMB (12 mM); and 500 µL of PBS.Yang et al. reported 65 mM to be the optimal concentration of H 2 O 2 in their work (Yang et al., 2018).
Analytical merits of the method
An easy and quick colorimetric detection method was used to detect dopamine under the ideal experimental conditions.Based on the catalytic activity of the synthesized Co-HAp nanocomposite, the sensor's sensitivity to dopamine detection was examined.Various dopamine concentrations were used during the evaluation of the fabricated platform for its analytical merits.As seen, with the increasing concentration of dopamine, results the UV-Vis peak at 652 nm diminishes.Different concentrations of dopamine were taken in the range of 0.9-35 μM, as shown in Figures 5i, ii.In the absence of dopamine, it shows a peak at 652 nm in the UV spectrum and decreases linearly as dopamine concentration rises.Dopamine detection was performed in a linear range of 0.9-35 µM with an LOD of 0.51 µM and an LOQ of 1.7 µM with regression coefficient (R 2 ) value of 0.993.
Comparing the suggested sensor with different methods
As indicated in Table 2, the efficacy of the suggested sensor was evaluated by contrasting the findings with those of previous SCHEME 2 A schematic representation of the proposed mechanism shows the oxidation of the chromogenic substrate and its subsequent reduction as a result of dopamine addition to the reaction mixture.
investigations.The comparison shows the excellent performance the proposed sensor in terms of LOD and wide linear range.
Interference studies
Under optimized conditions, the fabricated platform was tested for the selective sensing of dopamine.For this purpose, glucose, K + , Mg 2+ , human serum (HSA), uric acid, nitrite and ascorbic acid, were examined as potential interfering species, as shown in Figure 6.The findings demonstrate that dopamine has a relatively low absorption value as compared to the other species.
Application of the fabricated sensor in physiological solution
In order to examine the use of the suggested platform, we applied the technique for dopamine detection in physiological solutions.For this purpose, a physiological solution was used to analyze dopamine.Physiological solutions of three different concentrations of dopamine were prepared for sensing through the fabricated platform.As the dopamine concentration increases, the intensity of the absorption peak at 652 nm decreases linearly.The colorimetric change was observed in a short time of only 2 min.As shown in Figure 7, the sensor is very efficient and exceptionally sensitive for detecting dopamine in biological samples.
Conclusion
In conclusion, we have successfully fabricated a new Co-HAp sensing platform for the colorimetric sensing of dopamine.All the characterizations confirmed the successful synthesis of HAp and Co-HAp.The synthesized mimic enzyme, with the synergistic effect of hydrogen peroxide, oxidized the TMB with a visible colorimetric change that was subsequently confirmed with a UV-Vis spectrophotometer.The addition of dopamine to the reaction mixture resulted in the reduction of the oxidized TMB to TMB red and the disappearance of the blue-green color.The fabricated platform (Co-HAp) was highly sensitive, quick, and selective for the sensing of dopamine as compared to the previously reported methods.The fabricated platform was used for the sensing of dopamine in physiological solutions.The proposed sensor has the potential to be used as a laboratory tool for the diagnosis, management, and monitoring of various neurological disorders at low cost and easy operation.
FIGURE 1 (i) FTIR spectra for HAp (A) and Co-HAp nanocomposite (B), indicating the presence of characteristic peaks in the synthesized materials.(ii) XRD pattern of the HAp (A) and Co-HAp nanocomposite (B), showing broadening of the peaks in doped HAp.SEM image of the HAp (iii) and Co-HAp nanocomposite (iv), indicating the spherical shape and nanoporous morphology of the synthesized platform.EDX analysis of the HAp (v) and Co-HAp nanocomposite (vi) shows the presence of Co along with other elements.(vii) TGA thermogram of the prepared HAp (A) and Co-HAp nanocomposite (B), demonstrating its thermal stability in the temperature range of 100-800 °C.
SCHEME 1Schematic representation of the synthesis of HAp and Co-HAp.
Figure
Figure 1ii displays the XRD pattern of the synthesized HAp and Co-HAp.Using the JCPDS database Card-No.(09-0432),(Palierse et al., 2022), the distinct diffraction peaks were compared based on the anatase HAp and Co-HAp phases.According to the standard data sheet the XRD pattern of the doped samples 1 (B) was found to be comparable to that of crystalline HAp (Ca 10 (PO 4 ) 6 (OH) 2 ).With the exception of a minor widening of the peaks at 32 °, 33 °, and 34 °2θ, the XRD profiles of the doped HAp did not change significantly from those of pure HAp and did not show any new peaks belonging to cobalt.The probable anionic replacement of OH ions in the HAp by chloride ions of cobalt chloride, in addition to the replacement of calcium by cobalt, might be the cause of this drop in crystallinity.Crystal lattice distortion, crystallite size, and percentage crystallinity are the factors that influence peak broadening(Khaliq et al., 2023).Using the Scherrer equation for the anatse phase of HAp and Co-HAp and the most intense peak, the average particle size was determined(Nachit et al., 2016).The average crystal sizes of the anatase phases of HAp and Co-HAp were estimated to be about 38 and 32 nm, respectively.
Figure
Figure1iiirepresents the SEM image of HAp and Co-HAp nanocomposite (iv).The HAp and Co-HAp nanocomposites show a spherical morphology.A decrease in particle size was observed with the doping of cobalt into the HAp structure, which increases its surface area.Both pure and doped HAp show nanoporous structures, which can be helpful in providing the necessary surface area for catalytic reactions.
FIGURE 7
FIGURE 7Physiological sample analysis by adding different concentrations of dopamine in physiological solutions.
Figure
Figure 4v depicts the optimization of time for the proposed sensing platform.The maximal efficiency of the Co-HAp nanocomposite was observed after 2 minutes.Under optimal
TABLE 1
EDX analysis by weight of the synthesized HAp and Co-HAp nanocomposite.
TABLE 2 A
brief comparison of our synthesized platform with the reported sensors for the colorimetric detection of dopamine. | 5,811.4 | 2024-04-17T00:00:00.000 | [
"Materials Science",
"Medicine",
"Chemistry"
] |
Algorithm of Ray Casting Volume Rendering Based on CUDA
Direct volume rendering is one of the methods for the visualization of 3D data set, It does not construct intermediate entity, directly generate 2D graphics on the screen by the 3D data set, which is better for parallel processing, but the amount of calculation is large, difficult to rendering with conventional graphics hardware. This paper mainly elaborated optimization and improvement about the algorithm of ray casting volume rendering for the visualization of 3D data set, within the framework of the CUDA, using the multi-core parallel computing ability of GPU. The vertex shader, pixel shader, calculation of the starting point of the sampling points of light, color and opacity tired and synthesis of image operations are completed by GPU. The vertex shader, pixel shader, calculation of the starting point of light, color and opacity of the sampling points accumulation and image synthesis operation are achieved by GPU. Compared with the GPU-based ray-casting algorithm, the algorithm takes full advantage of the characteristics of the CUDA parallel processing, can quickly draw a higher quality image, rendering speed has been raised about 15%.
Introduction
The visualization of 3D data set has experienced 20 years of development since its proposed, the algorithm can be divided into two types,surface rendering and volume rendering.Surface rendering render the geometric surface with traditional computer graphics technology which is structured by 3D data set; Volume rendering is directly generated two-dimensional graphic by the corresponding three-dimensional data set.The latter has the advantage of showing the complete picture of the object as well as the details internal, but the amount of calculation is huge.It is difficult to achieve real-time rendering requirements with traditional algorithm based on software.Acceleration technology based on software such as opacity early termination, empty voxels Ignore, viewing transformation optimize etc, constrained by hardware,the effect of acceleration is limited ,can't achieve real-time dynamic graphic display.With the development of hareware technology, direct volume rendering is implemented on the high-end graphics workstations based on programmable hardware,this improved rendering speed,howerve the cost of hardware is high.With the advent of CUDA, we can direct manipulate hardware resources of GPU with class C high-level language without requiring the aid of graphics application programming interface, some of the visualization work can be completed by GPU with exclusive technology of programmable GPU pipeline ,the parallel processing of data greatly improved processing speed.
At present, the domestic research on CUDA technology is at the stage of development , also made a series of direct volume rendering acceleration technology based on CUDA.This paper mainly elaborated optimization and improvement about the algorithm of ray casting volume rendering for the visualization of 3D data set based on CUDA,we storing volume data into graphics memory in the form of three-dimensional texture, the vertex shader, pixel shader, calculation of the starting point of light, color and opacity of the sampling points accumulation and image synthesis operation are achieved by GPU,these all accelerate rendering.
Algorithm design 2.1 The architecture of CUDA and ray casting
CUDA is a hardware and software system of a GPU as a data parallel computing device, which is a three-tier structure composed by grid, block, organized by form of grid thread.Each grid is composed of a plurality of thread blocks, each block is composed of a plurality of threads.
Ray casting is along a fixed direction emitting a light (usually the viewing direction), light across the image sequence, and in this process, the image is sampled to obtain the color information of the sequence, meanwhile according to the light absorption model the color values and opacity will be accumulated, until the light across the image sequence.we will get a pair of complete visualization by accumulating the color of all pixels.
In ray casting algorithm, the calculation process about ray along a fixed direction of emitted light is independent of each other, the calculation method is the same and has the advantages of huge parallelism.According to this,the screen is considered as a grid,each pixel considered as a thread, according to the the image characteristics,the screen is divided into several regions,which is considered as a block, then ray casting algorithm can be implemented based on cuda three-tier architecture.The algorithm includes two parts, one is CPU (Host)part; another is GPU (Device)part.CPU part reads 3D data file and stores data in memory, initialize OpenGL environment, and establish links between video memory.receives the final results from GPU and display the result of visualizaton.GPU part ergods ray and fnd the intersection of ray and the data set, accumulats color and opacity, returns the final result to memory The specific steps of the algorithm are as follows: CPU part (1) Read dimensional data set of raw format, store in the device memory as the form of one unsigned char array.
(2) Initialize OpenGL environment, establish PBO (OpengGl pixel buffer objects) and create a link to GPU memory ; setting model transformation matrix, set the viewing direction.
(3) According to the size of data set, calculate the required size of block and thread.
GPU part (1) Initialize CUDA environment, establish 3D array and 3D texture of CUDA, unsigned char array data set of device memory read into 3D Array, set the texture parameters and bind 3D Textrue and 3D Array .
(2) Set pre parameters, calculate corresponding screen pixel position, viewpoint coordinates according to the thread index (3) Calculate the the distance to the camera,the coordinate of the intersection lin about the ray enter the data set and lout about the ray leave the data set with the Ray -Box Intersection method, according to the ray direction.
(4) Sample along the ray direction from the front to the back according to a preset sampling interval, get the color and opacity of the sampling points and accumulate them.Set the opacity threshold is equeal to 1, if the transparency value close to 1.0, stop computing, otherwise continue to sample the next point, until the left point is reached.
(5) The eventually accumulated value of color and opacity are stored in the PBO Thus, the GPU work is completed, the data of PBO is directly displayed on the screen through the API of OpenGL environment.
Mapping of threads and light and designing of pre parameters
Mapping of threads and light is creating a link between any ray with corresponding thread of CUDA,which is the key of parallelization about ray casting algorithm.Two built-in CUDA variables gridDim, blockDim signify the scale of the grid and thread blocks.Two built-in CUDA variables blockIdx,threadIdx signify the thread block number and thread index.Through the combination of the two signify the parallel thread unit.BlockIdx, threadIdx contains two dimensions, namely X and y.Create the index of every ray through thread index according to the formula 1 below.
Inorder to facilitate the calculation, the value of dimensions of threadIdx is standarded to the interval [-1,1] according to the formula 2 ,3 below.
imageW signify the width of view plane,imageH signify the height of view plane.
Pre parameter design,first standard view plane size and eye position, as shown in figure 1.
Eye position is on the z axis with coordinate(0,0,6), view plane is on the z axis(0,0,3) and the size is 3*3.Second,set unit Ray-Box, the imaging range constraint
Intersection of ray and color and opacity accumulation
The key of ray-casting algorithm is intersection of ray , accumulating color and opacity values along the direction of the ray, and the composite image.First step is intersection operation of ray, for any light, using vector notation to describe it, set two basic parameters,origin coordinates and view direction,then any ray can be expressed as a formula 4 below, t of formula 4 represents the displacement of ray in the direction of view. . .
For seeking the intersection between rays and Ray-Box according to methods proposed by G. Scott Owen, calculate the two intersections of ray and Ray-Box,named lin and lout, then taking lin as the origin, moving along the line of sight, each time forward step distance.Every step forward, calculate the coordinates of a sample point on the ray in space according to the formula 5 below.
Then,according to the coordinates of point, calculate the corresponding texture coordinates, obtain the appropriate texture value based on texture coordinates which is the gray value of the point.Create a one-dimensional transfer function to indicate different content of image.The function has four components which is R,G,B, α,the first three parameters is the conversion between color and gray, the fourth is to control the transparency, through the function we can control the conversion between gray and color.
Finally is the accumulation of opacity and color.In here , we calculate the color and opacity according to formula 7 and formula 8 below with method from the front to the back.
( 1) (1 ) In the formula 7 and 8 ,C represent color value, α represent opacity.In the experiment, the early termination of opacity is adopted.The threshold is set to 1.If the opacity has been reached threshold,the voxels behind will be Ignored,this can save time.At last, the accumulated value is directly stored to the PBO.
Experimental results and analysis
The algorithm above is implemented in the hardware system and software system followed,cpu is Intel Core2 dual core T5500, memory size is DDRII 2GB, graphics card is NVIDIA GeForce 8400M GS with memory 512MB,operation system is Windows XP, development environment is VC++6.0,CUDASDK3.1 and OpenGL3.1.It also Compared with ray casting algorithm based on GPU.In here two different scale data were used to test, the parameters of data set are shown in table 1.The size of original image about the experimental results is 512*512.
Performance analysis
From the above experimental results, rendering speed of algorithm CUDA based on has increased by about 15%.Texture generation, ray traversal and accumulation of color and opacity spend more time.Table 3 and As can be seen from two tables above,the time CUDA-based of texture generation, ray traversal, accumulation of color and opacity has greatly improved than that GPU-based,the former time is about 1/4 of that of the latter.This embodies that the characteristics of graphics chip is single-instruction, multiple data stream processing,the advantages of CUDA architecture is parallel processing.But the use of texture in CUDA SDK3.1 has a shortcoming, texture data stored in memory can not be directly copied to 3D Array of CUDA, it must use a temporary storage space as an intermediary, texture data is first copied to temporary,then form temporary to 3D Array of CUDA.So quickly generate texture, storage need more time this affects the rendering,secondly,the algorithm CUDA-based consists of two parts,cpu parts and gpu parts, the data need to exchange between the two parts, it also has some impact on the speed.The final rendering speed increased by about 15%.Optimization of imaging effect, mainly lies in the data filtering using normalized linear filtering.
Copyright
The paper and researches is original and unpublished,they also have not been submitted to other conferences or journals before.
Conclusion
In this paper, a parallel CUDA architecture technology, programmable hardware technology, ray casting algorithm are used for visualization of 3D data set.Under the same hardware and software condition,the same 3D data
Fig. 1 O
Fig.1.Spatial relation o is the origin coordinates of ray, eyeRay.d*lin is the coordinates of the point where ray in, eyeRay.d*step is the forward distance of ray each time).Since the pos is in the Ray-Box, its coordinate values is located between[-1, -1, -1] and [1,1,1].Information of image color and opacity is stored in CUDA 3D Array in the form of 3D texture, texture coordinates is located between [0, 1].Convert coordinate of pos to texture coordinates of postexture according to formula 6 below.
Fig 2
Fig 2(a) reconstruction based on CUDA
Figure 3
Figure 3 is the reconstruction of engine,which size is 256*256*256.Figure a is the reconstruction based on CUDA,figure b is the reconstruction based on GPU.The resulting image of figure a is better than figure b,it shows more internal details of engine.
Figure 4
Figure 4 shows the resulting image with transfer function of color,this can indicate different content of image.Here , the transfer function is one-dimensional.It's a simple mapping between the texture gray information of volume data and color,the information is stored in 3D array of CUDA as mentioned above.11 control points are set in the transfer function, representing the different gray scale values corresponding to the corresponding color, intermediate values, using simple linear interpolation algorithm to calculate.
Fig 3 (
Fig 3(a) reconstruction based on CUDA 4 lists statistics of the two part of the reconstruction process above.Table3 Time of algorithm GPU-
Fig 4 (
Fig 4 (a) reconstruction of engine with color transfer | 2,920.4 | 2014-03-25T00:00:00.000 | [
"Computer Science"
] |
Overview of legal traceability of GPS positioning in Australia
Global Positioning System (GPS) position verification and legal traceability in Australia supports industry, trade, science and innovation and is trusted and recognized domestically and internationally. At the end of 2017, the Australia’s national datum was transitioned from the Geocentric Datum of Australia 1994 (GDA94) to the Geocentric Datum of Australia 2020 (GDA2020). As such, the datum for the legal traceability of GPS positions in Australia has also moved to GDA2020. This paper highlights the importance of legal metrology and measurement in terms of GPS positions in accordance with the National Measurement Act 1960 (Commonwealth of Australia). Here we provide an overview of the process of issuing the so-called ‘Regulation 13 Certificates’ for Continuously Operating Reference Stations (CORS) across Australia. The position verification methodology is detailed, including the quality control, metadata assurance, and dynamic management of the certificates as well as positional uncertainty determination of CORS with varying quality. A quality monitoring system of positions is also discussed along with how measurement traceability is ensured including short-term and long-term position monitoring schemes.
Introduction
The national measurement system in Australia ensures a basis for legally traceable, consistent and internationally recognized measurements. With the growing societal dependency on Global Positioning System (GPS), the need for the legal traceability of GPS positions with respect to the Geocentric Datum of Australia (GDA), currently GDA2020 (Hu and Dawson 2018;ICSM 2018), has become increasingly apparent. In the interest of ensuring consistency of positions derived from private and government Continuously Operating Reference Stations (CORS), Geoscience Australia maintains an appointment as a legal metrology authority in accordance with the National Measurement Act 1960 (Commonwealth of Australia) and provides legally traceable positions Hu 2019).
Geoscience Australia's role in the national measurement system is to operate the Australia Fiducial Network (AFN) to appropriate standards, for example, to meet the highest requirements of all kinds of applications (Beavan 2005;Firuzabdi and King 2011), and to ensure key CORS across Australia that are operated by other agencies such as state survey authorities are appropriately linked to the AFN (Dawson and Woods 2010;. Geoscience Australia can issue certificates of verification under Regulation 13 of the National Measurement Regulations 1999 in accordance with the National Measurement Act 1960. These are commonly referred to as Regulation 13 Certificates. Regulation 13 Certificates provide coordinates and their uncertainties with respect to the Recognized-Value Standard (RVS) of measurement of position in Australia (Hu andDawson 2013, 2018). In Australia, the GPS position of a station with legal traceability is defined as at the time of measurement and with the stated instrumentation of a GPS monument with respect to the Geocentric Datum of Australia.
The measurement traceability is ensured by comparing the computed solution against the RVS for position of the AFN stations, as well as weekly combined solutions computed by the International GNSS service (IGS) in the International Terrestrial Reference Frame (ITRF), currently ITRF2014, and the individual global analysis centres of the IGS. The validity and traceability of GPS is ensured via its link to the global Satellite Laser Ranging (SLR) and Very Long Baseline Interferometry (VLBI) observing networks through the ITRF. As the AFN is a reference for national geodetic networks, reliable and upto-date coordinates must be available for all the AFN stations. This requires that we not only detect and identify timing of position offsets at the AFN sites, but we also estimate the offset magnitude which is used for the position update .
Geoscience Australia is responsible for maintaining a consistent set of geodetic position and velocity estimates for the 109 AFN GNSS sites across Australia. In order to ensure the long term reliability and quality control of the legal traceability of the GPS position in Australia, a thorough site performance monitoring system has been initiated and carried out in addition to the meta-data management and routine analysis (Owen et al. 2018;Hu et al. 2011Hu et al. , 2019. This paper overviews Geoscience Australia's approach to the legal traceability of GPS positions, and the process of legal certification including the quality standards and the quality management system of the position verification process. The quality management includes our approach to monitoring the impact of: equipment configuration changes; antenna malfunctions; crustal deformation; and processing strategy and modelling updates. Some examples are given based on experience within the Asia Pacific Reference Frame (APREF) community . Finally, the structured maintenance and continual improvement program for the verifying laboratory are also discussed.
History of the recognized-value standard for measurement of GPS position
To align the Australian datum to the ITRF, which is a global reference frame, Australia adopted the Geocentric Datum of Australia (GDA). The first geocentric datum in Australia was GDA94 which originally consisted of 10 stations in the Australian Fiducial Network (AFN) as shown in Fig. 1; seven stations are located on mainland Australia, one station in Tasmania and two stations are on Macquarie and Cocos Islands. The Recognized-Value Standard (RVS) of measurement of position in GDA94 was determined from GPS campaign data observed in 1992, 1993 and 1994 and aligned to the ITRF1992 at epoch 1994.0. The GDA94 positions of the 10 AFN stations were estimated to have an absolute accuracy of about 2 cm at 95% confidence level in the horizontal components (Morgan et al. 1996;ICSM 2018).
To improve the consistency of GDA94 with the realization of ITRF2008, on 4 April 2012, the AFN was extended to include 21 sites as shown in Fig. 2. The coordinates of the updated 21 AFN stations were adopted directly from ITRF2008 then subsequently transformed to GDA94 using the transformation parameters published Dawson and Woods (2010). The main reason for the update was the large differences between the 1998 RVS and the ITRF2008 after Helmert transformation, which were up to 15 mm for horizontal and 60 mm for vertical components.
During this period, the scope of accreditation for the Regulation 13 certificate changed from 32 mm for horizontal and 54 mm for vertical components of least uncertainty to 7 mm for horizontal and 15 mm for vertical components, respectively. The least uncertainty means the smallest uncertainty of measurement that can realistically be expected under ideal conditions, and the change of the scope also reflects the precision of the updated recognized-value standard of measurement of position (ICSM 2018).
Due to the motion of the Australian tectonic plate, the above updated GDA94 coordinates have continued to diverge from ITRF92 coordinates. By 2020, the difference would be approximately up to 1.8 m in the horizontal components. Complementary to this, there have been many improvements and updated realizations of the ITRF. For instance, the differences between ITRF1992 and ITRF2014 causes about 9 cm change in ellipsoidal heights in Australia and parts of the Australian crust have deformed (ICSM 2018). Therefore, it is necessary to update the RVS and align GDA to the current ITRF2014. As such, the GDA94 RVS was updated in October 2017 to GDA2020 and the AFN was extended further to include 109 stations across the Australian plate as shown in Fig. 3. The 109 AFN stations were selected from the Australia Regional GNSS Network (ARGN) and AuScope CORS network based on the following criteria ): • are operated by Geoscience Australia or similar agency; • are located on the Australian Tectonic Plate, within Australia's jurisdiction • are on a high quality survey monument (such as a concrete pillar); and • have residual velocity less than 1 mm/year relative to the Australian rigid plate motion model.
The RVS GDA2020 coordinates and velocities for the 109 AFN stations are derived from the cumulative solutions of the long-term position time series which are the ITRF2014 coordinates and velocities and were mapped forward to the epoch of 2020.0 using the derived Australian plate motion model. The cumulative solutions are part of the products of the Asia-Pacific Reference Frame (APREF) project with more than 20 years data since 1996 . These coordinates and velocities can be found in the GDA2020 technical manual (ICSM 2018) and refer to National Measurement Act 1960 -Recognized-value standard of measurement of position determination 2017 F2017L01352 (https ://www.legis latio n.gov.au/Detai ls/F2017 L0135 2).
GPS position of a station in Australia with legal traceability is a set of point coordinates with stated instrumentation installed on a stable monument with respect to the Geocentric Datum of Australia (GDA2020) referred to the GRS80 ellipsoid at the epoch 2020.0.
The methodology of GPS position verification
We used CATREF software (Altamimi et al. 2002(Altamimi et al. , 2016Hu et al. 2019) to estimate station velocities while combining 1210 weekly solutions into a long-term solution. Only those stations having more than 2.5 years of observations are considered noting that velocity estimates can be biased due to unreliable estimated seasonal signals (Blewitt and Lavallée 2002). The coordinates and velocities of these sites are originally determined in ITRF2014 at epoch 2010.0, then propagated forward to epoch 2020.0 using the Australian plate motion model. This means the adopted coordinates of GDA2020 for a CORS corresponds to the position on January 1, 2020. The GDA2020 site velocity needs to be applied to compute the position of the site at another date.
Daily solutions of the APREF stations were processed using Bernese GNSS Software version 5.2 (Dach et al. 2012). We applied up-to-date models and procedures following the International Earth Rotation Service (IERS) standards 2010 (Petit and Luzum 2010) and IGS recommendations. We used the absolute satellite and receiver antenna phase calibration model (Schmid et al. 2007), Fig. 3 The distribution of the extended 109 AFN stations in 2017 of GDA2020 elevation cutoff angle of 10° for observation selection, VMF1 grids (Boehm et al. 2006) for tropospheric delay, and the FES2004 model (Lyard et al. 2006) for ocean tide loading. IGS final GPS satellite ephemerides and earth orientation parameters were used in the computations. The double difference carrier phase observables at 30-s epoch intervals were used for GPS data processing. Other measurement modelling and parameter estimation included (Hu et al. 2011;Hu and Dawson 2018;Hu et al. 2019) solid earth tide displacements and Ocean tide loading displacements (Lyard et al. 2006); receiver clock corrections as well as absolute antenna phase centre variation and offset corrections; troposphere zenith delays estimated at 1-h intervals for all stations. Quasi-Ionosphere-Free (QIF) integer ambiguity resolution strategy is used for routine analysis with elevation dependent observation weighting and minimum constraint condition for daily network solution in terms of the ITRF2014 using subset of the IGS14 reference stations (Dach et al. 2012).
The daily solutions are stacked into weekly solutions using Bernese software based on the daily normal equations. Then the weekly solution was transformed from ITRF2014 to GDA2020 using the approach recommended in the GDA2020 Technical Manual (ICSM 2018). Before generating the Regulation 13 certificates for the sites, the GDA2020 solutions are checked to ensure the quality of the computed solution to meet the following requirements: • metadata in the Software Independent Exchange Format (SINEX) solution file is consistent with the site log files; • ensuring that there are no excessive data deletions where at least 80% data accepted per station for data processing; • the Root Mean Squares (RMS) of daily coordinate repeatability of all user supplied stations for the weekly solutions must be less than 5 mm for horizontal components and less than 10 mm in vertical component (Dach et al. 2012); • checking the minimally constrained solution against the IGS14 reference frame and/or the IGS combined analysis for the corresponding time period, to ensure RMS less than 5 mm for horizontal components and less than 10 mm for vertical component before Helmert transformation; • checking the final GDA2020 solution against the RVS coordinates of 109 AFN stations to ensure RMS less than 7 mm for horizontal components, and less than 15 mm for vertical component.
Following the above process, we have issued Regulation 13 certificates for 450 CORS sites on the Australian plate as shown in Fig. 4. Based on the 2017 data set, Hu and Dawson (2018) detailed the results of the Australian CORS position verification analysis that has led to the creation of certificates of verification of the reference standard of measurement for position in accordance with Regulation 13 of the National Measurement Regulations 1999, National Measurement Act 1960. GDA2020 coordinates and uncertainties are also reported. An example of Regulation 13 certificate for the site WWLG is given in Appendix 1.
Uncertainty of GPS positions
Position uncertainties were calculated in accordance with the principles of the International Standardization Organization (ISO) Guide to the Expression of Uncertainty in Measurement (GUM 1995), with an interval estimated to have a confidence level of 95% at the time of verification. The combined standard uncertainty was converted to an expanded uncertainty using a coverage factor, k, of 2. Position uncertainties are divided into type A and type B sources. Table 1 summarizes the major type A and B uncertainty sources for GPS analysis of the position verification.
Type A uncertainty sources were evaluated by adopting an a priori sigma of 0.001 m for the precision (1 sigma) of the L1-frequency, one-way, phase observation, at zenith (Dach et al. 2012). The corresponding uncertainties of all parameters were determined, by standard error propagation theory, in the least-squares estimation process used in the GPS analysis. Since the formal (internal) precision estimates of GPS solutions are well known to be optimistic, a factor of 10 (i.e. variance scale factor of 100) was subsequently applied to the variance-covariance matrix of the computed GDA2020 coordinates (Hu and Dawson 2018). This factor was selected based on an empirical assessment of the routine APREF analysis and is also largely with consistent previous research, see for example Blewitt and Lavallée (2002), Altamimi et al. (2002Altamimi et al. ( , 2016 and Firuzabdi and King (2011).
Type B uncertainty sources, which in practice contribute to position uncertainty, cannot be estimated from the statistical analysis of short-period (i.e. 7-day) observations; these include environmental effects, such as longperiod station loading (deformation) processes (Johnson and Agnew 1995;Altamimi et al. 2002;Blewitt and Lavallée 2002;King and Williams 2009).
Quality control of GPS position verification
Knowing the long-term stability of each CORS is necessary for ensuring the internal consistency and stability of the national datum as well as the reliability of the legal traceability of GPS position Hu et al. 2019). The quality control of GPS position verification can be performed through long-term position time series monitoring and short-term positioning performance Hu 2019).
The use of the long-term position time series allows a better geophysical interpretation of the observed site motion, in particular to understand the residual signal or so-called non-linear motion which may be related to site stability or local geophysical phenomena (King and Williams 2009). The horizontal component is assumed to be primarily related to tectonic plate motion, while the height component is associated with local or regional uplift or subsidence (Feissel-Vernier et al. 2007) which can be caused by both geophysical and anthropogenic sources.
As part of the products of the APREF project, Geoscience Australia estimate weekly coordinates for more than 450 CORS within the Australian plate with issued Regulation 13 certificates of GDA2020 coordinates (Hu et al. 2011. Each weekly solution is derived based on daily 24-h data in terms of GPS week. The above CORS network is based on voluntary contributions from more than 10 companies and State and Territory Governments entities across Australia. Each site is operated in accordance with those belonging to the IGS network, with the same conflicting goals of inclusiveness and selectivity (Altamimi et al. 2002). This means that although guidelines exist for equipment changes, different institutes use different practices. For example, some antenna or equipment changes within the Australian network are not always communicated to Geoscience Australia, who ensures the daily network management. In addition, when erroneous behavior is detected at one of the sites, Geoscience Australia cannot do more than inform the site operator of the change in behavior and request proper action to be taken (Hu et al. 2011). This can create issues with data availability, metadata consistency and cause unknown coordinate changes.
Therefore, a set of quality control systems is designed to detect significant deviation from published position and velocities, whereupon the published coordinates and velocities may be updated if the new estimates differ from the adopted values by pre-specified tolerances. The quality control starts from the metadata checking, Receiver Independent Exchange Format (RINEX) header information validation against the site log files, and progresses to the solution quality checking through comparison with the official IGS weekly solutions using the residuals from the Helmert transformation (Owen et al. 2018;Hu et al. 2011Hu et al. , 2019. For short-term variations, we monitor the published Regulation 13 coordinates with weekly monitoring of solutions difference after routine analysis. The GDA2020 coordinates at epoch 2020.0 for the recognized value of 109 stations across Australia plate are used as reference for ensuring the measurement traceability as detailed in Hu and Dawson (2018). The differences between the computed and the reference coordinates were calculated and subsequently transformed to the North, East and up components. Taking GPS week 2063 as an example, we found that 3% of coordinates for the RVS stations exceeding the position uncertainty bounds. There are several reasons to explain this finding: (1) some RVS stations are determined from short position time series of just over 2.5 years with noiser velocity uncertainties; (2) some sites contain discontinuities caused by equipment changes including antenna type and antenna radome changes; and (3) site instability caused by local environmental effects (Beavan 2005;King and Williams 2009). It is well known that the vertical component is the most sensitive component to equipment changes in particularly antenna changes or site environment changes (i.e. trees growing over or near the antenna blocking the sky view). Figure 5 presents a typical RVS site with Regulation 13 certificate where its antenna radome had been damaged at the end of year 2018. The resulting damage caused a jump of the position time series for the vertical component as shown in Fig. 6. In this case, the issued Regulation 13 certificate was cancelled and removed from public access, and a new Regulation 13 certificate was re-issued after the site operator removed the antenna radome. For long-term quality control, we monitor the CORS site stability through the position time series analysis using CATREF software (Altamimi et al. 2002) based on weekly solutions of the routine analysis for the APREF CORS network. The weekly coordinates in the combined APREF solution are linked to the ITRF2014 by minimally constraining the coordinates of a set of IGS core stations to the IGS14 position . For the first step, we generate raw position time series which allow identification of discontinuities and outliers. For this purpose, we remove the original constraints from each weekly APREF solutions and calculate a cumulative solution estimating position and velocity along with the position offsets. The resulting position time series i.e. so-called modelled time series are actually the residuals between the cumulative solution and each weekly solution after removing 14 parameter transformation (Hu et al. 2011. As already shown by Kenyeres and Bruyninx (2004), the Helmert transformation absorbs the common network velocity, the reference frame changes and any periodic signals such as annual signal common to the whole network and it allows identification of outliers and offsets. From long-term time series, all offsets and discontinuities due to equipment changes, e.g., antenna problems are estimated in the final cumulative solutions to obtain a set of coordinates and velocities, these discontinuities are stored in a SINEX file for the combination of position time series (Kenyeres and Bruyninx 2004;Altamimi et al. 2016).
We use the CATREF software package to do the combination of weekly solutions and generate the position time series (Altamimi et al. 2002(Altamimi et al. , 2016. The output of the position time series contain information such as geophysical signals, mismodelled effects, outliers and discontinuities. Based on the raw time series the outliers and offsets are treated and discontinuities are set up. Each station was independently checked (e.g. Hu et al. 2011Hu et al. , 2019. There are many factors degrading the site stability and the reliability of the position series including site environment changes which can cause an inconsistency in the position time series and even change the position repeatability. We can set up them as offsets or outliers. If unaccounted for, these effects can degrade the quality of the estimated parameter such as velocities and coordinates in terms of biases or higher uncertainties. A typical example is the station TELO, one of the Victorian CORS network. The issued Regulation 13 certificate was cancelled as this site is no longer suitable for legal traceability because of instability caused by the building movement. As shown in Fig. 7, the position time series for this site in both the east and vertical components demonstrate large variability. The site operator has confirmed this special case and decided to relocate the CORS to a more stable installation. In this case, the issued Regulation 13 certificate had to be cancelled and removed from public access. We call this type of CORS sites as a non-conforming case which cannot meet the quality requirements of GPS legal traceability of position verification. We have found several sites appearing to exhibit strange behaviour in either horizontal components or the vertical component. There are many phenomena that contribute to non-linear motion at a site, these include local subsidence or hydrological instabilities related to the periodic circulation of underground water; thermal expansion of the GPS stations (e.g., Johnson and Agnew 1995;van Dam et al. 2001;Caporali 2003;Romagnoli et al. 2003); monumentation problems such as the antenna being attached to an unstable building. These are possible explanations of the long and short term variability of the position time series. Geophysical processes such as earthquakes may produce significant displacement offsets Fig. 7 Anomaly behavior in the east and vertical components of the position time-series at station TELO in Victoria, Australia, the position time-series generated after removing the plate motion model and outliers as well as offset estimation that should be estimated or corrected (Altamimi et al. 2016). This information is critical for the CORS users if they want to derive accurate relative position for the other new sites from the CORS. Feissel-Vernier et al. (2007) define the site stability to include possible geophysical instabilities, as well as equipment or the monument foundation (i.e. attached to a building), and found the level of stability to be consistently lower than 3.5 mm for horizontal component and 4 mm vertical component.
The weekly solutions and combined SINEX files are the starting point for long term monitoring the CORS performance and consequently the position time series of each site. The official APREF products consists of weekly position solutions in the SINEX format . We update the APREF weekly cumulative position and velocity estimation and associated position time series, where the detected position offset or outliers are taken into account. We also maintain a database of metadata management as well as discontinuities in SINEX format (Hu et al. 2011Owen et al. 2018). A flowchart of legal traceability of issuing regulation 13 certificate for a site is illustrated in Appendix 2.
Concluding remarks
Geoscience Australia recognizes the implications and promise of GPS technology and is progressing the adaptation of GPS methods to improve the Geocentric Datum of Australia since 1992 along with the constant improvement in our knowledge of terrestrial reference frames (Altamimi et al. 2002;Hu et al. 2019). We have advised and updated several new realizations of GDA, refining at each time the recognized values of GDA coordinates (Dawson and Woods 2010;ICSM 2018).
The rigorous realization of geodetic frames based on CORS networks requires continuous monitoring of the set of position and velocity estimates defining the particular datum. Where coordinate changes are noted, whether caused by human or natural effects, the position of the CORS is re-estimated. We monitor the GDA2020 coordinates and velocities through weekly solutions of the APREF GPS CORS network.
The final goal of the study is to achieve reliable traceability of the GPS position in Australia. We create a SINEX file including the IGS discontinuities for the APREF CORS network. We update the position time series weekly, however the related discontinuity or offsets requires several weeks of data after the event or even longer to identify changes in the coordinates. When estimating the site velocity, we introduce the offset and reestimate a set of coordinates for the site while assuming the velocity is identical before and after the offsets.
Our position and velocity estimation are compared with the IGS published weekly solutions to ensure the traceability of GPS position. We found excellent agreement both in the position and velocity estimation in 2-4 mm for horizontal components and 3-6 mm for vertical components . The compiled discontinuities SINEX file allows estimating the offsets along with the position and velocity estimation. The position time series analysis enables visualising the long-term behaviour of the station as a whole to detect and identify the hardware malfunctioning as well as monitoring the site stability. | 5,824 | 2020-09-14T00:00:00.000 | [
"Law",
"Engineering"
] |
Realization of Free‐Space Long‐Distance Self‐Healing Bessel Beams
A new approach for generating long‐distance self‐healing Bessel beams, which is based on a ring‐shaped (annular) lens and a spherical lens in 4f‐configuration, is reported. With this, diffraction‐free light evolution of a zeroth order Bessel beam over several meters is shown and available scaling opportunities that surpass current technologies by far are discussed. Furthermore, it is demonstrated how this setup can be adapted to create Bessel beam superpositions, realizing the longest ever reported optical conveyor beam and helicon beam, respectively. Last, the self‐healing capabilities of the beams are tested against strong opaque and non‐opaque scatterers, which again emphasizes the great potential of this new method.
Introduction
Bessel beams-electromagnetic wave packets with an amplitude envelope described by a Bessel function-are diffraction-free entities with superior self-healing capabilities. [1] As such, they are in principle ideal tools for optical communication through turbulent atmosphere. [2] However, until this day, Bessel beams can only be achieved with a few centimeters in length even under ideal conditions, which limits their use for most long-range applications. [3] Bessel beams were first suggested by Durnin in 1987 [4] and experimentally realized soon after. [5] At the time, people used the fact that Bessel beams are equivalent to a conical superposition of plane waves and hence are represented by a ring of infinitesimal thickness in the angular frequency domain. Therefore, the simplest way to achieve a finite energy approximation of a Bessel beam is to use an annular aperture that is placed in the back focal plane of a thin spherical lens. [5] However, this method is rather inefficient, as most of the incident power is obstructed by the annulus. A Bessel beam may also be created using an axicon, or conical lens element, which provides the conical superposition of the wave components in real space. [6,7] This offers a higher efficiency, but the Bessel beam still only exists in the focal region of the conical lens, such that for beams exceeding a few centimeters unrealistically small base angles or very large axicon diameters would be required.
Although experimentally generated Bessel beams are still far away from their theoretical counterparts, they already find use in various applications, such as optical particle manipulation, [8,9] atomic dipole traps, [10] nonlinear optics, [11] microscopy, [12] material processing, [3] and quantum communication. [13] Moreover, superpositions of Bessel beams are applied for conveyor beams, [14,15] helicon beams, [16][17][18] or general radially self-accelerating beams. [19] Due to their excellent self-healing capabilities, [20] they may take a superior role in free-space terrestrial and satellite communication. [2] Recent studies suggest that Bessel beams are well suited for stable long-range communication through turbulent media. [21] However, these theoretical considerations have to assume a technology which surpasses currently achievable propagation lengths by far.
In our work, we report on the generation of Bessel beams in free space that exceed current experimentally reachable propagation lengths by about two orders of magnitude; that is, we present diffraction-free light evolution over several meters at Laser Photonics Rev. 2019, 13,1900103 optical wavelength. To this end, we employ a sophisticated yet simple optical component, which resembles a cylindrical lens that is morphed to a closed ring-like form. Such devices have been used well before the invention of the laser for imaging purposes.
Here we use them in reverse, to focus the incoming light into a sharp ring-shaped focal plane. [6,22] After Fourier transforming this extremely narrow ring of light a high-quality Bessel beam of zeroth order is obtained. By adding a helical phase, also higher order Bessel beams are achievable. In a sense, this approach is similar to the annular slit by Durnin et al.; however, without the inherent loss of an annular aperture. Moreover, it overcomes the technical limitations of a diffractive ring-lens as reported by Xin et al. [23]
Experimental Section
For our experimental implementation, we fabricated two prototype specimen: one with a single ring-lens for the generation of a single Bessel beam, and one with two concentric ring-lenses for generating a superposition of Bessel beams. Both were realized in-house by means of diamond machining. Figure 1 shows an illustration of the constructed prototype characterization setup. All technical specifications for the optical components can be found in the supplementary material.
Note that this setting is not optimized for efficiency; coating the ring-lens and using a preceding beam-shaping component would render the filter-mirror [M] and the associated optics [B 1 , L 4 ] unnecessary. Moreover, [B 2 ] was merely used to ensure normal incidence on the SLM. With this, we achieved a ring of radius R = 1.48 mm and full width at half maximum (FWHM) of w = 4.9 µm. Importantly, ring-width and shape are strongly influenced by the illumination pattern incident on the ring-lens as well as residual aberrations. Those, in turn, can alter the propagation dynamics of the ensuing Bessel beam, as we discuss in the Supporting Information. From numerical simulations, the above values are expected to provide a Bessel beam of zeroth or-der, which is 1.73 m in length and has an FWHM of the central lobe of 30.8 µm. For comparison, a Gaussian beam of similar width would reach 6.8 mm in length.
Long-Range Propagation
We experimentally recorded the propagating Bessel beam with a movable CCD camera and measured the width of the beam's central intensity maximum. As shown in Figure 2, it increases from 29 to 33 µm over a total range of 2.5 m.
A Gaussian beam of the same initial width would over the same distance broaden by a factor of about 600. Even a Gaussian of ten times that beam waist would still broaden by a factor of six. Note that the recorded experimental Bessel beam is longer than our theoretical prediction. This was accomplished by shifting the Fourier-lens [L 6 ] with respect to the ring plane. This introduces an additional phase term to the beam, which allows to trade parts of the diffraction-free nature for an increase in propagation range.
Optical Conveyor Beams
In a second experiment, we demonstrate the realization of an optical conveyor beam, that is, a beam that exhibits a set of evenly spaced focus-like high intensity peaks along the propagation direction. Such beams are potentially useful in particle manipulation, [15] material processing, [24] or laser-assisted plasma generation. [25,26] To this end, we fabricated a device with two concentric ring-lenses in order to prepare a superposition of two Bessel beams of zeroth order. With a setup similar to that described in Figure 1 (see Supporting Information) we achieved ring radii of R 1 = 1.48 mm and R 2 = 2.61 mm, respectively. We measured the peak intensity of the beam in three distinct Laser Photonics Rev. 2019, 13,1900103 Realization of a long-range optical conveyor beam. In the upper plot a cross section of the peak intensity along the optical axis for two superimposed zeroth order Bessel beams is shown. The bottom inset shows an iso-intensity graph for the central interval with 50% of the peak intensity considered as its iso-surface. The beam contains about 100 high intensity peaks over a propagation distance of 110 cm. intervals using an electronically controlled linear stage, as shown in Figure 3, and achieved 18 high intensity peaks per 20 cm interval. Note that the conveyor beam is shorter than the single Bessel beam presented in Figure 2. This is because the Bessel beam associated with the larger ring inherently exhibits a shorter propagation range and ceases to exist after about 110 cm.
Helicon Beams
When adding different helical phase orders to the individual Bessel beams from before, a helicon beam is formed. [16] Helicon beams are a special case of radially self-accelerating beams [19] and exhibit continuous spiraling trajectories. To this day, they are mostly generated using holographic techniques, which limits the number of achievable rotations to a mere handful. In our experiment, we prepare the phase orders n 1 = −1 and n 2 = +2 to obtain a helicon beam with a threefold rotational symmetry. The beam profile is recorded as before. From the collected volume data, an iso-intensity graph was calculated which is shown in Figure 4. The beam exhibits 36 full revolutions over a propagation range of 120 cm. This, together with the higher efficiency obtainable with our ring-lens device, makes helicon beams finally accessible to various fields of research and industrial applications.
Self-Healing
A peculiar feature of Bessel beams are their extreme selfhealing properties, allowing them to reestablish the initial field Laser Photonics Rev. 2019, 13, 1900103 Figure 4. Generation of helicon beams. Shown is the iso-intensity graph for the three main lobes of a helicon beam built from two Bessel beams with orders n 1 = −1 and n 2 = +2. The iso-surface corresponds to 33% of the peak intensity. The entire beam contains 36 full revolutions over a propagation distance of 120 cm. distribution even after rather devastating perturbations. [27,28] Importantly, in an experiment, this property will scale with the number of side lobes in the amplitude distribution of the Bessel beam. Our single beam from the beginning for instance supports 267 of those lobes within the FWHM of its apodization function (see supplementary material for more details), whereby excellent self-healing capabilities are anticipated. To assess the self-healing capacities of our beam experimentally, an opaque obstacle in form of a small screwdriver was placed 2 cm behind the Fourier lens, blocking a major part of the beam, as shown in Figure 5a. Although the beam experiences heavy distortions right after the obstacle (Figure 5b), after about 1 m of propagation it is essentially restored (Figure 5c). Most impressively, even the peak power remains at 96.9 % of the unperturbed case. In order to further quantify our results, the intensity distribution of each propagation step (with obstacle) was cross-correlated with the undistorted case (without obstacle). The results are shown in Figure 5d and indicate a 98.4 % recovery of the beam after 120 cm propagation.
We envision free-space optical communication through turbulent media as one of the most important applications of our method for generating high-quality long-range Bessel beams. In this vein, it is imperative to demonstrate that our generated beams are not only resilient against solid obstacles but also against non-opaque scatterers such as water-droplets. The latter type of scatterers poses a much greater challenge to the self-healing mechanism, as light is not simply removed but instead interferes with other parts of the beam profile along the propagation direction. To mimic such a rain-like scenario in the laboratory, we prepared a special sprinkler-container and placed it above the beam-line. The container holds 0.5 L of water and has 60 randomly positioned holes of 1 mm diameter along a path of 3 cm width and 12 cm length. It was placed 15 cm after the Fourier lens and 90 cm before the CCD camera, such that there is sufficient distance for the self-healing to occur. While pouring water through the beam, we recorded the beam-profile and subsequently determined the point stability of the central high intensity lobe as shown in Figure 6. Although the outer regions of the beam are subject to heavy distortions, the central lobe is always clearly visible and remarkably stable throughout the entire measurement. The heaviest average fluctuations we have observed are 3.9 µm in horizontal and 3.3 µm in vertical direction, respectively. Both values are around 10 % of the central lobe's FWHM.
Conclusion
The quality and range of experimental Bessel beams strongly depends on their lateral extent and, thus, on an appropriately large aperture. Our ring-lens technology excels in this regard. For one because the ring-lens is not the aperture limiting component, but moreover because the setup is easily adaptable to a reflective configuration in which the Fourier lens is replaced by a parabolic mirror. This way, the beam diameter could easily be increased from currently a few centimeters into the range of meters, which would further amplify the already extraordinary propagation and self-healing characteristics of our beams. With this, the utilization of Bessel beams in long-range optical communication through turbulent atmosphere and satellite applications is now in close reach. Nonetheless, we still foresee many exciting questions and prospects for future research: For example, is it possible to omit the Fourier transforming lens in large-scale set-tings and replace it by simple beam propagation into the far-field? Moreover, is it possible to overcome the inherent diffraction limit of the ring-lens using a metamaterial hyperlens-design, [29,30] such that the ring-width reaches the sub-wavelength regime? Last, what happens when a secure single-photon-based communication channel is implemented and quantum effects start to play a role? The answer to these and other questions are now in reach experimentally.
Supporting Information
Supporting Information is available from the Wiley Online Library or from the author. | 2,979.2 | 2019-09-13T00:00:00.000 | [
"Physics"
] |
Visualization of Protein Folding Funnels in Lattice Models
Protein folding occurs in a very high dimensional phase space with an exponentially large number of states, and according to the energy landscape theory it exhibits a topology resembling a funnel. In this statistical approach, the folding mechanism is unveiled by describing the local minima in an effective one-dimensional representation. Other approaches based on potential energy landscapes address the hierarchical structure of local energy minima through disconnectivity graphs. In this paper, we introduce a metric to describe the distance between any two conformations, which also allows us to go beyond the one-dimensional representation and visualize the folding funnel in 2D and 3D. In this way it is possible to assess the folding process in detail, e.g., by identifying the connectivity between conformations and establishing the paths to reach the native state, in addition to regions where trapping may occur. Unlike the disconnectivity maps method, which is based on the kinetic connections between states, our methodology is based on structural similarities inferred from the new metric. The method was developed in a 27-mer protein lattice model, folded into a 3×3×3 cube. Five sequences were studied and distinct funnels were generated in an analysis restricted to conformations from the transition-state to the native configuration. Consistent with the expected results from the energy landscape theory, folding routes can be visualized to probe different regions of the phase space, as well as determine the difficulty in folding of the distinct sequences. Changes in the landscape due to mutations were visualized, with the comparison between wild and mutated local minima in a single map, which serves to identify different trapping regions. The extension of this approach to more realistic models and its use in combination with other approaches are discussed.
Introduction
Understanding the processes leading to a protein folding into its native (functional) state is one of the important problems in molecular biophysics. In the 1960s, Anfinsen hypothesized that a protein in its native state and under physiological conditions would adopt such a structure with the lowest possible energy [1]. Though this hypothesis turned out to be correct, no explanation was offered to explain the large range of characteristic folding times, which may vary from milliseconds to seconds. In what became known as the Levinthal Paradox, in 1969 Levinthal argued that, due to an exponentially large number of states, a random search for the native structure would take cosmological times [2]. The solution to this paradox came from the energy landscape theory [3][4][5][6][7], which embeds the statistical nature of the folding process. The folding happens in a very high dimensional space, but in one of the possible descriptions, the complex landscape theory is projected along the reaction folding coordinate. The effective folding landscape topology is like a funnel, which has an energy gradient toward the native state region. This theory explained quantitatively the data for the folding of several proteins [8][9][10][11][12][13][14], and the funnel topology is correlated with the thermodynamics and kinetics of folding [15]. Many aspects of the folding funnel can be inferred from this approach, such as analysis of conformational maps [16,17], folding mechanisms involving mutants [18], and topological features in the transition state [19].
In other approaches, local minima are individually addressed and go beyond one-dimensional representation [20,21]. Visualization of distances between local minima is a very appealing way of showing the underlying structure of the funnel. However, visualizing the local minima poses a significant challenge owing to the multidimensional nature of the system. Among the motivations to investigate the funnel details and its visualization is the potential help in understanding the role of metastable states, kinetic routes and conformational changes associated with protein function [22][23][24]. The visualization of potential and free energy surfaces is not essential for calculating any dynamic or thermodynamic properties, but it can certainly help in providing insights as to what those properties might be [20,25,26]. Methods such as Principal Component Analysis (PCA) have been used in funnel visualization for isobutyryl-(ala) 3 -NH-methyl (IAN) [27], where disconnectivity graphs were used to visualize the overall organization of the landscape [28]. The potential energy surface is represented in terms of local minima and the transition states that connect them, providing a convenient coarse-grained representation of the corresponding landscape [29]. This method has been applied to a wide number of systems. For example, Lennard-Jones clusters present multi-funnel characteristics [30][31][32]. Disconnectivity graphs are able to reveal the effects of gatekeepers in the potential energy surface by raising the energies of low-lying minima relative to the global minimum [33]. The diferences in folding efficiencies can also be inferred in proteins with and without frustration for structure based models [34]. Disconnectivity graphs can also be extended for the visualization of free energy landscape, maintaining the description of barriers faithfully [26,35,36]. When rate constants are associated with the rearrangements mediated by each transition state, a kinetic transition network can be defined [37,38]. So the kinetics and thermodynamics of complex transitions can be modeled in terms of transitions between the relevant conformational substates [39][40][41], in which kinetic transition networks are constructed from geometry optimization and molecular dynamics simulations. These examples show that this method overcomes the fundamental limitations of reactioncoordinate-based methods. Most of these approaches emphasize the kinetic path between probed states, and are able to indicate, for example, the funnel aspect of the landscape against a hub-like hypothesis [41].
In this paper we focus on the structural organization of conformations, looking at the difference of contacts in each conformation. We propose a suitable conformation metric that reflects the underlying landscape in which the kinetics takes place. The method is tested in a 27-mer protein lattice model, folded into a 36363 cube, which has been extensively used in protein folding studies [3,42,43], and in particular for visualization methods [44]. We restricted the visualization to local minima of regions from around the transition-state to the native state. These partially folded states are the relevant ones in the study of metastable states and function-related conformation changes. The data obtained from computational simulations in a lattice model were projected on a 2D or 3D plot with the Force-Scheme method [45], which allowed us to map the connectivity of conformations (local minima). The choice of a metric is essential in order to reach a sensible connection between the original data and the projection, and it must efficiently distinguish between pairs of conformations. From the analyses, we noted that distinct sequences lead to different patterns, from which folding routes could be established and the effects from mutations could be probed.
Results and Discussion
The simulation of the folding dynamics probes the conformations associated with local minima within given time intervals. We are interested in mapping the partially folded states, associated with conformations from the transition-state to the native configuration. The transition state was inferred from the free energy as a function of degree of nativeness (see Supporting Information) for the protein-like sequences A, Af, B, C and D. Conformational states are characterized by the energy and nonbonding contact points for each monomer of the sequence. The dataset thus generated is multidimensional, and its visualization requires dimension reduction projection methods. A crucial point for the projection is to establish a metric for the distance between two conformations. We tried several possibilities, including the Minkowski family of metrics [46], of which the Euclidean distance is one example. These did not lead to physically plausible results since the computation of such metrics considers that lack-ofcontact comparisons define similar elements. In the lattice case, the absence of contact (''0'' comparisons) occurs when two conformations do not present contacts. In this scenario a binary distance is a better choice, i.e., only contacts (''1'' comparisons) are relevant.
The measure between two conformations i and j has to satisfy commutativity and null distance to itself, i.e., The structural measure or distance shown to be most effective was the ratio between the dissimilarity (D i,j ) and similarity (C i,j ) between i and j, which is equivalent to the ratio between the Jaccard index and the Jaccard distance [47], defined as (1) conditions, but compares the overall conformation, which may not properly account for local details. One could argue that this topological distance, which could capture static features of the conformation space, may not cope with details of folding. Folding process is an intrinsically dynamic process, which is also the basis of the the discontinuity graphs discussed in the Introduction. Moreover, two structurally similar conformations could differ in terms of the dynamics for folding. We therefore incorporated in the simulations a dynamic measurement defined by where n(i,j) is the number of local minimum intermediates required to go from i to j conformations. M d (i,j) corresponds to the minimum calculated over all the paths going from i to j (or vice-versa). The measurement is normalized upon dividing by the largest distance encountered. This approach resembles the method using to determine kinetic transition networks [48][49][50]. In subsidiary simulations we noted that using an effective distance M ef (in Eq. (4)), which takes into account the dynamic measurement, yields essentially the same results as with our initial measurement defined in Eq.2. Therefore the use of the latter appears to embed the underlying landscape of the system.
Visualizing the folding funnel
The protein funnel was obtained by projecting the multidimensional local minima, distributed according to the effective metric distance, onto a 2D surface. The 5 sequences investigated, viz. A, Af, B, C and D, are described in detail in the Methods. Figure 1 shows the funnel representation of sequence A, in which the minima are colored according to conformation energy in Figure 1a, or according to the reaction coordinate Q in Figure 1b. The steep convergence to the native state either in energy or Q representation is an indicative of the principle of minimal frustration associated with this sequence. The important information is the relative distance between two given points, and the axes were removed because the directions do not have any special meaning. Different regions in the 2D representation can be associated with different partially folded motifs, as shown in Figure 1a. As expected, different time intervals sample different minima, thus yielding varying local minima resolution, but the overall funnel pattern was maintained (see Figures S3, S4, S5 and S6 in the Supporting Information). The pattern preservation for distinct time intervals (in MCs) ensures that the sequence possesses a unique ''signature'', with clusters of conformations becoming denser as the number of time intervals decreases (probing more fluctuations). For a 30 MCs interval, in particular, a more refined energy distribution can be visualized with the identification of higher energy conformations when compared with local minima in simulations with larger time intervals. Figure 2 shows that the funnel landscape obviously depends on the protein sequence, with a unique native structure being represented by a unique funnel landscape. The sequence D, in particular, has a doubly degenerate native state, where the two lowest-energy conformations differ from each other by 5 native contacts. The existence of these two native states is reflected in two clusters of points in Figure 2d. For this sequence, a change from one region (native state) to the other native state requires unfolding (i.e. the need to move towards the periphery in the projection).
Note that, for sequences that are difficult to fold (Figure 2a and 2c), the number of conformations with intermediate energy (in the green light blue region) increases considerably, in comparison with the easily-foldable sequences (A and B) (Figure 1 and 2b). By the same token, the sequences with non-efficient folding funnels take a much longer average time to fold, as shown in Figure S2 in the Supporting Information.
In order to generate a 3D visualization for the funnel, the 2D representation was taken for the x and y axes, while the energy was taken as the z axis, with the lowest energy value corresponding to the native state. Color encodes the reaction coordinate Q, which is the degree of nativeness. Figure 3 shows the 3D picture of the funnel for the sequence A, while the figures for the other sequences are given in Figures S7 and S8 in the Supporting Information. It must be stressed that the result of the projection method is independent of the initial condition of the states in the 2D representation. The native conformation converges to the center of the funnel without any constraint or external force. The global minimum of the system, or native configuration, in the center of the 2D representation reinforces the funnel-like structure of the landscape.
Folding routes
The 2D and 3D visualizations of the folding funnels appear to confirm that the strategy proposed here is suitable for describing the folding process, but they do not suffice to ensure that the choice of the distance metrics is robust. The latter can be probed by analyzing the folding routes, for in a good funnel representation the folding route has to be represented by a sequence of small steps in the effective funnel representation. Figure 4a shows two routes generated from first passage time simulations, which show mostly small steps between successive minima. The details of this representation can be seen in different folding routes, which probe very distinct regions of the phase space (associated with different partially folded motifs). Also worth mentioning is that the Figure 1. Visualization in 2D of the conformation space for sequence A. Each point represents one conformation (local minimum) and the distance between points refers to the projection of their effective distance. The axis directions do not have any special meaning and have been removed. In (a) the color is associated with the conformation energy. In (b) the color is associated with the reaction coordinate Q, where Q~28 represents the native state. doi:10.1371/journal.pone.0100861.g001 routes do not directly cross the empty regions, but go around them through neighboring connected states. Figure 4b shows that, for sequence A, the distances between two subsequent local minima in the 2D representation are almost always very small, which means that no drastic changes occur in conformation from one minimum to the next. This confirms the robustness of the approach presented here.
Analysis of a mutation
The 2D projection was also used to explore a mutation in sequence A, where two monomers were exchanged to yield a less stable sequence (see Table 1 in the Methods). The effects from the mutation can be evaluated by mapping the data of the two sequences in the same projection. Due to mutation a set of conformations is no longer energetically favorable for the folding. This can be seen in Figure 5a where the whole region on the left is missing for the mutated sequence (green points). One thousand (1000) folding routes were calculated for each sequence, with examples shown in Figures 5b and 5c. In contrast to the wild sequence (A), for the mutated sequence (Af) the routes normally probe a significant part of conformational space before reaching the native state, with 95% of the pathways occurring on the righthand part of the projection. The mutation stabilizes a different set of local minima, which hinders the folding process and causes a considerable increase in the average folding time (as seen in Figure S2). Note that most of the minima in the mutated sequence do not coincide with those of the wild sequence, thus indicating that they are structurally different, even though they have the same native state.
Conclusions
Visualization was based on the assumption that the distance between two conformations was the ratio between the Jaccard index and the Jaccard distance taking into account all non-bonded contact points. The suitability of the approach could be confirmed by comparing the funnels and folding routes for 5 sequences, where much larger folding times were estimated for sequences known to be difficult to fold. Furthermore, a doubly degenerate sequence yielded a funnel with two native states, as expected. Since the methods employed are entirely generic, this approach is a potential tool to be used in association with other methods that efficiently probe the energy landscape, such as diffusion-mapdirected MD (DM-d-MD) [51], disconnectivity graphs [20] and metadynamics [52]. The method was tested in a simple lattice model, in which the minima were sampled with variable time intervals. It will be straight forward to apply this methodology to realistic models and more meaningful sampling methods, such as those used by Wales [20,21,25]. In particular, our method may be helpful to probe details of folding trajectories and effects of mutation in the study of metastable states. As applications, previous work using disconnectivity graphs analyzed the potential energy landscapes of proteins involving gatekeeper residues [33,53,54]. By probing the gatekeeper residue contacts using our method we expect to be able to shed light into the nature of these peculiar conformational states.
Model
In this lattice model, a globular protein is modeled as a simplified heteropolymer made up of 27 monomers (or beads) covalently bonded. The monomers are placed on the vertices of a cubic lattice. These models are capable of accounting for several features of protein folding [42], where the most compact (folded) structure is a 36363 cube. One contact is defined for two monomers that are at nearest-neighbor distances but not connected covalently. In the lattice model the maximum number of contacts is 28: The energy of the system is given by E~n l E l zn u E u , where n l is the number of (non-covalent) contacts of like monomers and n u is the number of contacts between distinct monomers. The folding kinetics is performed with the Metropolis algorithm in a Monte Carlo simulation with typical motions in polymers [42]. Here we use a low hydrophobicity regime with E l~{ 3 and E u~z 3 in arbitrary units. This regime was chosen to mimic the folding behavior where the sequence evolves toward its native state without going through a hydrophobic collapse [43,55]. Five sequences were chosen for the analysis, which exhibit very distinct features, as indicated in Table 1. For each conformation, the free energy was calculated as a function of the parameter Q (See Figure S1 in the Supporting Information). The data collected for the projection is restricted to conformations from around the transition state (Q TS {1) to the native state (Q~28): The simulation temperature was set to 1:1 T f , in order for the conformational space to be visited as thoroughly as possible, thus avoiding the sequence having to spend long times in its native state. Local minima were obtained within time intervals segmented along the Monte Carlo trajectories. 4 time intervals were used: 30, 100, 300 and 1000 Monte Carlos steps (MCs). For each interval, the total time was set so that 10 7 minima were obtained. The conformation at each local minimum was stored in a 27|27 binary matrix representing all the contacts. The conformational matrix is symmetrical and an element c i,j is 1 if there is a contact between monomers i and j and 0 otherwise.
Metric
The projection of these multidimensional data was performed using a metric based on the conformational similarity (Jaccard index) and dissimilarity (Jaccard distance), referred to as the structural measurement: M s (Eq. 2). We also tested a dynamic measurement in which the number of intermediate minima for going from one conformation to the other was taken into account. This latter metric was named dynamic measurement M d (Eq. 3). Using these measurements one may calculate a normalized effective distance between any two conformations, Projection Our goal is not to develop a technique for dimensionality reduction. We want to visualize the similarity between conformations according to our metric. Since the information of structures occurs in a multidimensional space, there is a need for projection into a lower dimension. As with any projection technique, we can create the projection in up to three dimensions [56]. The choice of two dimensions is simply for the ease of data interpretation. 3D projections are very difficult to interpret due to occlusions and overlaps which, in most cases, do not bring real gain compared to 2D [57].
The projection onto a 2D plot was made using the distance matrix with the Force-Scheme method [45], where the objects are initially placed in random positions, and then attraction and repulsion forces between the objects take the system to equilibrium according to a chosen heuristics. Here, the system was initialized with the conformation energies, which proved more efficienct for convergence of the method. After the first placement of the objects, iterations within the Force-Scheme method are performed to preserve similarity in the original space into the projected space. In the first iteration, for each projected point y i [Y , (where Y is the input dataset) a vector is calculatedṽ v i,j~( y j {y i ),Vy j =y i : Then y i is moved in theṽ v direction by a step D, defined as: where k is the number of previous iterations. After an iteration, each object should be moved closer to its similar ones until the system converges. The number of iterations may be defined arbitrarily or the scheme may be stopped when a threshold is reached. Here the process was stopped when the difference in distances for a given object between two consecutive iterations was below a threshold of 10 {4 : In order to build the 3D funnel, the points in the 2D projection are shifted along a perpendicular axis according to their energies, thus generating a 3D structure where the lowest-energy states are placed on the bottom. We also performed tests with one of the most precise projection techniques in terms of distance preservations, referred to as Classical Multidimensional Scaling (MDS) [56]. The results were similar to those produced by the Force-Scheme in terms of distributing the points on the plane according to the similarity between conformations, with the final shape of the funnels also being very similar. The MDS technique, however, is much more costly in computational time, and in some cases ordinary microcomputers lack the power to obtain the funnels. Therefore, we opted for the Force-Scheme approach, which is much faster and allows one to process thousands of conformations in a few minutes with a simple PC. Zscore is calculated according to methodology described by Dima et al. [58].`Sequence design by Shakhnovich et al. [59] which has been used in other studies [42,43]. 1 This sequence was obtained through a permutation of two monomers in A, which results in three frustrated contacts in the native structure. doi:10.1371/journal.pone.0100861.t001 | 5,238.8 | 2014-07-10T00:00:00.000 | [
"Biology",
"Computer Science"
] |
Platforms for Non-speakers Annotating Names in Any Language
We demonstrate two annotation platforms that allow an English speaker to annotate names for any language without knowing the language. These platforms provided high-quality ’‘silver standard” annotations for low-resource language name taggers (Zhang et al., 2017) that achieved state-of-the-art performance on two surprise languages (Oromo and Tigrinya) at LoreHLT20171 and ten languages at TAC-KBP EDL2017 (Ji et al., 2017). We discuss strengths and limitations and compare other methods of creating silver- and gold-standard annotations using native speakers. We will make our tools publicly available for research use.
Introduction
Although researchers have been working on unsupervised and semi-supervised approaches to alleviate the demand for training data, most state-ofthe-art models for name tagging, especially neural network-based models still rely on a large amount of training data to achieve good performance. When applied to low-resource languages, these models suffer from data sparsity. Traditionally, native speakers of a language have been asked to annotate a corpus in that language. This approach is uneconomical for several reasons. First, for some languages We thank Kevin Blissett and Tongtao Zhang from RPI for their contributions to the annotations used for the experiments. This material is based upon work supported by the Defense Advanced Research Projects Agency (DARPA) under Contracts No. HR0011-15-C-0115 and No. HR0011-16-C-0102. The views, opinions and/or findings expressed are those of the authors and should not be interpreted as representing the official views or policies of the Department of Defense or the U.S. Government. 1 https://www.nist.gov/itl/iad/mig/lorehlt-evaluations with extremely low resources, it's not easy to access native speakers for annotation. For example, Chechen is only spoken by 1.4 million people and Rejiang is spoken by 200,000 people. Second, it is costly in both time and money to write an annotation guideline for a low-resource language and to train native speakers (who are usually not linguists) to learn the guidelines and qualify for annotation tasks. Third, we observed poor annotation quality and low inter-annotator agreement among newly trained native speakers in spite of high language proficiency. For example, under DARPA LORELEI, 2 the performance of two native Uighur speakers on name tagging was only 69% and 73% F 1 -score respectively. Previous efforts to generate "silver-standard" annotations used Web search (An et al., 2003), parallel data (Wang and Manning, 2014), Wikipedia markups (Nothman et al., 2013;Tsai et al., 2016;, and crowdsourcing (Finin et al., 2010). Annotations produced by these methods are usually noisy and specific to a particular writing style (e.g., Wikipedia articles), yielding unsatisfactory results and poor portability.
It is even more expensive to teach Englishspeaking annotators new languages. But can we annotate names in a language we don't know? Let's examine a Somali sentence: "Sida uu saxaafadda u sheegay Dr Jaamac Warsame Cali oo fadhigiisu yahay magaalada Baardheere hadda waxaa shuban caloolaha la yaalla xarumaha caafimaadka 15-cunug oo lagu arkay fuuq bax joogto ah, wuxuu xusay dhakhtarku in ay wadaan dadaallo ay wax kaga qabanayaan xaaladdan" Without knowing anything about Somali, an English speaker can guess that "Jaamac Warsame Cali" is a person name because it's capitalized, the word on its left, "Dr," is similar to "Dr." in English, and its spelling looks similar to the English "Jamac Warsame Ali." Similarly, we can identify "Baardheere" as a location name if we know that "magaalada" in English is "town" from a common word dictionary, and its spelling is similar to the English name "Bardhere." What about languages that are not written in Roman (Latin) script? Fortunately language universal romanization (Hermjakob et al., 2018) or transliteration 3 tools are available for most living languages. For example, the following is a Tigrinya sentence and its romanized form: An English speaker can guess that "ዓብደልፈታሕ አል-ሲሲ" is a person name because its romanized form "aabedalefataahhe 'ale-sisi" sounds similar to the English name "Abdel-Fattah el-Sissi," and the romanized form of the word on its left, "ፕረዝደንት," (perazedanete) sounds similar to the English word "president." Moreover, annotators (may) acquire languagespecific patterns and rules gradually during annotation; e.g., a capitalized word preceded by "magaalaa" is likely to be a city name in Oromo, such as "magaalaa Adaamaa" (Adama city). Synchronizing such knowledge among annotators both improves annotation quality and boosts productivity.
The Information Sciences Institute (ISI) developed a "Chinese Room" interface 4 to allow a nonnative speaker to translate foreign language text into English, based on a small set of parallel sentences that include overlapped words. Inspired by this, RPI and JHU developed two collaborative annotation platforms that exploit linguistic intuitions and resources to allow non-native speakers to perform name tagging efficiently and effectively.
Word recognition. Presentation of text in a familiar alphabet makes it easier to see similarities and differences between text segments, to learn aspects of the target language morphology, and to remember sequences previously seen.
Word pronunciation. Because named entities often are transliterated into another language, access to the sound of the words is particularly important for annotating names. Sounds can be exposed either through a formal expression language such as IPA, 5 or by transliteration into the appropriate letters of the annotator's native language.
Word and sentence meaning. The better the annotator understands the full meaning of the text being annotated, the easier it will be both to identify which named entities are likely to be mentioned in the text and what the boundaries of those mentions are. Meaning can be conveyed in a variety of ways: dictionary lookup to provide fixed meanings for individual words and phrases; description of the position of a word or phrase in a semantic space (e.g., Brown clusters or embedding space) to define words that are not found in a dictionary; and full sentence translation.
Word context. Understanding how a word is used in a given instance can benefit greatly from understanding how that word is used broadly, either across the document being annotated, or across a larger corpus of monolingual text. For example, knowing that a word frequently appears adjacent to a known person name suggests it might be a surname, even if the adjacent word in the current context is not known to be a name.
World knowledge. Knowledge of some of the entities, relations, and events referred to in the text allows the annotator to form a stronger model of what the text as a whole might be saying (e.g., a document about disease outbreak is likely to include organizations like Red Cross), leading to better judgments about components of the text.
History. Annotations previously applied to a use of a word form a strong prior on how a new instance of the word should be tagged. While some of this knowledge is held by the annotator, it is difficult to maintain such knowledge over time. Programmatic support for capturing prior conclusions (linguistic patterns, word translations, possible annotations for a mention along with their frequency) and making them available to the annotator is essential for large collaborative annotation efforts.
Adjudication. Disagreements among annotators can indicate cases that require closer examination. An adjudication interface is beneficial to enhance precision (see Section 4).
The next section discusses how we embody these requirements in two annotation platforms.
Annotation Platforms
We developed two annotation tools to explore the range of ways the desiderata might be fulfilled: ELISA and Dragonfly. After describing these interfaces, Figure 1 shows how they fulfill the desiderata outlined in Table 2. Annotation Panel. For each sentence in a document, we show the text in the original language, its English translation if available, and automatic romanization results generated with a languageuniversal transliteration library. 7 To label a name mention, the annotator clicks its first and last tokens, then chooses the desired entity type in the annotation panel. clicking a token in the document will show its full definition in lexicons and bilingual example sentences containing that token. A floating pop-up displaying romanization and simple definition appears instantly when hovering over a token. Rule Editor. Annotators may discover useful hueristics to identify and classify names, such as personal designators and suffixes indicative of locations. They can encode such clues as rules in the rule editor. Once created, each rule is rendered as a strikethrough line in the text and is shared among annotators. For example (Figure 1, if an annotator marks "agency" as an organization, all annotators will see a triangular sign below each occurrence of this word.
ELISA
Adjudication Interface. If multiple users process the same document we can consolidate their annotations through an adjudication interface (Figure 3). This interface is similar to the annotation interface, except that competing annotations are displayed as blocks below the text. Clicking a block will accept the associated annotation. The adjudicator can accept annotations from either annotator or accept the agreed cases at once by clicking one of the three interface buttons. Then, the adjudicator need only focus on disputed cases, which are highlighted with a red background.
Dragonfly
Dragonfly, developed at the Johns Hopkins University Applied Physics Laboratory, takes a more word-centric approach to annotation. Each sentence to be annotated is laid out in a row, each column of which shows a word augmented with a variety of information about that word. Figure 4 shows a screenshot of a portion of the Dragonfly tool being used to annotate text written in the Kannada language. The top entry in each column is the Kannada word. Next is a Romanization of the word (Hermjakob et al., 2018). The third entry is one or more dictionary translations, if available. The fourth entry is a set of dictionary translations of other words in the word's Brown cluster. (Brown et al., 1992) While these tend to be less accurate than translations of the word, they can give a strong signal that a word falls into a particular category. For example, a Brown cluster containing translations such as "Paris," "Rome" and "Vienna" is likely to refer to a city, even if no translation exists to indicate which city. Finally, if automated labels for the sentence have been generated, e.g., by a trained name tagger, those labels In addition to word-specific information, Dragonfly can present sentence-level information. In Figure 4, an automatic English translation of the sentence is shown above the words of the sentence (in this example, from Google Translate). Translations might also be available when annotating a parallel document collection. Other sentence-level information that might prove useful in this slot includes a topic model description, or a bilingual embedding of the entire sentence. Figure 4 shows a short sentence that has been annotated with two name mentions. The first word of the sentence (Romanization "uttara") has translations of "due north," "northward," "north," etc. The second word has no direct translations or Brown cluster entries. However, its Romanization, "koriyaavannu," begins with a sequence that suggests the word 'Korea' with a morphological ending. Even without the presence of the phrase "North Korea" in the MT output, an annotator likely has enough information to draw the conclusion that the GPE "North Korea" is mentioned here. The presence of the phrase "North Korea" in the machine translation output confirms this choice.
The sentence also contains a word whose Romanization is "ttramp." This is a harder call. There is no translation, and the Brown cluster translations do not help. Knowledge of world events, examination of other sentences in the document, the translation of the following word, and the MT output together suggest that this is a mention of "Donald Trump;" it can thus be annotated as a person.
Experiments
We asked ten non-speakers to annotate names using our annotation platforms on documents in various low-resource languages released by the DARPA LORELEI program and the NIST TAC-KBP2017 EDL Pilot . The genres of these documents include newswire, discussion forum and tweets. Using non-speaker annotations as "silver-standard" training data, we trained (Lample et al., 2016). The lexicons loaded into the ELISA IE annotation platform were acquired from Panlex, 8 Geonames 9 and Wiktionary. 10 Dragonfly used bilingual lexicons by (Rolston and Kirchhoff, 2016).
Overall Performance
The agreement between non-speaker annotations from the ELISA annotation platform and gold standard annotations from LDC native speakers on the same documents is between 72% and 85% for various languages. The ELISA platform enables us to develop cross-lingual entity discovery and linking systems which achieved state-of-the-art performance at both NIST LoreHLT2017 11 and ten languages at TAC-KBP EDL2017 evaluations . Four annotators used two platforms (two each) to annotate 50 VOA news documents for each of the five languages listed in Table 2. Their annotations were then adjudicated through the ELISA adjudication interface. The process took about one week. For each language we used 40 documents for training and 10 documents for test in the TAC-KBP2017 EDL Pilot. In Table 2 we see that the languages with more annotated names (i.e., Albanian and Swahili) achieved higher performance.
Silver Standard Creation
We compare our method with Wikipedia based silver standard annotations on Oromo and Tigrinya, two low-resource languages in the LoreHLT2017 evaluation. Table 3 shows the data statistics. We can see that with the ELISA annotation platform we were able to acquire many more topically-relevant training sentences and thus achieved much higher performance. Native-speaker Annotation Non-speaker Annotation Figure 5: Russian Name Tagging Performance Using Native-speaker and Non-speaker Annotations. Figure 5 compares the performance of Russian name taggers trained from Gold Standard by LDC native speakers and Silver Standard by nonspeakers through our annotation platforms, testing on 1,952 sentences with ground truth annotated by LDC native speakers. Our annotation platforms got off to a good start and offered higher performance than annotations from native speakers, because non-speakers quickly capture common names, which can be synthesized as effective features and patterns for our name tagger. However, after all low-hanging fruit was picked, it became difficult for non-speakers to discover many uncommon names due to the limited coverage of lexicon and romanization; thus the performance of the name tagger converged quickly and hits an upper-bound. For example, the most frequently missed names by non-speakers include organization abbreviations and uncommon person names. Table 5 shows that the adjudication process significantly improved precision because annotators were able to fix annotation errors after extensive discussions on disputed cases and also gradually learned annotation rules and linguistic patterns. Most missing errors remained unfixed during the adjudication so the recall was not improved. | 3,320.8 | 2018-07-01T00:00:00.000 | [
"Computer Science",
"Linguistics"
] |
Population attributable fraction: comparison of two mathematical procedures to estimate the annual attributable number of deaths
Objective The purpose of this paper was to compare two mathematical procedures to estimate the annual attributable number of deaths (the Allison et al procedure and the Mokdad et al procedure), and derive a new procedure that combines the best aspects of both procedures. The new procedure calculates attributable number of deaths along a continuum (i.e. for each unit of exposure), and allows for one or more neutral (neither exposed nor nonexposed) exposure categories. Methods Mathematical derivations and real datasets were used to demonstrate the theoretical relationship and practical differences between the two procedures. Results of the comparison were used to develop a new procedure that combines the best features of both. Findings The Allison procedure is complex because it directly estimates the number of attributable deaths. This necessitates calculation of probabilities of death. The Mokdad procedure is simpler because it estimates the number of attributable deaths indirectly through population attributable fractions. The probabilities of death cancel out in the numerator and denominator of the fractions. However, the Mokdad procedure is not applicable when a neutral exposure category exists. Conclusion By combining the innovation of the Allison procedure (allowing for a neutral category) and the simplicity of the Mokdad procedure (using population attributable fractions), this paper proposes a new procedure to calculate attributable numbers of death.
Background
There are two mathematical procedures to estimate the number of deaths attributable to a risk factor such as obesity, smoking or alcohol consumption. Number of attributable deaths is the number of deaths in a population that could be avoided if the effects of the risk factor were eliminated from the population. The two procedures are Allison et al [1] and Mokdad et al [2]. Both procedures are under the assumptions of no confounding and no effect modification. Both can be applied to risk factors with polytomous exposure categories. The Allison procedure [1], originally developed for obesity attributable deaths, is rather complex. It involves 12 steps, uses hazard ratios, and requires calculating hazard rates by using a mathematical process to solve for an unknown quantity. The Mokdad procedure [2] on the other hand, is simpler. It involves only 6 steps, uses relative risks, and does not require solving for any unknown quantity.
In general, a common belief is that the more complex the procedure, the more accurate the results. Allison et al further stipulated that their procedure accounts for "complications", because it can estimate attributable deaths for body mass index (BMI) along a continuum (ie, for each unit of BMI), and can adjust for time using hazard ratio (HR) that the relative risks (RR) cannot achieve [1]. As a result, Mokdad et al used the Mokdad procedure to estimate the attributable numbers for tobacco and alcohol, and then they reverted to the more complex Allison procedure to estimate the attributable number for obesity.
A detailed read of Allison et al's paper revealed that two steps in the Allison procedure are not well-documented. First, while the equation for the overall number of deaths attributable to obesity and overweight (ω) is given in their paper, the equation to calculate the number of deaths attributable to each individual BMI category is missing. It is therefore unclear how data in their table three can lead to results in their table four. Second, it is said in their paper that the hazard (λ) can be obtained by numerically solving a complex equation for λ. Τhe actual method, however, is not given. For the less sophisticated users, the Allison procedure is not user-friendly.
There are several questions arising from looking at these two procedures: How are the Allison and Mokdad procedures mathematically and practically different? What are the best aspects of each procedure? Can the underlying equations be combined or modified to take advantage of the best aspects of both?
This paper compares the Allison and Mokdad procedures for the estimation of annual attributable number of deaths, both mathematically and using real data. The paper also "recovers" the missing Allison equation to calculate the individual number of deaths attributable to each BMI category, develops a similar and simpler equation using the logic of the Mokdad procedure, compares estimated number of attributable deaths under the HR and RR models, and looks at several options for numerically solving the equation for λ. This paper also proposes a modified Mokdad procedure that can achieve the same results as the Allison procedure.
Methods
Mathematical derivations from first principles from population attributable fraction (PAF), defined as the proportion of deaths in a population that can be attributed to the causal effects of a risk factor or set of factors, were used to demonstrate the relationship and differences of the Allison and Mokdad procedures. The missing mathematical equation to estimate the number of deaths attributable to each exposure category was derived for the Allison procedure. A similar equation was created for the Mokdad procedure. The two procedures were then "taken apart" and the logics behind the two procedures were examined and compared. Based on this, a new procedure (modified Mokdad) was developed combining the innovation of the Allison procedure and the logic of the Mokdad procedure. Finally, estimation methods under the hazard ratio and relative risk models were compared. Some options for solving for λ were described. Real datasets provided by Allison et al were used to illustrate the practical differences of the two procedures (Allison and the new procedure provided in this paper). Table 1 is a conversion table of the notations used in Allison et al, Mokdad et al, and this paper. The 12 steps in the Allison Procedure and the 6 steps in the Mokdad procedure to calculate the attributable number of deaths are summarized in Additional file 1, Appendix S1, using their original notations.
Results
1. Mathematical proof that the Allison procedure and Mokdad procedure differ in a neutral exposure category (Q) Based on Levin [3], and using notations in Table 2, where PAF is population attributable fraction; p is the probability of death in the population, or P(D); p 0 is the probability of death among the nonexposed, or P(D|E 0 ). Then using equation T1 (Table 2), is another frequently quoted form of PAF [3].
Extending the same methodology above for the case of a dichotomous exposure variable to the case of a polytomous exposure variable, and using equation T2 (Table 3), is Mokdad et al [2] (see Additional file 1, Appendix S1, equation A3), as 1-∑f i is P 0 ( Table 1).
From equation 3, Table 3, therefore is a modified form of Allison et al [1] (see Additional file 1, Appendix S1, equation A1), as shown below. From equations A2 (Additional file 1, Appendix S1) and 5, and given p = M/N, Comparing the Mokdad equation (6) with the Allison equation (A1), the Mokdad procedure allows only for a single nonexposed category E 0 and a number of exposure categories E 1 ... E i ...E k (see Table 3), while the Allison procedure in addition allows for a neutral (i.e., neither nonexposed nor exposed) category, in this case the category Q (underweight) (see Table 4).
The methodology of Allison et al leads to a modified Levin equation as shown below.
The original Allison equation (Additional file 1, Appendix S1, equation A1) was written for a three category exposure (R, reference group or the nonexposed E 0 ; O, overweight and obese groups; and Q, the underweight group) ( Table 4) and is where ω is number of deaths attributable to exposure categories E 1 ... E i ... E k combined, M is total number of deaths, N is total number in population, P(R) is f 0 = 1- From equation A1, and using the notations of this paper (Table 4), ,
Comparing equation 8 (derived from Allison et al) with Levin's original equation for PAF (equation 1)
which is identical to the Mokdad procedure, the Allison procedure subtracts out a certain weighted proportion of deaths associated with the neutral category (the underweight Q) from the attributable deaths to the exposure (the overweight and obese). In other words, the Allison procedure allows for a neutral exposure category (neither nonexposed nor exposed), while the Mokdad procedure does not.
Fraction of population exposed to an exposure category (i) P i f i Fraction of population exposed to a neutral category (e.g., underweight) P(Q) Let p = probability of death in population f = fraction of population exposed p 0 = probability of death in the nonexposed R = relative risk It follows that for a dichotomous exposure variable, based on the second row of the Τherefore, the missing equation is where ω i is number of deaths attributable to each exposure category i.
Equations 9 and 10 are new equations we created for the Allison procedure. It is worth noting that the underweight (Q) category disappears in equations 9 and 10, as it does not play a role in the calculations of attributable numbers to obesity (O). However, equations 9 and 10 are still difficult to use because the probabilities of death in the exposed and reference groups can only be estimated through a complex process: The next section shows that the Mokdad procedure can be modified to do the same calculations as the Allison procedure, but more simply.
Development of an equation for number of deaths attributable to each exposure category for the modified Mokdad procedure
The Allison procedure directly estimates the number of attributable deaths, ω. The Mokdad procedure indirectly estimates ω by first estimating PAF. Using the logic of the Mokdad procedure, we develop a new, modified Mokdad procedure to estimate ω as follows: From equation 10, and using notations in Table 4, Let f i = fraction of population exposed to exposure category i R i = relative risk for exposure category i compared to non exposed It follows that for a polytomous exposure variable Let f 0 = fraction of population nonexposed = 1 -Σf i -f q f q = fraction of population underweight It follows that when a neutral category is involved It then follows that Equations 11 and 12 are new equations we created for exposure category-specific PAF and attributable number, respectively. Because the Mokdad procedure uses the PAF approach, the neutral category Q reappears in equations 11 and 12, because Q is part of the total population. P(Q) is easy to obtain from health surveys. Equation 12 is expected to yield identical results as those of equation 10 (Allison's), because it is derived from equation 10. Equation 12 is simpler to use as it needs only total deaths (M), fractions of exposure in each category ((P(O i ), P(R), P(Q)) which are readily available from health surveys and the relative risks (RR i , RR q ). Additionally, equation 10 is difficult to use because it also requires total population (N), probability of death in each of the exposure categories (P(D|O i )), and probability of death in the reference group (P(D|R)). The probabilities of death are difficult to obtain.
Difference in the estimated number of attributable deaths under the hazard ratio and the relative risk models
Relative risk (RR) is an estimate of hazard ratio (HR). HR is the ratio of hazard rates (instantaneous incidence rates) in the exposed to the nonexposed at a point in time [4]. RR is the ratio of average risks of disease or death in the exposed to the nonexposed over a period of time. where h is hazard ratio (HR) and λ is hazard rate in nonexposed.
In theory, as pointed out by Allison et al, RR estimates "without adjustment for time can bias results (though the bias may be small)" [1]. The question is, in practice, does it matter whether RR or HR is used? Table 5 is a theoretical comparison of HR and RR using equation 13, based on hazard rates of 0.01 and 0.10, and HR of 1, 3, 5, 7. When hazard rate is low (e.g., 0.10 or below), HR and RR are close to each other. The lower the hazard rate (e.g., 0.01), the closer together the HR and RR. From the real data for Alameda County Health Study provided by Allison et al, when HR was 1.39, the RR was 1.38766116; when HR was 0.98, the RR was 0.98008466 (Table 5). There is no practical difference in the real setting in the estimated number of attributable deaths under the HR and RR models.
Options for numerically solving an equation for the hazard of death in the nonexposed (l)
There are commercial packages available for solving an equation for an unknown quantity; packages such as MATHEMATICA and MAPLE [5]. However, for these packages there is a steep learning curve for beginners, and packages can be quite expensive [6]. We looked into two simpler non-commercial options which one can easily program at no cost.
The first option is Newton's method (Additional file 1, Appendix S2). Applying Newton's method to the Alameda County Health Study data, provided in Allison et al's table three, gave an estimated λ of 0.008651. The second option is Taylor series (Additional file 1, Appendix S3). Applying Taylor series to the same data gave the same estimated λ of 0.008651. The two options gave virtually the same answers, with an error margin of less than 0.000001.
Comparison of the Allison procedure and modified Mokdad procedure with real datasets
We used the real dataset from the Alameda County Health Study provided by Allison et al [1] to compare the results using the Allison procedure and the modified Mokdad procedure, under both the hazard ratio (HR) and the relative risk (RR) models (Additional file 1, Appendix S4).
From our Additional file 1, Appendix S4, it can be seen that the results using the Allison procedure and the modified Mokdad procedure, under the hazard ratio (HR) and the relative risk (RR) models, are very similar to each other. The Allison procedure is a HR approach and the Mokdad procedure is a RR approach. Therefore the results of the Mokdad procedure using RR are closer to the Allison procedure than the Mokdad procedure using HR. However, the Mokdad procedure using HR to approximate RR provides good enough estimates of attributable number of deaths, and it avoids the use of equation 13 which involves estimation of RR that involves complex estimation of λ.
Discussion
The procedures recommended by Allison et al [1] and Mokdad et al [2] can both be applied to estimate the number, as well as fraction, of a single outcome (such as death) attributable to a risk factor (such as increased body mass index, BMI) that is polytomous (e.g., overweight, obese, and even stratified by BMI unit).
Although not specifically mentioned in the two original articles [1,2], both procedures can be applied to one or more risk factor combinations (such as BMI and smoking) as long as the risk factor combinations are expressed in independent (i.e., nonoverlapping) exposure categories. Furthermore, both procedures are under the assumptions of no confounding and no effect modification by the risk factors of interest and other covariates (such as age or sex).
The Allison procedure can be applied to the situation when there is a nonexposed category, one or more exposure categories, and one or more neutral (neither nonexposed nor exposed) categories. Allowance of a neutral exposure category is a benefit of the Allison procedure from a causal inference perspective, because in reality the population cannot always be dichotomized into nonexposed and exposed. The Mokdad procedure cannot allow for a neutral category. This paper proposes a modified Mokdad procedure that can achieve the same results as the Allison procedure, but through a simpler way.
The Allison procedure involves twelve steps, while the Mokdad procedure involves only six steps (Additional file 1, Appendix S1). The reason why the Allison procedure involves more steps is because it attempts to directly estimate the attributable number of deaths (equation A1), and this necessitates the estimation of the probabilities of death in the nonexposed, various exposure and the neutral categories. This in turn necessitates the calculation of the hazard rate in the nonexposed, λ, which requires substantial mathematical skills. The Mokdad procedure, on the other hand, first calculates the population attributable fraction (equation A3), and then obtain the attributable number of deaths by multiplying the PAF with the total number of deaths (equation A2). Our paper (equations [1][2][3][4] shows that in the derivation of the equation for PAF in the Mokdad procedure, the probabilities of death cancel out each other in the numerator and denominator, leaving only fractions of exposure and relative risks as necessary input parameters for the estimation of PAF. This greatly simplifies the calculation process in the Mokdad procedure. The Mokdad procedure, however, breaks down if a neutral category (such as underweight) that is neither nonexposed (such as normal weight) nor exposed (such as overweight and obese) exists. Also, while the Mokdad procedure can calculate the overall number of deaths attributable to a risk factor with multiple exposure categories, it does not calculate the number attributable to each individual exposure category. The Allison procedure, on the other hand, can estimate the individual exposure category attributable numbers (although the exact equation was not given in Allison et al [1].) By combining the innovation of the Allison procedure [1] (i.e., allowing for a neutral category which is neither nonexposed nor exposed) and the simplicity of the Mokdad procedure [2] (i.e., calculating attributable numbers indirectly through population attributable fractions), this paper proposes a new procedure (modified Mokdad) to calculate population attributable fractions and numbers (Table 6). This paper also "recovers" the missing equation in Allison et al that calculates the individual exposure category attributable numbers (equation 10), and develops a similar equation using the Mokdad Table 5 Comparison of hazard ratio (HR) and relative risk (RR) using the notations of Allison et al [1] Hazard rate in nonexposed [3], from a two-category risk factor (nonexposed, exposed) (equation 1) to a three-category risk factor (nonexposed, exposed, and neutral) (equation 8).
Both our proposed procedure (Table 6) and the Allison procedure allow for one or more neutral categories (such as underweight). The numbers of death associated with the neutral categories are excluded from the calculation of the number of death attributable to the risk factor under study. The Allison procedure involves twelve steps while the proposed procedure involves only eight steps. The proposed procedure, using the logic of the Mokdad procedure, does not require calculation of the probabilities of death in the various categories. Therefore no solving for the hazard λ is required. The proposed procedure is expected to produce similar results as the more complex Allison procedure. Slight discrepancies in the results, as shown in the real examples provided in this paper, are due to rounding errors in the additional steps in the Allison procedure to estimate probabilities of death in various nonexposed, exposure and neutral categories, and solving for λ, the hazard in the nonexposed, all of which are not required in the proposed procedure. Discrepancies will also occur depending on whether relative risks or hazard ratios are used, but this is expected to be small when the event (e.g., death) is rare (section 4). If one insists to use the Allison procedure instead of the proposed procedure, this paper discusses a number of options for solving for λ which could be helpful (section 5). Table 6 Eight steps in our proposed new procedure (modified Mokdad) to calculate number of deaths attributable to a risk factor with multiple exposure categories, allowing for one or more neutral categories Let ω be the number of deaths attributable to a risk factor (e.g. overweight and obese); and ω i be the number of deaths attributable to a specific exposure category i (e.g. overweight) of the risk factor.
Using the notations in Table 2 | 4,632.6 | 2010-08-31T00:00:00.000 | [
"Mathematics"
] |
Hierarchical System Decomposition Using Genetic Algorithm for Future Sustainable Computing
: A Hierarchical Subsystem Decomposition (HSD) is of great help in understanding large-scale software systems from the software architecture level. However, due to the lack of software architecture management, HSD documentations are often outdated, or they disappear in the course of repeated changes of a software system. Thus, in this paper, we propose a new approach for recovering HSD according to the intended design criteria based on a genetic algorithm to find an optimal solution. Experiments are performed to evaluate the proposed approach using two open source software systems with the 14 fitness functions of the genetic algorithm (GA). The HSDs recovered by our approach have di ff erent structural characteristics according to objectives. In the analysis on our GA operators, crossover contributes to a relatively large improvement in the early phase of a search. Mutation renders small-scale improvement in the whole search. Our GA is compared with a Hill-Climbing algorithm (HC) implemented by our GA operators. Although it is still in the primitive stage, our GA leads to higher-quality HSDs than HC. The experimental results indicate that the proposed approach delivers better performance than the existing approach.
Introduction
Software tends to improve continuously by making changes repeatedly to meet ever-changing requirements and business conditions rapidly. Although these changes positively contribute to software maintenance or evolution, they can also cause software erosions that constantly decay the internal structure of a software system, violating the software architecture principles, dropping system performance, or shortening the useful lifetime of the system. Such problems occur repeatedly while performing a software project and are constantly dealt with until the end of its lifetime to meet rapidly changing requirements. Thus, a method that can easily understand and maintain software architecture to avoid software erosions is in demand all the time.
Software architecture provides a high-level view of a software system. The software architecture of a system makes it easy for developers to understand the system during its development and maintenance [1][2][3]. As one of the most important and frequently used views of software architecture [1,4], the module view plays a significant role in understanding a system, especially its static structure. In general, it is built by breaking down the system in terms of the functionality of the system into subfunctionalities of subsystems until the latter are small enough to manage and understand [3]. Thus, naturally, the module view is presented as a Hierarchical Subsystem Decomposition (HSD) [4][5][6]. The hierarchical representation is especially useful for understanding large-scale software systems [5,7]. Despite having great advantage in understanding a software system, HSD is difficult to manage well enough. For example, a modification of a system could not be properly reflected to the HSD architecture due to lack of time and resources. This causes inconsistency and inaccuracy between the HSD architecture and the actual architecture of the software system. Even if the HSD architecture is properly updated according to the modification of a system, it could gradually deteriorate during the continuous modification. These cases fail to lead to understanding of a system in a detailed level by software developers; hence the need to recover HSD from software artifacts such as source codes or design specifications.
The recovery aims to increase understanding of a software systems structure by finding an HSD with the quality properties required by experts. The "objectives" of recovery may be domain-specific or system-specific. They may also vary according to the purpose or circumstances of recovery. For example, when a legacy system may have its source code only, the quality structural properties of an HSD could be the objectives of recovery. If the system has a well-managed change history, the history could provide an objective to minimize the change impact among subsystems.
There are several approaches that can be applied to recover an HSD of a system [5,[8][9][10]. The approaches represent the given objectives for their criteria. Likewise, their recovery techniques recover a quality HSD in terms of the criteria. Nonetheless, the existing approaches have a limitation in representing the objectives of recovery for their criteria. The approaches use specific criteria embedded in their techniques. That makes the approaches able to adopt only specific kinds as well as a limited number of objectives that are compatible with their criteria. Even if they are compatible, it often means significant work for the users of the approaches to represent the objectives for the specific criteria, especially when the objectives are many and complex. Thus, a minor change of objectives or the introduction of new objectives could impose a heavy burden on the users. As such, a new approach that is flexible in adapting objectives to its criteria is needed.
In this paper, we propose an approach for recovering the HSD of a software system that is based on a genetic algorithm that has been applied to find optimal solutions in many large and complex problems [11][12][13][14][15][16]. In order to apply the genetic algorithm for recovering HSD, we propose a chromosome expression to present HSD architecture and also design crossover and mutation operators to create new HSDs with good traits from the HSDs in a population. In our experiment, we validate the proposed approach based on two open source software systems using the fitness functions with various weighting values to evaluate the HSD quality. The experimental results are compared and analyzed with the existing Hill-Climbing algorithm [12].
A broad spectrum of programs, algorithms, policies [17,18], and processing methods associated with information technology that helps to create a better world for mankind are essential parts in Future Sustainability Computing (FSC), which aims to deal with the potential issues concerning the sustainability of current computing techniques as well as information technologies and their environments [19][20][21]. To better understand and provide a possible solution for FSC, this study discusses hierarchical system decomposition utilizing a genetic algorithm while considering various types of information processing technologies and computational frameworks used for the cloud/cluster/mobile computing including optimization, machine learning, prediction, and meta-heuristics in addition to decision support systems and system security and stability, etc.
The rest of this paper is organized as follows: Section 2 introduces a general genetic algorithm used in this work and several metrics used for defining the fitness functions of a genetic algorithm in this study; Section 3 proposes an approach to recovering HSD using a genetic algorithm; Section 4 presents the evaluation of our approach; Section 5 describes the existing studies for recovering subsystem decomposition; Section 6 discusses threats to validity; and Section 7 concludes this paper with a summary as well as future research directions.
Genetic Algorithm (GA)
As one of most well-known global search algorithms [22,23], genetic algorithm (GA) imitates the survival and reproduction of the fittest individuals in nature. As shown in Figure 1, a typical existing GA was rewritten to be able to take a new form that is appropriate for the HSD. In this GA, there are four parameters: the number of chromosomes in the population (npop); the probability of crossover (p c ), the probability of mutation (p m ), and the number of iterations for termination (nIter). A search by the GA starts from the initial population consisting of npop chromosomes created by an initialization algorithm. The search repeats nIter iterations of reproduction of the fittest individuals. In each iteration, a new population is created from the current population. To create a new population, new chromosomes are created by stochastically applying crossover and mutation to them according to p c and p m . Among two old and two new chromosomes, two fittest chromosomes are selected for the new population. After a new population has been selected, the algorithm runs repeatedly until nIter c eventually becomes smaller than nIter t to find the optimized chromosome.
Metrics for Fitness Functions
Coupling, cohesion, and complexity are three commonly used structural properties to assess the quality of modules (e.g., functions, classes, and subsystems) and software systems [24][25][26][27][28]. In this study, we also use them to evaluate subsystems and define fitness functions for evaluating HSDs. Unfortunately, there is no measure that can evaluate HSDs as they are. Nonetheless, there are several complexity, cohesion, and complexity measures that can be adopted for evaluating subsystems in the existing literature. We adopt the measures and define quality measures for HSDs.
In this section, we define the notations used for describing the measures used in this paper. We then describe the complexity, cohesion, and coupling measures used for defining the fitness functions of GA.
Proposed Notation for HSD
Software system S consists of basic entities and dependencies between them, S = {BE, D}. BE is a set of basic entities of S. Edge d∈D represents the dependency of basic entities be 1 and be 2 , (be 1 , be 2 )∈D through the inheritances and references between them. In Figure 2, the system has nine basic entities and dependencies between them. For example, basic entity e2 has a dependency on basic entity e1 by inheriting e1 or referencing the data or functions of e1. The HSD of a system, S, has a hierarchical structure of {SB, R}. SB is a set of subsystems in S, including root(S) as the root of the HSD. In addition, sb in SB has a set of basic entities be(sb)⊂BE. In other words, the basic entities in be(sb) are grouped in sb, and they have grouping relationships with each other at the level of sb. In Figure 2, there are five subsystems, SB = R, S1, S2, S3, S4, and root(S) = R. Each subsystem contains several basic entities. For example, S1 and S2 have be(S1) = e3, e4, e5, e6, e7 and be(S2) = e3, e4. The basic entities in be(S1) have grouping relationships at the level of S1, and those in be(S2) have grouping relationships at the level of S2.
The relationship r∈R denotes that a subsystem is part of another subsystem called part-of relationships between subsystems. In the relationship, subsystem sb can have its parent subsystem, prt(sb)∈SB. In other words, there are part-of relationships between sb ∈SB and its children cld(sb) ⊂ SB by the edges of R between them. So, be(cld(sb)) ⊂ be(sb). In Figure 2, cld(S1) = {S2, S3}, prt(S2) = S1, and be(S2) ⊂ be(S1). Not only be(S2) ⊂ be(S1), but also S2 ⊂ R. Thus, a descendant of subsystem sb ∈ SB, dsc(sb), is part of sb. On the other hand, subsystem sb has ancestors ast(sb). For example, ast(S3) = {S1, R}.
According to part-of relationships, e∈BE belongs to one or more than one subsystem. So, basic entities can be grouped in more than one subsystem, with the part-of relationships representing the level of a subsystem wherein its basic entities have grouping relationships. For example, in Figure 2, e3 and e4 are grouped together in S2. e3 and e5 are grouped in the higher level, S1. Therefore, e3 and e4 have a more homogeneous or closer grouping relationship than that of e3 and e5.
In be(sb), there are several basic entities that do not belong to cld(sb), and the basic entities belong to sb directly, be dir (sb) ⊂ be(sb). In Figure 2, be dir (R) = {e1, e2}.
Complexity
A complex subsystem makes it hard for a developer to understand the subsystem. A subsystem consists of basic entities and its child subsystems; one should understand them to understand the subsystem. If there are too many constituents, the subsystem might be hard to understand. As shown in Equation (1), the complexity of a given subsystem sb, Cpx(sb), is defined by the number of its constituents, which is one of the most frequently used concepts of complexity measures [24,25].
The larger the value of Cpx(sb), the more basic entities and child subsystems there are. So, a high value of Cpx(sb) has a negative effect on an HSD.
The complexity of a given HSD, S, is computed by taking the maximum value among the complexity values of all subsystems, as presented in Equation (2).
If the complexity of an HSD is too high, there is at least a very large subsystem. Likewise, the HSD should be restructured by breaking down large subsystems or moving some of its basic entities and child subsystems to proper locations.
Cohesion
A subsystem needs to consist of a cohesive set of basic entities. To measure the cohesiveness, we use a similarity measure that is used in hierarchical clustering [5,29,30]. The similarities between basic entities are measured by the similarities of their feature vectors. The features of a vector for a basic entity represent other basic entities on which the basic entity has dependencies by inheritance and reference relationships. Thus, the vectors of two basic entities are similar, and they are interacting with similar basic entities. That makes them cohesive.
In this work, we use binary vectors representing whether a basic entity has dependencies of features or not. To measure the similarity between two vectors, Jaccard coefficient [5] is used. For the given basic entities be 1 and be 2 , as shown in Equation (3), the Jaccard coefficient Jcr between them is given by: 0 ≤ Jcr(be 1 , be 2 ) = 1/2·a The value of a is the sum of the features present in both. The values of b and c represent the sum of features present in just one of two vectors. The larger the value of the Jaccard coefficient, the more similar the two vectors. Thus, Jcr(be 1 , be2) = 0 means that there are no features represented in both vectors. In addition, when they have the same vectors, it has 1.
As provided in Equation (4), the cohesion of subsystem sb∈SB is computed by averaging the similarity between the basic entities in the boundary of the subsystem.
A large value of Coh(sb) means that the target subsystem consists of similar basic entities, and that it is cohesive.
There are two versions of cohesion for HSD S, as shown in Equations (5) and (6). The first cohesion is given by the sum of cohesion of all subsystems in S.
The second version is given by: ρ is between 0 and 1. It can be used to give weights to cohesion values of subsystems. In our study, as presented in Equation (7), ρ is given by: ρ makes the cohesion of a higher-level subsystem in the hierarchy more important than that in a lower level. It also has a recursive form to make the cohesion of a subsystem affected by its ancestors. So, even if a subsystem has high cohesion without ρ, the low cohesion values of its ancestors make it lose up to half of its cohesion by ρ.
Coupling
A subsystem of a system that is loosely coupled with other parts of the system is more manageable and understandable according to the SE principles [31]. In many previous studies [24][25][26][27][28], the coupling of a module is measured by counting the number of modules interacting with it. We define the coupling measure for a subsystem in a similar point of view.
The coupling of subsystem sb, Cpl(sb), is caused by the external dependencies of its basic entities on other basic entities or subsystems "outside". In our coupling measure, the boundary of the outside of sb is defined as the boundary of prt(sb). Thus, as shown in Equation (8), the basic entities and subsystem counted as the targets of external dependencies of sb are given by: Ex be (sb) = be dir (prt(sb)) Ex sb (sb) = cld(prt(sb)) − {sb} (8) This is to reflect the "top-down decomposition" of an HSD. A hierarchy of subsystems is built by decomposing a system into subsystems and the subsystems into sub-subsystems. Thus, a subsystem that is loosely coupled with other subsystems should be decomposed to make its offspring loosely coupled with each other, and the decomposition need not concern the coupling of the outside.
Cpl(sb) is computed by the ratio of the number of external dependencies on basic entities and subsystems in the outside of sb, ExDep be (sb) and ExDep sb (sb), to the number of dependencies of basic entities in sb on the basic entities in prt(sb), Dep(sb), ExDep be (sb), ExDep sb (sb), and Dep(sb) are given by Equation (9): Dep(sb) = (be 1 , be 2 ) ∈ D be 1 ∈ be(sb), be 2 ∈ be(prt(sb))[be 1 be 2 ] As provided in Equation (10), Cpl(sb) is given by: A low value of Cpl(sb) means that sb is loosely coupled with other parts and is decomposed well in view of coupling. As shown in Equation (11), the coupling of a given HSD S is computed by aggregating the coupling values of the subsystems in S.
Cpl(S) uses the same weighting schemes used in Coh 2 (S). In contrast to Coh 2 (S), low Cpl(S) means high quality of HSD S.
In this way, a new concept of the metrics for individual metrics (i.e., complexity, cohesion, and coupling) essential for the GA that will be applied to HSD has been presented clearly by mathematically redefining them in terms of software architecture, allowing automated HSD. The actual form of utilization of such concept and corresponding notations to perform HSD is described in detail in the following section.
Recovering the Hierarchical Subsystem Decomposition Using GA
To achieve automated HSD recovery that is flexible to the criteria, we propose an approach to recovering HSD using GA. In Section 3.1, we describe how to represent and initialize a chromosome. In Section 3.2, we propose the GA operators (crossover and mutation) with fitness functions used in this study.
Chromosome Representation and Initialization Algorithm
We define a representation of a chromosome that encodes the characteristics of an HSD. The two most important characteristics of HSD are grouping relationships between basic entities and part-of relationships between subsystems. It is because a good subsystem is designed to have basic entities and subsystems that are homogeneous to each other and heterogeneous with the outside of the subsystem to achieve modularity [31,32].
We propose an initialization algorithm that aims to generate a chromosome of an HSD "arbitrarily." The initialization algorithm consists of two phases. In the first phase, it creates subsystems having only basic entities. It is accomplished by selecting an arbitrary number of basic entities and allocating them in a new subsystem. In the second phase, part-of relationships between the subsystems are assigned until only one subsystem without its parent is left. The subsystem becomes the root of a created HSD by the initialization algorithm.
With this initial algorithm, a chromosome for HSD is generated, and the calculation through crossover and mutation will be performed to find the optimal decomposition structure with the fitness function. To achieve this, the crossover, mutation, and fitness functions have to be redefined to perform HSD. The definitions and corresponding execution processes are described in the following section.
Crossover
The method with its execution flow of crossover is proposed based on the concept and notations defined for HSD in Section 2, representing in a pseudocode form to present them clearly.
A crossover operator for HSD recovery should be able to create new chromosomes that inherit good grouping and part-of relationships of their parents. Thus, the crossover in our approach aims to remove the chromosome subsystems causing poor grouping and part-of relationships and rebuild it based on the good groupings of another chromosome. Figure 4 presents our crossover algorithm. It consists of two sub-algorithms: selection and regrouping. First, selection is applied to two given chromosomes of a system for crossover. For each chromosome, several supergenes of subsystems are selected arbitrarily and removed to leave parts of it. Figure 5 presents the pseudocode of the selection algorithm. An example of selection for a chromosome is shown in Figure 6, and subsystem S1 is selected for removal from the HSD. Next, regrouping regroups each chromosome's basic entities that lose their grouping and part-of relationships based on those of the other chromosomes. After the regrouping of two given chromosomes, two new chromosomes inheriting the grouping and part-of relationships of their parents are created. Figure 7 shows the pseudocode of the grouping algorithm. Figures 8 and 9 present an example of regrouping in the HSD and chromosome versions. The example shows the status of the target HSDs and chromosomes at three points of Figure 7. The first point shows the status of two given chromosomes before regrouping progresses. At point 2, part of a chromosome is created by removing the basic entities of a dirty subsystem. In addition, point 3 shows the result of regrouping after replacing the basic entities of a dirty subsystem of a chromosome with part of the other chromosome created at point 2.
Mutation
The method with its execution flow of mutation is proposed based on the concept and notations defined for HSD in Section 2, which is represented in a figure to present them clearly.
We define three mutation operators that cause small change to HSDs: (1) moving an entity (Mae), (2) moving a subsystem (Mas), and (3) modularizing (Mdz). The Mae mutation operator moves basic entities from one subsystem into another subsystem. The Mas mutation operator moves a subsystem into another subsystem. The Mdz mutation operator creates a new child of a subsystem by grouping basic entities belonging to the subsystem directly. Figure 10 shows examples of the mutation operators. When a mutation operator is applied to an HSD during search by the GA in our approach, an operator is selected stochastically based on the number of feasible cases.
Fitness Function
The fitness function is proposed by using the notations for cohesion and coupling for HSD introduced in Section 2.
We define two fitness functions for experimental purposes: FF 1 and FF 2 . They are on the structural properties, and their metrics are introduced in Section 2.2. The first fitness function, FF 1 , is given by Equation (12): The higher FF 1 is, the higher the quality of a given HSD. We use FF 1 , which is designed to produce HSDs having similar structures with dendrograms produced by hierarchical clustering algorithms for architecture recovery [5,8]. According to the formula of Coh 1 , FF 1 is high when an HSD has many subsystems that are highly cohesive. Since a subsystem has a small number of basic entities, it commonly becomes highly cohesive.
As shown in Equation (13), FF 2 is the aggregation of three structural properties: complexity, cohesion, and coupling.
This fitness function is designed to search HSDs with low complexity, high cohesion, and low coupling. A high value of FF 2 means that HSD has high quality. The weight variables w 1 , w 2 , and w 3 can be used to emphasize some properties among them or balance them. The change of weight values would change the objectives of a search. Therefore, a recovered HSD would have different structural characteristics according to the weight values.
If complexity is emphasized more than the others, the size of a subsystem would be reduced by decreasing the number of its basic entities by moving some basic entities to other subsystems or creating new subsystems consisting of such. If there are too many offspring in a subsystem, the offspring would also be reduced.
More weight on cohesion would lead to a rich hierarchy, similar to an HSD by FF 1 . Unlike Coh 1 of FF 1 , Coh 2 with ρ gives more weight to higher-level subsystems of a hierarchy. That makes FF 2 with more weight on cohesion prefer wide hierarchies than deep ones because the Coh 2 of low-level subsystems is highly likely to be reduced by ρ.
If coupling is emphasized, an HSD would have a relatively small number of subsystems. This is because a large subsystem can encapsulate a large number of interactions between basic entities. On the other hand, the number of offspring of a subsystem is reduced because a large number of offspring can cause a large number of interactions between them as well as high coupling. Cpl with ρ prefers deep hierarchies for the same reason as the Coh 2 mentioned above.
Evaluation
In this section, we evaluate our approach to show how well it performs in finding an HSD intended by the given criteria with limited cost. First, we introduce our tool, HireGA (Hierarchical subsystem decomposition recovery by GA), which implements our approach. Then, we present the overview of the evaluation process, and the result of evaluation is finally described.
HireGA Tool
HireGA supports our approach to recovering HSDs of a system written in Java. Figure 11 depicts the overview of HireGA. The tool takes the source code of a subject system and a configuration as input. The configuration consists of fitness function and parameters related to a GA setting. The parameters consist of npop, p c , p m , and nIter, which are introduced in Section 2.1. HireGA consists of four parts: static analyzer, GA controller, data collector, and decoder. The static analyzer conducts static analysis on the source code to figure out the basic entities and dependencies among them. The GA controller creates the initial population based on the results of analysis by the static analyzer and explores the search space in order to find an optimal HSD. The decoder decodes the chromosome detected by the GA controller into an HSD. The given fitness function evaluates chromosomes (i.e., HSDs) during the search. The data collector outputs an operational profile on crossover and mutation describing how much and how frequently they give improvement to chromosomes. The profile consists of two kinds of data: success rates and improvement degrees. They are given by Equations (14) and (15)
Overview of the Evaluation Process
In Figure 12, we present the overview of our evaluation process, which consists of 4 parts: data generation, analysis of the structural characteristics of search results, analysis of operational profile, and performance evaluation.
In the data generation part, we replicate N times HSD recoveries for a subject system and a configuration. The two subsystems used in this evaluation, JMSN and DNSJava, are presented in Table 1 together with the number of basic entities and their description. For each subject system, we use 98 configurations, 7 fitness functions, and 14 GA settings, as presented in Tables 2 and 3. As a result, N detected HSDs, and their operational profiles are produced for each subsystem and configuration. The fitness functions in Table 2 have different objectives from each other. For example, fitness functions 5, 6, and 7 in Table 2 target loosely coupled subsystems; the others, fitness functions 1, 2, 3, and 4, target highly cohesive ones. In this experiment, we run HireGA on a limited cost. We regard the number of fitness computations as the cost of a search because the computations take the most time in a search. As shown in Table 3, the limitation is achieved by keeping nPop · nIter = 10,000 for all GA settings. In part 2, we evaluate the ability of our approach to detect an HSD that reflects the objectives of a given fitness function. If our approach has such ability, it will detect HSDs having different structural characteristics led by different fitness functions. We analyze and compare the structural characteristics of the detected HSDs for each subject system and its 7 fitness functions in part 1. As the target configuration of analysis, the best-performing GA setting is selected for each fitness function among the 14 settings in Table 3. The performance of searches using a given configuration is determined by the average fitness values of N detected HSDs for the configuration. The N detected HSDs for the selected 7 configurations are analyzed by the 12 analysis metrics in Table 4. The first 5 metrics are structural metrics that show the structure of a given HSD, such as the number of subsystems. The remaining 7 metrics are quality metrics that show the complexity, cohesion, and coupling of a given HSD in various ways. Table 4. Analysis metrics to analyze the structural characteristics of a detected HSD.
Metrics Description
An sb The number of subsystem in a given HSD In part 3, the crossover operator in our approach is evaluated. The crossover operator is the most significant operator contributing to the performance of a GA. Thus, the efficiency of crossover in search is important to guarantee the performance of our approach. To evaluate our crossover operator, we analyze operational profiles to figure out the contribution of our crossover and mutation operators to the search ability of our approach. We select 7 configurations for a subject system: 7 fitness functions in Table 2 and a GA setting for each fitness function. For the 7 configurations, we analyze the operational profiles collected in part 1. This analysis would also give clues to decide which values for a GA setting, nPop, p c , p m , and nIter, are more suitable for our approach.
In part 4, we evaluate the search ability of our approach by comparing it with that of an approach using a Hill-Climbing algorithm (HC) implemented by our mutation operators. For each fitness function in Table 2, we produce N detected HSDs by the HC. Then, the search results are compared with the search results obtained by our approach in part 1. If our GA has better performance than the HC, we can conclude that the search concept of a GA, i.e., exploring population by crossover and mutation, is more efficient than an HC's search concept of exploring from random solution(s) only with mutation.
Analysis Results on the Structural Characteristics
The results of searches for the fitness functions in Table 2 are shown in Figures 13 and 14. In Figures 13 and 14, 1 to 14 of the x-axis represent the IDs of the GA settings in Table 3. In addition, point 15 of the x-axis represents the ID of an HC. The y-axis represents the fitness values measured by fitness functions denoted on the top of each chart. The average of fitness values for each GA setting is indicated by a horizontal bar. The GA settings that give the highest average fitness for each fitness function are presented in Tables 5 and 6. For the GA settings, the measured values of analysis metrics in Table 4 are presented in Tables 7 and 8 for JMSN and in Tables 9 and 10 for DNSJava. Figure 13. Fitness values of the detected HSDs by GAs in Table 3 for JMSN. Table 3 for DNSJava. Table 8. Values of quality metrics of GAs in Table 5 Table 9. Values of structural metrics of GAs in Table 6 for DNSJava. Table 10. Values of quality metrics of GAs in Table 6 for DNSJava. The GAs in Tables 5 and 6 are grouped into three in order to analyze the influence of changes of fitness functions on the structural characteristics of the detected HSDs. A change in the fitness function used means a change in the objectives of a search. The grouping allows us to fix some objectives and analyze the influence of the changes of other objectives. The three groups are as follows: GAs having different cohesion and coupling (GA 1 , GA 3 , GA 6 ), GAs having different complexity with high cohesion (GA 2 , GA 3 , GA 4 ), and GAs having different complexity with low coupling (GA 5 , GA 6 , GA 7 ).
GAs Having Different Cohesion and Coupling
In Tables 5 and 6, GA 1 uses only the cohesion measure as fitness function FF 1 . GA 3 and GA 6 use fitness function FF 2 , which aggregates complexity, cohesion, and coupling. Cohesion is more emphasized in GA 3 , and coupling is more emphasized in GA 6 . So, the degree of emphasizing coupling increases from setting GA 1 to GA 6 .
The three values of each analysis metric for GA 1 , GA 3 , and GA 6 are normalized by the largest value among the three values. For example, in Figure 15b, the max value of Cpx is 0.3 of GA 1 , and the normalized value of Cpx of GA 6 is (0.229/ 0.3) ≈ 0.763. The normalized values for GA 1 , GA3, and GA6 are presented in Figure 15a,b. This way of normalization is consistently used throughout our evaluation. Table 5 for JMSN.
In Figure 15a, the HSDs by GA 1 have relatively wide and rich hierarchies. As the coupling is emphasized, the number of subsystems and their offspring decreases. Such change coincides with the characteristics of fitness functions described in Section 3.2.3.
In Figure 15b, most of the cohesion metrics decrease, and most of the coupling metrics increase through GA 1 to GA 6 . This coincides with the change of their fitness functions. GA 2 has the lowest Cpx because complexity and cohesion are emphasized together. Cpx ave tends toward high values, which is probably due to the increase in the number of basic entities belonging to a subsystem directly by emphasizing coupling. The increase overwhelms the decrease of the number of offspring, and that causes the increase of An cst and Cpx ave .
The analysis on DNSJava shows a very similar result to the result of JMSN. The measured structural and quality metrics of GA1, GA3, and GA6 for DNSJava are presented in Figure 16a,b. Table 6 for DNSJava.
GAs Having Different Complexity with High Cohesion
The measured structural and quality metrics of GA 2 , GA 3 , and GA 4 for JMSN are presented in Figure 15c,d.
In Figure 15c, HSDs with more subsystems are detected as complexity is weighted. As w 1 increases from 0 to 2, cohesion also increases by sacrificing complexity a bit in Figure 15d. Using complexity as part of a fitness function helps detect more cohesive HSDs than the ones detected by GA 2 . According to the definition of complexity, it forces the GA to explore HSDs with relatively rich subsystems having a small number of basic entities. So, the number of subsystems increases by making wide hierarchies. Such change can increase cohesion according to Section 3.2.3. On the other hand, coupling increases for the same reason.
When w 1 increases from 2 to 6, the deep hierarchies become more preferable. A large number of offspring in subsystems cause high complexity and high coupling. Thus, the number of offspring is reduced by making deep hierarchies, and this in turn lowers not only complexity but also coupling and cohesion. The change coincides with the characteristics of fitness functions and the change of fitness functions.
The measured structural and quality metrics of GA 2 , GA 3 , and GA 4 for DNSJava are presented in Figure 16c,d. According to Figure 16c, the decrease of complexity is accomplished by making the detected HSDs deep and rich. The number of offspring decreases by making deep hierarchy, and that can cause a decrease of coupling and cohesion as mentioned above.
GAs Having Different Complexity with Low Coupling
The measured structural and quality metrics of GA 5 , GA 6 , and GA 7 for JMSN are presented in Figure 15e,f. Since coupling is emphasized, search without weight on complexity detects the HSDs having a relatively small number of subsystems. More weight on complexity, w 1 = 2, makes the number of subsystems grow and the size of the subsystems shrink. Such causes high coupling according to the definition of Cpl. If coupling and cohesion are considered together, HSDs having subsystems with low coupling would be created instead of subsystems with high cohesion. So, complexity is decreased by not only increasing the number of subsystems but also making the hierarchy deeper to increase coupling as minimally as possible, but at the expense of cohesion. As a result, both cohesion and coupling deteriorate.
In the change of w 1 from 2 to 6, complexity is decreased by increasing the number of subsystems in the detected HSDs. Unlike the change of w 1 from 1 to 2, the depth and width of the hierarchies show little change. The increase in the number of subsystems causes an increase of coupling and cohesion. Nonetheless, cohesion shows little change because the cohesion values of the subsystems might be very low by emphasizing coupling.
The measured structural and quality metrics of GA 5 , GA 6 , and GA 7 for DNSJava are presented in Figure 16e,f. The analysis on DNSJava shows very similar results to the result of JMSN, but the change of analysis metrics for DNSJava is less drastic. In our manual inspection of the resulting HSDs of GA 5 , they commonly have a large subsystem contributing to coupling. So, the increase of w 1 decreases Cpx a lot by reducing the number of basic entities in the large subsystems. Such causes relatively small changes to the rest of the HSDs and the analysis metrics.
As we have shown in the analyses on structural characteristics for JMSN and DNSJava, the search results are varied according to fitness functions. Therefore, we can conclude that our approach can reflect different objectives of the fitness functions and produce HSDs that reflect the difference of objectives.
Analysis Results on the Operational Profile of Mutation and Crossover
As shown in Figures 13 and 14, nPop = 50 and nIter = 200 always show better performance than nPop = 100 and nIter = 100, respectively. In most of the cases, the GA settings of p c = 0.5 and p m = 1.0 give the best performance. The performance of a GA is heavily dependent on its crossover and mutation operators. Likewise, the efficiencies of crossover and mutation contributing to finding a quality HSD affect the best-performing GA setting and vice versa. Therefore, in this analysis, we figure out the search efficiencies of crossover and mutation by analyzing the operational profiles of the detected HSDs by several GA settings. The analysis also leads us to the best GA settings based on the analysis result.
As the targets of the analysis, we choose the GA settings of p c = 0.5 and p m = 0.5 because they give equal chances of application to crossover and mutation. We only present the details of analysis on the results of recovery by our approach for DNSJava and the GA settings having nPop = 50 and nIter = 200. That is because the analysis on JMSN or the other GA settings gives a similar result to the presented analysis and in consideration of space.
Analysis on nPop and nIter
The average success rates for 10 runs with the fitness function in Table 2 are shown in Table 11. We divide a running of GA in 3 phases by dividing 200 iterations into 3 and record the success rates of crossover and mutation for each phase separately. The numbers on the top of Table 11 represent the IDs of the fitness functions in Table 2. In Figure 17a, we present charts of the normalized values of data in Table 11. In addition, Table 12 and Figure 17b show how much the fitness values are improved by applying operators. Table 11. Success rates of operators in nPop = 50 and nIter = 200 for fitness functions in Table 2 of DNSJava. Table 2 of DNSJava. In these charts, the success rates and improvement degrees of crossover are relatively high in phase 1 but decrease sharply through phases 2 and 3 for all cases. The mutation rates also decrease, but relatively gradually compared with crossover. Therefore, crossover plays the role of improving the quality of HSDs in the early phase of search. On the other hand, mutation plays the role of rendering relatively small changes during search.
Regardless of nPop = 50 or 100, the limited number of genetic traits in the population is "consumed" by crossover, and new chromosomes with "good" traits and high quality are created rapidly in the early phase of search as shown in Figure 18, which depicts the change of fitness of the fittest HSD of each iteration. After the early phase, the fitness increases relatively slowly, and mutation contributes to the increase with higher efficiency than crossover. Thus, if the limit of cost is not expanded, it is better to increase the chances of mutation improving the chromosomes by creating new good traits. In other words, a larger number of iterations show better performance.
Analysis on p c and p m
In the analysis in Section 4.4.1, the efficiency of crossover is considerably reduced in the early phase of search. In the GA of Figure 1, mutation is applied after crossover is applied or skipped. Thus, the mutation could be affected by the low efficiency of a GA in the late phase of search. Table 13 presents the success rates of combinations of operators for GAs whose setting is nPop = 50, p c = 0.5, p m = 0.5, and nIter = 200 for the fitness functions in Table 2. There are three combinations: crossover-only, mutation-only, and both. Crossover-only occurs when crossover is applied to an HSD but mutation is skipped. Mutation-only is the opposite. The combination "both" is a case wherein crossover and mutation are applied to an HSD. The normalized values of the data in Table 13 are drawn as a chart in Figure 19. Table 13. Success rates of the combination of operators in nPop = 50 and nIter = 200 for fitness functions in Table 2 of DNSJava. Table 13.
Combination of Operators
In Figure 19, the success rates applying both crossover and mutation decrease drastically as the search proceeds, along with those of crossover-only, but mutation-only maintains relatively high success rates. We can conclude that applying mutation after applying crossover is inefficient in the late phase of search because of the low success rate of crossover. Nonetheless, applying both crossover and mutation and crossover-only is relatively efficient in the early phase. In the analysis in Section 4.4.1, the efficiency of crossover is also relatively high in the early phase. So, it is important to set p c properly to balance its efficiency in the early phase and its inefficiency in the late phase of search.
In the same perspective, mutation can also offset the improvement by crossover with poor operation. Still, it can be easily inferred that the effect of failure of mutation is far smaller than the improvement degree by crossover in the early phase based on Figure 17b. In the late phase of search, mutation plays a refining role, as analyzed in Section 4.4.1. Therefore, it is better to keep p m high during search.
The values that we used for p c and p m are (0.0, 0.5, 1.0) and (0.1, 0.5, 1.0). Among the settings, 0.5 for p c and 1.0 for p m are the most suitable values according to our analysis.
Performance Evaluation by Comparing Our Approach with Search by HC
An HC starts its search from a point of a search space and terminates when it reaches the (local) optimal solution or a given termination condition is met. To compare with our GA fairly, the HC used in this evaluation is implemented by the mutation operators of our approach to exploring neighbors. Likewise, the number of evaluations of HSDs by fitness functions is given as a termination condition for the HC. The HC is (re)started at random positions when one run of the HC is terminated before the termination condition is met. We maintain nPop • nIter = 10,000 throughout all the GA settings. This restricts the cost for a run to 10,100 evaluations at most, 10,000 for a search, and 100 for initializing a population of 100 chromosomes. When the number of evaluations by the HC reaches 10,100 during the search, the search is finished. Then, the fittest HSD is taken as an optimal solution among the detected HSDs by multiple runs of the HC.
We apply the HC to JMSN and DNSJava with the fitness functions in Table 2. The search results for the systems are presented in Figures 13 and 14 together with the search results of our GA. The 15th datum in each chart represents a search result by the HC.
The performance of the HC in DNSJava is much worse than that in JMSN. As the size of a target system grows, the number of neighbors of an HSD also grows. So, the larger a system is, the higher the cost incurred by the HC to find a fitter HSD among neighbors than a given one. Such reduces the search space that can be covered by the HC at a given cost. In addition, the impact of change to an HSD by mutation in a large system is weaker than that in a small system. Therefore, the efficiency of mutation becomes lower as the size of a system grows, whereas the efficiency of our crossover becomes relatively higher.
During our experiment, we realized that HC cannot finish just one run of an HC in DNSJava. This means that an HC starts from only one point in the search space and searches as "deep" as possible. In contrast to this search strategy, an HC can (re)start multiple points to search "widely" but relatively "shallowly." Therefore, the higher performance of GA than HC is likely caused by poor search strategy as well as the efficiency of crossover.
Thus, we take the strategy with various numbers of restarts and run the HC with them. We choose seven numbers for restarts: 1, 2, 5, 18, 20, 50, and 100. Figure 20 presets the search results of GAs in Table 3. The data of 15 to 21 represent HCs with the seven restart numbers, as presented in Table 14.
Among the results of the HC with the fitness functions in Table 2, we present only the result of FF 2 (6, 1, 6) because it shows relatively high or similar performance among all results and relatively high efficiency of mutation during our evaluation. Table 3 and HCs with various restart numbers. Despite the changes of the restart number for the HC, the GA in our approach always shows better performance than the HC. We conclude that the search strategy of a GA is more adequate for recovering HSDs than HCs. Our crossover operator also contributes to the outcome. Especially, as a software system for recovery expands, HC is expected to require much more resources (i.e., time) to perform similarly to GA.
The architecture of an extensive software system can be often interpreted by employing a technique Hierarchical Subsystem Decomposition (HSD), but as it is possible that the contents obtained from such a task have not been managed properly, they would not reflect the current picture. Constant update or revision can cause many changes in the software so that recovering or regenerating HSD contents is not an easy task.
This paper proposes an approach that can work through the past contents and reproduce an optimal solution that suits the design criteria well. As the objectives of such a task can be domain/system-specific or depend on the environmental conditions, the proposed approach focuses on complying with each objective effectively by utilizing a genetic algorithm creating a fitness function to work through all across search areas to deliver an optimal HSD form. The approach was subjected to the tests for evaluation where two open-source software systems having a set of 14 fitness functions in the algorithm. The architectural characteristics of the HSD types recovered by using the proposed approach varied depending on the objective given, and the result showed that the crossover type was more effective during the early search stage and the mutation enhanced the performance a little throughout the entire search process. The performance of the genetic algorithm used for the approach was analyzed in comparison with the Hill-Climbing algorithm and produced a better result in quality-wise even considering that it was in the early stage of development. The same result was expected for the proposed approach when compared with other existing approaches.
Recovery Approaches for Subsystem Decomposition
There have been several studies published regarding the recovery of subsystem decomposition from the design or source code of a system.
Mitchell et al. [33] proposed a famous bunch tool for architecture recovery. The bunch tool transforms a system into a directed Module Dependency Graph (MDG) consisting of nodes representing basic entities and edges representing dependencies between entities. In addition, it decomposes MDG into several MDG partitions using a hill-climbing or simulated annealing algorithm. The bunch tool uses Modularization Quality (MQ) as a fitness function that represents the balance between the internal and external edges of MDG partitions. The MDG partitions present a flat subsystem decomposition that consists of disjoint partitions of basic entities and no part-of relationships between its subsystems.
Seng et al. [34] proposed an approach to improving subsystem decomposition. They adopted the grouping genetic algorithm (GGA) [35], which is a specialized genetic algorithm designed to solve a grouping problem with GA. They used the aggregation of five structural metrics as a fitness function: cohesion, coupling, complexity, cycles, and bottlenecks. GGA searches only the flat subsystem decomposition that fits the given fitness function. An issue with this approach, and other existing approaches, is that they cannot provide the advantages of hierarchy, which is useful for understanding a system, especially when it is large and complex [8].
Jeet et al. [36] proposed an approach that is based on a combination of a nature-inspired black-hole algorithm and a genetic algorithm, which is named the Genetic Black Hole Algorithm (GBH). The combination of evolutionary algorithms tested on various software systems and resulted in better solutions.
The studies detailed in [5,29,30] adopted hierarchical clustering algorithms, which apply a greedy strategy to cluster similar basic entities. They used hierarchical clustering algorithms to produce a dendrogram presenting similarities among basic entities. Dendrograms can be used to recover the hierarchical structure diagrams HSDs by clustering basic entities in various levels of similarity.
The authors in [9,10] proposed recovering subsystem decompositions using concept analysis. Concept analysis represents several attributes of basic entities, such as global variables in use and interacting basic entities, to build a concept lattice, which consists of partially ordered sets of concepts. Basic entities are clustered as subsystems at a proper level of the concept lattice. A concept lattice represents "part-of-relationships" between sets of basic entities sharing the same attributes.
The studies detailed in [5,9,10,29,30] are similar to our approach in that they can be used for recovering HSDs; however, they use technique-dependent forms of criteria to find high-quality HSDs. When there are criteria provided by an expert, the criteria should be transformed to specific forms suited to their techniques. This makes the approaches inflexible for recovering HSDs using various (different) forms of criteria, because the transformation is highly difficult to perform. This is especially true when complex criteria are given. In practice, hierarchical clustering algorithms require measures defining similarities between basic entities. Concept analysis is also needed to represent the relationships among basic entities as concepts. However, when the criteria are complex, such as the multi-modal criteria used in Seng et al.'s work [34], transforming the criteria to a similarity measure or to concepts is almost impossible to accomplish.
Although the existing studies are helpful in performing subsystem decomposition, they cannot provide the advantages of hierarchy that is crucial for understanding a system, especially when it is large and complex.
GA Approaches Based on Tree Representation in Other Domains
In Koza's work [37], GA is used to create a LISP (List Processor: high-level programming language) expression, which for example is a minimal-length-expression used to fulfill a given task. The author represented a LISP expression as a tree structure, which is referred to as a syntax tree, and uses the tree structure as the individual for the GA instead of a chromosome, and crossover and mutation operators that are specific to the domain are provided. The crossover takes two syntax trees as parents, and a subtree from each tree is inherited by their children. Subsequently, the parents exchange their subtrees by each removing a subtree and attaching the subtree to the other parent in place of the removed one. As an example of syntax trees (i.e., expressions) of LISP: T1: (OR (NOT D 1 )(ANDD 0 D 1 )) T2: (OR (OR D 1 (NOT D 0 ))(AND (NOT D 0 ))(NOT D 1 )) S1: (NOT D 1 ) S2: (AND (NOT D 0 ))(NOT D 1 )). T1 and T2 are given as the targets for crossover, and the subtrees (i.e., sub-expressions) in two syntax trees, S1 and S2, are selected for crossover. Crossover between T1 and T2 results in two new legal syntax trees: T1': (OR (AND (NOT D 0 ))(NOT D 1 )(AND D 0 D 1 )) T2': (OR (OR D 1 (NOT D 0 )) (NOT D 1 )).
In contrast, a mutation is performed to produce a small change in a syntax tree. There are three mutation operators and they are highly specific to the domain. For example, the editing operator simplifies an expression based on editing rules such as (AND X X) → X.
A tree structure is one of several feasible forms used to represent a hierarchical structure. An HSD can be represented by a tree structure: nodes for subsystems and basic entities, and edges representing where subsystems and basic entities belong. Therefore, the GA is the easiest alternative considered for use in recovering HSDs [38]. In R. Lutz's work [39], a complexity metric based on information theory was proposed, and a GA was applied to validate the metric without guarantee of the adaptability of the GA to the recovery of HSDs. To address this issue, crossover is adapted to resolve discrepancies between syntax trees and HSDs, and mutation operators suited to HSDs are applied [37].
However, the GA studied in [37,39] did not adapt well to the task of exploring HSDs because it does not consider the characteristics of each HSD, which should be considered to produce quality HSDs. First, the characteristics of an HSD are far different when using a syntax tree. The hierarchy of syntax trees represents the order of processing of sub-expressions. In an HSD, grouping relationships of basic entities (i.e., which entities are grouped together in which subsystem level, and part-of relationships among subsystems) are the most important characteristics. Second, in contrast with syntax trees, there are dependencies between the nodes of tree structures in HSDs. A node representing a basic entity should appear in a tree at least once, but only once. This means that there should be nodes representing all basic entities and that the grouping relationships between basic entities should be unique. Therefore, in the crossover for syntax trees, attaching a sub-tree to another syntax tree is equivalent to adding a sub-expression to an expression, and the result of the attachment is considered to be a legal expression. However, applying crossover to HSDs can produce illegal HSDs, which have duplicate and/or missing nodes representing basic entities. The gap between the GA and the characteristics of the HSD recovery problem requires a repair procedure after crossover [38].
The inconsistency between the GA proposed in [37,39] and the characteristics of the problem (recovering HSDs) causes a serious issue in that the actual effect of the crossover performed on HSDs is highly difficult to grasp [38]. This makes it challenging to apply the GA to HSD recovery in various circumstances. For example, when a very large-scale system requires architecture recovery, scalability can be a problem. In this case, other information, such as the change history of the system, can be used to reduce the search space by reducing the arbitrary elements in the GA and by guiding the operators to proceed in an indicated and more efficient search. In some cases, the system expert can modify several elements of GA, such as the initialization of the algorithm, and crossover and mutation operations, to fit them to a specific problem. Therefore, to apply GA in various circumstances, the GA should be developed by analyzing the characteristics of the HSD, and these characteristics should be reflected to the elements of the GA. This was one of the primary reasons why we began this study, which was to develop such a GA.
Threats to Validity
(Implementation of a GA) In several studies, various approaches to implement a GA are proposed [11,34,39,40]. Among all feasible implementations, we selected one of the simplest implementations (provided by [11]) to avoid distortions associated with using a more complex and advanced GA.
(Weights of FF2) During our experiment, we used more values for the weight variables of FF2 than we presented in the evaluation. Although we did not present the analysis on all weights because of space considerations, the conclusions for the omitted weights were the same as those reached in the analysis presented.
(Decrease of crossover efficiency in the late stages of a search) The decrease in efficiency results from two causes: a small population size or the inherent convergence problem of GAs [11]. The former can be resolved by increasing the population size, but this increases the cost of a search. Regarding the convergence problem, several techniques have been proposed to resolve it. For example, a GA can use a selection strategy to compose a new generation during the search, and this new generation is one of the most influential elements toward achieving GA convergence. Our approach can adapt other selection strategies, such as roulette wheel selection [11], instead of the greedy selection strategy that we actually used in this study.
(Implementation of HC) We use the simplest implementation of HC, the first-ascent HC, which moves the search to the first discovered neighbor with higher quality [40] for the same reason that we use a simple implementation of a GA. There could be other implementations of HC that provide higher performance in recovering an HSD. If there is such an implementation, it would be worth using, because HC is simpler to understand and implement than GA. A study on various implementations of an HC algorithm for HSD recovery is one of our planned future studies.
Conclusions
In this paper, we have proposed an approach to recovering an HSD, which provided an architecture-level understanding of a software system. A method of understanding the software system at the architecture level was provided in this study along with an HSD scheme useful for reducing the possibility of software erosions. When an HSD of a software system is recovered, the objectives of recovery are given by experts. The objectives of recovery may be domain-specific or system-specific. They may also vary according to the purpose and circumstances of recovery. Thus, an approach to recovering an HSD of a software system should be able to adopt various objectives of recovery. To achieve this, we have proposed a recovery approach using GA, which is able to adopt various objectives easily as a fitness function and find a quality HSD in a vast search space.
To apply a GA in our approach, we provided a chromosome representation that represents grouping relationships among basic entities and part-of relationships among subsystems as the characteristics of an HSD affecting its quality. The initialization algorithm was designed to create arbitrary HSDs for initializing the population of a GA. Crossover was proposed to create new HSDs inheriting the grouping and part-of relationships from existing HSDs in the population. Mutation operators were provided to create a new HSD by rendering a small change to the grouping and the part-of relationships of an HSD in the population. In addition, fitness functions were designed to have different objectives, which were presented by the structural properties of an HSD and metrics assessing the properties.
In our evaluation, we evaluated the ability of our approach to recover different HSDs according to different objectives of recovery. We applied our approach to two open source software systems with fitness functions having different objectives of recovery. Then, we analyzed the search result to figure out whether our approach could find different HSDs according to the various fitness functions. As a result, our approach recovered HSDs having different structural characteristics intended by the fitness functions and their objectives of recovery.
Afterward, we evaluated how much our crossover operator contributed to the search efficiency of our approach, since crossover is a key feature significantly influencing the performance of a GA. We analyzed the contributions of our crossover and mutation operators to the search efficiency of the GA. We also compared our approach with an approach using an HC. As a result, crossover provided great efficiency to a search by contributing to the early-phase search, whereas mutation played the role of refinement throughout the entire search. On the other hand, the comparison of our approach and an approach using an HC showed us that the GA was the better strategy for the recovery. According to the analysis and comparison, our crossover operator had a significant role during the search. The performance of our GA was attributed to the efficiency of our crossover operator.
Our approach can be applied to recover an HSD intended by various objectives given as fitness functions. The change history or design specification of a system could be used for the establishment of the objectives of recovery by experts. For example, the change history of a software system could be used for establishing an objective to recover an HSD with low change impact among subsystems. In different objectives, our approach may show different behavior and require different calibration. Therefore, we plan to repeat the evaluation of our approach in various objectives. hierarchical system decomposition (HSD) briefly and evaluating the performance of idea from a single perspective in terms of complexity, cohesion, and coupling measures. By contrast, this paper presents a specific notation for utilizing the genetic algorithm for HSD along with the concrete concepts for those three aspects followed by a new proposal for the crossover and mutation methods, which are considered to be the key to genetic algorithms, within the framework of HSD. In addition, for the performance evaluation, an additional target system was used in the experiment for verification compared to the conference proceeding paper in which only one target system was used. Further, the reliability of the verification method was enhanced by describing the changing process of fitness functions and quality metrics in detail. Meanwhile, the level of completeness and quality of the paper has been largely increased by supplementing the related study section and analyzing possible threats to validity.
Conflicts of Interest:
The authors declare no conflict of interest. | 14,578.2 | 2020-03-11T00:00:00.000 | [
"Computer Science",
"Engineering",
"Environmental Science"
] |
Offloading of diabetes‐related neuropathic foot ulcers at Swedish prosthetic and orthotic clinics
This study aimed to assess (1) the use of different offloading interventions in Sweden for the healing of diabetes‐related plantar neuropathic forefoot ulcers, (2) factors influencing the offloading intervention choice, and (3) the awareness of current gold standard offloading devices.
| INTRODUCTION
Diabetes-related foot ulcers (DFUs) affect 19%-34% of persons with diabetes during their lifetime 1 and precede 50% of lower-limb amputations in Sweden. 2 Excessive repetitive mechanical trauma, increased peak pressure, and shear stress on the feet in combination with peripheral neuropathy commonly cause DFUs on the plantar surface. 3 DFUs are associated with premature death and lead to a cascade of negative effects for the affected persons, with an impact on their quality of life and accumulated costs for the health care systems. [4][5][6][7][8] However, there is evidence that multidisciplinary treatment, including offloading interventions that redistribute plantar pressure, is effective in shortening healing time. 3 The International Working Group on the Diabetic Foot (IWGDF) 3 recommends that a total contact cast (TCC) and nonremovable knee-high walkers be used as the gold standard treatments for plantar neuropathic forefoot DFUs that are not affected by severe ischaemia and infection.
Removable knee-high offloading devices and ankle-high offloading devices are recommended as second-and third-choice treatments.
The fourth recommendation by IWGDF is that if none of the abovementioned devices are available or all of them are contraindicated, the use of felted foam in combination with appropriate footwear should be provided.
There is a gap between clinical practice when clinicians select offloading interventions for the treatment of DFUs and the IWGDF recommendations, which are based on scientific evidence. [9][10][11] Studies in Europe, the USA and Australia have shown that practitioners underutilise nonremovable knee-high devices, [9][10][11] although the majority of them agree that these devices are the gold standard treatment. 9,10 To our knowledge, no study has presented the type and frequencies of offloading interventions that practitioners provide to patients with plantar forefoot DFUs in Sweden, which may differ compared to other countries depending on differences in culture, reimbursement systems, etc. Furthermore, there is a lack of studies that investigate the factors that practitioners consider important when selecting offloading interventions to heal plantar forefoot DFUs.
| AIM
This study aimed to assess (1) the use of different offloading interventions in Sweden for healing plantar neuropathic forefoot DFUs, (2)
| Study design
The study was an observational cross-sectional study that collected answers from a quantitative survey.
| Questionnaire and pilot testing
A questionnaire developed and used by Raspovic and Landorf in an Australian study 9 was translated into Swedish with the authors' approval. Six out of 12 questions (questions 1 and 4-8) were selected from the original questionnaire. Question number 8, originally concerning whether practitioners considered nonremovable offloading to be the gold standard in offloading, was separated into two questions concerning whether TCC and a nonremovable kneehigh walker, respectively, were considered the gold standard in offloading. The original English questionnaire was translated into Swedish based on the recommendations of the World Health Organization. 13 First, the first authors (IG and EDS) translated the questionnaire from English to Swedish. An expert panel including a In question 1, the practitioners were asked to specify the percentage of patients for whom 14 predefined offloading interventions were used. The practitioners were free to add additional interventions or combinations of interventions. In questions 2-5, the practitioners were asked to rank on a 5-point Likert scale (from 0 = never to 4 = always) how often 28 predefined practitioner-, patient-, intervention-, and wound-related factors were considered in the provision of offloading interventions. In questions 6-7, the practitioners were asked if TCC or a nonremovable knee-high walker was considered the gold standard offloading device.
| Participants and procedure
An e-mail invitation was sent to 39 operational managers of 51 P&O clinics from all 21 regions in Sweden. The invitation included study information and a link to the questionnaire. The operational managers were asked to share the electronic questionnaire with a practitioner who fulfiled the inclusion criterion of having at least 12 months' of experience providing offloading interventions to patients with DFUs.
Two follow-up emails, including a reminder to participate, were sent to the operational managers 7 and 11 days after the initial e-mail. The participating practitioners gave informed consent to participate by checking a box in the electronic questionnaire. According to Swedish law, no ethical approval for the study was needed because the study did not entail any physical intervention, did not affect the participants in any physical or psychological manner, and did not collect personal data or data about the participants' criminal offences.
| Statistical analysis
The data were exported from SurveyMonkey into International Business Machines Corporation (IBM) Statistical Package for the Social Sciences version 27.0 (Armonk, NY: IBM Corp; 2020). Descriptive statistics were calculated for all questions. Spearman's correlation coefficient was calculated to estimate the correlation between the two questions on the practitioners' awareness of gold standard offloading devices (TCC and nonremovable knee-high walker).
| RESULTS
A total of 35 out of 51 practitioners answered the questionnaire, yielding a response rate of 69%.
| Use of offloading interventions
In addition to the 14 predefined devices, the practitioners reported using six other offloading interventions or combinations of offloading interventions (Table 1). Modified off-the-shelf footwear with insoles was provided by 86% of the practitioners (to a mean of 59% of patients) and modified off-the-shelf footwear without insoles was provided by 49% of the practitioners (to 30% of patients). Postoperative shoes, categorised as ankle-high interventions, were provided by 71% of the practitioners (to 12% of patients). Removable knee-high walkers and removable casts were provided by 49% of the practitioners (to 9% of patients), and 20% of the practitioners provided TCCs (to 8% of patients). No practitioner provided offloading wound dressings or nonremovable knee-high walkers.
| Factors considered in the provision of offloading interventions
The median ranking of "often" was registered for 14 of the 28 factors that practitioners considered when providing a patient with an offloading intervention ( Table 2). Wound-related factors were considered "often" or "always". Patient-and practitioner-related factors were considered "sometimes" or "often". Intervention-related factors showed a more heterogeneous pattern where secondary complications, patient tolerance, and gait instability were "often" considered, medical backup and bulkiness/weight were "sometimes" considered, and cost, appearance, and time to apply the intervention were "seldom" considered.
| Awareness of gold standard offloading devices
On the question about whether practitioners considered TCC to be the gold standard treatment, 26% answered yes, 37% were unsure and 37% answered no ( Table 3). The average percentage of patients GIGANTE ET AL. provided with TCC among the answers (yes, no, unsure) ranged from 0% to 3% (Table 3). On the question of whether practitioners considered nonremovable knee-high walkers to be the gold standard treatment, 26% answered yes, 23% were unsure, and 51% answered no ( Table 3). None of the practitioners provided nonremovable kneehigh walkers (Table 3). Spearman's correlation coefficient between the answers to the two questions on the awareness of gold standard offloading devices was 0.812 (p < 0.001; Table 4). Thus, practitioners who agreed that TCC is the gold standard for offloading also tended to agree that a nonremovable knee-high walker was the gold standard for offloading.
| DISCUSSION
The practitioners provided a great variety of offloading interventions to treat plantar neuropathic forefoot DFUs. They mainly provided modified off-the-shelf footwear with insoles, while TCCs and nonremovable knee-high walkers were highly underutilised. Practitioners most often considered wound-related factors in the selection of interventions but also considered practitioner-, patient-and intervention-related factors. The majority of the practitioners were unaware or unsure of the current gold standard devices for neuropathic forefoot DFUs.
The international IWGDF guidelines recommend nonremovable knee-high devices (TCCs or nonremovable knee-high walkers) as the first-choice offloading treatment, removable knee-high devices (removable casts or walkers) as the second choice, removable anklehigh devices as the third choice, and footwear combined with felted foam as the fourth-choice treatment. 3 The IWGDF recommends against using footwear to heal plantar forefoot DFUs unless none of the aforementioned, more effective, offloading devices are available.
However, as all of these more effective devices are available in Sweden, footwear should not be used for healing plantar forefoot lower priority to the use of shoes, insoles or removable orthoses. 14 In the current study, only 20% and 0% of the study practitioners provided TCCs and nonremovable walkers, respectively, 49% provided removable casts or knee-high walkers, 71% provided a removable ankle-high device and no practitioner-provided footwear in combination with felted foam. Furthermore, 86% of the practitioners provided footwear that the IWGDF recommends against T A B L E 2 Frequencies of practitioner-, patient-, intervention-, and wound-related factors taken into consideration when P&O practitioners provide offloading interventions to treat plantar neuropathic forefoot ulcers in patients with diabetes. Whether an intervention will be tolerated 3 (3-4) 3.2 (0.7) 2-4
Median (quartile 1-3) Mean (SD) Min to max
How long an intervention will take to apply 1 (1-2) 1.5 (1.0) 0-4 Whether an intervention will cause gait instability 3 (2-4) 3.0 (0.9) 1-4 Wound-related factors In the Eurodiale study, which included 14 diabetic foot centres in 10 European countries, the centres provided TCCs to an average of 18% of patients with plantar fore-or midfoot DFUs. 11 The results of the current study and the study by Raspovic and Landrof 9 indicate that the choice of offloading intervention is a complex process and that several factors and barriers need to be considered. Notably, compared to the Swedish P&O practitioners, the Australian podiatrists considered to a higher degree whether the suggested offloading intervention would restrict wound care.
This factor is relevant only if a nonremovable device is used.
Because nonremovable devices were underutilised by P&O practitioners in Sweden, this factor was naturally less important. In the present study, a gap was found between the offloading interventions prescribed and the interventions recommended in guidelines, 3,14 a pattern that has been observed in other studies. [9][10][11] This gap between recommendations and practice may partly be because many DFU offloading studies on which the guidelines are based have a narrower perspective than the perspective of practitioners. That is, practitioners tend to consider more factors when choosing offloading devices, such as practitioner-and patientrelated factors, than those considered in studies on these devices.
Thus, future studies need to take a broader perspective on offloading, including patients' tolerance and preferences as well as the impact on gait and daily activities, to better reflect the clinical situation of patients.
There was a strong correlation between the answers (yes, no, unsure) to the two questions regarding awareness of the gold standard offloading treatments, meaning that practitioners who considered TCC the gold standard treatment also considered nonremovable T A B L E 3 Knowledge and provision of gold standard offloading devices.
Practitioners considering TCC the gold standard, % (n)
knee-high walkers to be the gold standard treatment. However, overall, the awareness of the gold standard of offloading was low; only 23% of the practitioners were aware that both TCC and nonremovable knee-high walkers are gold standard treatments for plantar forefoot DFUs compared to 83% of podiatrists in Australia 9 and 41.9% of centres treating DFUs in the USA. 10 Despite this low awareness among the practitioners in the current study and although they provided interventions almost opposite to the guidelines, the practitioners reported that they "often" considered evidence-based practice when prescribing offloading interventions. The reasons for this discrepancy among practitioners need to be further investigated to increase the use of the recommended offloading interventions. | 2,689.4 | 2023-01-18T00:00:00.000 | [
"Medicine",
"Engineering"
] |
Gauged Galileons From Branes
We show how the coupling of SO(N) gauge fields to galileons arises from a probe brane construction. The galileons arise from the brane bending modes of a brane probing a co-dimension N bulk, and the gauge fields arise by turning on certain off-diagonal components in the zero mode of the bulk metric. By construction, the equations of motion for both the galileons and gauge fields remain second order. Covariant gauged galileons are derived as well.
Introduction
Many recent investigations have involved-either directly or indirectly-the presence of galileons, which are higher-derivative scalar fields that both have secondorder equations of motion and are also invariant under a novel "galilean" symmetry: π(x) → π(x) + c + b µ x µ . Originally, this symmetry arose in the scalar sector of the decoupling limit of the Dvali-Gabadadze-Porrati (DGP) model [1], where it may be thought of as a small-field consequence of the nonlinearly-realized five-dimensional Poincaré symmetry. This galilean symmetry has since been abstracted and studied in its own right [2] (for a review of recent developments, see [3]). Galileons have rather nice properties and structure; it is non-trivial that there exist terms invariant under the galilean symmetry which also have second order equations of motion for the field π. Second order equations of motion guarantee that the theory does not propagate extra ghostly degrees of freedom which are common in other higher-derivative theories. Further, choosing to consider only these terms is consistent from an effective field theory viewpoint; the fact that they have fewer derivatives than other terms invariant under the galilean shift symmetry means that there exists a regime where galileons are the dominant terms and the others can be consistently neglected [4][5][6]. Additionally, due to their symmetry properties and the fact that they shift by a non-trivial total derivative under the symmetry (they are Wess-Zumino terms [7]) , galileon theories are radiatively stable-they are not renormalized at any loop order in perturbation theory [4,8].
The properties of galileons are simple to state, but the theories possess a rich and interesting phenomenology. Galileons have been used to address issues in both the early universe through inflation [9][10][11] and alternatives to inflation [12], as well as in the late universe where they have been investigated as a possible source of cosmic acceleration [13][14][15][16]. Galileons also make an appearance in ghost-free massive gravity, where they describe the interactions of the longitudinal polarization of the graviton in the decoupling limit [17,18] (for a review see [19]).
Many applications require that galileon theories be covariantized. This is possible, but retaining their second-order equations of motion requires introducing non-minimal coupling between the fields and curvature, generically destroying the shift symmetry of the field [20][21][22]. Appropriate non-minimal terms arise naturally in the probe brane construction [23]; this construction also elucidates the origin of the second-order equations of motion-the galileon terms descend from Lovelock invariants of the induced brane metric and from Gibbons-Hawking-York (GHY) boundary terms associated to bulk Lovelock invariants. The Lovelock terms are of course the only terms that may be added to Einstein gravity while maintaining second order metric equations of motion [24], and this property is passed down to the galileons through the probe brane construction.
The probe brane construction has been extended to curved brane backgrounds, on which fields are invariant under intricate nonlinear symmetries inherited from the isometries of the bulk [25][26][27][28]. The brane construction has also been generalized to higher co-dimension [4]; this generalization leads to a multi-galileon theory where the fields possess an internal global SO(N ) symmetry, which is inherited from the symmetries of the higher codimension bulk. Related multi-galileon theories were discussed in [29][30][31][32].
Recently it was shown by Zhou and Copeland [33] that it is possible to couple galileons to gauge fields while retaining second-order equations of motion. In this note we generalize the higher dimensional probe brane construction of [4] to recover the SO(N ) gauged galileon theories of [33] from a purely geometric setup.
Gauging the Galileons
An N -galileon theory contains N scalar fields π I , indexed by I = 1, · · · , N , which have second order equations of motion and a galilean and shift symmetry on each field: π I (x) → π I (x) + c I + b I µ x µ , where b I µ and c I are constants. There may also exist global internal symmetries under which π I transforms in a linear representation [4,32]. These global symmetries can be promoted to local ones [33]. Here we will restrict to the case where the galileons transform as a fundamental un-der SO(N ), the case which naturally follows from a codimension N brane construction [4], since our goal here is to demonstrate a brane perspective for deriving these gauged multi-galileons.
On flat space, there are two multi-galileon Lagrangians which respect the SO(N ) global internal symmetry for N ≥ 2 [4,32], The symmetries of these terms come in three sets [4] The first is ordinary Poincaré invariance for the scalar fields, the second is the galilean and shift symmetry, and the third is the internal SO(N ) symmetry (for which ω IJ is the infinitesimal antisymmetric parameter). As considered in [33], we may promote the global SO(N ) symmetry to a local one by minimal substitution, ∂ µ π I → D µ π I = ∂ µ π I + A I µJ π J . Here A I µJ is an anti-symmetric matrix, which is just the gauge connection in the fundamental representation of SO(N ), where the generators of SO(N ) are given by This minimal coupling procedure gives gauge invariant actions with second order equations of motion both for π I and A I µJ . The gauging, however, eliminates the galilean symmetry (this is similar to the situation that occurs when covariantizing the galileons).
The presence of second order equations of motion after the naïve gauging is not surprising, as was pointed out in [33]. Due to the structure of the spacetime index contractions, there will never be more than two derivatives on a π I and the highest derivatives on A µ enter through expressions of the form D λ F µν , where F µν is the field strength Since F µν contains only first derivatives on A µ , the equations of motion for A µ are at most second order. Note that the minimal coupling prescription is ambiguous. For instance, one could have changed the ordering of derivatives in the action, say ∂ µ ∂ ν π I → ∂ ν ∂ µ π I , and gauging the Lagrangians before and after this replacement would give different results, since the gauge covariant derivatives do not commute, We can't say one choice is "more minimal" than the other. Requiring second order equations of motion does not pin down the Lagrangian uniquely, since there is freedom to add non-minimal terms (even beyond those resulting from commuting derivatives) which do not lead to higher order equations. One of the virtues of the brane construction will be to pick out a particular set of nonminimal couplings. We may also consider coupling to gravity through naïve covariantization, ∂ µ → ∇ µ . This maintains second order equations of motion for π I , but there also arise terms of the form ∇ λ R µνρσ in the equation of motion of L 4 . As R is second order in derivatives of the metric, the equation of motion is third order in the metric. Adding a nonminimal coupling can remove these third order derivatives and those in the metric equations of motion [20]; for example the following has second order equations of motion for both the scalars and metric, This lagrangian can also be obtained from the probe brane construction [4]. As in the case of gauging, and as seen from [22], the choice of non-minimal terms in (7) is not unique; there are other possible non-minimal couplings which still give second order equations of motion for all the fields. No choice is singled out by the procedure of minimal coupling followed by the addition of non-minimal terms to cancel higher-order pieces of the equations of motion, and a virtue of the brane construction will be to single out a specific choice of non-minimal terms.
Covariantizing the galileons in this way breaks the galilean symmetry, but preserves the global SO(N ), which can then be gauged by replacing ∇ µ → D µ = ∇ µ + A µ . The resulting gauge and diffeomorphism invariant Lagrangian has second order equations of motion for the scalars, the metric, and the gauge fields [33].
The Higher Dimensional Brane Construction
In this section, we briefly review the probe brane construction of the multi-galileons. For a more detailed treatment, we refer the reader to [4,25].
The probe brane construction was originally developed [23] for single field galileons arising via a codimension one brane probing a flat bulk. The action is constructed from diffeomorphism scalars formed from the induced metric and extrinsic curvature of a 3-brane floating in the 5D bulk. Symmetries of the action are inherited from Killing vectors of the bulk [25] and the unique co-dimension one Lagrangians which have second order equations of motion are the 4D Lovelock invariants and the Gibbons-Hawking-York boundary terms for the 5D Lovelock invariants (and a tadpole term).
Extending the probe brane construction to higher codimension allows for the construction of multi-galileon theories [4]. We begin with a D-dimensional bulk with coordinates X A and metric G AB (X). The position of a 4-dimensional brane living in the bulk is given by embedding functions X A (x), where x µ are coordinates on the brane. Tangent vectors to the brane have components e A µ = ∂X A ∂x µ and the induced metric on the brane is There are also N ≡ (D − 4) vectors normal to the brane indexed by I, with components n A I , which satisfy The normal and tangent vectors are used to construct the N extrinsic curvature tensors, where ∇ A is the bulk covariant derivative, as well as the twist connection, which is the connection on the normal bundle, it has an associated curvature R I Jµν . Requiring the action to be invariant under reparametrizations of the brane restricts the action to be a diffeomorphism scalar constructed from these geometric ingredients, Here∇ µ is the world-volume connection, which acts on 4D spacetime indices with the Levi-Civita connection of the induced metric, and on normal indices with the twist connection. We fix the reparametrization symmetry of the brane worldvolume coordinates by choosing that is, we take the 4 worldvolume coordinates to coincide with the first 4 coordinates used in the bulk. The N remaining functions π I are the physical degrees of freedom for the brane.
Given a Killing vector K A of the bulk metric G AB , the induced metric and extrinsic curvature (and hence the action (12)) are invariant under δ K X A = K A . However, generically this destroys the gauge choice (13) by sending and we must restore the desired gauge via a brane reparametrization δ g X A (x) = ξ µ ∂ µ X A (x) with ξ µ = −K µ so that the combined gauge-preserving π I symmetry acts as and becomes a global symmetry of the gauge fixed action. Symmetries that have a K I component are nonlinearly realized and are thus symmetries of the bulk that are spontaneously broken due to the presence of the brane. Generic choices of the action (12) will not give second order equations of motion for the π I . For a 4-dimensional brane the unique terms that give second-order equations of motion are the 4-dimensional Lovelock terms and possible Gibbons-Hawking-York boundary terms for the higher dimensional Lovelock terms, whose specific form depends on the dimensions of the brane and the number of co-dimensions. A four dimensional brane has two 4D Lovelock terms-the cosmological constant and the induced Ricci curvature, The possible GHY terms for the 3-brane depends on the number of co-dimensions [4,34,35]. However, as was shown in [4], in the end no new possibilities for actions are generated beyond those given by (16), so we need only consider these two. The galileons are obtained by taking the bulk metric to be fixed and flat, G AB (X) = η AB . The induced metric isḡ µν = η µν + ∂ µ π I ∂ ν π I . Evaluating the actions (16) gives relativistic DBI versions of the SO(N ) symmetric galileons. A small field limit then reproduces (1). The flat metric has maximal symmetry and all of these symmetries are realized in the galileon theory. The Poincaré transformations along the brane become the 4D Poincaré transformations, the rotations in the extra dimensions become the internal SO(N ) symmetry, translations in the extra dimensions become the shift symmetry, and the (small field limit of) boosts into the extra dimensions become the galilean symmetry. The small field limit may be viewed as either an expansion in derivatives or in fields, both produce the same result [23,25]. From an algebraic perspective, the small field limit may be thought of as Wigner-İnönü contraction of the Poincare algebra along the co-dimension directions, that is, sending the speed of light in the directions away from the brane to infinity [7].
We now show how to obtain gauged symmetries from the previously discussed probe brane description. To gauge the symmetries, we simply turn on zero modes for the background metric G AB (X).
For example, to couple to gravity, we take the background metric to be [23] We have turned on the 4D part of the metric and allowed it to depend only on the 4D coordinates x µ . The induced metric now becomesḡ µν = g µν + ∇ µ π I ∇ ν π I . Evaluating the actions (16) gives relativistic DBI versions of the covariant SO(N ) symmetric galileons. A small field limit then reproduces precisely (7) and the canonical kinetic term [4]. The non-minimal terms in L 4,cov needed to make the equations of motion second order come out automatically, and a unique such term is produced. The metric (17) breaks the higher-dimensional Poincaré invariance. All that survives is the SO(N ) rotations and translations in the extra dimensions. This is reflected in the fact that the only symmetries left in (7) are SO(N ) rotations and shifts on the fields. The extended galilean symmetry is lost. The zero mode metric g µν and the scalars π I inherit a diffeomorphism transformation under the zero mode of higher-dimensional diffeomorphisms which preserves the ansatz (17), and this yields the diffeomorphism invariance of the 4D theory.
To gauge the SO(N ) internal symmetry, we will turn on zero modes of off-diagonal components of the background metric, corresponding to Killing vectors of the extra dimensions. We take a bulk metric of the form seen in Kaluza-Klein reductions The ξ I i (y)'s are Killing vectors of δ IJ (depending on y I , the coordinates in the extra dimensions), I denotes the components of the Killing vector in the extra-dimensional space and i labels the various Killing vectors. The coefficient functions A i µ (x) are arbitrary functions of the 4D coordinates which will be the gauge fields from the perspective of the brane. The induced metric on a 4dimensional brane, calculated in the gauge (13), is now given bȳ We want to gauge only SO(N ), so we will turn on only those Killing vectors corresponding to rotations in the extra dimensions. 1 The index i can then be taken to be the anti-symmetric index set [JK] which runs over N (N − 1)/2 values. The components of the Killing vectors ξ [JK]I are given by Now we have ∂ µ π I + A i µ ξ I i (π) = ∂ µ π I + 1 2 A JK µ ξ I [JK] (π) = ∂ µ π I + A I µJ π J = D µ π I , and we recover the covariant derivatives, so the induced metric indeed takes the form conjectured in [33], Evaluating the action (16) now gives a relativistic DBI version of the gauged SO(N ) galileons, whose small field limits reproduce the gauged galileons of [33]. For example, the gauged kinetic term comes from the cosmological term − √ −ḡ and expanding to quadratic order in π, we find The Einstein-Hilbert term yields, in the small field limit, whose equations for both the gauge field and scalar are second order. Note that a specific set of non-minimal couplings has been produced. 2 This lagrangian agrees with equation (18) of [33], up to integration by parts and addition of non-minimal couplings which have secondorder equations of motion. The ansatz (18) breaks the boost symmetries of the brane into the extra dimensions, and this is reflected in the fact that the galilean symmetry of the 4D theory is spoiled by gauging. The zero mode vectors A µ and the scalars π I inherit a gauge transformation under the zero modes of higher-dimensional diffeomorphisms which preserve the ansatz (18), and this yields the gauge invariance of the 4D theory.
To recover the gauged and covariant galileons of [33], we turn on both the zero mode gauge fields and the zero mode metric, The induced metric is now given bȳ The covariant derivative D µ ≡ ∇ µ + A µ now acts covariantly on both gauge indices and spacetime indices. The calculation of L 4 = − √ −ḡR now gives, in the small field limit, the gauged and covariant galileons of [33] with a specific set of non-minimal couplings, ensuring that the equations of motion for the metric, gauge fields and scalars are all second order, This lagrangian is equivalent to equation (41) of [33], again up to integration by parts and possible addition of non-minimal couplings which retain second-order equations of motion. Note that (26) is the natural fusion of (7) and (23).
Conclusion
Multi-galileon theories which are invariant under an internal global SO(N ) symmetry arise naturally from a co-dimension N probe brane construction, in which the bulk is a fixed isotropic manifold. By allowing parts of the bulk metric to become dynamical, we have shown that the SO(N ) symmetry can be gauged while retaining second-order equations of motion. While we have focused on the SO(N ) case for concreteness, some generalization is fairly straightforward. By exploiting the embedding of SU (N ) into SO(2N ), it should be possible to couple galileons to SU (N ) gauge fields using the same setup with a co-dimension 2N bulk. Additionally, we have restricted to the case where gauge fields transform in the fundamental representation, but it should be possible to generalize to some cases of gauge fields in other representations of other groups. The procedure would be to embed in a co-dimension M bulk, such that the group G we wish to represent is a subgroup of SO(M ), and the representation of G we wish to have can be found within the restriction of the fundamental of SO(M ) to G. Then, one would turn on only the gauge fields corresponding to G.
We expect that gauged galileons will have a rich and interesting phenomenology, possibly both for cosmology and for particle physics. It is possible that galileon theories may arise in beyond the standard model model physics, in particular, their non-renormalization theorem makes it very tantalizing to consider connections to long outstanding problems such as the hierarchy problem. In cosmology, galileons may arise in the dark sector. In either case, such applications will require an understand of the interplay between galileon theories and gauge fields. It is also possible that gauged galileons may allow for interesting defect solutions, in contrast to their un-gauged counterparts [6]. | 4,669 | 2011-12-29T00:00:00.000 | [
"Physics"
] |
Dark Energy and the Refined de Sitter Conjecture
We revisit the phenomenology of quintessence models in light of the recently refined version of the de Sitter Swampland conjecture, which includes the possibility of unstable de Sitter critical points. We show that models of quintessence can evade previously derived lower bounds on $(1+w)$, albeit with very finely-tuned initial conditions. In the absence of such tuning or other rolling quintessence fields, a field with mass close to Hubble is required, which has a generic prediction for $(1+w)$. Slow-roll single field inflation models remain in tension. Other phenomenological constraints arising from the coupling of the quintessence field with the Higgs or the QCD axion are significantly relaxed.
Introduction
A class of effective field theories (EFTs), while otherwise consistent, do not admit a UV-completion within a theory of quantum gravity. Such EFTs are said to lie in the Swampland [1]. Delineating the boundaries of the Swampland is an important task that could point to observable consequences of a theory of quantum gravity. Progress has been made by studying the general properties of string compactifications (e.g. see [2][3][4][5][6] or [7] for a recent review).
Alternate formulations of the dSC have also been proposed in [43,46,47]. We study a particular refinement of the dSC (henceforth referred to as RdSC) which has been proposed in [48,49]. The refinement allows (1.1) to be violated if the second derivative of the potential is sufficiently negative. Explicitly, the refined de Sitter conjecture states where c, c are constants of O(1).
[1] For a review of quintessence dark energy, see [22] The RdSC is motivated by its connection to the distance Swampland conjecture [3] through Bousso's covariant entropy bound [50]. The refined conjecture evades all the counter-examples to the dSC [19,47,51,52] since they involve tachyonic dS critical points [53,54]. Similarly, it also evades constraints arising from coupling of the quintessence field with the Higgs field, the pion and the QCD axion [20,43,45].
The original dSC had a firm prediction for the equation of state for the dark energy. In this note we investigate the implications of the refined de Sitter conjecture (RdSC) for dark energy phenomenology. We show that an arbitrarily fine-tuned initial condition can satisfy current and future constraints on w(z) for any values of c and c , but generic initial conditions retain a prediction of deviation from w = −1.
2 Elsewhere on the moduli space Dark energy observations allow a quintessence field in the current universe with a potential whose slope is of the order of the vacuum energy (in Planck units). Interesting constraints can be derived from considering the potential of the quintessence field along with other scalar fields like the Higgs and the pion (and potentially the axion) away from our current position on the moduli space.
If the quintessence does not couple to the Higgs, then the dSC is badly violated at the symmetric point of the Higgs potential [43]. A coupling of the Higgs with the quintessence, on the other hand, leads to larger-than-observed deviations in fifth force experiments, except for perhaps a very finetuned set up where this coupling vanishes around the Higgs minimum [20,43]. More recently it was shown that even this possibility is under tension from time-dependence of the proton-to-electron mass ratio [44].
This tension is relaxed when we consider the refined conjecture. Clearly, at the symmetric point on the Higgs potential the RdSC is satisfied irrespective of the coupling of the Higgs with the quintessence field. However, it is useful to see whether this is true along the entire relevant range of the potential. For illustration we take a toy potential, For values of h outside a small neighborhood of the origin (and away from the minimum), there is a slope in the h direction which satisfies the slope requirement in (R)dSC, where cv 2 /M Pl v. However, near the origin, the second derivative ∇ 2 h V causes the RdSC to be satisfied. The latter switches Finally, as the h field nears its minimum at v, neither of the above conditions is fulfilled, and the (R)dSC is satisfied by the slope in the φ direction, as long as λ > c. The norm of the gradient of the potential is sufficiently large even away from the minimum (and well into the region where |∇ h V | is large), if the following condition is met for the potential at h = v, An example scalar potential (e.g. the Higgs potential) with spontaneous symmetry breaking, with parameters exaggerated for clarity. In the blue region, the (R)dSC is satisfied by the h itself. In the orange region, the RdSC is satisfied along the h direction. In the green region near the minimum along the h direction, a "quintessence" field is needed to satisfy (R)dSC. If (R)dSC is satisfied at the minimum, then it is satisfied in all the green region (see text for details).
This is a very mild restriction on the potential. This analysis can be carried out in essentially the same way for the SM Higgs or the pion, as well as the QCD axion. In figure 1 we show a toy example of this phenomenon. We have chosen exaggerated numerical values for illustration. We see that the RdSC is satisfied everywhere along the potential with no cross coupling requirement between φ and h.
Finally, as noted in [49], the RdSC also resolves issues pointed out by [19] and, in a similar higher dimensional setting, by [20]. The general arguments in these papers only tell us about the presence of an extremum, and in particular a maximum / inflection point. Therefore, the RdSC can be easily consistent with these constructions if the smallest eigenvalue of the Hessian is sufficiently negative at these critical points.
Dark Energy
We now turn to an analysis of the implication of the refinement on dark energy phenomenology. The dSC allows one to place a lower bound on the equation of state parameter w ≡ p φ /ρ φ as presented in [23], and using current observations, an upper bound on c. We now show that the RdSC allows a class of models where the quintessence field satisfies the second clause of the refined conjecture, and we cannot put a bound on c using data. Further, one can arrange for w ≈ −1 today, albeit at the expense of very finely tuned initial conditions. For concreteness we illustrate this with an example potential for the quintessence field of the form, where V 0 = (2.23 meV) 4 is the value of dark energy in ΛCDM. This potential satisfies the second clause of the RdSC when b 2 ≥ c a 4 . The mass of the field can be conveniently written in terms of the Hubble parameter today, where Ω Λ = 0.692 is the dark energy fraction of the universe today. Given the dark energy content of the universe today, we see that this potential requires a 1. Then RdSC puts a lower bound on the curvature of the quintessence potential, b 2 > c , or |m 2 φ | > 3c Ω Λ H 2 0 . If we ignore quantum fluctuations, then it is clear that we can set the field at the top of the potential, and obtain w = −1 exactly. However, quantum fluctuations will destabilize the field from the maximum and fragment it to form domain walls. Therefore, we have to ensure that the field is sufficiently misaligned from the top to behave classically. The classical regime only requires a very tiny misalignment, (3.2) At the same time, if the field has a large misalignment from the maximum, its equation of state will deviate from w = −1.
There are two possible allowed regimes. The first is when b 1, the slope of the potential is large as soon as the field is misaligned from the maximum, and leads to large deviations in w + 1. In this case the initial misalignment has to be very finely tuned to be consistent with supernova observations. In the second case, near the boundary of the RdSC constraint, b 2 c a 4 , the mass of the field is order H 0 , and therefore it only starts rolling today for O(1) misalignment, making it consistent with observations.
In figure 2 we show the allowed parameter range in (a, b) space. We also show the contours of the maximal initial value of the field that is allowed by observations of Ω Λ [55] and w(z) [56]. As in [23], we choose the 2σ contours for w(z) in [56] and for Ω Λ from [57], and find the largest initial misalignment that is consistent with the measurement of Ω Λ and w(z). For a different procedure to extract w(z) constraints to put a bound on c, see [58,59]. For b 1, we see that the initial values have to be tuned, as noted above. In the region b ∼ 1, we can allow generic initial conditions. The trajectories of the fields for specific values of (a, b) are shown in Figures 3 and 4. We show the evolution of this system in (x, y) coordinates defined as as well as the evolution of the equation of state w. We have chosen examples where the initial conditions are allowed to be generic (Fig. 3) and where we need to tune the initial conditions to satisfy the supernova constraints (Fig. 4). We see that in either case, tuning the initial conditions allows us to push w as close to −1 as we want, evading the lower bound derived in [23]. However, in the absence of this initial condition tuning, a generic prediction on the equation of state can be estimated. If c 1, then it is generally hard to satisfy the current w(z) constraints with untuned initial conditions, and the least tuned initial conditions will typically have (1 + w) close to the constraints today.
Using the "slow-roll" approximation, 3Hφ V (φ), we can estimate the value of w for this parameter choice [60,61], It is interesting to compare this to a very similar looking bound derived in [23]. We emphasize however that in the current case this is not a hard bound but more a generic prediction. Due to this fact, we are unable to derive a robust bound on c using current data as was done for c in [23]. Further, we can trade off bounds on c with bounds on c , and with tuned initial conditions we can have both c and c to be O(1).
Conclusion and discussion
Swampland conjectures are in a very active phase of exploration. We have studied the implications of the recently proposed refined de Sitter conjecture [46][47][48][49]. The conjecture is motivated by an older distance Swampland conjecture and its connection with Bousso's covariant entropy bound. This conjecture circumvents a number of theoretical and phenomenological tensions arising from coupling of the quintessence field to other scalar fields in the standard model and beyond. The RdSC is also consistent with constructions which are claimed to be counter-examples to the earlier de Sitter conjecture.
As in dSC, the RdSC appears to be in tension with single-field slow-roll inflation if both c, c are strictly O(1) [62]. The fact that either | V | or |η V | are O(1) makes it impossible to naturally satisfy constraints on the scalar tilt n s −1 ≈ 2η V −6 V ≈ 0. In fact, recent analyses [55] show that O(1) values for either of η V or V are strongly ruled out. More extended inflationary models can potentially evade this tension; these models often come with detectable deviations from single-field case [25,28,63]. Unless the initial conditions are very fine-tuned, the conclusions of [23] for the future cosmology of the universe remain mostly unchanged.
It would be very interesting to try and identify models for such a scalar field. In the string axiverse we expect a plenitude of light scalars, and the positive dark energy of the universe in such a system can be made up of a number of light particles. What the RdSC adds to this picture then is that it hints towards an axion of mass comparable to the Hubble scale, with a misalignment from its minimum of O(M Pl ) such that it has just recently begun dominating the universe and appears briefly as dark energy, before eventually starting to oscillate as matter. | 2,898 | 2018-11-01T00:00:00.000 | [
"Physics"
] |
Higher-Spin Algebras, Holography and Flat Space
In this article we study the algebra generated by the holographically reconstructed cubic couplings for the type A minimal bosonic higher-spin theory on AdS$_{d+1}$, which were recently extracted from the free scalar $O(N)$ model. We demonstrate that it is equal to the unique higher-spin algebra for bosonic totally symmetric higher-spin fields in generic dimensions. This provides an explicit check of the holographic reconstruction and of the duality between higher-spin theories and the free $O(N)$ model in general dimensions, extending the result of Giombi and Yin in AdS$_4$. For completeness, we also address the same problem in the flat space for the cubic couplings derived by Metsaev in 1991, which are recovered in the flat limit of the AdS type-A cubic couplings. We observe that both flat and AdS$_4$ higher-spin Lorentz subalgebras coincide, hinting towards the existence of a full higher-spin symmetry behind the flat-space cubic couplings of Metsaev.
Introduction
In this paper we investigate the algebras underlying higher-spin theories on AdS-and flat-space, as implied by the available expressions for the metric-like action at cubic order.
Recently, a cubic order action for the type A minimal higher-spin theory on AdS d+1 was obtained employing its conjectured duality [1,2] with the free scalar O (N ) model [3,4]. 1 A non-trivial test of this duality would then be provided from the comparison of the rigid structure constants implied by the holographically reconstructed cubic action with the known expressions [10,11] for higher-spin algebra structure constants, which are unique in general dimensions [12]. This check of the duality is investigated in this work, with the outcome that the result is indeed positive, and with it extending to general dimensions the three-point function test [13] of Giombi and Yin in AdS 4 . 2 In particular, the test confirms that the holographically reconstructed cubic couplings solve the Noether procedure at the quartic order.
Re-winding a number of years, in [16,17] Metsaev constructed the complete cubic action for the higher-spin theory in flat space using light-cone methods. Motivated by the observation that it contains vertices which are not accommodated for in the original covariant classification [18][19][20][21][22] of cubic vertices in flat space, we study the corresponding rigid structure constants. This is particularly appealing, for the additional vertices in Metsaev's solution are lower derivative, including a two-derivative coupling of higher-spin fields with gravity [23,24]. We argue that this coupling can be considered as a minimal coupling, in accordance with the equivalence principle. 3 Indeed, in [26] a version of the Coleman-Mandula theorem was proven, stating that the flat space cubic interactions in the original covariant classification cannot give rise to a higher-spin theory with higher-spin generators satisfying non-trivial commutation relations with the isometry generators. 4 We extract explicit expressions for the Lorentz part of the structure constants, which we find matches with the Lorentz part of corresponding AdS 4 higher-spin algebra. The existence of these structure constants crucially relies on the presence of the additional lower derivative vertices, which are local in the light-cone gauge but (as we shall demonstrate) do not admit a standard covariant form. Of course, in spite of these promising results it remains to be seen whether other consistency conditions (such as those from higher-orders in the Noether procedure) permit the existence of a consistent unitary interacting theory. See [29][30][31][32][33][34] for recent works in this direction.
Outline
The article is organised as follows: In §2 we review the deformation of gauge transformations by higher-spin interactions and extract the rigid structure constants from the cubic vertices of [3]. §3 is devoted to the 4d cubic action of Metsaev [16,17] and to extracting the corresponding structure constants upon considering a formal covariantisation of the light-cone vertices. We then argue that these structure constants give a well-defined higher-spin symmetry, and that the two-derivative coupling of a higher-spin field to gravity can indeed be considered as a minimal coupling. In §4 we consider the aforementioned flat space cubic vertices in the framework of the spinor-helicity formalism. In §5 we discuss some consequences of such a higher-spin symmetry on higher order amplitudes. Concluding §6 summarises the results of the paper and presents few outlooks. §A gives a summary of higher-spin algebra structure constants.
The methods we employ to extract the structure constants were developed in previous works [26,[35][36][37][38][39], which we briefly review in this paper. In particular, we extract the structure constants from the first-order deformations of the gauge transformations induced by cubic interactions. This is reviewed in the following section.
Review: Noether approach to higher-spin theories
The Noether method to constructing interacting higher-spin gauge theories is a systematic perturbative approach, underpinned by the requirement of gauge invariance [40]: For a given spectrum, one begins with the free theory and adds interactions order by order in the weak fields in a way that is consistent with the gauge symmetries at each order.
(1. 8) In summary, in this paper we analyse this condition for the cubic action [3] established by holographic reconstruction for the type A minimal theory on AdS d+1 . The latter cubic action was not obtained via the standard Noether approach (1.2), and since higher-spin algebra structure constants are unique in generic dimensions, this study provides a non-trivial check of the holographic reconstruction and a test of the holographic duality. We also analyse this condition for the cubic couplings of the flat space theory [16,17], which was obtained by requiring closure of the Poincaré algebra on the light-cone.
2 The holographic cubic action and induced gauge symmetries
Review: AdS cubic couplings
In this section we review the construction of cubic interactions in higher-spin gauge theory on anti-de Sitter (AdS) space, in particular the completion of the theory at cubic order via holographic reconstruction [3]. The relevant results are recalled up to terms proportional to divergences and traces of the gauge fields. This traceless and transverse (TT) framework is sufficient for the purpose of extracting the corresponding putative higher-spin algebra structure constants (see e.g. [26]), which will be reviewed in §2.2. In the sequel all equalities therefore hold modulo terms proportional to traces and divergences, which we denote by TT = unless the context is clear.
Ambient-space formalism
The results for the cubic action are most conveniently expressed and obtained in the ambient space formulation (see [38,51] for further details). In this framework, AdS d+1 space with radius R is realised as a hyperboloid in an ambient (d + 2)-dimensional Minkowski space, From this point onwards we set R = 1. Symmetric spin-s fields ϕ µ 1 ...µs intrinsic to the AdS manifold are described in this framework by ambient avatars Φ M 1 ...Ms (X), 6 which satisfy homogeneity and tangentiality constraints In the above we introduced the generating function where U M is an ambient auxiliary vector. When τ = −2, the generating function (2.4) packages a tower of bosonic spin-s gauge fields with gauge symmetries providing an ambient description of the intrinsic gauge transformations Traceless and transverse cubic action The first non-trivial consistency condition (1.2) (i.e. 0 ) fixes the kinetic term of the higher-spin action. In the ambient space formalism, this reads 7 where the . . . are TT contributions which we disregard and for convenience we use a noncanonical normalisation. At cubic order, the TT part of a coupling can be encoded modulo total derivatives by a function of six building blocks (2.8b) 6 In particular, with pullback The most general ansatz for the TT part of the cubic action can then be expressed in the form where the function f (Y i , H i ) may be fixed by enforcing Noether consistency (1.2). The latter at step 1 yields the constraint, δ 12) and D is the differential operator 8 (2.14) In principle the coefficients k n s 1 s 2 s 3 in (2.11) may be determined from the higher order consistency conditions in (1.2). An alternative route was taken in [3], using the holographic duality between the type A minimal bosonic higher-spin theory and the free scalar O (N ) vector model. By matching the three-point Witten diagrams in the bulk theory to the dual CFT correlation functions of single-trace operators (figure 2.1), the cubic vertices for any triplet of spin {s 1 , s 2 , s 3 } have been determined with relative coupling constants 9 Assuming the holographic duality holds, the full TT cubic action thus reads From this result one may test the holographic duality by extracting the corresponding global symmetry structure constants and comparing with the known expressions [11]. This is carried out in the following sections.
Deformation of the gauge symmetries
In order to preserve the number of degrees of freedom, introducing interactions induces deformations in the gauge transformations. Cubic interactions may lead to O (ϕ) deformations, which can be seen from the 1 consistency condition in (1.2) In this section we determine such corrections induced by the holographically reconstructed cubic action (2.17). We employ the approach taken in [26], which used ambient space techniques to extract the deformations necessitated by the cubic structures (2.11). Given a cubic action, the idea is to extract the corresponding deformation of the gauge transformations from the consistency condition (2.18). The first step is to compute the variation of the cubic vertices off-shell under linearised gauge transformations. This is proportional to the equations of motion (by consistency), and for the holographic cubic action (2.17) this reads To satisfy (2.18), the above must then be compensated by the variation of the quadratic part of the action . (2.20) 9 Note the extra factors of √ si! compared to [3], due to the different choice of kinetic term normalisation.
The key to then solve (2.18) for the deformations is to rewrite (2.19) in a way that factorises the equations of motion, as in (2.20). This can be achieved by simply integrating by parts, which leads to 10 δ (0) where we introduced Without loss of generality, we can focus on δ (1) The deformation of the gauge transformation can then be read off from the above formulas, giving where we introduced the projector Π Φ to ensure the correct homogeneity degree and tangentiality conditions for the Fronsdal field Φ (c.f. [26]).
Gauge algebra structure constants
With the result (2.26) for δ (1) , as explained in §1.2 the deformed structure constants of the gauge algebra can be extracted through (1.5) 3 .
(2.27)
Referring the reader to [39] for further details, one obtains where Π E ensures the correct homogeneity and tangentiality conditions for a gauge parameter. 10 To this end, it is convenient to switch from encoding the vertex with basis (2.8) to the followinḡ
Holographic higher-spin algebra structure constants: Test of the duality
The formula (2.28) obtained in the previous subsection gives the lowest order commutator for the gauge algebra of a putative higher-spin theory dual to the free scalar O (N ) model. Higher-spin symmetry is in this context gauged, and the corresponding global (or rigid) higherspin symmetries can be obtained from evaluating the deformed bracket (2.28) on the gauge parameters E =Ē which satisfy the Killing equation In this section we show that these rigid structure constants indeed define a non-degenerate Lie algebra, coinciding with those of the Eastwood-Vasiliev higher-spin algebra [52,53]. The Eastwood-Vasiliev algebra has in fact been shown to be unique in generic dimensions, through consideration of the Jacobi identity at the quartic order [12]. This result may therefore be considered as a test of the holographic duality, demonstrating that the holographically reconstructed theory is indeed the same theory one would obtain by solving the Noether procedure up to the quartic order.
Killing tensors
We first review the solutions to the Killing equation (2.29) in the framework of ambient space. Combined with the tangentiality and homogeneity conditions on the ambient space gauge parameter E, it is straightforward to write down the Killing tensorsĒ, which read (2.31) Combined with the tracelessness of the gauge parameter, one can also conclude that the Killing tensors are completely traceless The generators of a putative underlying higher-spin algebra are the duals ofĒ M 1 N 1 ,...,MrNr , given by The . . . signify that the higher-spin generators, being contracted with traceless tensors, are defined as equivalence classes modulo traces: X · U , X 2 and U 2 . The T M 1 N 1 ,...,MrNr may thus be chosen to be traceless, with the symmetry of two row traceless O (d, 2) Young tableaux, (2.34) A generic killing tensor can therefore be parameterised by the following combination of null orthogonal auxiliary vectors w + and w − , where for ease of notation we have defined the following scalar contractions Computing the structure constants by evaluating (2.28) on Killing tensors (2.36) then boils down to an iterative application of the chain rule via where we further defined C ij := w + i · w − j . The . . . denote terms which may be neglected, as they are fixed by the symmetries of (2.34). Employing the operator identities (2.38), in the following we extract the corresponding global symmetry structure constants.
Structure constants
To extract the global symmetry structure constants induced by the holographic cubic action (2.17), we evaluate the bracket (2.28) on the gauge parameters corresponding to Killing tensors (2.36). By first considering the action of the projection operator Π E in (2.28), a simplification is given by noting that the lowest derivative part of the commutator [[Ē 1 , corresponding to the correct degree of homogeneity in X. The remaining higher-derivative terms must be dressed by factors of X 2 to match the degree of homogeneity, and therefore just give rise to trace terms which can be set to zero in the quotient (2.33). The gauge-algebra commutator on the Killing tensors is thus the lowest derivative monomial Evaluating the derivatives in above to obtain its explicit form is straightforward, but lengthy in general. One can proceed by expanding every term, performing all differentiations and re-summing. The final result can be expressed in terms of four basic traces which parameterise the most general decomposition of the trace of the triple tensor product of three two-row traceless window Young tableaux
Invariant form and cyclic structure constants
In order to compare our result obtained via holography with the known structure constants for higher-spin lie algebras (see §A), we require the cyclic structure constants f s 1 ,s 2 ,s 3 . These can be obtained with the knowledge of the corresponding invariant form κ s,s , Without loss of generality, the invariant form can be chosen to take the diagonal form (2.44) for some constants b s , which can be fixed uniquely up to an overall coefficient by enforcing cyclicity of f s 1 ,s 2 ,s 3 .
For example, they can be determined simply by considering the structure constants induced by the minimal gravitational coupling, which entails solving the equation for the coefficients b s contained in the definition (2.43). In this way we obtain where the overall constant has been fixed by normalising the 1-1-1 structure constants to the identity. This leads to the diagonal bi-linear form The corresponding cyclic structure constants are then where the sum over l ranges over the odd integers if s 1 + s 2 + s 3 is even, and over even integers if s 1 + s 2 + s 3 is odd. The coefficientsf (a,b,c) are defined by .
For simplicity the sum over n and m is extended up to infinity owing to the poles of the Γ-functions in the denominators of P (k i ,l) m,n . Therefore only finitely many terms contribute to the above sums for a given triplet of spin. 11 Although the result (2.48) is lengthy, the same complicated expression is obtained from expanding the known generating functions for the structure constants of the hs (so (d − 1, 2)) higher-spin lie algebra (see §A). In particular, since the latter algebra is unique in generic dimensions, 12 this verifies that the cubic couplings (2.17) obtained holographically give the same deformations of the gauge symmetries as those which would be obtained from the Noether procedure at quartic order independently of holography. This extends to general dimensions the tree-level three-point function test [13] of higher-spin holography by Giombi and Yin in AdS 4 .
Higher-spin cubic couplings in 4d flat space
For the remainder of this article we turn to higher-spins in flat space. Higher-spin cubic couplings which solve the Noether procedure up to the second non-trivial order (quartic) were first studied in the early 90s by Metsaev [16,17], using light-cone methods. 13 In the lightcone gauge, the Noether procedure reduces to requiring the closure of the Poincaré generators 11 The ranges can be straightforwardly recovered by solving the inequalities: (2.51d) 12 In AdS3 and AdS5 there are one-parameter families of higher-spin algbras, and in these dimensions the structure constants (2.48) coincide with the known expressions for parameters which correspond to the symmetries of the free scalar theory on the boundary. I.e. the test is also passed in these cases. 13 This postdated the original cubic classification of [23,24], solving the Noether procedure at the first nontrivial order.
deformed by cubic interactions. In this way the light-cone Lagrangian can be read off from the non-linear deformation of the light-cone Hamiltonian. The quartic order analysis of [16,17] remarkably led to the complete fixing of the flat space cubic action in four dimensions. In this section we analyse this cubic theory and explore a putative underlying higher-spin algebra in four-dimensional flat space, extending the discussion carried out in the previous section.
Light-cone gauge
We first review the gauge fixing of higher-spin fields in flat space to light-cone gauge. It is convenient to work with the light-cone coordinates The corresponding space-time derivatives are given by By introducing auxiliary variables u µ = (u + , u − , u,ū), the usual Fierz system which packages higher-spin fields 14 The leftover gauge symmetry can be completely fixed by requiring ∂ + u ϕ(x, u) = 0, for which the system (3.5) becomes By solving the divergence condition (3.7b) Recall the generating function ϕ(x, u) encodes spin-s fields together with the traceless constraint (3.7c) this enables the fields to be expressed in terms of the two physical helicities in four-dimensions which are encoded by a pair of complex conjugate scalar fields
Cubic vertices in light-cone gauge
We now apply the dictionary §3.1 for expressing physical quantities in light cone gauge to the most general cubic vertex. At first non-trivial order in the Noether procedure (which leaves the relative coefficients unfixed), gauge invariant parity preserving cubic vertices in flat space have been classified in a manifestly covariant form in [22,54,55]. Their general structure in TT (traceless and transverse, c.f. §2.1) gauge is given in generating function notation by with Y i and G defined as These are the flat space analogues of the AdS building blocks (2.8) and (2.12). The light cone gauge-fixing can be carried out directly at the level of (3.11), and is achieved for any term in (3.10) in combination with (3.9) simply by replacing where we have introduced (anti-)holomorphic light-cone momenta P (P ):
Parity violating vertices in light-cone gauge
In four-dimensions the epsilon tensor can be used to construct parity violating gauge-invariant vertices. The basic parity violating structures are given (up to integration by parts) by with i, j and k cyclically ordered. The most general parity violating vertex will then also depend on the above additional structures, multiplied by an arbitrary parity even structure built from (3.11) introduced in the previous section.
The light-cone gauge fixing in this case is obtained through the replacement Parity violating deformations of Metsaev's solution [16,17] are expected to have fixed overall coefficients, with the only dependence being on one parity violating parameter analogous to the θ parameter in the AdS 4 theory (see e.g. [56]). In this paper we shall only consider parity preserving cubic couplings.
Metsaev's cubic action
In the previous section, we considered the light-cone gauge fixing of cubic structures (3.10) obtained by solving the Noether consistency conditions with manifest covariance at first nontrivial order. At this order, the relative coefficients between the independent cubic structures are unfixed. On the other hand, purely within the light-cone framework, Metsaev fixed the cubic action completely almost 25 years ago [16,17]. In terms of the light-cone momenta P andP defined by (3.14), it takes the rather simple form with the sum running over all integer helicities ±s. 15 . Above we have introduced the coupling constant l to dress the higher-derivative interactions. We discuss a few notable properties of the action in the following. For given triplet of spins (s 1 , s 2 , s 3 ), in (3.18) there are as many structures as the number of positive combinations ±s 1 ± s 2 ± s 3 , giving rise to three distinct cases: There are two possible such combinations, one with 3s derivatives and another with s derivatives. The latter for s = 1, 2 reproduces the standard Yang-Mills self interaction and the Einstein-Hilbert minimal coupling, respectively.
2. s 1 = s 2 = s and s 3 = s Here the first unexpected feature emerges. In this case there are three different couplings, with 2s + s 3 , |2s − s 3 | and s 3 derivatives. This is in contrast to the expected number (two) of couplings, obtained from a covariant classification [20,55]. For example, here in the gravitational case s − s − 2 we have the expected couplings with 2s + 2 and 2s − 2 derivatives, but also a third with 2 derivatives. The latter may be considered as the gravitational minimal coupling and was referred as exotic in [57]. In the following sections, by examining corresponding deformations of the gauge symmetries, we will 15 It is interesting to note that the Γ-function coupling constant ensures strict light-cone locality: Whenever s1 + s2 + s3 < 1 or −s1 − s2 − s3 < 1, the coefficient of a putative "non-local" structure (i.e. with 1/∂ ± x ) vanishes identically.
argue that this two-derivative coupling may indeed be given the interpretation of minimal coupling.
3. s 1 = s 2 = s 3 This case gives even more surprises, where the number of independent couplings grows up to four, with s 1 + s 2 + s 3 , |s 1 + s 2 − s 3 |, |s 2 + s 3 − s 1 | and |s 3 + s 1 − s 2 | derivatives. This is to be compared with the covariant classification of cubic couplings, where only two couplings could be identified with s 1 + s 2 + s 3 and s 1 + s 2 + s 3 − 2s min derivatives. The additional couplings were also referred as exotic in [57].
In the following sections we study the "additional" vertices highlighted in 2. and 3. above and their possible implications.
Covariantising Metsaev's vertices
In this section we revisit the covariant classification of cubic vertices (c.f. §3.2), with the aim of accommodating the additional vertices reviewed in the previous section, which were discovered in the light-cone gauge. The latter were previously unaccounted for in the original covariant classification [18,20,22,55]. Furthermore, in this way we may apply covariant methods such as those in §2.2 and §2.3 for investigating a putative flat space higher-spin algebra.
In flat space, 16 the most general gauge invariant cubic structure can be parameterised by the building blocks for fixed external spins (s 1 , s 2 , s 3 ). Note that the structures (3.19) are polynomial in the oscillators only if k ≤ min(s 1 , s 2 , s 3 ). If k > min(s 1 , s 2 , s 3 ), the covariant expression is still formally gauge invariant and the light-cone gauge-fixing described in §3.2 can be formally applied.
The key observation, which was not considered in the original covariant classification, is that although the structures (3.19) for k > min(s 1 , s 2 , s 3 ) are formally non-polynomial in the oscillators ∂ u and∂ u , all non-polynomial dependence cancels out after gauge fixing to the light-cone. 17 More explicitly, employing the dictionary given in §3.2, on the light-cone the structures (3.19) for general k read 18 16 c.f. §2.1 for the AdS analogue. In particular, they differ by a factor e D where D is the differential operator (2.14) which generates corrections to the flat space result from non-zero curvature. 17 This essentially due to the factorised form of the light cone traceless condition whose solution are either holomorphic or anti-holomorphic in the variable u. 18 For simplicity we display the structures proportional to∂ s 1 , the remaining two can be obtained analogously.
which are indeed polynomial in ∂ u and∂ u . In the following we thus relax the constraint k ≤ min(s 1 , s 2 , s 3 ), and explore the structures in (3.19) which may accommodate consistent cubic interactions.
We first note that the structure f (k) s 1 ,s 2 ,s 3 is holomorphic only for k = s 3 or k = s 1 + s 2 , regardless if s 3 = s min or not. The non-holomorphic terms can either be removed by a local field redefinition when they are proportional to PP , or by a non-local but admissible redefinition (à la §3.5) when they are of the form P n /P m orP n /P m with both n = 0 and m = 0. The remaining terms which cannot be removed are non-local, but can be avoided by placing restrictions on the value of k as we discuss below.
Locality
Although in the above we relaxed the constraint k ≤ min(s 1 , s 2 , s 3 ), locality places restrictions on the range of k, which we consider here.
The only non-local terms which cannot be removed by admissible redefinitions, and which would give rise to a singular S-matrix, are those of the type 1 P nP m for n ≥ 0 and m ≥ 0 excluding the constant (for the details, see §3.5). This can be avoided requiring which upon cyclising the indices can be rewritten as With the locality condition (3.24) satisfied and allowing field redefinitions of the type described in §3.5, we are left with only holomorphic or anti-holomorphic terms which cannot be removed by a redefinition. Discarding couplings which do not give rise to (anti-) holomorphic structures gives the following list of covariant couplings for fixed spin: with a number of different local couplings equal to the number of unequal spins plus one. The two couplings which fall into the original covariant classification of [18,20,22,55] are given by
Covariant cubic action
Combining the above light-cone → covariant dictionary, we obtain the following (formal) 19 covariant rewriting of Metsaev's vertices up to the class of re-definitions given in §3.5: 20 (3.28) 19 We emphasise that this re-writing of Metsaev's vertices is strictly formal, and serves primarily as an auxiliary step to extract the higher-spin structure constants. On the other hand, it is possible to enlarge the functional space of polynomials φ µ(s) (x)u µ(s) and allow 1/Y poles, in spite of the lack of tensorial interpretation. Within such a non-tensorial functional space, (3.28) represents the covariantisation of Metsaev vertices. 20 The kinetic term is normalised with 1 2 s s! for convenience.
with j spanning the three values satisfying We now comment on the properties of (3.28) in contrast to the original covariant classification, in which additional vertices highlighted in §3.3 did not appear. As may be anticipated, the covariant form of these additional vertices contain poles in Y i . For example, the two-derivative gravitational coupling is given explicitly by with coupling constant fixed in (3.28). Notice that all spin-s two-derivative gravitational coupling constants are indeed equal : and spin-independent in accordance with the equivalence principle [25]. We emphasise that that all apparent singularities of the above non-polynomial solutions to the Noether procedure disappear upon gauge fixing to the light cone in 4d. This suggests that the non-local singular covariant form (3.30) might just be an artifact of choosing not to introduce auxiliary fields to solve for gauge consistency. Indeed, the reason why the above vertices were overlooked in the original treatment is that they do not admit a standard tensorial form. Taking into account these caveats, let us stress that we only use this rewriting as a formal trick to extract the structure constants using covariant methods (such as those in §2). As we demonstrated above, a non-singular formulation of the exotic vertices is currently only available in a Lorentz non-covariant frame (i.e. in light-cone gauge).
Light-cone locality
In this section we detail the class of field-redefinitions used to obtain the formal covariant re-writing (3.28) of the light-cone cubic action (3.18). Using the identity: we can see that the combination PP (up to total derivatives) is proportional to the equations of motion, and for this reason can be removed by a field redefinition. In particular, going on-shell in light cone gauge is equivalent to setting The above equation factorises in four dimensions, 21 and so it can be solved in two possible ways: 22 Notice that for both P = 0 andP = 0 no non-singular and non-trivial solution can be written down. Interestingly, the above observation implies that formally one can relax the light-cone locality condition which requires the vertex to be a polynomial in P andP (at least for any fixed triple of spins), since a non-singular branch in (3.34) can always be chosen if one of them is zero on-shell. Using this observation, we show that in this case there exists an enlarged class of field re-definitions which leave the S-matrix invariant. These are not globally defined as they are singular for generic on-shell configurations, however they are non-singular on one branch (3.34) of the on-shell surface at a time. In §3.4 this enlarged class of re-definitions enabled a formal covariantisation of the exotic light-cone vertices in a particular field frame.
In order to discuss these issues, we recall the important requirement that the S-matrix of a theory should be finite for generic on-shell configurations. Since combinations of the type PP vanish on-shell usually in the light cone gauge one has only holomorphic or anti-holomorphic local vertices. For fixed external spins, the S-matrix in this case is thus polynomial in the light-cone momenta P andP .
In four-dimensions, the factorisation property (3.34), which gives a factorised on-shell surface: permits a wider class of vertices: Consider a vertex of the type 23 P n P m , with m > 0, (3.37) which are non-polynomial in one of the light-cone momenta. There are two distinct cases to consider: n = 0 and n > 0. For n = 0, while such vertices yield a non-singular S-matrix on the branch {P = 0}, they are singular on the branch {P = 0} and are thus excluded. For n > 0, however, there is a crucial difference: Although this type of vertex is also singular on {P = 0}, on the branch {P = 0} they are proportional to the equations of motion: covariance. 22 On-shell this recovers in disguise the well-known holomorphic structures usually found in the spinor-helicity formalism (see e.g. [58] and also §4.). 23 The discussion proceeds in the same way for vertices of the form P n P m . (3.36) and can be removed by a field redefinition. This redefinition is not globally defined, but however finite on one branch of the on-shell surface. Motivated by this observation, it seems reasonable to allow non-local vertices of the type (3.37) for n > 0, which arise from choices of the field-variables which may not be globally defined on the full on-shell surface but still well-defined on either of the branches {P = 0} and {P = 0}. Let us stress that the above functional class, although enlarged compared to the generic case, allowing such singular redefinitions do not remove on-shell non-trivial local vertices. For example a vertex proportional to P n which lives on the {P = 0} branch of the stationary surface cannot be removed by a redefinition of the typeP /P on the same branch. In particular, multiplying or dividing byP is not allowed in this functional space for field configurations P ≈ 0.
Deformations of Gauge Symmetries from Metsaev vertices
With a covariant form (3.28) of the couplings [16,17] established, we can extract the corresponding deformations of the gauge transformations and their commutators by employing covariant formulas, as in the AdS case §2. We further extract the structure constants of a putative higher-spin algebra and discuss the result.
Gravitational coupling of higher-spins in flat space
We first consider the gravitational coupling of spin-s gauge fields, in particular in the view of the two-derivative s-s-2 coupling highlighted in §3.3. We extract the structure constants of the semi-simple (higher-spin Lorentz) subalgebra of the putative higher-spin algebra, and argue that the latter two-derivative coupling can be interpreted as a minimal coupling of higher-spin gauge fields to gravity.
To this end it is straightforward to apply the same techniques employed for the AdS case in §2.2. The deformed gauge bracket is where via integration by parts all derivatives are made to act on the gauge parameters, and Π ξ enforces tracelessness and the correct homogeneity degree in x.
To determine the would-be higher-spin algebra structure constants, we evaluate (3.39) on Killing tensorsξ, u · ∂ xξ (x, u) = 0. (3.40) In Minkowski space the Killing tensors are given by a set which transform as two-row Young tableaux: The generalised higher-spin Lorentz generators correspond to those with k = s − 1, while those with k < s − 1 acquire a natural interpretation in terms of generalised hyper-translation generators. For simplicity we restrict to the former, where the higher-spin gravitational coupling should give rise to the structure constants of the type f 2ss . These specify the transformation properties of the spin-s generators under the Lorentz part of the isometry.
Owing to the inclusion of the "additional" two-derivative couplings, the following subtlety must be considered. Since these vertices are singular in covariant form, the deformed structure constants will involve terms which are non-polynomial in the Y i variables. This leads to singular expressions when considering a contraction with polynomial type functions, such as those in (3.41): We adopt the prescription to simply drop the singular terms (which we justify below), and thus neglect them in the sequel. With this prescription we fix the bi-linear form as where the coefficients b s,k determined by requiring cyclicity of the corresponding 2-s-s structure constants. For the higher-spin Lorentz subalgebra (with k = s − 1), these are where, as for the AdS case in §2.3, the overall constant has been fixed by normalising the 1-1-1 structure constants to the identity. As a consistency check of our prescription for dealing with singular terms, (3.44) precisely reproduces the result obtained in the AdS 4 theory in (2.46) when normalising the kinetic term canonically. We also note that the bi-linear form is non-degenerate precisely due to the contribution of the lower-derivative exotic couplings. The fact that from (3.42) we obtain the same expression as for the f 2s 1 s 2 structure constants (2.48) in AdS 4 suggests that the additional two-derivative s-s-2 vertices can be interpreted as minimal couplings of spin-s gauge fields to gravity in flat space. Furthermore, this agreement suggests that the additional vertices we observe in the light-cone gauge should not be considered as true independent additional vertices. Indeed, there exists a unique combination of the standard local vertices and the exotic lower derivative ones: which admit an invariant bilinear form for the generalised Lorentz subalgebra of the putative higher-spin algebra. In AdS 4 , due to the non-commutative nature of covariant derivatives, gauge invariance fixes such lower derivative vertices in combination with higher-derivative vertices. The fact that they appear to be independent vertices in flat space could be related to the singular nature of the flat-limit. Therefore, in flat space they look singular and they need to be added by hand, but the singularity disappears upon considering a gauge fixing to the light cone gauge in four-dimensions. For any given triplet of non-zero spins, we thus end up with one abelian higher-derivative vertex and one lower derivative cubic vertex which is a linear combination of standard local vertices and exotic ones (quasi-minimal coupling). The relative coefficients can be fixed by the requirement that the higher-spin Lorentz subalgebra admits an invariant bilinear form and the solution to quartic consistency precisely fulfils this requirement. One can speculate that upon introducing auxiliary fields these vertices may be rewritten in local form. The analysis presented above may be interpreted as a hint that the corresponding theory has an underlying higher-spin symmetry, a possibility which we discuss further in the following section.
Is there a higher-spin algebra underlying Metsaev's vertices?
In this section we give the extension of the result in the previous section for the s-s-2 structure constants to the generic case of s 1 -s 2 -s 3 .
Since the computation is intrinsically four-dimensional, the following dimension dependent identity should be employed: m,n π (−1) m (k1 + k2 + l + 1)(k1 + k3 + l + 1)(k2 + k3 + l + 1) (3.48) , . (3.49) This result precisely coincides with the same structure constants (2.49) in the AdS 4 theory, and thus illustrates that our formal covariantisation (3.28) of the cubic vertices in [16,17] uncovers the full Lorentz part of the higher-spin symmetry. The latter can also be rewritten in terms of the Moyal product in the enveloping algebra construction for sl(2, C). We emphasise that the result (3.47) crucially relies on lower-derivative exotic couplings, whose covariant form might require the addition of auxiliary field to be reduced to a standard local formulation. The fact that the above higher-spin Lorentz structure constants precisely coincide with the AdS 4 higher-spin lorentz structure constants may also hint towards the existence of a welldefined relation between the theory in AdS 4 and in flat space. This is compatible with the existence of a contraction of the AdS 4 higher-spin algebra which naturally preserves its Lorentz part. In this respect it is important to make an analogous study of the hypertranslationtype global symmetries, which are not preserved in such a contraction. While we have not been able to determine the full list of structure constants for the hypertranslations, we have checked a number of lower spin examples. This would further clarify whether there exists an infinite dimensional extension of Poincaré algebra behind the cubic couplings (3.18). Should it prove that the Metsaev theory is governed by a flat space higher-spin algebra, two main open questions then arise: First if the algebra advocated above can be realised as a contraction of the AdS 4 higher-spin algebra. Second, if there exists an oscillator realisation of it based on a universal enveloping algebra construction. Some attempts in this direction can be found in [59], where some issues were also pointed out in looking for a proper way to factor the trace ideal. Some of the obstructions found in previous literature might be overcome via dimensional dependent identities, while they are expected to remain in d > 4.
Spinor-helicity Formalism
Recently there has been a renewed interest in the spinor-helicity formalism (see [60,61] for reviews on the subject) in the context of both massless [58] and massive [62] higher-spins, with progress so far restricted to cubic amplitudes. Owing to the tight relation between lightcone and spinor helicity formalism (see for instance [29,63] in the context of higher-spins), in this section we revisit this analysis in the light of our results presented in the preceding section. This is complementary to the recent work [64], which studied the relation between cubic vertices in the original covariant classification (i.e. not accounting for the exotic lower derivative vertices of Metsaev considered in the present work) and three-point spinor helicity amplitudes.
A key feature of the spinor-helicity formalism is that the on-shell conditions for massless fields can be solved without giving up manifest covariance. For example, in this formalism the solution to the massless scalar Klein-Gordon equation is where we introduce two-component spinors λ a andλȧ, with λ | η = λ a η b ab and [λ |η] =λȧη˙b ȧḃ . Here, we have the following action of the little group on the polarisation tensors (see e.g. [58] and references therein for further details): In the following we review the generalisation of the above setting to higher-spin fields. For convenience, in four space-time dimensions one works with the fundamental representations (1/2, 0) and (0, 1/2) rather then with the usual vector oscillators. Hence we need to solve the following Fierz system: while the traceless condition becomes automatic in this formalism due to the symmetrised form of the indices and to the antisymmetric nature of the αβ . As is standard in dealing with higher-spin fields, we introduce the generating function notation: ..αs,α 1 ...αs (x) χ α 1 · · · χ αs χα 1 · · · χα s , (4.4) in terms of which the Fierz system reads In this language the above equations can be solved in terms of a reference spinor which we denote by q α andqα. Going to momentum space and solving the mass-shell condition as one gets the following general solution to the Fierz system: Let us note that the dependence on the auxiliary spinor is exactly compensated by the left-over on-shell gauge invariance; no dependence on the auxiliary spinor remains at the level of the amplitude. Indeed it is straightforward to prove that: while the on-shell gauge invariance reads in this formalism: The problem of writing couplings modulo field redefinitions can be then posed at the level of the fields where the generic coupling has the form . (4.11) A key point of the above expression is the GL(1) invariance which must be imposed on the function C together with gauge invariance. In order to properly study the above problem it is useful to first determine the identities among the various λ i that are implied by momentum conservation.
To begin with, momentum conservation implies either 12 = 23 = 31 = 0 , where IJ = λ I,a λ J,b ab and [IJ] =λ I,ȧη J,ḃ ȧḃ . The above precisely reduce to holomorphicity of the amplitudes recovered in the light-cone gauge. Furthermore, due to over-antisymmetrisation one also gains the identity (4.14) At this point the possible building blocks are then given by In the case that theλ are proportional to each other, momentum conservation further implies 16) and similarly for the anti-holomorphic components. This identity, together with the divergenceless condition and GL(1) invariance, reduces the number of independent building blocks to the following: P 3 = 12 [12] , P 1 = 23 [23] , P 2 = 31 [31] , while any other building blocks can be expressed in terms of the above modulo Fierz identities. Making use of the following useful identities valid for any reference momentum q: together with their anti-holomorphic counterparts, we can now construct the couplings, classifying them depending on the helicity involved: i.e. (+++), (++−), (+−−) and (−−−). One observes that for each helicity combination only one particular function C gives a non-vanishing result.
• (+ + +): In this case the non vanishing coupling is recovered from • (+ + −): (4.20) • (+ − −): While in restricting attention to the above building blocks gauge invariance is manifest, the above results are in complete agreement with those found in the previous sections in the lightcone gauge. 24 In more detail, using the dictionary §3.2 to go from covariant cubic structures to the light-cone, combined with the above we can go straight from light-cone gauge to the spinor-helicity formalism. This is a one-to-one map, and thus resolves the puzzle regarding the mis-match between the original covariant classification of cubic vertices and three-point 24 To see this one needs to employ the identity S-matrix structures reviewed in [64]. Notice in particular the explicit non-local form of the vertices (4.21) and (4.20) involving one opposite helicity, as observed for the covariant counterparts of the exotic light-cone vertices in §3.4. Let us stress that in [64] it was observed that in mapping higher-derivative local higherspin couplings to the spinor-helicity formalism lower derivative structures appear as total derivatives, up to terms proportional to the equations of motion. This is to be expected, for in the ambient space formalism the minimal coupling is precisely generated when considering the radial reduction of such total derivatives [35]. On the other hand, in this work we point out the existence of additional lower-derivative couplings which are non-singular only in fourdimensions and which reproduce the spinor-helicity structures without being multiplied by vanishing factors as in the case of [64]. The price to pay is a mild non-local form of the corresponding covariant expressions for the vertices. Furthermore, the corresponding functional class has been argued ( §3.5) to be fully compatible with the existence of non-trivial couplings avoiding the triviality argument of [41].
Higher-spin algebras and higher-order amplitudes
It is illuminating to study in more detail the consequences beyond cubic order of a possible higher-spin symmetry (i.e. in the case that the structure constants (3.47) yield a well-defined higher-spin algebra) behind Metsaev's cubic couplings. Similar investigations have been made in [42,65,66]. Likewise, it turns out that the higher-spin symmetry places very strong constraints on the momentum dependence of any 4pt amplitude. 25 We study the action of a hypertranslation on a higher-spin field, as obtained from the cubic couplings (3.28) extending the discussion of [66] to the Metsaev case. We first consider the deformation generated by the 0-r 1 -r 2 coupling with r 1 + r 2 derivatives, 26 which possesses the standard covariant form f (0) i.e. it does not originate from the additional exotic vertices, see §3.4. The corresponding deformations of the gauge transformations for a spin-r 3 field are given by: which rotates the spin-r 3 field into a scalar φ 2 through a hypertranslation ξ (0) r 1 (w). The latter however vanish identically if r 3 > 0. This follows from the following identity for hypertranslations: In the absence of exotic couplings with lower derivatives, this has a very simple consequence: It implies that the four-scalar amplitude should rotate into itself under hypertranslations ξ, In more detail, consider the following spin-r hyper-translation transformations of a scalar field where g i are the coupling constants entering the cubic action of the theory. The condition The above identity is simply a different incarnation of Weinberg result [25], but follows here as a consequence of higher-spin symmetry as opposed to the consideration of a soft limit for the external particles. In particular, given the arbitrariness of the gauge parameter ξ, this is equivalent to For spins r = 1 and r = 2 this enforces charge conservation and equivalence principle. For r > 2 and g i = 0 however, the it implies that the scalar amplitude itself must be a distribution concentrated on a measure zero set of kinematical configurations which allow the above identity to be satisfied. The amplitude must then be concentrated on kinematical configurations which solve These are the configurations in which the particles do not interact and where least one of the Mandelstam variables vanishes: A 0000 = a(s, t) δ(u) + a(t, u) δ(s) + a(u, s) δ(t) , Indeed, for instance u = 0 implies that t = −s and p 1 = −p 4 and p 2 = −p 3 , i.e. triviality, together with analogous results for other channels. In this illustrative u = 0 example we end up with where b is an arbitrary vector. This is satisfied if the g i are equal and their colored/charged legs satisfy the appropriate antisymmetry conditions for odd spins. We thus see that higher-spin symmetry forces the theory to have trivial S-matrix at quartic order, and that the standard Weinberg result is recovered from higher-spin symmetry.
We now re-consider the above discussion, but this time including the effect of the exotic couplings which are present in the cubic Lagrangian (3.28). Exotic coupling may indeed provide a way out to the above argument. Since Y 3 annihilates the hypertranslation generators, it is sufficient to consider the deformation to gauge transformations coming from 0-r 1 -r 2 exotic cubic couplings with no Y 3 dependence, namely where the . . . are singular terms proportional to Y −1 2 , which are non-singular upon gauge fixing and do not contribute on global symmetries. Since in this case the transformation (5.12) is non-vanishing, the four-scalar amplitude should not be invariant under higher-spin hypertranslations as in (5.4). By global higher-spin symmetry, it must be compensated by the variation of amplitudes involving a single spinning external leg, where the transformation acts on the latter external field: Assuming that there is no k000 exotic structure like for the k00 case, the most general form for a planar k-0-0-0 amplitude in the s and u channels compatible with Poincaré invariance reads [42]: 27 14) with f k an arbitrary function of one variable and with no pole in the complex plane. 28 The hypertranslation transformation of this amplitude reads where we sum over the action of the symmetry transformation on all external legs and without loss of generality restrict to the part of the variation which generates a 4-scalar structure (as the other structures are related to this one by higher-spin symmetry).
The assumption that f k (t) does not contain poles in t then implies that the contributions with k > 0 in (5.15) contain a number of derivatives N k which is bounded from below N k ≥ 2k + r − 1. The contribution with r derivatives (those in (5.15) for k = 0) thus cannot be compensated by any other amplitude. One can then argue that the only way to obtain an amplitude which is consistent with higher-spin symmetry is to have extending the previous result (5.9). Notice that the above conditions only arise for amplitudes with the number of external legs being greater than or equal to four. At cubic level there is no non-trivial Mandelstam invariant and non-trivial cubic couplings are compatible with higherspin symmetries. By higher-spin covariance, this result suggests that any 4-point amplitude is 27 We restrict our attention to amplitudes which can be dressed with Chan-Paton factors owing to the fact that Metsaev's solution (3.18) admits such an extension. The case without Chan-Paton factors is slightly more general, but the conclusions presented below for the case with Chan-Paton factors continue to hold. 28 This assumption follows from factorisation and unitarity.
proportional to a sum of δ-function distributions, reminiscent of AdS Mellin-amplitudes that can be extracted from free theories [47,48]. We emphasise that a key assumption of the above discussion is the absence of r-0-0-0 exotic structures. This is motivated by the absence of r-0-0 exotic structures, but should be verified.
Conclusions
In this paper we studied the higher-spin algebra structure constants induced by the action at cubic order of the type A minimal bosonic higher-spin theory on AdS d+1 space implied by holography [3]. The explicit form of the structure constants for the deformed gauge symmetries are obtained, together with the associated normalisation of the bilinear form. We show that these structure constants coincide with the known expressions, which are unique in generic dimensions. This demonstrates that the holographically reconstructed cubic action solves the Noether procedure up to quartic order, and extends the tree-level three-point function test of the higher-spin / vector model duality by Giombi and Yin in AdS 4 [13] to general dimensions.
We also considered the same problem for the cubic order action of the theory in 4d flat space found by Metsaev in [16,17]. The couplings themselves where obtained by solving the quartic consistency in the light-cone formulation, where higher-spin symmetry is not manifest. Remarkably, the couplings include lower derivative vertices which were not captured in the original covariant classification of cubic structures [18-20, 22, 55]. These include two-derivative couplings of higher-spin fields to gravity, which we argued to be minimal. After extracting the explicit form of the higher-spin structure constants, we argue in favour of a well-defined higher-spin algebra behind the cubic couplings. The existence of such a higher-spin algebra crucially relies on the additional couplings couplings in flat 4d Minkowski space, initially found in [23,24].
We end with a few summarising remarks and outlooks: • Extending the dicussion of §5, there are indeed various examples in the literature where the implications of higher-spin symmetry on higher-order amplitudes have been considered. To date there are compelling arguments that both conformal higher-spin theories in flat space [66,69] and higher-spin theories in AdS [47,48] 29 have trivial S-matrix-like observables. These have been shown to be proportional to delta-function-like distributions concentrated on measure zero space for kinematic configurations. Together with the same story in flat space §5, this seems to point towards higher-spin symmetry being incompatible with a non-trivial S-matrix, at least in all known examples.
• It would be interesting explore the possibility of other covariantisations of the exotic light-cone vertices, which may avoid the formal singularities obtained through the covariantisation prescription in this note. It is conceivable that this would only be possible upon introducing infinitely many auxiliary fields. 29 The the context of the AdS/CFT duality, the analogue of the S-matrix in AdS has been argued to be the Mellin transform of the dual CFT correlators [70][71][72][73][74]. The Mellin transform of correlation functions in a free CFT are ill-defined, since they are power-functions in the cross-ratios. However they can be formally defined as a δ-function distribution [75].
• Finally, another interesting direction would be to check the results obtained in this note directly in the light-cone gauge, skipping the step of covariantisation of the vertices. Some ideas in this direction have been discussed in [76]. Indeed the analysis carried out in this work demonstrates that a well defined formulation of the exotic vertices is only available so far in a Lorentz non-covariant frame. It would also be interesting to investigate the relations between this case and the case of self-dual forms where a similar covariantisation problem arises [77].
A Higher-spin algebra structure constants Higher-spin algebra structure constants for symmetric tensor fields can all be obtained via universal enveloping algebra constructions [11,53]: 30 hs(I) = U(so(d, 2)) I . (A.1) The above is usually realised through the use of oscillators by embedding the isometry algebra of AdS space into sp(2N ) for appropriate choices of N . For all known higher-spin algebras involving totally-symmetric fields, the quotient operation is conveniently realised by a quasiprojector ∆, [11,53]. The quasi-projector ∆ can be defined as a non-polynomial element of the universal enveloping algebra of sp(2N ) which by construction projects out the ideal and picks a well-defined representative in (A.1). Below we summarise the list of known higher-spin algebras in various dimensions and for totally symmetric fields. Apart from one parameter families arising in d = 3 and d = 5, the higher-spin algebra for totally symmetric tensors is unique.
The metaplectic representation The starting point to treat all higher-spin algebras in Table 1 in a unified fashion is given by the metaplectic representation of sp(2N ) which is defined by: Dim.
with C AB the sp-invariant tensor, C AC C BC = δ B A , which is used to raise and lower the indices according to Q A = C AB Q B . Above we introduced the Moyal -product in the Weyl-ordering, whose integral representation is so that discarding boundary terms one recovers The sp(2N ) generators read and following [11], it is convenient to introduce Gaussian generating functions for universal enveloping algebra elements of the type Above, the auxiliary variables U AB are assumed to be factorisable: U AB = u A u B , so that U A[B U C]D = 0 and the tracelessness of the generators is automatic U A B U B C = 0. Finally, the above oscillator realisation admits a unique supertrace [78]: The -monoid The oscillator realisation introduced above enables the structure constants of the various higher-spin algebras to be encoded in terms of the unique star-product and trace operation defined in sp(2N ). To this effect it is useful to recall the Monoid structure of Gaussians under the star product, which reproduces the sp-product up to a Caley transform [79] F (S 1 ) F (S 2 ) = F (S 1 S 2 ) ↔ F (S) = 1 where by definition S ≡ S A B and S 1 S 2 is the standard matrix multiplication S 1A B S 2B C (recall that indices are raised and lowered with the invariant sp-tensor C AB ). The above monoid structure allows multiple -products to be evaluated without the need for any computation, except for simply inverting the Caley transform (A.10) For example: Quasi-projectors It is straightforward to restrict the universal enveloping algebra to its subalgebras. In the so(d, 2) case this simply amounts to the following choice of generators: However in doing so ideals emerge, which are generated by trace components. The simplest way to factor them out is to change the definition of the trace while keeping the same oscillator realisation induced by sp (2N ). This can be achieved by the following ansatz for the trace on the respective quotient [11,53]: . (A.13) The element ∆ g is defined by the requirement that it fixes a representative in the -product algebra.
Given the Howe dual pair (sp(2), so(d − 1, 2)) and considering the trivial representation for the sp(2), we have the ideal y i · y j ∼ 0. This ideal can be quotiented by the following quasi-projector [11]: 1−x y + ·y − . (A.14) The above Gaussian structure of the quasi-projector (first obtained in [53], but in a different form), makes it possible to extract all structure constants for the respective higher-spin algebras from the corresponding sp(2N ) structure constants.
hs[so(N )] structure constants Evaluating the determinant (A.11) (see [11]) gives: Tr e 2ρ y − α y + in terms of the contractions of eq. (2.41). Evaluating the integrals in the quasi-projector then gives the following bilinear form: | 13,689.2 | 2016-09-04T00:00:00.000 | [
"Mathematics",
"Physics"
] |
GW170608: Observation of a 19-solar-mass Binary Black Hole Coalescence
On June 8, 2017 at 02:01:16.49 UTC, a gravitational-wave signal from the merger of two stellar-mass black holes was observed by the two Advanced LIGO detectors with a network signal-to-noise ratio of 13. This system is the lightest black hole binary so far observed, with component masses $12^{+7}_{-2}\,M_\odot$ and $7^{+2}_{-2}\,M_\odot$ (90% credible intervals). These lie in the range of measured black hole masses in low-mass X-ray binaries, thus allowing us to compare black holes detected through gravitational waves with electromagnetic observations. The source's luminosity distance is $340^{+140}_{-140}$ Mpc, corresponding to redshift $0.07^{+0.03}_{-0.03}$. We verify that the signal waveform is consistent with the predictions of general relativity.
corresponding to redshift 0.07 +0.03 −0.03 . We verify that the signal waveform is consistent with the predictions of general relativity. * Deceased, February 2017. † Deceased, December 2016.
INTRODUCTION
The first detections of binary black hole mergers were made by the Advanced Laser Interferometer Gravitational-Wave Observatory (LIGO) (Aasi et al. 2015;Abbott et al. 2016a) during its first observing run (O1) in 2015 (Abbott et al. 2016b,c,d). Following a commissioning break, LIGO undertook a second observing run (O2) from November 30, 2016 to August 25, 2017, with the Advanced Virgo detector (Acernese et al. 2015) joining the run on August 1, 2017. Two binary black hole mergers (Abbott et al. 2017a,b) and one binary neutron star merger (Abbott et al. 2017c) have been reported in O2 data. Here we describe GW170608, a binary black hole merger with likely the lowest mass of any so far observed by LIGO.
GW170608 was first identified in data from the LIGO Livingston Observatory (LLO), which was in normal observing mode. The LIGO Hanford Observatory (LHO) was operating stably with sensitivity typical for O2, but its data were not analyzed automatically as the detector was undergoing a routine angular control procedure (Section 2 and Appendix A). Matched-filter analysis of a segment of data around this time revealed a candidate with source parameters consistent between both LIGO detectors; further offline analyses of a longer period of data confirmed the presence of a gravitational wave (GW) signal from the coalescence of a binary black hole system, with high statistical significance (Section 3).
The source's parameters were estimated via coherent Bayesian analysis (Veitch et al. 2015;Abbott et al. 2016e). A degeneracy between the component masses m 1 , m 2 prevents precise determination of their individual values, but the chirp mass M = (m 1 m 2 ) 3/5 (m 1 + m 2 ) −1/5 is well measured and is the smallest so far observed for a merging black hole binary system, with the total mass M = m 1 + m 2 also likely the lowest so far observed (Section 4). Individual black hole spins are poorly constrained, however we find a slight preference for a small positive net component of spin in the direction of the binary orbital angular momentum.
In combination with GW151226 (Abbott et al. 2016c), this system points to a population of black hole binaries with component masses comparable to those of black holes found in X-ray binaries (Section 5) and significantly below those seen in other LIGO-Virgo black hole binaries.
We also test the consistency of the observed GW signal with the predictions of general relativity (GR); we find no deviations from those predictions.
DETECTOR OPERATION
The LIGO detectors measure gravitational-wave strain using dual-recycled Michelson interferometers with Fabry-Perot arm cavities (Aasi et al. 2015;Abbott et al. 2016a). During O2, the horizon distance for systems with component masses similar to GW170608-the distance at which a binary merger optimally oriented with respect to a detector has an expected signal-tonoise ratio (SNR) of 8 (Allen et al. 2012;Chen et al. 2017)-peaked at ∼1 Gpc for LLO, and at ∼750 Mpc for LHO.
At the time of GW170608, LLO was observing with a sensitivity close to its peak. LHO was operating in a stable configuration with a sensitivity of ∼650 Mpc; a routine procedure to minimize angular noise coupling to the strain measurement was being performed (Kasprzack & Yu 2016). Although such times are in general not included in searches, it was determined that LHO strain data were unaffected by the procedure at frequencies above 30 Hz, and may thus be used to identify a GW source and measure its properties. More details on LHO data are given in Appendix A.
Similar procedures to those used in verifying previous GW detections (Abbott et al. 2017b) were followed and indicate that no disturbance registered by LIGO instrumental or environmental sensors (Effler et al. 2015) was strong enough to have caused the GW170608 signal. Calibration of the LIGO detectors is performed by inducing test-mass motion using photon pressure from modulated auxiliary lasers (Karki et al. 2016;Abbott et al. 2017d;Cahillane et al. in preparation). The maximum 1-σ calibration uncertainties for strain data used in this analysis are 5% in amplitude and 3 • in phase over the frequency range 20-1024 Hz.
The Advanced Virgo detector was, at the time of the event, in observation mode with a horizon distance for signals comparable to GW170608 of 60−70 Mpc. This was however during an early commissioning phase with still limited sensitivity, therefore Virgo data are not included in the analyses presented here.
SEARCH FOR BINARY MERGER SIGNALS
3.1. Low latency identification of a candidate event GW170608 was first identified as a loud (SNR ∼9) event in LLO data, via visual inspection of singledetector events from a low-latency compact binary matched filter ('template') analysis (Usman et al. 2016;Nitz et al. 2017b,a). Such events are displayed automatically to diagnose changes in detector operation and in populations of non-Gaussian transient noise artifacts (glitches) (Abbott et al. 2016f). Low-latency templated searches (Cannon et al. 2015;Messick et al. 2017;Adams et al. 2016;Nitz et al. 2017b) did not detect the event 3.1 Low latency identification of a candidate event Normalized energy Figure 1. Time-frequency power maps of LIGO strain data at the time of GW170608. The characteristic upwardschirping morphology of a binary inspiral driven by GW emission is visible in both detectors, with a higher signal amplitude in LHO. This figure, and all others in this Letter, were produced from noise-subtracted data (Section 4).
with high significance because LHO data were not analysed automatically. The morphology of the LLO event is consistent with a compact binary merger signal, as shown in Figure 1 (lower panel), but a noise origin could not be ruled out using LLO data alone. Consequently, LHO data were investigated and were determined to be stable at frequencies above 30 Hz (see Appendix A). A segment of LHO data around the event time was then searched with a filter starting frequency of 30 Hz, using templates approximating the waveforms from compact binary systems with component spins aligned with the orbital angular momentum (Bohé et al. 2017;Pürrer 2016). The fraction of SNR expected to be lost due to imposing the 30 Hz cutoff, as compared to the lower starting frequencies typically used in O2 data (Dal Canton & Harry 2017), is ∼1% or less. An event was found having consistent template binary masses and spins, times of arrival and SNRs in LHO and LLO. Based on this 2detector coincident event an alert was issued to electromagnetic observing partners 13.5 hours after the event time, with a sky localization (Singer & Price 2016) covering 860 deg 2 (90% credible region). GRB Coordinates Network Circulars related to this event are archived at https://gcn.gsfc.nasa.gov/other/G288732.gcn3.
Offline search
To establish the significance of this coincident event, a period between June 7, 2017 and June 9, 2017 was identified for analysis during which both LIGO interferometers were operating in the same configuration as at the event time. Times at which commissioning activities at LHO produced severe or broad-band disturbances in the strain data were excluded from the analysis. Standard offline data quality vetoes for known environmental or instrumental artifacts were also applied, resulting overall in 1.2 days of coincident LHO-LLO data searched.
Two matched filter pipelines identified GW170608, with a network SNR of 13. An candidate event is assigned a ranking statistic value, in each pipeline, that represents its relative likelihood of originating from a GW signal vs. from noise. One pipeline estimates the noise background using time-shifted data (Usman et al. 2016), finding a rate of occurrence of noise events ranked higher than GW170608 of less than 1 in 3,000 years. This limit corresponds to the maximum background analysis time available from time shifts separated by 0.1 s. The other pipeline uses different methods for ranking candidate events and for estimating the background (Cannon et al. 2015;Messick et al. 2017) and assigns the event a false alarm rate of 1 in 160,000 years.
A search for transient GW signals coherent between LHO and LLO with frequency increasing over time, without using waveform templates (Klimenko et al. 2016), also identified GW170608 with a false alarm rate of 1 in ∼30 years; the lower significance is expected as this analysis is typically less sensitive to lower-mass compact binary signals than matched filter searches.
Binary Parameters
The parameters of the GW source are inferred from a coherent Bayesian analysis (Veitch et al. 2015;Abbott et al. 2016e) using noise-subtracted data from the two LIGO observatories. Noise subtraction is a data processing step that removes several instrumental noise sources from the GW strain measurements (Abbott et al. (2017b) and references therein), thus increasing the expected SNR of compact binary signals in LHO data by typically 25% (Driggers et al. 2017). The likelihood integration is performed starting at 30 Hz in LHO and 20 Hz in LLO, includes marginalization over strain calibration uncertainties (Farr et al. 2015), and uses the noise power spectral densities (Littenberg & Cornish 2015) at the time of the event.
Two different gravitational-wave signal models calibrated to numerical relativity simulations of general relativistic binary black hole mergers (Mroue et al. 2013; Table 1. Source properties for GW170608: given are median values with 90% credible intervals. Source-frame masses are quoted; to convert to detector frame, multiply by (1 + z) (Krolak & Schutz 1987). The redshift assumes a flat cosmology with Hubble parameter H0 = 67.9 km s −1 Mpc −1 and matter density parameter Ωm = 0.3065 (Ade et al. 2016 (Pretorius 2005;Campanelli et al. 2006;Baker et al. 2006), are used. One waveform family models the inspiral-merger-ringdown signal of precessing binary black holes (Hannam et al. 2014), which includes spin-induced orbital precession through a transformation of the aligned-spin waveform model of Khan et al. 2016); we refer to this model as the effective precession model. The other waveform model describes binaries with spin angular momenta aligned with the orbital angular momentum (Bohé et al. 2017;Pürrer 2016), henceforth referred to as non-precessing. For their common parameters, both waveform models yield consistent parameter ranges. A selection of inferred source parameters for GW170608 is given in Table 1; unless otherwise noted, we report median values and symmetric 90% credible intervals. The quoted parameter uncertainties include statistical and systematic errors from averaging posterior probability samples over the two waveform models. As in Abbott et al. (2017a), our estimates of the mass and spin of the final black hole, the total energy radiated in GWs as well as the peak luminosity are computed from fits to numerical relativity simulations (Hofmann et al. 2016;Keitel et al. 2017;Healy & Lousto 2017;Jiménez-Forteza et al. 2017).
The posterior probability distributions for the sourceframe mass parameters of GW170608 are shown in Figure 2, together with those for GW151226 (Abbott et al. (Abbott et al. 2016c). In the top panel, we further compare GW170608 and GW151226's source-frame total mass (left) and source-frame chirp mass (right). All other known binary black holes lie at higher chirp masses than GW170608 and GW151226.
Binary Parameters
The probability that GW170608's total mass is smaller than GW151226's is 0.89. While the chirp mass is tightly constrained, spins have a more subtle effect on the GW signal. The effective inspiral spin χ eff , a mass-weighted combination of the spin components (anti-)aligned with the orbital angular momentum (Racine 2008;Ajith et al. 2011), predominantly affects the inspiral rate of the binary but also influences the merger. We infer that χ eff = 0.07 +0.23 −0.09 disfavoring large, anti-aligned spins on both black holes.
An independent parameter estimation method comparing LIGO strain data to hybridized numerical relativity simulations of binary black hole systems with non-precessing spins (Abbott et al. 2016g) yields estimates of component masses and χ eff consistent with our model-waveform analysis.
Spin components orthogonal to the orbital angular momentum are the source of precession (Apostolatos et al. 1994;Kidder 1995), and may be parameterized by a single effective precession spin χ p (Schmidt et al. 2015). For precessing binaries, component spin orientations evolve over time; we report results evolved to a reference GW frequency of 20 Hz. The spin prior assumed in this analysis is uniform in dimensionless spin magnitudes χ i ≡ c|S| i /(Gm 2 i ) with i = 1, 2 between 0 and 0.89, and isotropic in their orientation; this prior on component spins maps to priors for the effective parameters χ eff and χ p . The top panel of Figure 3 shows the prior and posterior probability distributions of χ eff and χ p obtained for the effective-precession waveform model. While we gain some information about χ eff , the χ p posterior is dominated by its prior, as for previous GW events (Abbott et al. 2016b(Abbott et al. ,c, 2017a, indicating that we cannot draw any strong conclusion on the size of spin components in the orbital plane ). The inferred component spin magnitudes and orientations are shown in the bottom panel of Figure 3. We find the dimensionless spin magnitude of the primary black hole, χ 1 , to be less than 0.75 (90% credible limit); this limit is robust to extending the prior range of spin magnitudes and to using different waveform models.
The measurability of precession depends on the intrinsic source properties as well as the angle of the binary orbital angular momentum to the line of sight (i.e. inclination). The inclination of GW170608's orbit is likely close to either 0 • or 180 • , due to a selection effect: the distance inside which a given binary merger would be detectable at a fixed SNR threshold is largest for these inclination values (Schutz 2011). For such values, the waveform carries little information on precession.
The distance of GW170608 is extracted from the observed signal amplitude given the binary's inclina- tion (Abbott et al. 2016e). With the network of two nearly co-aligned LIGO detectors, the uncertainty on inclination translates into a large distance uncertainty: we infer a luminosity distance of D L = 340 +140 −140 Mpc, corresponding to a redshift of z = 0.07 +0.03 −0.03 assuming a flat ΛCDM cosmology (Ade et al. 2016).
GW170608 is localized to a sky area of ∼520 deg 2 in the Northern hemisphere (90% credible region), determined largely by the signal's measured arrival time at LLO ∼7 ms later than at LHO. This reduction in area relative to the low-latency map is partly attributable to the use of noise-subtracted data with offline calibration (Abbott et al. 2017b).
Consistency with General Relativity
To test whether GW170608 is consistent with the predictions of GR, we consider possible deviations of coefficients describing the binary inspiral part of the signal waveform from the values expected in GR, as was done for previous detections (Abbott et al. 2016h,d, 2017a. Tests involving parameters describing the merger and ringdown do not yield informative results, since the merger happens at relatively high frequency where the LIGO detectors are less sensitive. As in Abbott et al. (2017b), we also allow a sub-leading phase contribution at effective −1PN order, i.e. with a frequency dependence of f −7/3 , which is absent in GR. The GR predicted value is contained within the 90% credible interval of the posterior distribution for all parameters tested.
Assuming that gravitons are dispersed in vacuum like massive particles, we also obtained an upper bound on the mass of the graviton comparable to the constraints previously obtained (Abbott et al. 2016b(Abbott et al. ,h, 2017a. Possible violations of local Lorentz invariance, manifested via modifications to the GW dispersion relation, were investigated (Abbott et al. 2017a), again finding upper bounds comparable to previous results.
ASTROPHYSICAL IMPLICATIONS
The low mass of GW170608's source binary, in comparison to other binary black hole systems observed by LIGO and Virgo, has potential implications for the binary's progenitor environment. High-metallicity progenitors are expected to experience substantial mass loss through strong stellar winds (Spera et al. 2015), implying that high-mass black hole binaries are not formed in such environments. GW170608's low mass suggests formation in a higher metallicity environment; however, formation at lower metallicity with comparatively lower mass progenitors is not excluded. Further discussion of the relationship between black hole masses and metallicity can be found in Abbott et al. (2016i).
At this lower boundary of the observed distribution of binary black hole masses, we can compare component black holes with those found in X-ray binaries. Xray binary systems contain either a black hole or neutron star which accretes matter from a companion donor star. Low-mass X-ray binaries (LMXBs) are X-ray binaries with a low-mass donor star which transfer mass through Roche lobe overflow (Charles & Coe 2003). The inferred component masses of GW170608 are consistent with dynamically-measured masses of black holes found in LMXBs, typically less than 10M (Özel et al. 2010;Farr et al. 2011;Corral-Santana et al. 2016).
Black holes in LMXBs are believed to form with near-zero spin and acquire spin as a byproduct of mass accretion (Fragos & McClintock 2015). For GW170608, we infer an effective spin probability distribution that is concentrated around zero with the 90% credible interval extending to small positive spin. Thus we can exclude highly negative anti-parallel spin components, while remaining broadly consistent with expected LMXB spins (Miller & Miller 2015;Fragos & McClintock 2015).
Binary black holes may form through many different channels, including, but not limited to, dynamical interaction (Rodriguez et al. 2016;O'Leary et al. 2016;Mapelli 2016) and isolated binary evolution (Belczynski et al. 2016;Eldridge & Stanway 2016;Lipunov et al. 2017;Stevenson et al. 2017b). While the inferred masses and tilt measurements of GW170608 are not sufficiently constrained to favor a formation channel, future measurements of binary black hole systems may hint at the formation histories of such systems (see Abbott et al. (2017a) and references therein). It may be possible to determine the relative proportion of binaries originating in each canonical formation channel following O(100) binary black hole detections Stevenson et al. 2017a;Zevin et al. 2017;Talbot & Thrane 2017;Farr et al. 2017a,b).
The detection of GW170608 is consistent with the merger populations considered in Abbott et al. (2016j,d) for which a rate of 12-213 Gpc −3 yr −1 was estimated in Abbott et al. (2017a).
OUTLOOK
LIGO's detection of GW170608 extends the range of known stellar-mass binary black hole systems at the lowmass boundary, and hints at connections with other known astrophysical systems containing black holes. The O2 run ended on August 25th, 2017; a full catalog of binary merger gravitational-wave events for this run is in preparation, including candidate signals with lower significance and systems other than stellar-mass black hole binaries (Abbott et al. 2017c). Estimates of the merger rate and mass distribution for the emerging compact binary population will also be updated.
With expected increases in detector sensitivity in the third advanced detector network observing run, projected for late 2018 (Abbott et al. 2016k), detection of black hole binaries will be a routine occurrence; studying this population will eventually answer many questions about these systems' origins and evolution. APPENDIX A. ANGULAR COUPLING MINIMIZATION GW170608 was observed during a routine instrumental procedure at LHO that minimises the coupling of angular control of the test masses to noise in the GW strain measurement. To maintain resonant power in the arms, the pitch and yaw angular degrees of freedom of the four suspended cavity test masses at each detector (Abbott et al. 2016a) must be controlled. This is achieved by actuating on the second stage of the LIGO quadruple suspensions. A feed-forward control is employed in order to leave the beam position of the main laser on the test mass unchanged while this actuation is applied. However, if this position differs from the actuation point, the angular control can affect the differential arm length, thus introducing additional noise in the strain measurement (Kasprzack & Yu 2016). As the beam position can drift over periods of hours or days, the angular feed-forward control must be periodically adjusted in order to minimize the coupling to strain.
During this procedure, high amplitude pitch and yaw excitations are applied to the test masses via actuation of the suspensions. Each of the 8 angular degrees of freedom is excited at a distinct frequency; the resulting length signals are observed via demodulation at each excitation frequency, revealing how strongly the corresponding degree of freedom couples to differential arm length. The feed-forward gain settings are stepped at intervals of approximately 45 s and the global minimum of angular control coupling to strain is determined from the resulting measurements. The frequencies of angular excitations are equally spaced between ∼19 Hz and ∼23 Hz, generating excess power in the differential arm motion, and thus in the measured strain around these frequencies. This procedure covers from ∼2 minutes before to ∼14 minutes after GW170608, shown in Figure 4 (left). During the period from −2 to 2 minutes substantial excess noise is visible at frequencies around 20 Hz. To characterize this noise we show amplitude spectral densities derived from 240 s of data both before the onset of the angular excitations and during the excitations around the event time in Figure 4 (right). No effect on the spectrum is visible above 30 Hz.
During the procedure, angular control gain settings are stepped abruptly; inspection of all such transition times shows no evidence for transient excess noise in the strain data outside the 19-23 Hz excitation band. The closest transition to the event time was 10 s before the binary merger, thus any transient noise associated with this transition Figure 1, it is not designed to show short-duration transient events. The strain amplitude is normalized to the interval between −6 and −2 minutes relative to the event time. See Appendix A for discussion of the feature around 20 Hz due to an angular control procedure. Right: Amplitude spectral density of strain data at both LIGO observatories for 240 s around the event time, (−2, 2) minutes on left panel, and for data before the start of the angular coupling minimization at LHO, (−6, −2) minutes. Excess noise is clearly visible around 20 Hz but data above 30 Hz are unaffected.
could not have affected the matched-filter output at the event time (template waveforms for GW170608-like signals have a duration between 2 and 3 s.) Furthermore, events from a single-detector matched-filter search covering other periods at LHO when this procedure was performed shows no anomalous features compared to other times. Thus, we find no evidence that the angular coupling minimization affected the recorded strain data at LHO around the event time at frequencies above 30 Hz. | 5,418 | 2017-11-15T00:00:00.000 | [
"Physics"
] |
Investigating the factors of enterprise social media strain: The role of enterprise social media’s visibility as a moderator
The significant effect of enterprise social media (ESM) usage has been extensively researched. However, recent studies and analysis have also emphasized the importance of understanding the negative aspects of ESM’s use. By applying uses and gratifications theory (UGT), this study proposes a research model that tests how employees’ ESM usage (hedonic, social, and information values) leads to ESM-related strain through perceived information overload. The study collected data from 315 Chinese employees using a survey method and analyzed the results using AMOS 21.0 software. Structural equation modeling (SEM) was applied to analyze the proposed hypothesis. The results indicate that perceived hedonic, social, and information values are significant predictors of perceived information overload. Such overload is also significantly associated with ESM-related strain. The results also indicate that ESM visibility strengthens the significant relationship between perceived information overload and ESM-related strain. Furthermore, managers can also train individuals to use ESM appropriately. We recommend that employees can better control and manage their ESM usage by recognizing the causes of excessive use.
Introduction
Enterprise social media (ESM) has introduced modern management practices to organizations, from the development of creative marketing plans to the transformation of connectivity, cooperation, and information exchange. An extensive literature shows that applying ESM to the workplace can improve individual job efficiency and productivity [1][2][3][4][5]. According to Leonardi and Meyer [4], ESM is a digital portal that allows individuals to share expertise with specific workers, to broadcast detailed information to everyone, and to edit, filter, and access the content of others without the interdependencies of time and space. Previous research has found that proper use of ESM by individuals can benefit both workers and organizations [6,7]. However, it can be detrimental if its use is excessive. Therefore, as ESM becomes more widely used in the workplace, workers may be overwhelmed with information, interaction, and social messaging, resulting in perceived overload and strain. For example, excessive use of ESM can lead to information overload and irritation, to individuals not concentrating on their task, and to errors in decisions. In particular, inappropriate and unreasonable use of ESM is becoming common among employees, with significant consequences for individuals and organizations. Therefore, to avoid and reduce the harmful consequences of inappropriate ESM use, it is critical to recognize the predictors and root causes of such activities-the motivation for our research. According to recent literature, individuals may utilize specific technologies or social media to satisfy their desires or needs; if these incentives or necessities are gratified, they will prefer to re-experience them [8,9]. ESM, as a social platform, encourages employees to participate in a variety of tasks such as posting, gathering information, and interacting with others [10,11]. These practices satisfy individuals' unique needs-interpersonal, informational, and hedonic -through ESM [12,13]. Alksasbeh, Abuhelaleh [14] found that, when social media addresses the expectations of students, it can increase their level of its use. Furthermore, several researchers have discovered that the fulfillment of needs can influence users' inappropriate usage [15,16]. Thus, as individuals feel that their requirements are addressed by social media, they increase their use of the technology to gain additional satisfaction. When the intensity of use reaches a certain level, information overload occurs [12,17]. "Information overload" is when the amount of information to which individuals are subjected exceeds the degree to which they can manage it efficiently [18]. As a result, the present study investigates the relationship between the gratification of needs-the needs satisfied by ESM usage-and perceived information overload.
However, the impact of perceived information overload on ESM-related strain is not independent of the ESM context. In comparison to other communication technologies, ESM provides a forum for open employee communication and cooperation [19]. It makes conversations between workers visible to all in the organization [20]. On the one hand, ESM visibility offers extremely visible communication among workers that can promote social bonding, social relationships, and information sharing [3]. On the other hand, ESM visibility offers uncontrolled and unorganized content that can overwhelm individuals' interpretive and analytical abilities, [21] resulting in information ambiguity. Therefore, highly visible information on the ESM platform may also interrupt employees' daily life and cause strain. For example, when individuals search for information using ESM to address problems, they may deal with a huge quantity of information [22] which contributes to ESM-related strain. However, although recent studies have clearly considered ESM visibility for its logical context, it has not been empirically examined much [21]. To address this research gap, this research explores the moderating role of ESM visibility on the relationship between perceived information overload, and ESM-related strain.
The purpose of this study is to examine the relationship between the gratification of needs and ESM-related strain through perceived information overload using data collected from Chinese employees. Based on uses and gratifications theory, this study also investigates the moderating role of ESM visibility in the relationship between perceived information overload and ESM-related strain. This study makes an important contribution to the current literature. It firstly highlights the relationship between the gratification of needs and perceived information overload to address the negative impact of social media usage. Secondly, it discusses the potential role of ESM visibility as a moderator. Thirdly, the results of this research may help managers and ESM designers to better understand the causes of ESM-related strain and perceived information overload. This study extends our understanding of the relationship between ESM usage (for hedonic, social, and information value) on strain and offers evidence that allows managers to design ESM usage guidelines for employees to control the negative consequences of unreasonable ESM usage.
Uses and gratifications theory (UGT)
Uses and gratifications theory (UGT) provides a useful framework for understanding why individuals use and choose a certain type of technology [23,24]. The basic assumption of UGT is that people are not automatically attracted to media content but instead utilize media to satisfy their different hedonic, emotional, and psychological needs [25]. UGT can also be used to describe user attitudes in computer-mediated communication (CMC) media contexts [26]. Unlike conventional media, such as print and television, CMC enables users to personalize their information to communicate with others. In the context of UGT, scholars argue that individuals initially use media or technology to fulfill their requirements and, after satisfaction, they continue to repeat the same experience [23,25]. UGT could thus provide a suitable theoretical framework for our study.
UGT has been commonly used to describe CMC media use in a wide range of contexts, such as online gaming [27], internet services [28], and email [29]. In recent technology, scholars have also applied UGT to social media technology such as Facebook, WeChat, Twitter, and Weibo [24,30]. For example, Gan and Li [30] investigated three types of gratification that would encourage people to continue using WeChat: hedonic, social, and utilitarian. Based on the UGT and ESM usage literature, we propose that ESM would provide workers with three types of gratification: social, information, and hedonic [8,23]. Social gratification refers to the use of ESM by individuals to establish social relationships with coworkers, allowing them to recognize the social value provided by ESM [23,30]. Information gratification is the fulfillment of information requirements. Employees may access a range of information from ESM technology-such as content, and information references-allowing them to understand the importance of the information provided by ESM, thereby gratifying their need for information [23,30]. Hedonic gratification allows employees to use ESM to obtain enjoyment and pleasure, thereby accessing its hedonic value [23,30]. Although several scholars have recognized different forms of gratification offered by ESM for individuals, and others have stressed the links between social media availability and the gratification of needs, very little research has been conducted on the connection between perceived information overload and ESM-related strain. To fill this research gap, we used UGT as the theoretical foundation to investigate the workplace link between the gratification of needs and ESM strain through perceived information overload.
ESM-related strain
Several studies have recently investigated ESM usage in the workplace with mixed findings [12,17,44,45]. For example, Chen and Wei [17] reported that social or work-related use of ESM causes overload and ESM-related strain among employees; this may have an adverse impact on their work efficiency. Cao and Yu [12] examined the link between excessive ESM use and work outcomes, observing that such excessive use has adverse effects on job performance. In contrast, Pitafi, Kanwal [44] reported that ESM use has a significant impact on employee work performance. Hence, theoretical studies have begun considering the negative aspects of ESM technology, such as ESM-related strain. Since each user communicates to other users throughout the ESM network, a flood of communication or information is produced. This high-speed flow of information may create ESM-related strain among employees that can include feelings of anxiety, pressure, helplessness, and tension. Ayyagari, Grover [46] reported that excessive use of technology may influence employees' work outcomes and, ultimately, contribute to strain. As a result, this study examines the ESM usage factors that cause ESM-related strain among individuals.
Users and gratifications theory and perceived information overload
"Hedonic ESM usage" applies to the use of ESM technology primarily for entrainment and pleasure [12]. "Perceived hedonic value" signifies the satisfaction and happiness that results from the ESM content and the connections thereby established with other individuals [24]. To realize hedonic value, employees may experience satisfaction, enthusiasm, tweets, and excitement with coworkers [23,47]. The intrinsically enjoyable existence of ESM encourages greater employee engagement and its extensive use [48,49]. Consequently, an ESM participant may be excited because they discover similar preferences or read interesting material shared by colleagues [8,50]; they may then share more content. In order to sustain a high level of pleasant experience, these workers can behave irrationally, expending much time and energy on ESM and exchanging ever more information [12,51]. Since each individual is sharing interesting items with other individuals on the ESM platform, there is a stream of information. This rapid flow of information can result in information overload. As a result, based on the literature, we suggest the following hypothesis: H1a: The use of ESM for perceived hedonic value has a positive effect on perceived information overload.
"Perceived social value" refers to the benefits that one may experience by efficiently establishing or managing personal relationships, seeking social support, and promoting social interaction with coworkers on the ESM platform [12]. ESM is a web-based platform that links people with family members, relatives, acquaintances, and coworkers at any time and from any location [4]. As the number of social connections increase, workers may receive a significant number of responses from their online friends through ESM [9]. In order to sustain these huge social networks for acquiring social support and a sense of belonging, individuals must respond as quickly as possible by ESM [12]; this type of action can result in information overload. According to UGT, as workers consider the social value provided by using ESM at workplace, their usage can be affected. In other words, they can increase their use of this technology as a result of the socially significant benefits they receive. Information overload can occur when this rate of use reaches its maximum. Sun, Wang [52] discovered that social importance greatly increases the degree to which people use ESM. Chen and Kim [15] have shown that the social importance perceived by employees is a significant factor in supporting ESM use. We thus propose: H1b: The use of ESM for perceived social value has a positive effect on perceived information overload.
ESM serves as a valuable channel for communicating and exchanging information [53,54]. "Perceived information value" refers to the benefits of a user obtaining important information from ESM [55]. Previous studies have reported a significant correlation between perceived information value and the actual use of technology [56]. ESM encourages individuals to post and share work-related information, allowing employers to satisfy their workers' information needs [31,57]. Additionally, employees' information value expectations can affect information exchange [24]. For example, an employee with a high perception of information value is more inclined to share information, ideas, and interactions on the ESM platform with colleagues. Consequently, they may expend much time and energy checking for notifications and sharing material on ESM [35,44]. The more valuable the information an employee obtains from the online community, the more likely they will share information in that community. According to recent research, information overload is exacerbated by individuals' communication, content-sharing, likes, updates, comments, and posts on an ESM platform [58]. Information value also motivates employees to share more information. Therefore, this study proposes the following hypothesis: H1c: The use of ESM for perceived information value has a positive effect on perceived information overload.
Perceived information overload and ESM-related strain
Much data has been generated with recent advances in information and communications technology; consequently, the phenomenon of information overload has become more readily recognized and encountered [12,17]. "Information overload" refers to people's assessment and interpretation of the types of items that are outside their ability to manage [59]. Scholars have used overload terms in several fields of research, including "work overload" [46], "information overload" [59], and" system-feature overload". As volumes of information cross a certain threshold, people may have problems locating and interpreting it [36] and thus make decision errors. Therefore, when an employee seeks additional information that is needed, their decision-making ability may suffer. The proliferation of ESM technology has resulted in massive amounts of information being immediately generated and disseminated [55]. ESM is a public platform, so employees can post, broadcast, exchange, and disseminate information rapidly at any time [19,60]. As a result, they are unable to process information efficiently, indicating that excessive use of ESM can result in information overload [12]. According to Wurman [61], information overload causes individuals to feel depressed, uneasy, and emotionally exhausted. Zhang, Zhao [59] also found that a high volume of information may contribute to social network exhaustion as the rapid production and dissemination of information on ESM has negative effects such as fear, frustration, and anxiety. This study therefore suggests the following hypothesis: H2: Perceived information overload has a positive effect on ESM-related strain.
ESM visibility as moderator
ESM visibility allows individuals to make their behaviors, information, and knowledge visible to their colleagues [31,62,63]. ESM visibility presents a quick way of identifying what other employees are doing and with whom they are communicating [3,35,36]. ESM visibility allows them to easily approach the broad social network. Such unregulated interactions may increase the possibility of information overload. Earlier studies have concluded that ESM visibility may contribute to information overload because there is a vast amount of information outside of an individual's control [12,22]. For example, notifications and a constant flow of work-or non-work-related information require employees to view and process a considerable amount of information with workmates. This unstructured flow of information can divert their focus from their work and lead to negative attitudes toward others.
Furthermore, the visibility of ESM leads employees to be anxious about the disclosure of their weakness and failures; they may believe that ESM usage requires a significant amount of time and effort, thus raising their loading perception [17,64]. ESM visibility thus amplifies the flow of information: it encourages employees to establish a better social presence on ESM, which necessarily requires workers to share and exchange a significant amount of information with other colleagues [65], leading to information overload. Although ESM allows workers to collaborate and fulfill their workplace and social needs [66,67], the visibility it facilitates causes an intensifying of demands, creating a feeling of overload and fatigue [68]. Therefore, this study proposes the following hypothesis: H3: ESM visibility moderates the significant relationship between perceived information overload and ESM-related strain, such that the higher the ESM visibility, the higher the relationship between perceived information overload and ESM strain.
Data collection procedures
In order to achieve its aim, this study collected data by surveying Chinese workers employed in several companies. Due to the increasing popularity of ESM technology in China, we decided to conduct the survey there. Moreover, ESM technology has been extensively adopted by many businesses as a cost-effective tool for their employees' work-related communication.
To capture accurate and valid responses, the current research focused on specific information about ESM usage in the workplace, making data collection through a survey difficult. Consequently, we collaborated with a well-known educational institution to ensure the reliability of our study. This organization is involved in several training programs for employees, especially about information systems. We selected organizations that had adopted ESM technology for their employee's work-related communication. Before the data collection, we also conducted several meetings with employees to ensure their use of ESM. Several additional strategies were used to identify valid responses; for example, we developed two questions to distinguish invalid answers. The nature of these questionnaires corresponds to the two elements in the questionnaire but contradicted the original interpretation of the items. If the participants gave the same answers to the two adjacent questions, then their replies were considered irrelevant.
If the time it needed to complete the questionnaire was quite small, it was declared a non-serious response and was removed. We also discussed the objective of study with the managers of the selected companies and assured employees that their feedback would remain confidential, only to be used in academic research. Before data collection, we designed the questionnaire and invited five PhD-level faculty members of the information systems department for review and suggestions. After this discussion and feedback, some items of the questionnaire were redesigned. We also conducted a pilot study on 57 participants; its results were found accurate, such that >0.700. The results of the pilot study motivated us to collect further data. In addition, authors has followed the ethical guidelines of Tianjin University of Commerce China. This study has approved by the ethical committee of Tianjin University of Commerce China. The ethical committee also has waved the consent of this study. Over August to November 2020, the survey questionnaire was distributed to employees. To encourage the response rate, we sent reminder emails and made phone calls to all the participants. We sent 450 questions and received 340 responses-a response rate of 75.55% within four months. After evaluating 315 complete questionnaires used in final data set, some were discarded because they were improperly completed or some entries were left blank. Furthermore, following the procedure of Armstrong and Overton [69], we used the chi-squared procedure to analyze possible nonresponse bias by comparing the first and last 25% of participants over all indicators. This found that the two groups did not vary substantially, indicating that nonresponse bias was not a major problem. The demographic details for the survey are shown in Table 1.
Research instruments
Previously validated instruments were used in this study to analyze the perceptions of participants. All the constructs were measured using a five-point Likert scale which ranged from "strongly agree" to "strongly disagree". Since this research is based on Chinese workers, the we adopted recommendations from previous studies [70] and used a back-translation mechanism to ensure the accuracy of all instruments. Firstly, we invited three experienced native-speaking Chinese translators to translate the original English version of the questionnaire into Chinese. We then approached three other Chinese professionals to translate the Chinese version of the questionnaire into English; this procedure was replicated several time times before the translation accurately reflected the original items. A total of ten constructs were used in this study, including the control variable. The details of all the measurement items follow.
ESM-strain. The outcome construct of ESM-strain included four items and was measured using items from Ayyagari, Grover [46]. The scale measures the overall strain with excessive use of ESM. The sample item of this scale is "I feel drained by activities that require me to use enterprise social media".
Perceived information overload. The scale of this overload consisted of four items and was devised using items from Zhang, Zhao [59]. This scale measured the overall information load with excessive use of ESM. The sample item of this scale is "I am often distracted by the excessive amount of information available to me on enterprise social media".
ESM visibility. ESM visibility was used as a moderator construct and consisted of three items. ESM visibility was measured using items from Leonardi [19]. The sample item of this scale is "Enterprise social media enable me to see other coworkers 'answers to other coworkers' questions".
Perceived social value. The scale of perceived social value consisted of three items and was measured using items from Ding, Yang [23] and Zhang, Li [56]. The sample item of this scale is "Sharing information with others using ESM can improve my relationship".
Perceived information value. The scale of perceived information value consisted of four items and was measured using items from Ding, Yang [23] and Zhang, Li [56]. The sample item of this scale is "I accumulate much knowledge through ESM users' shared information".
Perceived hedonic value. The scale of perceived hedonic value consisted of three items and was measured using items from Ding, Yang [23] and Zhang, Li [56]. The sample item of this scale is "I have fun interacting with ESM".
Control variables.
In order to analyze the actual effect of an independent variable on a dependent variable, we also controlled some constructs that may affect the outcome. Following the guidelines of previous studies, gender, age, education, and experience were used as control variables [71].
Results and analysis
Before being analyzed, all data was screened using SPSS software for gaps or outliers in the data set. We analyzed the data in two steps. Firstly, we tested reliability, standard factor loading, and the validity of all the instruments. Secondly, we applied structural equation modeling to analyze the hypothesis of the study.
PLOS ONE
Factors of enterprise social media strain: The role of enterprise social media's visibility as a moderator
Measurement model
AMOS and SPSS assessed the reliability, convergent validity, and discriminant validity of the proposed research model. By applying a two-step approach, we initially conducted confirmatory factor analysis (CFA) to test the measurement model and determine the reliability and validity of the research model before analyzing the structural relationship of the suggested hypotheses. Previous studies recommended that the values of Cronbach's alpha (CA) and composite reliability (CR) should be higher than the minimum suggested value of 0.700 [72][73][74]. The findings of Table 2 indicate that the CA values of all the constructs range from (0.770 to 0.895) and the CR values from (0.840 to 0.889)-higher than the suggested value of 0.700. The average variances extracted (AVE) of all the constructs are also shown in Table 2; they range from (0.611 to 0.715), higher than the recommended value of 0.500 [72,75,76]. Similarly, the literature recommends that the loading of all the items should be higher than 0.600 [72], and the results of Table 4 indicate that all the items have loadings higher than 0.600. Thus, all of the findings indicated that the research model has an appropriate degree of convergent validity and reliability.
We used several methods to analyze the discriminant validity of the proposed research model by observing the results of Tables 2-4. The results of Table 2 indicate that all constructs have MSV values higher than the ASV values [36]. We then applied the procedure suggested
PLOS ONE
Factors of enterprise social media strain: The role of enterprise social media's visibility as a moderator by Fornell and Larcker [72] to analyze the discriminant validity of the research model; Podsakoff, MacKenzie [77] suggested that the highest co-relation value between variables should be less than 0.700. Next, we compared the pair-wise square root of the AVE of all the constructs with inter-correlation in Table 3. The results indicate that all the AVE square-root values are higher than the inter-correlation values, suggesting an adequate level of discriminant validity for the proposed research model. In addition, we also considered the results of Table 4, which indicated that the value of each item on its assigned construct was higher than that of the other construct. Hence, we conclude that the research model also has an acceptable level of discriminant validity. Common method bias (CMB) may occur in the responses due to the existence of cross-sectional data [77]. For this research, we used a multipronged method to assess the probability of CMB. Firstly, we attempted to minimize the possibility of CMB at the participant level by using one reverse item to keep the participants focused when responding to the questionnaire. Secondly, to examine the probability of CMB in the data set, we used Harman's single-factor test [78,79]. The results indicated that there were six variables with eigenvalues greater than 1.0-the first factor only indicated 26.59% of the variance, which was less than the 40% threshold. Thirdly, the findings of Table 3 confirmed that all of the constructs have co-relation values smaller than 0.600 [80]. In addition, we used the approach developed by Liang, Saraf [81] to analyze the CMB concern. As a result, we examined the substantive factor loading and method factor loading for each variable. The findings revealed that the substantive factor accounted for 66.3% of the variance, while the method factor accounted for 1.3%, indicating that there was no likelihood of an issue with CMB in the existing study. Finally, we performed a variance inflation factor (VIF) procedure to analyze the possibility of CMB. The outcome revealed that
PLOS ONE
Factors of enterprise social media strain: The role of enterprise social media's visibility as a moderator VIF results are lower than the minimum suggested value of 3.3 [82], meaning that CMB is not a significant problem in this analysis. Altogether, the evidence demonstrated that there was no CMB problem in the current study. Prior to assessing the structural equation modelling, the fit values of the measurement model were analyzed using AMOS version 21.0 with a maximum likelihood estimation method for all variables [83]. The outcome indicated that the values of model fit (CFI = 0.910, TLI = 0.890, IFI = 0.911, NFI = 0.874, PNFI = 0.873, REMSA = 0.053, CMIN/DF = 3.083) were within the suggested range and satisfactory, as indicated in Table 5.
Moderation analysis
The existing study also analyzed the moderating effect of ESM visibility on the link between perceived information overload and ESM-strain. We proposed in Hypothesis 3 that ESM visibility strengthens the relationship between perceived information overload and ESM train. The findings indicate that the interaction term (perceived information overload × ESM visibility) has a significant relationship with ESM-strain (B = 0.113, t = 2.011; p < 0.05), validating H5a.
To fully understand the moderating effect of ESM visibility in our research model, we further used a graphic approach suggested in previous research [86]. According to Fig 3, ESM visibility strengthens the relationship between perceived information overload and ESM strain.
Discussion
This study investigated ESM-related stain using UGT as a theoretical foundation. It also examined the moderating role of ESM visibility in the link between perceived information overload and ESM-related strain. The empirical analysis validated the suggested hypotheses. Specifically, the results indicated that perceived hedonic, social, and information value have a significant effect on perceived information overload, supporting H1a, H1ab, and H1c, which accords with our assumptions. These findings reflect that, to obtain a high perception of information, social, and hedonic value, employees are likely to share information, ideas, and interactions on the ESM platform with colleagues. Previous studies also reported similar results [12,24,87]. For example, Cao and Yu [12] reported that excessive use of ESM has a significant effect on strain. Sun, Liu [8] also observed that perceived hedonic, social, and information value causes excessive usage of ESM by employees. The results also confirmed that perceived information overload has a significant effect on ESM-related strain, thus affirming H2. ESM is an open platform where employees can easily share and exchange work or non-work-related information; this high frequency of information causes anxiety and depression among employees. Previous scholars have also found that information overload has a negative impact on individual work performance [12,51]. Thus, Chen and Wei [17] also found a curvilinear relationship between information overload and ESM-related strain.
Furthermore, the findings also show that ESM visibility strengthens the relationship between perceived information overload and ESM-related strain-H3 is also supported by
PLOS ONE
Factors of enterprise social media strain: The role of enterprise social media's visibility as a moderator current data set. ESM visibility forces individuals to establish a better social presence on ESM, which can necessarily require workers to share and exchange a significant amount of information with other colleagues [65], leading to information overload. ESM visibility also allows employees to view the communication content of others, even if they are not directly involved in that communication [3,20]. Chen and Wei [17] reported that communication visibility significantly strengthens the relationship between ESM use and information overload.
Theoretical implications
The current study can make numerous theoretical contributions. Firstly, prior studies have been generally based on the positive role of ESM [3,44,66]. For example, Pitafi, Kanwal [44] observed that it has a positive effect on individual work performance through task interdependence. Cao, Ali [88] also argued that ESM use enhances team performance. In contrast, the present research investigates the use of ESM and establishes an empirical link between ESM use (hedonic, social, and information values) and ESM-related strain through perceived information overload. These findings further advance our theoretical understanding of the relationship between ESM use and related strain through perceived information overload, clarifying that information overload causes ESM-related strain. A second contribution is that the study's results also indicate that information overload has a significant effect on ESM-related strain. These results extend those studies that address ESM-related strain in the workplace [36]. We have also highlighted the role of perceived information overload, which exerts a great negative influence on the strain.
Thirdly, the present study investigates the role of ESM visibility as a moderator and found that ESM visibility reinforces the relationship between perceived information overload and ESM-related strain. Previous studies have highlighted the significant role of ESM visibility [3,65], with Engelbrecht, Gerlach [65] reporting that ESM visibility may have a positive effect on knowledge-sharing because employees can learn from the communication activities of colleagues. Visibility allows individuals to view the historical communication of other employees at any time. Nevertheless, due to nature of ESM technology, ESM visibility also causes information overload.
Finally, this study has some contributions to UGT literature. Although several researchers have used UGT to analyze the formulation mechanism of social media acceptance and continued usage behavior, its impact on the formation process of ESM-related strain has received little consideration. Even if ESM usage in the workplace is not extreme, employee output can suffer as a result of perceived information overload.
Managerial implications
The results of this study have several implications and suggestions for managers. Its findings show that ESM usage (hedonic, social, information value) has a positive effect on perceived information overload since social media is commonly utilized by individuals in corporations for socialization rather than work-related collaboration [17]. Furthermore, the current study indicates that perceived information overload has a significant impact on strain. We suggest that managers acquaint themselves with the features of ESM before applying it within the organization, and guide workers in the logical use of ESM technology for information-sharing. Managers can also apply some policies to control employee's ESM use. An example is formulating some ESM guidelines that are consistent with corporate culture and specifying when and how an individual can use ESM. Organizations should designate a certain time for workers to use ESM for suitable enjoyment and entertainment, allowing them to better apply themselves to their respective duties.
In addition, this study shows that ESM visibility strengthens the connection between perceived information overload and ESM-related strain. ESM designers can implement some relevant technological features when developing and improving ESM technology. They should specifically strengthen the technological functions relating to ESM visibility or incorporate certain configuration options so that users can easily track and manage their actions and minimize possible excessive usage. For example, designers may include optional features in ESM to limit interaction requests at specific times, thereby reducing the duration and intensity with which workers maintain social queries during work time. ESM developers may also include some screening features that help users to better locate required information on ESM, reducing the amount of time they waste in searching non-related content on ESM during work time.
Furthermore, managers may also provide trainings to their employees to use ESM appropriately. We recommend that individuals can better control and manage their ESM usage by recognizing the causes of excessive use. Accordingly, employees may limit their use of ESM for non-work-related activities during working hours. They should also monitor the extent and duration of their use of ESM at work, as well as using alternative methods of communication such as telephone or face-to-face communication, thus minimizing their dependence on ESM.
Limitations and future directions
Although the current study has numerous implications, there are some limitations that we note here for future researchers. Firstly, the participants of study were Chinese employees and it focused on ESM users. Future scholars may apply the same conceptual model to other countries and compare the results. Nevertheless, China is an ideal country for this study, as ESM is widely used by Chinese employees for work-related communication [8]. In addition, the role of ESM in Chinese culture to satisfy individual social, information, and hedonic desires can vary from Western societies. As a result, future studies may enhance the generalizability of research by incorporating diverse cultural contexts. Another sampling issue is that self-reporting of data by users is used in this study, which is considered subjective [89]. Future scholars can concentrate on various data sources, such as objective data from the technical department on the practical application of ESM.
Thirdly, the existing study is based on a cross-sectional method to demonstrate the impact of visibility allowed by ESM on workers' excessive usage. A longitudinal method could more deeply reveal the changes in ESM usage behavior over time, which could be more significant.
In addition, the current study does not investigate the mediating effect of perceived information overload. Future studies could use another moderator and also examine the mediating role of information overload.
Finally, the present research is an empirical investigation into ESM. While ESM can only be accessed by employees inside the corporation, they also use external social media tools (Twitter, Facebook, Whatapp, WeChat) to not only communicate with workmates but also with friends and relatives [45]. As a result, future studies should also explore how various overload experiences are induced by external social media technology.
Conclusion
The objective of this study is to investigate the link between the gratification of needs and ESM-related strain through perceived information overload using data collected from Chinese employees. The current study supported all the suggested hypothesis. Specifically, the results indicate that perceived hedonic, social, and information values are significant predictors of perceived information overload. Such overload is also significantly associated with ESM-related strain. The results also indicate that ESM visibility strengthens the significant relationship between perceived information overload and ESM-related strain. | 8,180.2 | 2022-03-08T00:00:00.000 | [
"Business",
"Computer Science"
] |
Using multivariable Mendelian randomization to estimate the causal effect of bone mineral density on osteoarthritis risk, independently of body mass index
Abstract Objectives Observational analyses suggest that high bone mineral density (BMD) is a risk factor for osteoarthritis (OA); it is unclear whether this represents a causal effect or shared aetiology and whether these relationships are body mass index (BMI)-independent. We performed bidirectional Mendelian randomization (MR) to uncover the causal pathways between BMD, BMI and OA. Methods One-sample (1S)MR estimates were generated by two-stage least-squares regression. Unweighted allele scores instrumented each exposure. Two-sample (2S)MR estimates were generated using inverse-variance weighted random-effects meta-analysis. Multivariable MR (MVMR), including BMD and BMI instruments in the same model, determined the BMI-independent causal pathway from BMD to OA. Latent causal variable (LCV) analysis, using weight-adjusted femoral neck (FN)–BMD and hip/knee OA summary statistics, determined whether genetic correlation explained the causal effect of BMD on OA. Results 1SMR provided strong evidence for a causal effect of BMD estimated from heel ultrasound (eBMD) on hip and knee OA {odds ratio [OR]hip = 1.28 [95% confidence interval (CI) = 1.05, 1.57], p = 0.02, ORknee = 1.40 [95% CI = 1.20, 1.63], p = 3 × 10–5, OR per standard deviation [SD] increase}. 2SMR effect sizes were consistent in direction. Results suggested that the causal pathways between eBMD and OA were bidirectional (βhip = 1.10 [95% CI = 0.36, 1.84], p = 0.003, βknee = 4.16 [95% CI = 2.74, 5.57], p = 8 × 10–9, β = SD increase per doubling in risk). MVMR identified a BMI-independent causal pathway between eBMD and hip/knee OA. LCV suggested that genetic correlation (i.e. shared genetic aetiology) did not fully explain the causal effects of BMD on hip/knee OA. Conclusions These results provide evidence for a BMI-independent causal effect of eBMD on OA. Despite evidence of bidirectional effects, the effect of BMD on OA did not appear to be fully explained by shared genetic aetiology, suggesting a direct action of bone on joint deterioration.
Introduction
Although osteoarthritis (OA) is a major cause of morbidity worldwide, effective pharmacological treatment remains elusive. It may be possible to develop novel therapeutic approaches based on understanding of risk factors. Several large population-based studies have identified positive relationships between bone mineral density (BMD) and hip and knee OA. 1 Mendelian randomization (MR), which is commonly used to explore causal relationships, 2-4 has recently obtained evidence for a causal role of BMD on hip and knee OA risk. 5 Body mass index (BMI), a risk factor for OA [6][7][8] and positively associated with BMD, 9 may bias MR estimates for the relationship between BMD and OA. Funck-Brentano et al. addressed this by excluding instrument(s) associated with BMI. 5 An alternative approach, yet to be applied in this context, is the use of multivariable MR (MVMR) to estimate the direct causal effect of the exposure on the outcome when the instrument(s) are associated with multiple risk factors. 10 Alternatively, rather than a causal effect of BMD on OA, shared biological pathways may contribute to both traits. Consistently with this possibility, a genetic correlation between lumbar spine (LS)-BMD and OA has been observed. 11 Genetic correlation may give rise to bidirectional causal relationships in MR analysis.
As well as the relationship between BMD and OA, relationships with BMI could be characterized by bidirectional relationships. A causal effect of BMI on BMD is well established; the skeleton adapts to the increased load placed upon it by increasing BMD. Alternatively, a causal pathway between BMD and BMI is plausible via the metabolic effects of bone turnover. Murine osteocalcin knockouts have increased fat mass and are insulin-resistant; 12 in humans, higher BMD is associated with lower circulating osteocalcin, which may mediate the positive association between BMD and fat mass. However, an MR analysis found no evidence of a causal pathway between femoral neck (FN) or LS-BMD and BMI in children. 9 To provide a more complete understanding of the relationship between BMD and OA, we tested bidirectional relationships between BMD, OA and BMI ( Figure 1) using one-sample (1S) and two-sample (2S) MR, and aimed to determine the direct (i.e. unconfounded) causal pathways between these variables using MVMR.
Key Messages
• Mendelian randomization (MR) analyses suggest that bone mineral density (BMD), assessed from heel ultrasound scans, is a risk factor for osteoarthritis, independently of adiposity.
• Evidence for reverse causality (i.e. a causal effect of osteoarthritis on BMD) may reflect the shared biological pathways contributing to bone and joint development.
• Latent causal variable (LCV) analysis provides evidence for a direct causal effect of BMD on osteoarthritis, which is not fully explained by genetic correlation between these two traits.
• This paper illustrates the utility of methods such as LCV analysis and multivariable MR when examining causal pathways in situations in which complex relationships exist, such as those between BMD, body mass index and osteoarthritis.
Individual-level analyses
Individual-level analyses were performed in the UK Biobank population. UK Biobank is a UK-wide population-based health research resource consisting of $500 000 people, aged 38-73 years, who were recruited in 2006-2010. 13 Participants provided a range of information [e.g. demographics, health status, lifestyle/physical activity (PA) measures] via questionnaires and interviews; anthropometric measures and blood samples were taken (data available at www.ukbiobank.ac.uk). A full description of the study design, participants and quality-control methods has been published. 13
MR
A summary of all MR analyses performed, along with the source of each of the instruments, is presented in Table 1 and of the assumptions of MR and how we tested these in Figure 2. We examined causal relationships with hip and knee OA separately, given the availability of separate genome-wide association studies (GWAS) for these outcomes, which have no overlap in terms of the most strongly associated single-nucleotide polymorphisms (SNPs).
One-sample MR 1SMR analyses were performed in the UK Biobank population using the instrumental-variable regression ('ivreg') function of the Applied Econometrics with R package. 14 Exposures were instrumented by an unweighted genetic risk score (GRS), generated as the sum of the dosage for exposure-increasing alleles (data sources provided in Table 1). Analyses were adjusted for age at BMD/BMI assessment, sex, genotyping chip and 40 principal components. Continuous exposures (eBMD/BMI) were standardized before analysis. Effect estimates for binary Figure 1 Diagram summarizing hypothesized relationships between bone mineral density, body mass index and osteoarthritis Thicker arrows represent stronger hypothesized relationships. Diagram does not take account of temporality of relationships due to the uncertainty in the temporal sequence, e.g. OA may first cause an increase in BMI due to reduced PA, leading to further OA through greater joint loading; however, it is equally possible that BMI leading to an increase in joint loading is the initiating event. BMD, bone mineral density; BMI, body mass index; OA, osteoarthritis. outcomes (hip/knee OA) were generated from a linear two-stage least-squares regression and represent the increased probability of having OA per unit increase in the exposure. We generated an estimate of the odds ratio (OR) per standard deviation (SD) increase in the exposure, for comparison with 2SMR results, by first regressing the instruments on the exposure, generating predicted values of the exposure, and then regressing the predicted values of the exposure on the binary outcomes using a logistic-regression model. The standard errors for this estimate are likely to be underestimated. 15 Two-sample MR To maximize the sample size, and thus statistical power, we performed 2SMR using summary-level data from published GWAS. 2SMR analyses were performed using the TwoSampleMR R package, version 0.4.22. 16 SNPs that explained a greater proportion of the variance in the outcome than the exposure. 17 The proportion of variance explained by each SNP was calculated using the p-value and sample size and the 'get_r_from_pn' function of the 'TwoSampleMR' package for continuous variables and the 'get_r_from_lor' function for dichotomous variables. The 'get_r_from_lor' function requires the case prevalence in the study population to be specified, which was calculated as the number of cases divided by the total sample size (15% for knee OA and 8% for hip OA). Seven, four and two eBMD SNPs were excluded for analyses with hip OA, knee OA and BMI outcomes, respectively. Two BMI SNPs explained a greater proportion of variance in hip OA risk, 1 for knee OA risk and 15 for eBMD. One knee OA SNP was excluded due to a greater r 2 for eBMD. All Steiger-filtered SNPs are listed in Supplementary Tables S2-S7 (available as Supplementary data at IJE online). Estimates were generated using inverse-variance weighted (IVW) random-effects metaanalysis of the Wald ratios for each SNP.
Multivariable MR
As we hypothesized that BMI is a confounder of the BMD-OA relationship (i.e. a common causes of both phenotypes), we determined the independent effect of BMD on OA outcomes by performing 1S MVMR including GRS for both BMI and BMD as instruments. Both instruments were regressed on each exposure to generate a predicted value for each exposure. The predicted values for each exposure were then included in a multivariable regression to generate the effect of one exposure on OA when conditioning on the other exposure. Analyses were adjusted for sex, genotyping chip and 40 principal components (PCs). Sanderson-Windmeijer conditional F-statistics were calculated as measures of instrument strength in MVMR analyses. 18 Sensitivity analyses MR-Egger regression was performed to generate an estimate of horizontal pleiotropy in the two-sample analyses. 19 Weighted median regression determined the robustness of IVW estimates as weighted median estimates are valid as long as 50% of the information is derived from valid instruments. 20 We repeated the 2SMR analyses restricted to eBMD SNPs also associated with FN-BMD (p < 5 Â 10 -8 ) in the GEFOS FN-BMD meta-analysis, 21 to determine whether FN-BMD has a stronger effect than eBMD on hip or knee OA risk. We also performed a latent causal variable (LCV) model, as described by O'Connor and Price, 22 to determine whether there is a true causal effect of BMD on OA, independently of the genetic correlation. Full methods are described in the Supplementary Information (available as Supplementary data at IJE online).
Results
Confirming observational relationships between BMD, OA and BMI in UK Biobank A total of 334 061 individuals in UK Biobank with genotype data also had measurements of eBMD, BMI, covariates and hospital-diagnosed hip OA; 341 920 had data for knee OA. The mean (SD) age of those with hip OA was 61.7 (6.0), of those with knee OA was 60.2 (6.9) and of controls was 56.2 (8.1) years (Supplementary Table S8, available as Supplementary data at IJE online). Fifty-seven per cent of people with hip OA were female compared with 50% with knee OA and 54% of the controls. Both hip and knee OA cases were heavier than controls, with mean BMI 28.9 (5.0), 30.3 (5.4) and 27.1 (4.6) kg/m 2 , respectively. Descriptive statistics were virtually the same when restricting to individuals with complete data for eBMD, BMI and OA who were included in the multivariable MR analyses (Supplementary Table S8, available as Supplementary data at IJE online). . In one-sample analyses, IV1 was tested by calculating the F-statistic, which is a measure of instrument strength. A >10 threshold is used to determine sufficient instrument strength. 2 IV2 was tested by determining the association between the instruments and potential confounders of the exposure-outcome relationship. In two-sample analyses, to satisfy IV1, we ensured that all instruments were robustly associated with the exposure by only including SNPs associated with the exposure at genome-wide significance. To address IV3, MR-Egger regression was performed to generate an estimate of horizontal pleiotropy (intercept) and a pleiotropy-robust estimate of the causal effect (slope). Weighted median regression was performed to determine the robustness of IVW estimates as weighted median estimates are valid even if 50% of the SNPs are not valid instruments. 20 BMD, bone mineral density; BMI, body mass index; OA, osteoarthritis; SNP, single-nucleotide polymorphism.
MR analyses provide evidence for bidirectional causal pathways between BMD and OA A summary of MR results is presented in Figure 3. In 1SMR, eBMD was causally related to both hip and knee OA, with an SD increase in eBMD related to a 29% [95% confidence interval (CI) ¼ 5, 58] increased odds of having hip OA and 39% (95% CI ¼ 19, 63) increased odds of having knee OA ( Table 2). The F-statistic confirmed sufficient instrument strength (F > 3000). Univariable 1SMR results were unaltered by using individual SNPs rather than PRS as instruments, but F-statistics were lower, as was the effect estimate for the causal effect of eBMD on knee OA (Supplementary Table S9, available as Supplementary data at IJE online). The BMD risk score was related to BMI but was not related to PA or hormone-replacement-therapy use (Supplementary Table S10, available as Supplementary data at IJE online). In 2SMR analyses, IVW provided evidence for a causal effect of eBMD on hip OA [OR per SD increase ¼ 1.09 (95% CI ¼ 1.03, 1.16)], which was relatively consistent (in magnitude) across the three MR methods ( Figure 4 and Supplementary Figure S1, (Figure 4 and Supplementary Figure S4, available as Supplementary data at IJE online). The knee, but not hip, OA GRS was related to BMI, potentially invalidating instrumental-variable assumption 2 (IV2) (Supplementary Table S10, available as Supplementary data at IJE online).
BMI is a strong causal risk factor for BMD and OA with weaker evidence for bidirectionality 1SMR provided evidence that BMI has a strong causal effect on hip and knee OA, with an SD increase in BMI associated with a 68% (95% CI ¼ 41, 100) increased odds of Figure 4). There was strong evidence, from 1SMR, that the causal pathway between BMI and knee OA was bidirectional, with weaker evidence for hip OA (Table 2). Additional adjustment for total weekly PA (assessed using the International Physical Activity Questionnaire) did not attenuate these relationships. 2SMR, however, provided weak and inconsistent evidence (across the three methods) of a causal effect of hip OA on BMI only (Figure 4 and Supplementary Figures S7 and S8, available as Supplementary data at IJE online). We could not perform bidirectional 1SMR for BMD-BMI as the FN-BMD SNPs were identified by weightadjusted GWAS, meaning the instrument for FN-BMD may be inversely related to weight and thus BMI. 23 2SMR using summary statistics from the eBMD GWAS, not adjusted for weight, did not identify a causal effect of eBMD on BMI (Figure 4 and Supplementary Figure S9, available as Supplementary data at IJE online). There was robust evidence for a causal effect of BMI on eBMD in 1SMR, with an SD increase in BMI causing a 0.07SD (95% CI ¼ 0.04, 0.11) increase in heel BMD (Table 2). This estimate was like that from 2SMR and the effect size was consistent for IVW, weighted median and MR-Egger analyses, although the MR-Egger intercept did reveal some evidence of horizontal pleiotropy (Figure 4, Supplementary Table S11 and Supplementary Figure S10, available as Supplementary data at IJE online).
Multivariable MR identifies an independent causal effect of eBMD on OA Overall, the 1S and 2S analyses provided consistent evidence that BMI is a confounder of the relationship between BMD and hip/knee OA (i.e. a common cause of both phenotypes, Figure 3). We therefore used 1SMVMR to examine the causal effect of eBMD on OA after accounting for BMI. Following adjustment for BMI, eBMD was found to be an independent causal risk factor for both hip and knee OA with a similar magnitude of effect to that observed in MR analyses not accounting for BMI. BMI had a stronger effect than eBMD for both hip and knee OA (Table 2). Sanderson-Windmeijer Fstatistics were >1000 for both instruments. Results were generally consistent when using individual SNPs as instruments, although the evidence for a causal effect of eBMD on knee OA was weakened, as was the instrument strength estimated by F-statistics (Supplementary Table S9, available as Supplementary data at IJE online).
MVMR provided evidence for a BMI-independent causal effect of OA on eBMD [b hip ¼ 1.23 (95% CI ¼ 0.45, 2.01), b knee ¼ 5.84 (95% CI ¼ 2.69, 8.99), Table 2]. The causal effect of BMI on BMD was independent of hip OA [b ¼ 0.08 (0.04, 0.12)]. When conditioning on knee OA, an inverse effect of BMI on BMD was observed [b ¼ -0.19 (95% CI ¼ -0.39, 0.00)]. This is unlikely to be bias due to conditioning on a common outcome (i.e. collider bias), as genetically predicted OA is not a common outcome. 18 The BMI-independent causal effect of knee OA on eBMD was not observed when using individual SNPs as instruments (Supplementary Table S9, available as Supplementary data at IJE online).
LCV analyses provide evidence for a non-pleiotropic causal effect of BMD on OA To determine whether shared underlying genetic aetiology fully explained the observed causal effect of BMD on OA, we performed LCV modelling using weight-adjusted summary statistics for both FN/LS-BMD and hip/knee OA. The LCV analysis identified evidence for genetic correlations between BMD (measured at both the FN and LS) and OA at both the hip and knee (rho ¼ 0.16-0.23, Supplementary Table S12, available as Supplementary data at IJE online). There was evidence for a partial causal effect of BMD at both sites on OA at both sites, independently of genetic correlation and weight, with the largest magnitude of causal effect observed for FN-BMD and knee OA, with a genetic causality proportion of 0.64.
Discussion
We have found strong evidence for a causal effect of BMD on hip and knee OA using 1SMR, which was relatively consistent with 2SMR. MVMR confirmed that the effect of BMD on OA is independent of BMI. Our results also suggest that there is a bidirectional causal effect between OA and eBMD. We have confirmed strong causal effects of BMI on eBMD, hip and knee OA, with no causal effect of eBMD on BMI. Finally, we have found some evidence of a positive causal effect of knee and hip OA on BMI. The observed causal effect of BMI on eBMD in this adult population is consistent with a previous analysis of a paediatric population (mean age 10 years), in which a causal effect of BMI on FN-BMD was observed. 9 As seen in this current analysis, Kemp et al. found no evidence for a causal effect of BMD on BMI. 9 The strong causal effect of BMI on both hip and knee OA corroborates previous MR analyses. 5,24 The causal effect of eBMD on hip and knee OA that we observed is consistent with previous MR analyses identifying causal effects of FN and LS-BMD on hip and knee OA 5,24 and our recent analyses showing that generalized high bone mass (BMD Z-score >3.2 at the hip or L1) is related to greater worsening of osteophyte and joint space narrowing (JSN) severity at both the hip and knee. 25,26 Taken together, these findings suggest that bone parameters in general have a causal effect on OA, regardless of the site or method of measurement. However, the magnitude of the effect of eBMD on OA was larger in 2S analyses restricted to SNPs associated with FN-BMD. There are two potential explanations for a stronger effect of BMD on OA when restricting to FN-BMD loci. First, FN-BMD measured by dual-energy X-ray absorptiometry may be a more accurate representation of the biological pathways between bone and cartilage, compared with eBMD, which represents a combination of speed of sound and broadband ultrasound attenuation. Alternatively, since the FN primarily comprises cortical bone, whereas heel BMD is predominantly trabecular, 27,28 these findings may reflect the fact that cortical bone is more strongly related to OA pathogenesis compared with trabecular bone. For example, cortical BMD might be expected to correlate more strongly with subchondral plate sclerosis compared with trabecular BMD, which is implicated in the progression of OA. 29 Inconsistently with the results of our analysis, some previous studies have provided evidence to suggest that high BMD is related to reduced progression of OA, 30,31 although this could be explained by index-event bias, in which conditioning on OA leads to spurious associations between OA risk factors. 32 We have found some evidence for reverse causality in the relationship between eBMD and OA. The positive direction of effect is as expected from artefactual elevation, rather than loss of bone mass due to reduced PA. However, as we do not expect BMD measurements at the heel to be artefactually elevated by features of OA, the observed causal effect of OA on eBMD in 1SMR may reflect biological pleiotropy (i.e. the same underlying biological pathways may be contributing to both phenotypes). Consistently with shared biological mechanisms contributing to both BMD and OA, Hackinger et al. identified a genetic correlation between LS-BMD (but not FN) and OA. 11 By performing a cross-phenotype meta-analysis between OA and LS-BMD, the authors identified a number of known loci, as well as a novel locus in the SMAD3 gene. 11 SMAD3 is part of the transforming growth factor b (TGFb) signalling pathway, which regulates osteoblast differentiation. The first discovered OA loci, growth differentiation factor-5 (GDF5), is a ligand for this pathway. 33 The canonical Wnt signalling pathway is involved in the regulation of osteoblasts and mutations in this pathway can lead to high or low BMD; e.g. activating mutations in low-density lipoprotein receptor-related protein 5 (LRP5, the receptor involved in Wnt signalling activation) cause high BMD. 34 This signalling pathway has been implicated in OA pathogenesis; 35 increased levels of a Wnt signalling inhibitor, DKK1, were associated with reduced progression of hip OA in a population of Caucasian women. 36 However, we did find stronger, more consistent, evidence for an effect of eBMD on OA, as opposed to vice versa. This could reflect the stronger instrument for BMD, but our LCV analyses using the full set of summary statistics provided further evidence for a causal pathway between BMD and OA, not driven by genetic correlation (or confounding by weight as evidenced by the MVMR), suggesting that bone may still have a direct effect on OA, e.g. via increased joint loading or through related structural alterations in the subchondral bone, such as denser subchondral trabecular bone, which has been linked to the progression of JSN. 37
Strengths and limitations
We have utilized the largest data sets possible to maximize the power to detect causal effects. We have ensured that there is no overlap between our exposure and outcome populations. We have examined individual-level data in UK Biobank to perform 1SMR to strengthen evidence.
However, we were unable to use eBMD instruments for 1SMR as they were identified in the same population used for analysis; reassuringly, F-statistics suggested that our FN-BMD instrument was of reasonable strength. We did not use the largest available meta-analysis as the source of the BMI instruments due to significant sample overlap with UK Biobank. 38 However, the Locke et al. Europeanonly meta-analysis, which we used for our instrument source for both 1S and 2SMR, still included >300 000 individuals and identified 77 loci; the PRS generated from these SNPs had a strong F-statistic suggesting that the magnitude of effects identified in 1SMR analyses are unlikely to be explained by bias due to weak instruments. Our OA outcomes for 1SMR were based on hospital diagnosis; it is unclear how this phenotype relates to radiographic features of OA, such as JSN, which are commonly used as clinical trial outcomes. Using a severe phenotype as the outcome means reduced power in GWAS and leads to fewer genome-wide significant loci and a greater chance of weak instrument bias (as highlighted by the much smaller F-statistics for the OA instruments). The OA outcomes from the GO consortium included a range of definitions of hip and knee OA, including hospital diagnosis, radiographic evidence and self-reported OA definitions. Heterogeneity in phenotype also reduces the power to detect loci in GWAS. The ORs from 1SMR are estimates and standard errors (SEs) are likely underestimated, 15 so caution should be taken when interpreting these effect sizes. There may be additional risk factors related to the genetic variants that we did not include in our MVMR models. The UK Biobank population is limited by a latent population structure even after restricting to White Europeans and adjusting for PCs, 39 which may confound estimates generated by 1SMR. The UK Biobank population examined was White British and all instruments were derived from predominantly White European populations, meaning that we were unable to examine causal effects in non-European populations, limiting generalizability to other ethnicities for whom the prevalence of, and therefore risk factors for, osteoarthritis may differ. [40][41][42][43] The prevalence of OA is higher in men from UK Biobank compared with women, despite evidence in the general population suggesting a higher prevalence of knee OA in women. 44 This could be explained by selection bias, as women and healthy individuals (i.e. free of OA) are more likely to participate in UK Biobank. Although we adjusted for sex in our analyses, it is possible that there are other variables related to participation in UK Biobank that we could not account for in our analyses. Individuals with OA have a higher risk of premature mortality than the general population, 45 which could cause further selection bias if those with severe OA are less likely to survive to participate in UK Biobank. However, this selection bias is unlikely to explain the observed positive causal effect of OA on BMI, but may explain the positive causal effect estimate for OA on eBMD.
Conclusions
We have found evidence for a BMI-independent causal effect of BMD on hip and knee OA and some evidence for a bidirectional causal effect, which we hypothesize to reflect shared underlying genetic aetiology. We have confirmed strong causal effects of BMI on BMD and hip and knee OA, and have found novel evidence for a causal effect of knee OA on BMI, which did not appear to be mediated by pain-associated reductions in PA. Further analyses are required to determine the shared pathways contributing to both BMD and OA, and to determine the mechanisms by which higher BMD causes OA.
Notes
Oncology & Metabolism and Healthy Lifespan Institute, University of Sheffield, Sheffield, UK; 42 Institute of Biomedical Sciences, academia Sinica, Taipei, Taiwan.
Supplementary data
Supplementary data are available at IJE online.
Ethics approval
UK Biobank received ethical approval from the Research Ethics Committee (REC reference : 11/NW/0382). | 6,140.4 | 2021-03-26T00:00:00.000 | [
"Medicine",
"Biology"
] |
Morphological & Biochemical Effects of Aqueous Extract of Neem (Azadirachta indica) on Liver of Adult Albino Rats
Introduction: Neem (Azadirachta indica) is an important medicinal plant which is traditionally used all over the world as household remedy in diabetic and hypertensive patients and also for the cure of various dermatological ailments. Aims & Objectives: To evaluate the gross and biochemical effects of aqueous extract of neem leaves on the liver of adult albino rats. Place and duration of study: The study was undertaken at the Anatomy department of Shaikh Zayed Postgraduate Medical Institute, Lahore. Material & Methods: 45 Albino rats of both genders were used and equally divided into group A (Control), group B (low dosage) and group C (high dosage), each containing 15 animals randomly. The rats of group A received distilled water, while group B and C received 40 mg/kg and 100mg/kg of aqueous extract of neem respectively for 20 days using an orogastric tube. At the end of complete dosing schedule, rats were sacrificed and the livers were dissected out for examination. Animals were evaluated for gross (appearance of rats, appearance of liver, body weight of animal, weight of liver, relative tissue body weight index) as well as for biochemical (Serum ALT levels) parameters. Results: The body weight and weights of livers of experimental groups were decreased as compared to control group and it was statistically significant with p-value < 0.001. Similarly biochemical parameters were markedly impaired in group B and group C. Conclusion: The present research work demarcates that the higher doses of neem extract induce remarkable gross and biochemical effects on liver.
INTRODUCTION
Neem is related to the Mahogany tree of Meliaceae family. Its botanical name is Azadirachta indica A. Juss (Syn: Melia Azadirachta). 1 Neem is native of India and harvested in tropical and sub tropical countries with widespread distribution throughout the world. 2 Azadirachta indica is evergreen tall tree (up to 20-30 meters). It has white flowers that develops into bunches of small fruits which are swollen and look like olives. Its complex leaf pattern resembles to that of walnut with spreading branches like crown. 3 It starts fruiting from 3-5 years and has a life expectancy up to 200 years. Neem shows an adaptation to wide range of temperature. It can survive in hot temperature as high as 44 °C and tolerates cold from 0°C -4°C to an altitudes up to 500 m. Usually grows in those areas where annual rainfall is from 450 mm to 1200 mm. 3,4 Literally each and every part of the plant is blessed with medicinal properties and is commercially available. It is mother of every therapeutically used plant that has been utilized extensively many decades back and still been utilizing for medicinal and ritual purposes. It's low cost and easy availability has enabled many individuals to pick up advantage from this dynamic plant. 5 The neem tree contains about 140 bioactive ingredients and it is rich in proteins. Azadirachta indica is bitter in taste because of the presence of complex compounds called "Triterpenes". Its tree contains about 40 different types of active agents known as Tetranoterpernoids or Limonoids. 6,7 Several chemical constituents have been isolated from neem leaves using different solvent systems like water (hot or cold), ethanol, methanol, acetone, petroleum, ether and hexane. The most active and well known chemical compounds found in neem are Azadirachtins. 8,9 It is highly oxidized triterpenoid. Azadirachtins rapidly dissociates in the presence of light and moisture. But Azadirachtin is stable when composed in oily medium together with neem natural compounds. 10,11,12 The accurate mechanism of neem is not well recognized but possible pharmacokinetics of neem leaves are involvement with mitochondrial bioenergetics which leads to inhibition of electrochemical proton gradient which is the main energy generated in mitochondria. This inability to utilize oxygen is presented as cytotoxic hypoxia and that ultimately leads to metabolic acidosis and hyperpnoea 13 . Neem also interacts with receptors and changes membrane permeability and integrity. 13,14 It has been proven that Azadirachta indica leaf and bark extract inhibit prostaglandin synthetase as compared to acetyl salicylic acid. 15,16 It has been documented that neem is an effective hypoglycaemic agent as it increases insulin secretion and promotes the utilization of glucose by peripheral receptors as it hinders the effects of epinephrine on metabolism of glucose. 16,17 They have tremendous fungicidal and bactericidal properties along with the quality of regulating growth in insects. In addition one of the major ingredients of neem leaf aqueous extract is Nimbidin which has anti-inflammatory, anti-pyretic, anti-arthritic, hypoglycaemic, antiulcer and antitumor effects. 17,18 Various researches have been carried out to evaluate the effects of neem on different animals. The Azadirachta indica leaf aqueous extract caused damage to seminiferous tubules of male mice at a dose of 200mg/kg destruction and also distortion of spermatogenesis. 19 Similar effects are also observed in testis of monkeys. 20 A study was conducted to show the effects of Vepacide, a neem based pesticide on the biochemical enzymes like aspartate aminotransferase (AST) and alanine aminotransferase (ALT) in serum and different tissues of rats of both genders. Albino rats were given low (80mg/kg), medium (160mg/kg) and high (320 mg/kg) dosage of coconut oil containing vepacide orally for 90 days. Biochemical profile suggested that by increasing the dose of vepacide caused an increase in the ALT and AST enzymes of serum, kidneys and lung tissues whereas the levels of these enzymes were lowered in liver tissue. As the neem is extensively used by people as selfmedication without proper medical advice and its use is increasing day by day and the possible consequences lead to an evaluation of effects of chronic therapy with the drug. Therefore, the present research is being conducted with an aim to observe the consequences of neem leaves aqueous extract on morphology of adult albino rats.
MATERIAL AND METHODS
45 adult albino rats of both strains (weighing about 180-200g) were obtained from Veterinary Research Institute, Lahore. They were divided in three groups; A (control), B (experimental low dose) and C (experimental high dose). Each group consisted of 15 rats. The weight of each rat was carefully recorded in a Performa. For identification, the rats were marked with permanent pointer and were placed in different cages for 21 days. A 12 hours light / dark cycle was maintained. 21 The animals were allowed free access to food and water. . ALT levels of all three groups were measured in 1cc of blood in each rat prior to the administration of the neem extract or the start of experiment. Fresh leaves of neem were collected locally and then its extract was obtained from the PCSIR, Laboratories Complex, Lahore. Water extraction was performed by refluxing the powdered leaves with distilled water. 22 Aqueous extract of neem leaves was given to the animals by orogastric intubation. The control group A containing 15 rats and were not given any extract except for equivalent proportion of 0.1% distilled water, 20ml per kg of weight of body by orogastric intubation for 20 days. The experimental group B containing 15 rats, each of which received 40mg per kg body weight of neem extract by orogastric intubation for 20 days (for example: if the weight of rat was 180g, 7.2 mg dry neem powder dissolved in 2ml of distilled water was given. It means each ml contains 3.6 mg of powder.) Then the experimental group C contained 15 rats. All the animals of this group were given 100 mg per kg body weight of neem extract by orogastric tube for 20 days (For example: if the weight of rat was 180g, 18 mg dry neem powder was dissolved in 2ml of distilled water. It means each ml contains 9 mg of powder). Hence the dose was adjusted according to weight of the rat. At the end of study, the rats of all groups were weighed properly and recorded in performa. For ALT determination, blood samples were collected from each rat through tail and allowed to clot. All the rats were euthanized by giving morphine 0.3-0.5mg/kg intraperitoneally, as an analgesic agent. The anaesthetic agent sodium pentobarbital was administered intraperitoneally with dose of 45mg/kg. 13 The animals were put in a supine position with their belly facing up and limbs fixed to the dissection board. A midline incision was made with a pair of scissors from groin to chin and extended laterally. Liver was made free from surrounding structures and placed on a blotting paper to make it free of blood and fluids. After recording the weight of the liver, it was washed with normal saline to remove blood and fixed with 10% formalin for 48 hours in appropriately labelled tissue bottles.
Statistical analysis:
The overall data was calculated and compared with the help of computer software Social Package of Statistical Sciences (SPSS) version 24. Qualitative variables like gross appearance were described by using frequencies and percentage for each group. Comparison for these qualitative variables among groups was performed by using CHI-SQUARE test. Quantitative variables like weight of animal before sacrificing, weight of liver, relative tissue weight index (RTWI) & serum ALT levels were described by using mean ± S.D. Comparison for these quantitative variable was performed by using ANOVA, Tukey's test for post-hoc analysis was used where required. P-value ≤ to 0.05 was considered statistically significant.
RESULTS
The animals of control group A were active and healthy looking throughout the experiment. Eating habits of this group was normal. Statistically this parameter is constant.
At the end of experiment, 3 animals in group B and 7 in group C looked apparently abnormal, as compared to control group A. These animals were lazy and sluggish in response. There was statistically significant difference in the gross appearance of rats among three groups with p-value 0.003. There was significant decrease in the weight of animals of group B & C as compared to control group A. When comparison was made for liver weight, the animals of group A had higher weight for liver than that of experimental groups. (Table-1, Table-2). Similarly the relative tissue body weight index was significantly reduced in group B and group. (Table-3) The external surface of all 15 livers of control group A was smooth and their colour was reddish brown. But the gross appearance of livers of 3 rats of experimental group B and 7 rats of experimental group C showed haemorrhagic areas on their external surface, randomly affecting all lobes as compared to smooth surface of control group A. (Table-1 , Fig-1,2,3) The serum was rapidly separated by centrifugation of clotted blood. Sera were stored at -20°C until assayed for the biochemical parameter. Alanine amino transferase were determined on fully automatic chemistry auto analyzer, Dimension, RXL from Siemens, USA. There was slightly increase in ALT levels from experimental group B and group C with p-values 0.002 and <0.001 respectively. Chi-square = 45.0 p-value < 0.001
DISCUSSION
Generally Neem has been utilized as a vital piece of our lives for quite a long time as an insecticidal, 23 disinfectant, prophylactic, antipyretic, antiparasitic, antiarthritic, 21,22 antifungal and hypoglycaemic agent. 23,24 People use it as an alternative treatment for a variety of health ailments and skin problems. 25 Neem leaf extract is common supplement which is easily available in drug stores and raw leaves are often consumed by many people in our society. It is usually considered as safer in wide range of doses. Now it has been proved that it is toxic at higher doses especially when used for longer duration. Its harmful effects on liver, kidneys, lungs and male and female reproductive organs have also been well documented. It is considered very dangerous to the children when used in any dose. 25,26 Azadirachtin is the most bioactive tetranortriterpenoid which is the bitter component of neem 9,10,11 and has been proved lethal. Its unique quality is due to its anti-feedent properties against insects. It acts as systemic poison, moult inhibitor, causes delay in post embryonic development, antifertility effects, chitin and enzyme inhibition. 16,26 The present work was designed to evaluate the harmful effects of neem on morphological and biochemical parameters of liver as it is the main metabolizing organ for neem and its constituents. It has been suggested that neem based formulations affect membrane alterations and influence the oxidant defence mechanism. 27 In the present study the gross appearance of rats of control group A were normal after the administration of neem extract but the 3 animals of group B and 7 animals of group C were lazy, weak and not responding to the particular stimuli, at the end of experiment. The difference in appearance was significantly different among different groups with p-value 0.003. In this study the average body weight of experimental groups B and C showed gradual weight loss and feed intake after neem consumption as compared to control group A. The difference was significant with p-value <0.001, suggesting toxic stress and growth retardation in treated rats. The result of this present research work showed overall weight loss which coincides with the findings of previous studies by Rahman and Siddique that showed decreased appetite and weight loss of albino rats at higher doses of neem extract. 28 The result of the present study revealed that the gross appearance of livers of 3 rats of experimental group B and 7 rats of experimental group C had haemorrhagic areas on external surface randomly affecting all lobes as compared to smooth surface of control group A. These results showed that by increasing the dose of neem, the haemorrhagic areas were also increased. It was possibly due to the toxic stress in treated rats especially on blood and blood forming elements. 27,28 These findings correlate with the study performed by Samuel, Thomas and others, who observed the dose related prolongation of PT and APTT values by oral administration of crude neem leaf acetone-water extract on albino rats. They suggested that it may be due to impaired liver function which could influence directly or indirectly the synthesis of coagulation factors. 28 When the comparison of average liver weights was made among three groups, it was observed that the average liver weight was decreased in experimental group B which was low dose group and further reduced in high dosage that was group C. This clearly showed that the decrease in animal body weight and liver weights resulted in decreasing the RTWI values in experimental groups. The possible mechanism of toxicity due to neem based formulation is time, dose and tissue specific inhibition in glutathione-s-transferase, reduced glutathione and UDP-glucuronyl transferase activity in liver, lungs, kidneys and brain. Also there is decrease in cytochrome P-450 reductase activity in liver and brain. 28 In the present study, the average ALT level for group A was significantly lower than of group B and C with p-values 0.002 and <0.001, the level of group B was significant from group C with p-value <0.001. This finding correlates with the studies of Rahman and Siddiqui, who proved that the exposure to neem based pesticides caused an increase in aspartate and alanine aminotransferase in serum, kidneys and lungs whereas these enzymes decreased in liver. 26,27 This might be due to increased permeability of plasma membranes followed by the necrosis of cellular tissues caused by the neem.
CONCLUSION
This examination demonstrated that organization of aqueous neem leaf extract in high dosages for longer time caused noteworthy negative effect on the morphology and biochemical parameters of liver of adult albino rats. . Although neem is commonly used as non-allopathic medicine but dosing still is not standardized. There is a need to evaluate safer dose and duration of usage of neem in general public. | 3,641.4 | 2018-09-02T00:00:00.000 | [
"Medicine",
"Biology"
] |
A Minimal Subset of Features Using Feature Selection for Handwritten Digit Recognition
Many systems of handwritten digit recognition built using the complete set of features in order to enhance the accuracy. However, these systems lagged in terms of time and memory. These two issues are very critical issues especially for real time applications. Therefore, using Feature Selection (FS) with suitable machine learning technique for digit recognition contributes to facilitate solving the issues of time and memory by minimizing the number of features used to train the model. This paper examines various FS methods with several classification techniques using MNIST dataset. In addition, models of different algorithms (i.e. linear, non-linear, ensemble, and deep learning) are implemented and compared in order to study their suitability for digit recognition. The objective of this study is to identify a subset of relevant features that provides at least the same accuracy as the complete set of features in addition to reducing the required time, computational complexity, and required storage for digit recognition. The experimental results proved that 60% of the complete set of features reduces the training time up to third of the required time using the complete set of features. Moreover, the classifiers trained using the proposed subset achieve the same accuracy as the classifiers trained using the complete set of features.
Introduction
Handwriting recognition is the ability of recognizing handwritten text from a scanned file, image, touch-screen or other tools and converting it into an editable text [1]. Handwriting recognition is a quite complex problem. The ideal goal How to cite this paper: Alsaafin, A. and The recognition of handwritten digits is not a new problem. It is a problem that has continued to be an active topic in research for several reasons [5]: The problem is suitable for image processing and pattern recognition using a small number of classes. Availability of standard benchmark datasets reduces the effort and time spent in preprocessing the data. A lot of work and research have been done in this field that can be cited and built on. The practical applications encourage more research to be done.
Improvements in classification accuracy using existing approaches continue to be achieved using new approaches.
In addition, handwritten digit recognition is an active subject in OCR (Optical Character Recognition) applications and pattern classification/learning research [6]. OCR applications like postal mail sorting, bank check processing, and form data entry require high accuracy and speed techniques to achieve a satisfactory performance. The aforementioned reasons motivate us to proceed with this study.
In this paper, we employ a Feature Selection (FS) method in order to select a subset of relevant features using the MNIST dataset. The FS method used in this paper is called Feature Importance. This paper also implements various classification techniques in order to study their suitability for digit recognition. It presents a statistical analysis that compares models of different algorithms, The goal behind using classifiers of different algorithms is to find out the most suitable and efficient algorithm for digit recognition in terms of training time, accuracy, and confusion matrix. In addition, the purpose of introducing this comparison is to limit the algorithms that can be used to solve such a problem and devote more effort to develop the suitable techniques by doing more experiments and research.
The rest of this paper is organized as follows. Section 2 reviews related literature. Sections 3 and 4 include description of the dataset, FS, and classification techniques used in this paper. The stages of the performed experiment and the A. Alsaafin, A. Elnagar Journal of Intelligent Learning Systems and Applications obtained results are presented in Section 5. Section 6 includes some discussion before concluding this paper in Section 7. LeCun et al. [7] reported different accuracies obtained from various classifiers.
Related Work
The highest accuracy, 99.30%, was obtained via boosted convolutional neural network (CNN) that is trained using distorted data. Simard et al. [8] enhanced distorted sample generation as well as the implementation of CNN. This led to a small improvement in test accuracy, 99.60%. Liu et al. [6] used gradient direc-Journal of Intelligent Learning Systems and Applications tion features to implement several classification methods. The test accuracies that have been achieved are 99.42% by polynomial classifier, 99.58% by SVM classifier, and over 99% by many other classifiers. Holmstrom et al. [9] compared different statistical and neural classifiers based on PCA features. However, the PCA feature does not perform satisfactorily. Liu et al. [10] claim that training classifiers using MNIST dataset without using feature extraction methods shows inferior performance. This motivates us to compare the performance of various classification models trained with a subset of features against the complete set of features available in the MNIST dataset.
1) MNIST dataset
In this paper, we use MNIST dataset. The MNIST is a dataset developed by LeCun, Cortes and Burges for evaluating machine learning models on the handwritten digit classification problem [11]. It has been widely used in research and to design novel handwritten digit recognition systems. The MNIST dataset contains 60,000 training cases and 10,000 test cases of handwritten digits (0 to 9). Each digit is normalized and centered in a gray-scale (0 -255) image with size 28 × 28. Each image consists of 784 pixels that represent the features of the digits. Some examples from the MNIST dataset are shown in Figure 1. The MNIST dataset is balanced over the ten classes (0 -9). Figure 2 shows the percentage of each class in the MNIST dataset. 2) Feature selection methods Features that are used in any machine learning method have a huge influence on the obtained results. Having Irrelevant or partially relevant features can negatively affect performance of many models, especially linear algorithms like logistic regression [12].
Feature selection (FS) is a process of selecting features that contribute most to the prediction variable. An FS method selects a subset of relevant features and discards irrelevant/redundant features. The benefits of performing FS before data modeling are [12]: Reduces Overfitting: Less redundant data implies less chance to make decisions based on noise. Improves Accuracy: Less misleading data implies modeling accuracy improves. Reduces Training Time: Less data implies that algorithms train faster.
In general, there are two types of FS methods: filter method and wrapper method. The filter method selects a subset of the most relevant features based on the characteristics of the dataset. This type of FS methods is independent of the classification algorithms unlike the wrapper methods that use a predetermined classifier in order to evaluate the selected subset of features. The wrapper method is more computationally expensive than the filter method [13]. In this research, we selected a wrapper FS method called Feature Importance. The Feature Importance is a method that estimates the importance of features using bagged decision trees like random forest and extra trees [12].
Our goal in this project is not to improve machine leaning algorithms but compare the performance of different algorithms using a subset of features Journal of Intelligent Learning Systems and Applications against the complete set of features. Therefore, we use a simple FS method that serves our goals.
Classification Techniques
Classification is a process of finding a model/function that recognizes, describes, and differentiates two or more classes. The purpose of using a classification model is to predict the class of an object where its class label is unknown. Classification has a predictive nature-for a given set of features, the goal is to attempt to predict the value of another feature [14]. A classifier model is used to classify the actual data into defined classes. Ultimately, patterns need to exist in the data that can be exploited. The classification techniques used in this paper are explained below.
1) Multinomial Logistic Regression:
A Logistic Regression is a binary classifier that can distinguish between two classes. Multinomial Logistic Regression (MLR) is a logistic regression that is designed to solve multiclass/multinomial classification problems. In other words, Multinomial Logistic Regression is a model that predicts the probabilities of different possible outcomes of a categorically distributed dependent variable, given a set of independent variables [15]. It is used when the dependent variable is nominal and falls into one of many (i.e. three or more) classes/categories (e.g. the MNIST dataset has nine classes).
MLR classifier is commonly used as an alternative to naive Bayes classifier since it does not assume statistical independence of the random features that used as predictors. The MLR classifier is simple and requires a little time to learn, however, it becomes slow when use a very large number of classes to learn.
2) Decision Tree:
A Decision Tree (DT) classifier is a flow-chart-like structure, where the topmost node represents the root node of the tree, each intermediate node denotes a test on an attribute, each branch represents the outcome of a test, and each leaf node indicates a class label. The paths from root to leaf represent classification rules.
DT is a simple model to understand and interpret. However, computation can get very complex if many outcomes are liked (i.e. a decision tree reaches a great depth). The DT classifier becomes biased in favor of attributes with more levels/depth; however, it can be easily combined with other decision techniques in order to make better and accurate decisions [16].
3) Random Forest:
A Random Forest (RF) classifier [17] is one of the most effective techniques that is used for predictive analytics. It is an ensemble learning classifier that combines decisions from a sequence of base classifiers in order to make a decision. RF is correct for decision trees' habit of overfitting to their training set [18].
Decision Trees that have great depth tend to overfit their training sets thus have low bias but very high variance. RF classifier reduces the variance by averaging is an additive technique that is developed by Friedman [19]. It combines predictions from a sequence of base classifiers in order to make a prediction. Typically it is constructed of a number of simple/weak DT models where the output is the sum of the decisions of these base learners. The purpose of BT method is to achieve better predictive performance by minimizing the loss function when adding trees. Generally adding more trees makes the model more resistance to overfitting. Therefore, keep adding trees until no further improvement is observed can lead to build an efficient model that gives very accurate predictions. 5) Convolutional Neural Network: A Convolutional Neural Network (CNN) is a deep learning method that can be used for the purpose of feature extraction as well as classification. Hence, CNN acts as a feed-forward network that extracts topological properties from images [20]. It extracts features from raw images (i.e. contain the intended pattern) in its first layers, and then classifies the pattern with its last layers [20].
CNN is a powerful technique for image processing as well as natural language processing. The main advantage of CNN is the high accuracy in its results. However, it requires high computational cost. In addition, it needs a lot of data to be trained. The complexity of the CNN slows down the training process thus it is necessary to use a good GPU to overcome this problem.
Experimental Results
As previously mentioned, we use the MNIST dataset to perform the experiment of this study. In addition, we use a FS method to be examined with different classifiers: MLR, DT, RF, and BT. At the last stage of this study, we compare the same classifiers used at the previous stage in addition to CNN. In this section, we describe the tools that used throughout the experiment of this study, describe all the experiment stages, and finally present and discuss the obtained results.
1) Hardware and Software Tools Since we consider some factors (e.g. training time) that are affected by the tools used in the experiment, it is worth mentioning the hardware and software that have been used throughout this research. For the hardware tools, we used a laptop of MacBook Pro, 2.8 GHz Intel Core i7 with 4 GB RAM. All the setup and computations of this research have been performed on top of the CPU. For the software tools, GraphLab Create 1 is used for the implementation of this paper. GraphLab Create is a machine learning framework that is python-based libraries. It is designed to handle the major properties of real world data: scalable, variety, and complexity. The biggest advantage of using GraphLab is that it supports scalability, which allows using large datasets. It also supports different data Journal of Intelligent Learning Systems and Applications source such as JSON, CSV, HDFS/S3 and many more. Graphlab is a complete framework that has rich libraries for data transformation and manipulation. It is worth mentioning that we import some libraries from scikit-learn for the purpose of implementing the FS method used in this paper.
2) Pre-processing
Since the MNIST dataset is divided into standard training and testing sets, 80% and 20% respectively, we divide the MNIST training set into two proportions, 80% as a training set and 20% as a validation set. Therefore, the MNIST dataset is divided into three sets: training set (60%), validation set (20%), and testing set (20%). The training set is used to train the classifiers for the purpose of recognizing handwritten digits while the test set is used to assess the trained classifiers after fine-tuning them using the validation set.
Before training the classifiers, we select subsets of various sizes (i.e. number of features) using Feature Importance method. Table 1 The last step of this study is to compare the performance of the classifiers using the subset of features (60%) against the complete set of features. As shown in Table 2 and Table 3 a) Train and fine-tune the classification techniques In order to have a fair and valuable comparison, all the classification techniques are trained using the complete set of features. The MLR, DT, RF, and BT classifiers are trained as same as they trained in the Study I. A one layer CNN has been trained to classify an input to one of the ten classes (0 -9) available in the dataset. Similar to the other classifiers, the CNN is fine-tuned by increasing the number of iterations in order to fulfill the training criteria. b) Compare the classification techniques Table 3 presents the training accuracy, validation accuracy, test accuracy, and training time for all techniques after fine-tuning them based on the validation set. In this comparison, we did not consider the time required for testing since we use single classifiers (i.e. not combined classifiers) that do not require more than few seconds to predict any new input. As shown in Table 3, the MLR, DT, and RF have low test accuracy compared with the BT and CNN. However, they outperform the BT and CNN in terms of training time. The DT has the lowest training time (71.11 sec), although it has the lowest test accuracy (90.94%) among the four techniques. The CNN outperforms all the techniques in terms of test accuracy (98.36%). The BT has the highest training accuracy (100%) and the second highest test accuracy (97.91%). However, the BT requires the longest time to be trained among all other techniques (1396.03 sec). Figure 4 shows the confusion matrix of the BT and the CNN since they have the highest test accuracy compared with other techniques. The confusion matrix shows a more detailed breakdown of correct and incorrect classifications of each class on the MNIST test set. It composed of target (actual) label, predicted label, and the count of predicting the target label as the predicted label. As shown in Figure 4, the BT and CNN has misclassified digit "5" 19 times, mostly it was classified as digit "3". The CNN outperforms the BT in classifying all the digits except digit "6" and "8". The highest misclassified digit using the BT classifier is digit "7" that has been classified as digit "2" 15 times, while digit "6" is the most misclassified digit using the CNN, 16 mistakes. Using the CNN, digit "6" was often classified as digit "0". However, the CNN was the best in classifying digit "0", only 3 mistakes. On the other hand, the BT classifier was the best in classifying digit "1", 9 mistakes. In general, we suggest that most of the mistakes made by the BT and CNN were due the same shape that some digits have (e.g. "0" and "6"). Figure 5 shows the common and individual mistakes of the CNN and BT when classify digit "7" as digit "2".
A. Alsaafin, A. Elnagar Journal of Intelligent Learning Systems and Applications
Discussion
It is worth mentioning that the results obtained in the previous section can be optimized using different methods. We convinced of these results since our target was to find the most suitable classification techniques that can be used to solve the problem of recognizing handwritten digits. From this experiment, we can conclude that the CNN and the BT are the most suitable classification techniques among all other techniques used in this research or similar to them. Since the BT classifier requires a long time to be trained using the complete set of features, we suggest training the BT using a subset of 60% of features using FS method. This reduces the required time, complexity of the trained model, and required storage for digit recognition while still provides very similar performance to the model trained with the complete set of features.
As we discussed in the related work section, Liu et al. in [10] claim that training classifiers using the complete set of features available in the MNIST dataset gives unsatisfactory performance. However, the obtained results in the implementation section (Study I) show that the models trained with the complete set of features have accuracies not less than the accuracies obtained for the models trained using FS. Therefore, there is about 40% of features exist in the MNIST dataset are irrelevant/redundant features.
It is worthless to compare our results to other results since most of the related work has some limitations. Most of the related work available in the literature does not specify the type of the reported accuracy: training, validation, or testing accuracy. However, it is very important to identify the type of the obtained accuracy since the training accuracy could reach 100% if we kept fine-tuning the model until perfectly fit the training dataset. Although this leads to overfitting problem that causes a reduction in the test accuracy. We can notice from Table 3 that the BT classifier achieves 100% training accuracy without affecting the test accuracy. This is because the BT is an ensemble classifier that tends to be robust to overfitting, however, it will eventually overfit.
Many studies fine-tune and assess models using the same data (testing set).
This leads the test results to be overly optimistic, which means that the model is trained to give a good estimation on the test results. Therefore, assessing the models using the test set will be worthless since the model will give bad results once used to predict new data. To avoid this problem, the model should be fine-tuned using the validation set and then assessed using the test set.
The last point of this discussion is that some studies that use un-standard datasets do not present the distribution of the instances/examples available in the dataset. Imbalance datasets lead to have a high test accuracy with many false positive predictions. Therefore, the accuracy metric is not enough to evaluate the classification techniques since it does not capture everything. The confusion matrix is a very useful metric that shows the details of the predictions made by the trained models.
Conclusion & Future Work
This paper proposed a subset of 60% of features to train classification techniques for digit recognition. In addition, it has implemented and compared models of different algorithms: linear, non-linear, ensemble, and deep learning to study their suitability for digit recognition. The obtained results show that the proposed subset of features is not only minimized but also reduces the required time for training the model. Moreover, it reduces the computational complexity and lowers the required storage for digit recognition. The BT and CNN proved their suitability for digit recognition in terms of accuracy. However, the BT classifier requires using a subset of features since it needs a long time to be trained.
We evaluated the performance in terms of accuracy, training time, and confusion matrix.
In future, we plan to examine more FS methods with different classification techniques. Combining various FS methods is another area to be explored. Based on our observations from the confusion matrices of the BT and CNN, some digits are well recognized by some classifiers while misclassified by other classifiers. This observation motivates us to try FS with a combined machine learning techniques in order to enhance the required time as well as accuracy. | 4,866.4 | 2017-09-26T00:00:00.000 | [
"Computer Science"
] |
From Modelling Turbulence to General Systems Modelling
: Complex adaptive and evolutionary systems can, at least in principle, be modelled in ways that are similar to modelling of complex mechanical (or physical) systems. While quantitative modelling of turbulent reacting flows has been developed over many decades due to availability of experimental data, modelling of complex evolutionary systems is still in its infancy and has huge potential for further development. This work analyses recent trends, points to the similarity of modelling approaches used in seemingly different areas, and suggests a basic classification for such approaches. Availability of data in the modern computerised world allows us to use tools previously developed in physics and applied mathematics in new domains of scientific inquiry that previously were not amendable by quantitative evaluation and modelling, while raising concerns about the associated ethical and legal issues. While the utility of big data has been repeatedly demonstrated in various practical applications, these applications, as far as we can judge, do not involve the scientific goal of conceptual modelling of emergent collective behaviour in complex evolutionary systems.
Introduction
The remarkable expansion of the internet and computer technologies has brought unprecedented opportunities for accumulating data, implying that hypotheses and theories that previously had purely qualitative characteristics can now be quantified and tested [1]. These changes should bring an extension of methods developed in exact sciences and applied mathematics into areas of scientific enquiry associated with social and other related complex systems. These methods, of course, are not limited to basic statistics and conventional data processing, and they involve a number of different approaches associated with modelling complex systems.
One of the most complex mechanical systems we know is turbulence [2]. Turbulent reacting flows combine the complexity of randomness and coherence with a large number of species and reactions. Modelling turbulent reacting flows was perhaps one of the first attempts of quantitative simulations of a fairly complex system. While these efforts had their successes and failures, they certainly led to the development of advanced modelling tools (recently reviewed in [3]), and some of these tools have generic properties, i.e., can be used to simulate and analyse different complex systems. This conceptual similarity, which is shaped by the nature of complexity, has been occasionally discussed in publications [4]. While the author of this work has introduced and, in cooperation with his colleagues, developed new effective approaches to modelling reacting flows (e.g., conditional methods [3]), which also allow for general systemic applications, this work is not restricted to the consideration of conditional models. We explore a broad scope of conceptual issues associated with modelling complex systems in conditions of emerging revolutionary trends of extending quantified knowledge from mechanical and physical into social and psychological domains.
The key impetus for developing conceptual ideas examined here is, perhaps, summarised best by the following quote:
The New Age of Data Collection and Its Implications for Modelling Complex Systems
The low cost associated with electronic accumulation and storage of data has created a situation where companies and organisations store virtually all data available to them. While this is, indeed, the simplest and safest policy, it does not automatically make organisations more knowledgeable and/or more efficient. As noted by Eisenhauer [6], large volumes of data can obscure or hide information that is really useful or important. To become useful, data need to be processed and properly categorised. This often requires not only knowledge of data processing and database maintenance but also knowledge of the real world that these data represent.
These trends become even more evident when dealing with data used for the expansion of scientific knowledge and research. On one hand, the availability of data characterising complex social systems opens up, at least in principle, the possibility for the development and validation of new quantitative theories and methodologies in areas which were not previously amenable to quantitative analysis. Numerous methods of applied mathematics can now be used for quantitative analysis in areas that previously could be studied only qualitatively. On the other hand, data are just data, and the formal use of mathematical tools without developing proper physical understanding of the processes can lead to illusory successes. For complex systems, these data are usually represented by a sparse set of points in spaces of extremely large dimensions; these points reflect both systemic dependencies and effects of numerous unpredictable factors that can be seen and treated as random. For such systems, conventional tools of data analysis (such as correlation and principal component analyses) may or may not produce useful results. Indeed, even if we observe a correlation or an apparent dependence between two factors, this, generally, does not mean that one of these factors is caused by another since, for example, this co-dependence might be induced by a third factor that remains undetected. A relationship derived purely from data does not, by itself, represent a scientific theory and, if interpreted as a theory confirmed by these data, can lead us to erroneous conclusions. A rational theory needs to be based on the development of understanding, a reasonable hypothesis, application of logic and analysis, construction of a model, performing simulations, and validation and adjustment of the model in comparison with data from the real world. Although important, data are just one of many components needed for creating knowledge. Crossdisciplinary fertilisation of mathematical models and methods is very promising but it cannot be mechanistic and must be based on a deep understanding of both similarities and differences associated with different fields of study.
The new trends in application of mathematical methods to social systems of high complexity can be detected in the domain of internet research and advertising, but the exact scope and extent of these applications often remain unknown to the public; algorithms created and tuned to influence human choice are most effective when people are unaware about them. The few cases published in the literature [7,8] point to gathering extensive data on many people, followed by cross-disciplinary analysis and its subsequent application using individually tailored messages. While these examples are often associated with numerous ethical and legal failures, the question of whether these examples have nevertheless introduced any new science or conceptual understanding remains.
Kaiser [7] examined an apparent success in using data collected about voters by the infamous company "Cambridge Analytica" and characterised this approach not only as "effective" but also as "revolutionary". While this approach may indeed have some intellectual achievements from the practical (although certainly not moral) perspective, does it involve any breakthroughs in science? Experts in public relations know that wellthought and accurately targeted communications are likely to have a more pronounced effect on an individual. The claimed success of Cambridge Analytica had two major factors: (1) delivering personalised messages on a massive scale using modern means of communication and monitoring, and (2) people receiving messages remained unaware that they were placed within a virtual informational environment tailored specifically for them. From the systemic perspective, this is like the spreading of a virus, which is especially fast and effective (for the virus, of course) when we are unaware of its existence. However, after being exposed to various informational infections a few times, people in democratic societies will start learning from their mistakes and acquire some degree of intellectual immunity (and science has an obligation to help). While some people may like personalised messages and services, the second of the two factors listed above is not only unethical but also potentially dangerous for systemic stability.
While this new style of advertising campaigns may involve some intellectual achievements, its apparent successes are more related to the unorthodox breaking of unwritten, ethical and, possibly, legal rules than to principal scientific advances. Determining responses of individuals to personalised advertising does not involve predicting the collective dynamic of propagation of information between different groups of people, which is the essence of complex behaviours in evolutionary systems. The emerging practice of group-orientated messages and services seems to induce polarising trends, which are rather alarming and can become destabilising in democratic societies. The collective dynamic of complex systems needs to be studied, understood, and ultimately conveyed to the public, while the science of modelling complex systems must play a principal role in these advances. In physics, such collective trends correspond to the difference between single-particle and joint multiparticle distribution functions, or between average properties of a single element and complex behaviour of the whole system involving many elements.
In this work, we are more interested in modelling complex systems as a whole rather than in the autonomous modelling of individual elements of these systems. In this context, our goal is to establish fundamental links between modelling approaches used in different domains of science and suitable for complex systems, as well as a broad categorisation of these approaches.
Solving Turbulence
In the early 1920s, Arnold Sommerfeld, who was interested in the stability of fluid flows and turbulence, decided to give this problem to his most talented student, Werner Heisenberg. While the student subsequently won the Noble Prize in Physics, the problem of turbulence remains unresolved. It was investigated not only by Heisenberg but also by other most brilliant scientists (Kolmogorov, Batchelor, Obukhov, and many others), who had occasional success, but the full solution of the problem does not seem any closer than it was a century ago. Richard Feynman once reportedly characterised turbulence as "the most important unsolved problem of classical physics". Difficulties with solving the problem of turbulence became commonly known. When, as a junior student, I asked senior fellow students for advice about exciting directions of research, the answer was rather sobering "whatever you do, stay away from turbulence-many tried and all failed". These difficulties that have lasted over a century reflect our first encounter with complexity. Turbulence is a system of a large dimension that lies at the borders of order and chaos, possessing something we can call a chaotic order. We learned how to describe deterministic mechanical systems and can deal with purely random behaviour (such as that of molecules in thermodynamic objects), but we have difficulties with combinations of both [9]. Turbulence is random and diffusive, and yet it has coherent structures that tend to persist for surprisingly long times. Whatever we can say about turbulence is valid and invalid at the same time. The Kolmogorov theory of inertial interval seems to be correct [10]. It is correct but it needs refinements to account for fluctuations of the dissipation. These refinements need further refinements to account for intermittency (Kuznetsov and Sabelnikov [11]) and so on. Is Kolmogorov's law of inertial interval correct or incorrect? As in all complex systems, the answer depends on perspective.
While turbulence is a complex system and has similarities with other complex systems, there are noticeable differences between turbulence and what we call complex evolutionary systems-biological, social, economic, and technological systems (most complex adaptive systems are evolutionary of have evolutionary origins). Turbulence does have some inheritance [12] (when large eddies breakup into smaller eddies, the latter inherit some properties of the former), but this inherence is not reliable enough and significant enough to originate any substantial evolutionary processes. Eddies merely keep breaking up until they are dissipated by viscosity. Turbulence is ubiquitous and can be observed at different scales in a lab, in rivers and lakes, in the atmosphere, on other planets, and in the interplanetary space (see Figure 1). It is a mechanical phenomenon, and experiments with turbulence are repeatable, at least on the lab scale. This, however, does not usually apply to complex evolutionary systems; one cannot run one set of economic reforms and then the other one in the same conditions and make an objective comparison. The models developed for turbulence are expected not only to emulate the observed effects but also possess at least some predictive capabilities, while predicting phenomena in complex social systems is often very difficult if not impossible.
[10]. It is correct but it needs refinements to account for fluctuations of the dissipation. These refinements need further refinements to account for intermittency (Kuznetsov and Sabelnikov [11]) and so on. Is Kolmogorov's law of inertial interval correct or incorrect? As in all complex systems, the answer depends on perspective.
While turbulence is a complex system and has similarities with other complex systems, there are noticeable differences between turbulence and what we call complex evolutionary systems-biological, social, economic, and technological systems (most complex adaptive systems are evolutionary of have evolutionary origins). Turbulence does have some inheritance [12] (when large eddies breakup into smaller eddies, the latter inherit some properties of the former), but this inherence is not reliable enough and significant enough to originate any substantial evolutionary processes. Eddies merely keep breaking up until they are dissipated by viscosity. Turbulence is ubiquitous and can be observed at different scales in a lab, in rivers and lakes, in the atmosphere, on other planets, and in the interplanetary space (see Figure 1). It is a mechanical phenomenon, and experiments with turbulence are repeatable, at least on the lab scale. This, however, does not usually apply to complex evolutionary systems; one cannot run one set of economic reforms and then the other one in the same conditions and make an objective comparison. The models developed for turbulence are expected not only to emulate the observed effects but also possess at least some predictive capabilities, while predicting phenomena in complex social systems is often very difficult if not impossible.
Turbulence is essentially the first complex system that science encountered and, therefore, our initial idea of "solving turbulence" in the same way as we solve other mechanical problems was naïve. Complex systems do not have a single ultimate solution; they have many solutions and approaches that should be used under different circumstances. Complex systems permit and, perhaps, demand analyses and modeling using different perspectives. Two seemingly contradictory statements about complex systems may be correct at the same time, depending on qualifications and perspectives.
Turbulent Reacting Flows
While turbulence is complex, its complexity is to some extent limited. Common models of turbulence are never completely right but neither are they completely wrong in most cases, typically producing a substantial but limited error of 20-40%. The situation changes when we have chemical reactions involved [14]. These reactions are very sensitive, and inaccuracies in modelling can easily give predictions that differ from reality by orders of magnitude [15,16]. Complex evolutionary systems often have complex elements; the elements of human society are humans, which are also complex. Turbulence is a mechanical system, and its elements are notional fluid particles, which are not complex (at least in comparison with humans). The presence of reactions changes this; common combustion of, say, petrol, involves hundreds of species and thousands of chemical reactions which Turbulence is essentially the first complex system that science encountered and, therefore, our initial idea of "solving turbulence" in the same way as we solve other mechanical problems was naïve. Complex systems do not have a single ultimate solution; they have many solutions and approaches that should be used under different circumstances. Complex systems permit and, perhaps, demand analyses and modeling using different perspectives. Two seemingly contradictory statements about complex systems may be correct at the same time, depending on qualifications and perspectives.
Turbulent Reacting Flows
While turbulence is complex, its complexity is to some extent limited. Common models of turbulence are never completely right but neither are they completely wrong in most cases, typically producing a substantial but limited error of 20-40%. The situation changes when we have chemical reactions involved [14]. These reactions are very sensitive, and inaccuracies in modelling can easily give predictions that differ from reality by orders of magnitude [15,16]. Complex evolutionary systems often have complex elements; the elements of human society are humans, which are also complex. Turbulence is a mechanical system, and its elements are notional fluid particles, which are not complex (at least in comparison with humans). The presence of reactions changes this; common combustion of, say, petrol, involves hundreds of species and thousands of chemical reactions which take place in every fluid particle. Turbulent reacting flows are complex systems comprising complex elements.
Do turbulent reacting flows have evolutionary properties? In the present conditions, these properties do not seem particularly relevant (although one may note that genetic replication can be interpreted as a specific complex form of a chemical reaction). We do not know how the first replicators appeared on Earth, but evolutionary complexity must have emerged from a more basic form of complexity somewhere at the border between order and chaos. The latter provides for variability, while the former allows for some structures to exist, at least for some limited time. It seems that, as the primary and ubiquitous mechanical phenomenon, turbulence in conjunction with chemical reactions could have played a role in creating first replication mechanisms.
Turbulent Combustion Models
One of the simplest models used in reacting flows is the plug-flow reactor, which assumes an average uniform mixture that gradually evolves from the inlet of the reactor to its outlet [17]. This evolution is specified by the following ordinary differential equations for the average mass fractions: This model may work as a reasonable estimate for slow reactions but is rather crude when handling intensive combustion processes.
Another group includes quasi-laminar models that evaluate special variations but neglect turbulent fluctuations. These models are governed by the conventional reactive scalar equations [17].
Using average quantities is inaccurate for most realistic combustion processes. The plugflow reactor and quasi-laminar models can be referred to as average models or, more precisely, unconditionally averaged models.
Conditional models do take into account at least some of the turbulent fluctuations and are reasonably accurate but only for some classes of combustion processes that do not involve more complex phenomena such as extinctions and reignitions. CMC (conditional moment closure), which is the most widely known example of the conditional models is given by the following equations [18]: for Q i = Y i |Z , i.e., for the expectation of reactive scalar Y i , conditioned on a given value of the mixture fraction Z (see Klimenko and Bilger [18] for the complete form of the CMC equation). Here, N Z = (∇Z) 2 Z is the conditional scalar dissipation, u is velocity, and W is the source term. The models associated with the stationary frame of reference are referred to as Eulerian, while models connected to moving fluid are called Lagrangian. The conditional and unconditional (quasi-laminar) models specified above are Eulerian. The conditional models are intermediate in their complexity and accuracy between quasi-laminar models and Lagrangian PDF models considered below.
The Lagrangian PDF (probability density function) models are used for Monte Carlo simulations of the probability density functions of reactive scalars. These models are formulated using stochastic differential equations.
The superscript k marks different particles, and each of the particles possesses a number of properties Y 1 , Y 2 , . . . Here, w (k) (t) denotes the independent Wiener process. The Lagrangian PDF methods were analysed in the seminal work of Pope [19]. The PDF models also allow for Eulerian implementations called "stochastic fields" [20]. The synergy of the PDF and conditional methods resulted in the MMC (multiple mapping conditioning) approach, which involves adding stochastic equations for the so-called reference variables [21,22], to the system in Equation (4) and conditioning [22] of the mixing operatorM [23] not only on physical coordinates x = (x 1 , x 2 , x 3 ) but also on the reference variables x 1 , x 2 , . . . MMC models are often implemented combining Eulerian simulations of dynamic properties and sparse Lagrangian simulations of reactive components. Note that using Markov families of larger dimension due to additional (i.e., reference) stochastic variables such as those in Equation (5) allows us to represent a wider spectrum of effects.
In this section, we use notations involving ensemble averages, but the modelling approaches introduced above can also be used in conjunction with LES (large eddy simulation) filtering [24]. In LES methods, all fluctuations are divided into resolved fluctuations, which are fully simulated, and sub-filter (sub-grid) scales, which are modelled. Cascade interactions between different scales is an important problem in turbulence. As the filtering scale decreases and approaches the Kolmogorov scales (the scales of the smallest vortices present in a turbulent flow), LES models approach DNS (direct numerical simulations), a complete emulation of the turbulent reacting flows without modelling assumptions.
Transplantation of Models
As noted above, turbulent combustion has highly sophisticated modelling tools that have been developed over many decades. Quite often, these tools are general and may be used for modelling other complex systems, which are not necessarily directly related to combustion and may range from physical to social systems. Here, we need to distinguish ontological and epistemological sides of this problem. Modelling of complex systems is a less developed but rapidly growing area of scientific enquiry and engineering application.
From the epistemological (methodological) perspective, we are interested in modifying and adapting modelling tools and methods to simulate other processes, whether these processes are physically related to combustion or not. The existence of some broad similarities is sufficient.
From the ontological (physical) perspective, any complex reality is formed, in one way or another, by numerous chemical kinetic processes that must be consistent with the fundamental laws of thermodynamics. This complex systemic reality is an emergent property of many reactions; it is not reduceable to these reactions but ultimately must be consistent with the fundamental properties of the reaction models.
Despite large differences between different complex systems, we find that these systems tend to have at least some physical similarities, which enables the application of similar modelling methodologies to these systems. In the rest of this article, we try to address the ontological and epistemological sides of the problem examining both methodological and physical implications of using models of varying levels of complexity that are applicable to both combustion systems and general systems.
Classes of Systems Models
This section presents major classes of models that can be effectively used for modelling general complex systems introduced by Gell-Mann [25] and many others [26]. If we account for terminological differences (i.e., agents in fluids simulations are conventionally called particles, whether they represent a physical particle or not), classes of general systems models have their analogues in the different types of models used in simulating turbulent combustion.
Historical Classification
The most common models used for modelling general systems is system dynamics (see Forrester [27]), which involves more conceptual causal loop diagrams and more specific stock and flow diagrams that help analyse complex systems while dividing them into familiar types of interactions between the elements involved. From the mathematical perspective, the models of system dynamics correspond to a system of ordinary differential equations, which may be both linear and nonlinear. We can easily see that the models of system dynamics (6) use the same class of equations as those used by the plug-flow reactor (1). The models based on ordinary differential equations, such as system dynamics, generally originated in the 1960s and 1970s where typical computer power was sufficient only for solving ODEs. The recent decades have been marked by the emergence of more detailed, yet more computationally expensive models that pertain to partial differential equations, account for motions in physical space, and/or simulate spatially inhomogeneous processes. While quasi-laminar (2) models can be used in general systems modelling, this is not common. The most comprehensive methods associated with systems modelling are the agent-based models [28]. These models involve movements of agents (deterministic and stochastic) and interactions between them. It is easy to see that these models correspond to the PDF models (4) of turbulent combustion, where modelling agents are represented by notional particles. For general systems modelling, the mixing operator M needs to be generalised to reflect various possible types of interactions between agents (particles). Interactions of elements in evolutionary systems are often competitive. It would be reasonable to call more general versions of M the "interaction operator" [4,12]. These generalisations are considered in the next section. A class of models called cellular automata [28] can be seen as a special case of agent-based models where agents interact with their neighbours without moving in physical space. Cellular automata usually presume relatively simple interaction algorithms, illustrating a principle rather than modelling realistic physical objects. Overall, cellular automata are more historical than the conceptual category; models called agent-based can also have non-moving agents.
While conditional and LES-type models are well developed for (and extensively used in) combustion modelling, application of these approaches to general systems seems quite promising but uncommon at present. Lastly, we note that comprehensive DNStype modelling resolving the smallest details of the processes under consideration is very difficult for turbulent combustion but practically impossible for complex systems due to their complex, multidimensional, multiscale, and hierarchal nature. The correspondence of turbulent combustion and general system models is summarised in Table 1. Table 1. Relations between turbulent combustion and general system models.
Models for Turbulent Reacting Flows Models for General Complex Systems
Average and quasi-laminar models, plug-flow reactor System dynamics and other models dealing with direct emulation of overall performance of the system PDF Monte Carlo models, Largangian particle implementations, mixing Agent-based models with interaction between moving agents; particles are called agents Eulerian implementation of stochastic simulations (e.g., stochastic fields) Stationary agents and/or cellular automata, where agents do not move and usually represented by stationary cells
Models for Turbulent Reacting Flows Models for General Complex Systems
Conditional models and conditional/PDF models Elements of conditional methods are used occasionally but the methodology is not well developed for general systems LES and similar models with direct simulation of large scales and modelling small scales Reproducing large scales in conjunctions with a simplified treatment of processes at small-scales is promising, especially in conjunction with conditional models DNS or complete simulation of all (from large-scale to small-scale) features Modelling of all details is usually impossible for general complex systems
Conceptual Classification
First, we must distinguish average models that simulate the overall properties of complex systems and agent-based models that reproduce properties of complex systems by emulating behaviours of multiple elements. System dynamics is, perhaps, the bestknown approach to average modelling, but the average category can involve many other methods. For example, the behaviour of complex systems may be reproduced by using neural networks or other forms of AI [20].
Among the agent-based models, we distinguish global, Eulerian, Lagrangian, and combined models. In global models, agents interact globally without any constraints imposed by localisation. This can simulate homogeneous conditions (e.g., homogeneous turbulence) or correspond to instantaneous interactions at long distances. The other types of models (i.e., Eulerian and Lagrangian) involve localisation in physical space (or any other localisation space, for example, using the mixture fraction space in MMC), where agents interact only with their neighbours. In Eulerian models, the agents are stationary, for example, representing nodes, cells, or locations, as in cellular automata models. In Lagrangian models, the agents (or particles, as agents are called in fluids applications) move and, consequently, interact with different neighbours. The motion can involve both directional and random components.
Lastly, global, Eulerian, and Lagrangian models can be combined. For example, propagation of information within a population involves a network of stationary, Eulerian agents (individuals or groups of individuals), and each of these agents possesses a set of properties characterising human behaviour, while information is represented by Lagrangian agents, which move between Eulerian nodes and possess a different set of properties. In the context of modelling reaction flows, MMC, which was mentioned in Section 3.3, combines Eulerian and Lagrangian characteristics into a single model.
The major classes of models used in simulating complex systems can be summarised as follows:
Modified and hybrid models
Conditional, multiscale, multilevel, etc.
The group of modified and hybrid models is added to the classification to account for various modifications and combinations of the models. Conditional models either involve some kind of incomplete (conditional) averages or, as applied in MMC, enforce some conditional properties on the behaviour of the agents. Different models can be used at different scales and different levels of systemic hierarchy. Elements of a complex system are often complex systems on their own. These elements can be represented by agents, and each of these agents is linked to another model associated with the complex behaviour of the subsystems. Each of these subsystems can have elements that are also complex systems. In addition, we may distinguish, as conventionally applied in turbulence, macroscopic and microscopic processes.
While possibilities of combining different models in simulating complex systems are limitless, we need to raise a voice of caution. Complex systems are usually far too complex for modelling everything down to the smallest details as achieved in DNS. Practical success can be achieved by modelling certain features of interest associated with a complex system. A reasonable traffic model does not need to involve modelling mood for each driver. If a complex system consists of elements represented by complex subsystems, we face two types of complexity: (1) the emergent complexity associated with collective actions of elements, and (2) the complexity inherited from each element. While modelling a complex system, we are (and should be) predominately concerned with the emergent complexity, since this is the central element of systemic analysis, allowing us to achieve a practical outcome with a reasonable investment of resources. The complexities associated with each element often partially negate each other to form common statistical properties. In physics, for example, statistical thermodynamic properties often become independent of the detailed characteristics of the elements (molecules). While full-scale multilevel modelling can be avoided in many cases, this avoidance cannot be always guaranteed, and model hybridisation may become a necessity.
Major Features of Complex Systems and Models
This section outlines conceptual similarities and major differences that need to be bridged between combustion and systems modelling. In this context, a number of features need to be considered from both ontological and epistemological perspectives.
Modelling Multiscale Processes
Turbulent combustion processes have a wide range of scales, typically from 10 −5 to 10 −1 s for turbulence and from 10 −9 to 1 s for chemical kinetics [17]. Interactions of different scales are one of the major problems that turbulent combustion models need to deal with. For example, the Flamelet model [11,29] is very effective in dealing with fast localised reactions interacting with slower and larger turbulence, while MMC implements PDF treatment of smaller scales combined with conditional modelling at larger scales [22].
While complex systems usually involve multiscale interactions of systemic hierarchies, multiscale modelling is less common due to a dramatic increase in complexity. Modelling strategies for complex systems usually involve significant simplifications focusing on a particular time scale and on a particular level in the hierarchy of emergent systems. The range of nine orders of magnitude would correspond to the range of scales between 1 h and 1 M years. While modelling such a range is neither practical nor feasible, studying interactions between more close and more related scales is to become common in the future.
Complex systems are hierarchal when, for example, elements of a system are also systems consisting of more minute elements. Multiscale and multilevel properties of complex systems are usually closely related to each other. Modelling multilevel hierarchal systems can be difficult and often requires simplifications.
Conserved Properties and Information
Speaking about properties of systems and their elements, we need to distinguish conservative properties, which are preserved (or changed due to transport, sinks, and sources), and information, which is not preserved and can be easily replicated or destroyed. The stock and flow diagrams of system dynamics [30] distinguish stocks that change due to flows and feedback loop information that can be easily multiplied or divided by a constant.
The quantities used in chemical kinetics are conservative and are either preserved (e.g., energy and elements) or changed only due to reactions. The conventional mixing operator M is, therefore, conservative. While it can be easily generalised to handle information (for example, competitive mixing considered by Klimenko and Pope [4]), this addresses the problem epistemically but not ontologically. The emergence of information from conservative properties is a principal step that needs explanation; information used by biological and social systems must be based upon some chemical or physical transformations. It appears, however, that information emerges even in conventional reacting flows, specifically in premixed combustion.
In turbulent premixed combustion (whose accurate modelling remains one the most difficult unresolved problems of the 20th century), the two major states (fresh and burned) are separated by a very thin transitional region, which we can overlook for this discussion. A fluid particle in the fresh state y f transits to the burned state y b when and only when it receives a temperature boost from another burned particle [4].
where we can assume that y f = 0 and y b = 1; in combustion, such a variable y is conventionally called the reaction progress variable. The possibility of extinction is not considered in this model. If we have many particles in a uniform container, then where <y> and 1 − <y> specify the average fractions of the burned and fresh particles, and the constant c is proportional to the probability of interaction between two particles. Equation (8) It is obvious that transformation (7) replicates 1 bit of information from particle "l" to particle "k", which underpins the wide systemic applicability of the logistic equation given by (8). Therefore, while chemical kinetics deals with conserved properties, it can emulate replication of information-the principal element of all complex evolutionary systems-even under conditions of common combustion processes. The distinction of conserved properties, information, and signals is blurred.
Emergence of Chaotic Order
One of the most principal assumptions known in modern physics is the hypothesis of molecular chaos of Ludwig Boltzmann. The main implication of this hypothesis is Boltzmann's H-theorem that aligns kinetic equations with the second law of thermodynamics. Considering systems of notional particles or agents [31,32], the hypothesis of chaos can be expressed by P(y (k) , y (l) ) = P(y (k) )P(y (l) ), where P(y (k) ) is probability of particle k having properties y (k) . This hypothesis imposes severe constraints on the complexity of the system, restricting system behaviour to basic thermodynamics-like randomness and prohibiting hierarchal multiscale dependencies. Further research into the particle systems indicates that dependencies violating particle chaos emerge under some conditions [31]. This generally is not desirable in conventional combustion simulations but can be instrumental in simulating complex systemic effects.
Emergence of Intransitivity
Chemical kinetics is always compliant with the laws of thermodynamics, which enforces transitive total preorder of the states of the system as determined by increasing entropy (or decreasing Gibbs free energy-Gyftopoulos and Beretta [33]). Therefore, kinetic systems usually relax towards equilibrium or partial equilibrium without oscillations. This is in contrast with many complex systems where cyclic behaviours are quite common [12,34].
We need to note that thermodynamic constraints mentioned above are unbreachable only for closed systems. External interference can, at least in principle, reduce systemic entropy. Yet, oscillations are not common for chemical reactions, even in open systems. In this context, the example of periodic evolution (in open systems, of course) of the Belousov-Zhabotinsky reaction [35] clearly indicates that neither thermodynamics nor reaction kinetics prohibits cyclic behaviour. While transitive competitions can be accounted for by effective thermodynamics [12], the emergence of intransitivity [34,36] is an important step in developing the complexity of evolutionary systems [37]. Intransitivity should be expected to emerge naturally in open systems.
Complex Topologies, Networks and Emergence of the Small World
The models we primarily consider above are formulated either for homogeneous conditions or for plain physical space. Numerical methods use uniform rectangular grids to represent the topology of plain spaces. Modern models, however, may need to deal with more complex topological connectivity. For reacting flows in porous media, this connectivity can be expressed by a complex network of pores of different sizes [38]. The models developed for turbulent combustion [39] can also be useful for evaluation of reactive transport in porous medium; however, in the case of particle-based models, they need to replace random walk in the open physical space with random walk on graphs that reflect the structure of the porous medium [40,41].
A general expression for Markov random walk on graphs is given by the following recurrent expression: for probabilities p (n) = (p 1 , p 2 , . . . , p k ) (n) of particle location at nodes 1, 2, . . . , k for the timestep n. Here, T is the stochastic matrix (positive elements summing up to unity for each column), which specifies Markov transition probabilities [41]. For complex general systems, network structure appears to be even more important [40]. In such networks, localised links between nodes may be combined with long-distance connections. For example, using models (7)-(8) for simulation of a simple epidemic may need to take into account not only local travel, which may be approximated by a conventional Brownian-type random walk (such as Equation (4a)), as well as occasional long-distance flights. Considering the representation of connectivity by a graph, these flights correspond to occasional connections between remote nodes. For such graphs, we face a new phenomenon called the small-world effect when the number of nodes n r in a graph grows exponentially n r~e xp(cr) with the distance r from a selected central node [42]. This is in contrast with n r~r 2 for a localised grid on a two-dimensional surface. This small-world effect results in an exponentially fast propagation of epidemic, making the modern interconnected world more capable of and more susceptible to the fast propagation of information and viruses.
Concluding Remarks
Modelling reacting flows has been developed over many decades offering a spectrum of modelling methodologies with a wide range of complexity and refinement. The success of this development is largely determined by the availability of experimental data and the repeatability of experiments. While turbulence and realistic combustion kinetics are complex systems, modelling tools in combustion can always be checked against experiments or more detailed simulations.
Over many decades, studies of complex evolutionary systems could not enjoy the same level of verifiability and quantification and often had to resort to more qualitative analysis and observation. While reacting flows commonly allow for a definite formulation of the problem, this is usually not the case with complex socioeconomic systems, where even formulation of the problem is subject to a substantial degree of ambiguity, and proper experimental validation is often impossible. It is not a surprise that modelling of general complex systems was limited to more basic (yet still very useful, of course) approaches, such as system dynamics.
The age of the internet brought new conditions of effective communication networks and availability of social data, which will have far-reaching implications for further technological development and social dynamic. Some of these implications are undoubtedly positive and some are not. The availability of data opens possibilities for the quantification of social science and expansion of applications and methods previously used only in physics and mathematics to a much wider spectrum of problems. More effective and experiment-tested types of models can be applied to various complex evolutionary systems addressing numerous social and environmental challenges that humankind has to face [43]. These opportunities, however, are often forfeited in favour of ad hoc applications of available data to achieve immediate political and economic gains. While it is not raw data but the ability to construct a suitable model or theoretical framework using these data that can be successful, publicly available information indicate noticeable intensification in using data for political and economic gains. This intensification, however, does not extend to the main issue associated with complex evolutionary systems-the patterns of collective emergent behaviours.
Without introducing any new models, we demonstrate a broad consistency of models for reacting flows and general complex systems and examine two types of issues: physical (ontological) and methodological (epistemological). It is arguable that, while having numerous emergent properties, complex systems involve, at some basic level, transport and reaction; therefore, they must be consistent with the laws controlling reacting flows. Here, we discuss that reacting systems do allow for the emergence of multiscale behaviour, information, chaotic order, and cyclic intransitivity. Previously, we demonstrated the emergence of cooperation in general intransitive systems with intransitive competition. Note that a reduction of complex behaviour to the level of chemical reactions is usually either impossible or impractical and certainly is not proposed or advocated here.
Hence, further progress in modern methodologies for modelling complex systems (which involve not only physical but also social, economic, and technological processes) is likely to implement, explicitly or implicitly, the extensive set of methods developed in combustion modelling in conjunction with necessary adjustments and adaptations of these models to more general environments. | 9,500.4 | 2022-05-26T00:00:00.000 | [
"Physics",
"Engineering",
"Computer Science"
] |
Implementation of Kahoot as a Digital Assessment Tool in English Formative Test for Students of SMP Negeri 2 Temanggung in the Academic Year of 2019/2020
test Abstract ___________________________________________________________________ The paper aims to investigate the applicability of the Kahoot as a digital assessment tool. The participants of this study were 32 students of 8G at SMP Negeri Temanggung in the academic year 2019/2020. Kahoot was implemented in English formative test. This is a mixed method with survey technique to collect data and qualitative descriptive analysis to analyze the data. The results of the study indicated that the students thought that Kahoot was enjoyable, informative, useful, and fine. Those were analyzed in descriptive qualitative method. The finding reveals five results: Students good perception on using Kahoot in their English formative test. They can actively participate on the test and get the result directly from the teacher after all the students finished the test. The validity, reliability, and practicality of the test can be seen from the content that has been designed by the researcher, the practical usage of the application is also become the important aspect to be analyzed. Due to COVID -19 the students and teacher cannot do the
INTRODUCTION
The digital technology usage for educational purposes, including second and foreign language learning is expanding fast. Most of students in this era would expect to use digital devices to carry out an Internet browse when they write a paper to look for a suitable website to practice a language skill. Teaching, learning, experience, testing and evaluation, and teachinglearning process have an important role in education.
According to Kaya (2003) the process of teaching-learning is always in a cycle with planning, implementation and assessment. This is also in line with Basol (2015) that testing and evaluation are required in each area where teaching occurs, since assessment and evaluation are essential components in teaching. In other words, it can be concluded that testing and evaluation are the measurement of the successful teaching and learning process. Increasing studies showed that each part of the education process is closely related to the measurement and evaluation is the factor that makes measurement and evaluation an indispensable element (Yıldız & Uyanık, 2004). The process cannot be separated to each other.
The idea that digital technologies can help transform education and specifically assessment is not a new one. New technologies and tools have long been seen to open up new possibilities due to their potentially beneficial characteristics or affordances, such as offering more personalized, instantaneous or engaging assessment experiences. In many cases this potential has been realized and demonstrated benefits. However, the literature suggests that the use of digital technologies has yet to be 'transformative' and is often used via traditional assessment methods or within pockets of innovation that are not widespread.
Assessment is generally recognized as one of the most crucial elements of an educational experience. The assessment used by teacher is formative assessment. Good (2011) explains that formative assessment is used to gather information related to appropriate learning content, context, and learning strategies and to fill the existing gaps between the students' current performance and the targeted learning goal. It is also seen as one of the hardest to reform. However, there is an increasingly demonstrated need for assessment reform, particularly if it is to keep up with other theoretical, cultural and technological developments affecting teaching and learning. Current assessment methods, especially the heavy emphasis and priority afforded to high-stakes summative assessment, are often described as outdated, ineffective and at worst damaging.
The evaluation, on the other hand, is the process of decision-making based on the assessment results. The concept of evaluation incorporates assessment (Şaşmaz Ören, 2014, p.277). That is why, the teacher as a decision maker should decide on what to assess and how to evaluate this later. The assessment includes the materials given by the teacher. The teacher then design the way he assess students formative test. That is why the researcher would like to address the above problem into research in title The Implementation of Kahoot as a Digital Assessment Tool in English Formative Test for the Students of SMP Negeri 2 Temanggung in the Academic Year 2019/2020. Digital assessment tools provide teachers with instant feedback and make them do individual or group assessments in a lively and competitive environment (Yılmaz, 2017). Digital assessment in education is important in terms of feedback, control of the learning rates that vary from individual to individual, and learning quality to be achieved at the end of the assessment process. Continuous measurement and evaluation activities should be carried out in digital education in order to avoid problems in the aforementioned issues (Balta and Türel, 2013). Hague & Payton (2011) propose some suggestions for the teachers on the use of the digital technologies in the learning and teaching process. These recommendations are; be informed about the technological tools to be used, identify supplementary resources to be needed, and prepare contingency activities for the students against the possibility of encountering any problems. Bennett (2002: 14) argued that the 'incorporation of technology into assessment is inevitable'. However, as has been demonstrated by the introduction of many new 'innovative' technologies, the view that educational reform through technology is 'inevitable' and predetermined is usually tempered by the challenges in implementation and complexity of change in education systems. Indeed, Bennett goes on to acknowledge that 'it is similarly inevitable that incorporation will not be easy' (ibid).
Research had shown that formative assessment (or assessment for learning), as distinct from summative assessment (or assessment of learning), is a powerful tool that benefits learning and student achievement (Black and Wiliam at al. 1998)). Nicol and Macfarlane-Dick (2006) developed further ideas about the importance of 'self-regulated learning,' which identified an important role for students in their own assessment. However, even as evidence grows on the benefits of feedback through formative assessment and more teachers employ these methods, it still remains in the shadow of highstakes summative assessment's level of influence and unshakeable prioritization on national and international stages.
METHODS
Mix methods research is an approach to inquiry that combines or associates both qualitative and quantitative form (Crasswell, 2007). These methods were adopted in this study to investigate the applicability of the Kahoot as a digital assessment tool in Junior High School. The quantitative data is presented and counted by using SPSS Program. The qualitative research is a study in which events and phenomena are revealed in a natural environment by using data collection methods such as qualitative research, interview, observation and document analysis (Mason, 2002).
The writer will use the results of students' formative assessment of junior high school students as the main data sources. Furthermore, the researcher will analyze the results of students' English formative test that would be conducted by the teacher.
The participants of the research are 30 students of 8G Class and 1 teacher in SMP Negeri 2 Temanggung in the academic year 2019/2020. The researcher chooses one class only because of the availability of the gadget they use at home in this distance learning during COVID-19.
The data collection technique that was used is questionnaire. According to Brown (2001) questionnaire is any written document that provide respondents with a sequence of questions or statement in which they are to respond either by writing out their answers or choosing from an already existing or given answers. To know the students' perception on using Kahoot, interview by using Whatsapp application that was addressed to the teacher to know her perception on using Kahoot, and observation during the assessment process. The data was also from the result of the students' formative test that had been done by them at home.
In order to investigate the applicability of Kahoot in Junior high school 2 Temanggung, the implementation was performed for forty minutes in English formative test. After the application was performed, opinions of the students were gathered by using questionnaires asking personal information questions and open-ended and closed-ended questions. In addition, opinions of the teacher were obtained by using questionnaires asking open-ended questions and personal information. The questionnaire for students contained questions about the students' general thoughts about the Kahoot, the difficulties when using it, and likes and dislikes about it. In the questionnaire prepared for the teacher, there were questions about his general opinions on the Kahoot application, and her likes and dislikes about it
RESULTS AND DISCUSSION
This section discusses findings of data analysis to answer research questions. The focus of data analysis is on the students' perception, validity, reliability, and practicality, applicability, effectiveness, and implications of Kahoot in English formative test.
Students' perceptions on Kahoot
Feldman (2011) states that perception is the process by which organism interpret and organize sensation to produce a meaningful experience of the world. Perception generally consists of an observation on certain situation or environment.
It can be a mental image, concept or awareness of the environment elements through physical sensation or physical sensation interpreted in the light of experience and captivity for comprehension. Based on the finding from questionnaire and interview, it showed that most of the students have good perception on using Kahoot, whether for the teacher as the practitioner or the students as the participant. The data were collected by using online observation through application, students" questionnaire, interview, and the learners" achievements. Those were analyzed in descriptive qualitative method, though there was quantitative analysis on students" questionnaire by using SPSS Program.
Cetin (2018) found that Kahoot was enjoyable, informative, useful, perfect, and fine. It implies that Kahoot is not only a digital assessment but also as a tool for assessning students' test. The simplicity in using Kahoot makes the students comfortable to work with the application because they can access Kahoot through their smartphone and all the process of registration can be done easily.
Perception generally consists of an observation on certain situation or environment. Based on the finding from questionnaire and interview, it showed that most of the students have good perception on using Kahoot, whether for the teacher as the practitioner or the students as the participant. The data were collected by using online observation through application, students" questionnaire, interview, and the learners" achievements. Those were analyzed in descriptive qualitative method, though there was quantitative analysis on students" questionnaire by using SPSS Program. The result of the students' perception about the Kahoot application was as follows: there are 13 students who gave high responds on the use of Kahoot kahoot. It means that 52% of the students agree that Kahoot is a good application they used in English formative test. 11 students or 44% of the students gave medium respond or the researcher can say that they have good perception using kahoot. There was only one student who had low perception on using Kahoot. Turan & Goktas (2015) also reported that students like Kahoot application, as well as the fact that lesson are applied. From the statements above it can be concluded that Kahoot is a digital assessment tool that is appropriate to be used by junior high school students at SMP Negeri 2 Temanggung. The table showed that most of the students have good perception on using Kahoot. 13 students gave really high score, 11 students gave medium score, and only 1 student who gave low score on the questionnaire.
The validity, reliability, and practicality of the students' formative test on using Kahoot.
Validity refers to the evidence base that can be provided about appropriateness of the interferences, uses, and consequences that come from assessment (McMillan, 2001). The first characteristic of good test is validity, the students' formative test on Kahoot is valid; it measured the students' understanding of the use Present continuous tense and simple present tense. It based on the material that was given by the teacher previously. It did not measure something else. According to brown (2010), a valid test of reading ability actually measures reading ability and not 20/20 vision, or previous knowledge of a subject, or some other variables of questionable relevance. Mousavi (2009) refers validity as the degree to which a test looks right, and appears to measure the knowledge or abilities to measure based on the subjective judgment of the examinees who take it. The test, hence, will be able to measure what is claimed to measure.
A test is seen as being reliable when it can be used by different researchers under stable conditions, with consistent results and the results are not varying. Reliability reflects consistency and replicability over time. According to Brown (2003) the function of the test is to measure a person's ability, knowledge, and performance. In addition, reliability is seen as degree to which a test is free from measurement errors, since the more measurement errors occur the less reliable the test (Fraenkel & Wallen, 2003;McMillan & Schumacer, 2001, 2006Neuman, 2003). In the same way, Meer and Fraser (2004) ask how far the same test would produce the same result if it was administered to the same students under the same conditions. This helps the researcher and teacher to make comparisons that are reliable. The more errors found in the assessment the greater the unreliability would be and visa versa.
To check the reliability of the test, the teacher gave them the second chance to do the test and there consistency of the test results. Their results are almost similar to the first test that was done by them. Most of the students remember their answers and when they were asked to do the test again, the answer was just the way they did on the first test rather than reading through the question carefully.
The use of Kahoot application in formative test can be economic, effortless, and efficient. It was easy to design, easy to be administered because the result of the test can be downloaded directly after doing the test it is along with the score, point, and the result was easy to be interpreted. It is in line with Brown (2004) he said that the test that is practical it needs to be within the means of financial limitations, appropriate time constrains, easy to administrator, score, and interpret.
The applicability of Kahoot
Kahoot is a tool for using technology to administer quizzes, discussion or surveys (Play: 2014). Thomas (2014), Kahoot allows fast and easy access and is recommended for educators.
He stated that creating activities with Kahoot is beneficial because they can be used to review content of the lesson. The way the teacher implemented Kahoot in the English formative assessment by listing students' phones by asking them the brand of their phone to make sure that it can meet the requirement of Kahoot application. From the total number of 32 students of 8G, there is only 1 student who can't participate on this study because of the l of the gadget. The teacher designed the way he will access students' formative assessment along with the guidance that can be understood easily by both teacher and students.
The effectiveness of Kahoot to assess students' formative test in English
Effectiveness is another consideration in analyzing test. Braskamp and Engberg (2014) states that an effective assessment should have these criteria; having a clear purpose and readiness for assessment, involving stakeholders throughout the assessment process, teacher have to know about what they are going to assess and the way the will assess.
Moreover, an effective assessment is an assessment which considers the effect of the assessment on students' learning behavior and outcome, it also provide feedback afterward (Swanson;. The results from this study both accord with many of the previous studies in the field and stand in contrast with others. As will be explained, this effectiveness likely to explain why completion of the online formative test was found to be more effective in term of getting the direct result and feedback and the efficiency of time allotment.
The difference between Kahoot and traditional formative assessment is ease of grading. The students' online scores could be accessed online at any time. And since each student's quizzes were graded and their cumulative score tallied automatically, minimal administrative effort was required to transfer the learners' scores into the teacher's grade book. On the other hand, checking the paper-based homework was dependent upon the student coming to class and bringing their textbook. As this did not always happen, some time needed to be allowed for the submission of late work. Also, time was needed for data entry and to transform the raw scores into a final percentage. Due to these differences, while the deadline for the formative assessment by using Kahoot could be set as the last week of class, the pencil and paper homework was due a few weeks earlier.
Moreover, it can be suggested that pedagogical tools like Kahoot have the potential to enhance and improve high-stakes examination scores at Junior High School 2 Temanggung. Students who use Kahoot in doing the assessment felt positive about their experience. The results of this study also suggest that creating a fun and engaging environment also supports improved academic performance. Students will learn what excites them. If a student cares about what she or he is introduced to, she or he will be motivated to learn.
The implications for teacher, learners and school teaching curriculum with Kahoot
The impact of Kahoot use on students' English formative test, first it was easy to use and students' comfortableness of using Kahoot in a classroom context could be guaranteed. The competitive formative test could motivate other than frustrate their test. Kahoot was welcomed by the students, teacher, and school. The implementation of Kahoot in English formative test can also be implemented by other subjects.
The good impact for teacher, it can be efficient in the term of time and need less time for doing the analysis. The teacher can maximize their time for making questions and providing teaching materials for the students. The result of the test can be used as their documents so that in another time they can improve well in using the application.
The implications of using Kahoot can be divided into three; For teacher that is the main object of the research, by the end of the research the researcher hope that the application can be used by students. It can help students to be more aware of using the application.
The teacher as the practitioner can use this application. The application will also be familiar for the teacher. This is in line with the government's plan to educate and facilitate the teacher with the new platform of teaching and learning tools.
Hopefully by implementing this digital assessment tool, the school stake holder will facilitate students with good internet connection so that the teacher and students are able use digital assessment tool as a part of their assessment.
CONCLUSION
Based on the result of the study, there are five implications that can be drawn below. The English test is designed based on the need of students and the availability of the gadget, the test would be done should be valid, reliable, and can be used easily, the implementation should be done well, the application should be effective in the usage, and there must be further respond from the teacher as a doer and the school stakeholder as the facilitator. Moreover, the results of English formative test after undergoing the tryout are regarded to be appropriate and feasible to be implemented in the English assessment in junior high school. The first implication to this fact is that the English assessment can also be used by the English teacher but also the other English teacher who substitute the English teacher.
The research findings showed that the students' perception on using Kahoot in English formative test have encouraged enjoyment and described students' achievement. It shows that the students' ability to the English test. The second implication to this fact is that the teacher should use the English test aspects to assess all of subjects to encourage the children's enjoyment and give information on students' achievement.
The English teacher should be creative in conducting assessment. Using various types of assessment is helpful to describe the students' ability. Moreover, creating fun atmosphere during the test makes them relaxed during the test.
The researcher hopes that this designed English assessment can be used for other researcher as input for the same study. Besides, the digital tool of English assessment can be varied and developed to be more creative by using other appropriate tools.
Assessing young learners is different from assessing adult; therefore in designing a test the students of English department should consider the characteristics of young learners.
Other English teachers are expected to create various types of digital assessment in assessing their students. It is not only to measure the students' achievement but also makes them feel relax during the test.
The test developers are expected to create test which is not monotonous and stressful. Consider young learners have special characteristics different from adult, therefore the test developers need to create fun and varied test. | 4,893 | 2020-12-23T00:00:00.000 | [
"Education",
"Computer Science"
] |
Xenon Dynamics in Ionic Liquids: A Combined NMR and MD Simulation Study
The translational dynamics of xenon gas dissolved in room-temperature ionic liquids (RTILs) is revealed by 129Xe NMR and molecular dynamics (MD) simulations. The dynamic behavior of xenon gas loaded in 1-alkyl-3-methylimidazolium chloride, [CnC1im]Cl (n = 6, 8, 10), and hexafluorophosphate, [CnC1im][PF6] (n = 4, 6, 8, 10) has been determined by measuring the 129Xe diffusion coefficients and NMR relaxation times. The analysis of the experimental NMR data demonstrates that, in these representative classes of ionic liquids, xenon motion is influenced by the length of the cation alkyl chain and anion type. 129Xe spin–lattice relaxation times are well described with a monoexponential function, indicating that xenon gas in ILs effectively experiences a single average environment. These experimental results can be rationalized based on the analysis of classical MD trajectories. The mechanism described here can be particularly useful in understanding the separation and adsorption properties of RTILs.
■ INTRODUCTION
Room-temperature ionic liquids (RTILs) are a well-known class of materials characterized by a low melting point, low vapor pressure, and high chemical and thermal stability. 1−3 Due to their peculiar physicochemical characteristics, IL solutions are ideal solvents for many reaction, separation, and extraction processes. 4−7 Several studies have pointed out their utility in gas capture 8−10 and separation, highlighting that the absorption capability strongly depends on the local liquid structure 11 and mechanism of gas confinement. 12−14 In light of these applications, a detailed understanding of the relationship between the IL local structure and the dynamic properties of gaseous species dissolved plays a central role and may help to design task-specific materials. Spectroscopy of noble gases, 15,16 (especially xenon) loaded in nano/microstructured materials, has been used to probe the structure and the diffusion processes in porous media, 17 zeolites, 18 polymers, 19 and nanochannels. 20 Moreover, imaging NMR and diffusion measurements of thermally polarized and/or hyperpolarized xenon gas in free or confined spaces have been performed both at high and low magnetic fields 21,22 and are also widely used for medical applications. 23−26 Xenon gas, despite its chemical inertness, is particularly precious as an anesthetic in cardiovascular medicine and to treat drug addiction. Therefore, to improve its availability and reduce its cost, new MOF materials 27 and zeolite membranes 28 with tailored porosity have been proposed for selective xenon extraction and recycling. Among these new materials, the best adsorption and separation capacity is achieved when the pore size matches with the xenon kinetic diameter. Similarly, noble gas solubility 29 in ILs strongly depends on the free volume that in turn is correlated with the nanostructure. The higher solubility of xenon compared with other nonpolar gases 30 is due to its larger polarizability and subsequently, much stronger interactions with the IL.
In this article, we study the translational dynamics of xenon gas dissolved in some RTILs based on 1-alkyl-3-methylimidazolium chloride, [C n C 1 im]Cl (n = 6,8,10), and 1-alkyl-3methylimidazolium hexafluorophosphate, [C n C 1 im][PF 6 ] (n = 4, 6, 8, 10) (see Scheme 1 for molecular formulae). In particular, we consider the effect of the alkyl chain length of the IL cation on the motion regime of xenon gas. The diffusivity of xenon atoms, IL cations, and anions was independently examined by means of multinuclear pulsed gradient spin echo (PGSE) 129 Xe, 1 H, and 19 F NMR spectroscopy. Variable diffusion time experiments 31 allowed us to study the diffusion motion in the time range of milliseconds to seconds. Moreover, xenon spin−lattice relaxation times T 1 are measured using 129 Xe inversion recovery experiments to evaluate atomic dynamics in the picosecond timescale. The experimental data were analyzed following a conventional methodology suitable to evaluate free or restricted motion in liquids and in the gel phase. Finally, the dynamics of xenon gas loaded in n-alkanes [C n H 2n+2 ] n = 6, 8,10, liquid at room temperature, is also investigated, and the results of the pure liquids vs RTIL are compared. Our approach provides the characterization and the comparison of Xe@RTILs and Xe@alkanes' dynamics for the first time.
■ EXPERIMENTAL METHODS
Materials. All ionic liquids and alkanes were acquired from Aldrich and used without further purification. The chemical structures of all ions are depicted in Scheme 1.
NMR Sample Preparation. NMR "medium wall" tubes with a 5 mm external diameter and a 3.46 mm internal diameter were acquired from Wilmad. The tubes were filled roughly to the same height (5 cm) with ionic liquids and a short/thin capillary tube containing DMSO-d 6 (60−100 μL) was manually inserted ( Figure S3). The samples were dehydrated overnight at 343 K under dynamic vacuum (mechanical pump, usually less than 20 Pa, i.e., 1.4 × 10 −1 Torr). Afterward, the tubes were connected to a vacuum system and degassed several times by freeze−thaw technique and less than 8 Pa pressure (6 × 10 −2 Torr). Xenon gas was initially contained in a known reservoir volume (28.29 mL), with an initial pressure of 150 Torr (20 kPa). The volume was then put in contact with the NMR tube using Wilmad connectors. The volume of these necessary connectors (15 mL) was measured prior to sample preparation by nitrogen gas expansion using the same tubes. Xenon gas was then frozen in the tube using liquid nitrogen, and the tube was flame-sealed. Then, the sample was let to equilibrate for a week. The final nominal pressure of xenon gas was around 3.5 atom for all of the samples (see the Supporting Information for details and Figures S1−S3). Xenon solubility in ILs has been investigated in a wide range of temperature and up to 0.3 MPa pressure. 29 NMR Spectroscopy. 129 Xe NMR experiments were carried out on a Bruker DRX 500 spectrometer equipped with a 5 mm broadband inverse probehead. The 129 Xe PGSE experiments were performed using a stimulated-echo pulse sequence. The acquisition parameters were δ = 3 ms and Δ = (t d − δ/3) = 0.5−2 s in a step of 0.1 s for Xe@C 10 19 F experiments were performed on a Bruker AVANCE spectrometer operating at 500.13 MHz proton frequency equipped with a four nuclei switchable probe (QNP). PGSE data were acquired using the bipolar pulselongitudinal eddy current delay (BPP-LED) pulse sequence with the following parameters: 16 scans; relaxation delay of 10 s; δ = 3 ms and Δ = 1 s for IL samples; and δ = 2 ms and Δ = 0.02 s for alkane samples. A pulsed gradient unit capable of producing magnetic field pulse gradients in the z direction of 53 G/cm was used. The temperature was set and controlled at 305 K.
MD Simulations. We have used the software package Gromacs 32 to run MD simulations of several IL systems. The force field (FF) used features the charge distribution developed by Canongia-Lopes and Padua (CL&AP FF), 33 while the internal parameters are based on the Amber 34 FF implementation in Gromacs. All bonds were constrained by the LINCS algorithm. 35 The leap-frog integrator was used with a time step of 1 fs and a cutoff of 10 Å for the van der Waals and short-range electrostatic interaction. The particle−mesh Ewald (PME) 36 technique was used to handle long-range electrostatic interactions with an interpolation order of 4. Simulations were run in the NPT ensemble using the Berendsen thermostat 37 and the Parrinello−Rahman barostat 38,39 with applied isotropic periodic boundary conditions. Boxes were built starting from previous simulations 40,41 of the butyl systems, changing gradually the alkyl chain length after expanding the box to avoid overlap. Each system was then quickly relaxed to the volume under NPT conditions and equilibrated for 12 ns. The equilibration run was followed by a production run of 60 ns; configurations were saved every picosecond for further analysis. A first set of simulations was run with a box containing 500 ion pairs of [C n C 1 im][X] (n = 2, 4, 6, 8, 10 and X = Cl − , PF 6 − ) plus a Xe atom; for all systems, these boxes were simulated at 350, 400, 450, and 500 K and pressure of 1 bar. Inspection of the cation and anion diffusion coefficients revealed that some short-chain systems at the lower temperature were in a glassy state rather than a liquid state. In a second set of simulations, the temperature was set to 400 K and the pressure to 1 bar. The boxes of this second set contained 250 ion pairs plus a Xe atom. Three independent runs were produced for each system to estimate, together with the results of the first set of simulation at the same temperature of 400 K, the error associated with the diffusion coefficient of xenon. The systems studied by MD simulations are reported in Table S1.
Additional simulations were run for Xe@hexane and Xe@ decane to estimate the diffusion of xenon in the liquid alkane. The FF parameters for the two alkanes were the same used for the hydrophobic part of the alkyl chain of the imidazolium salts. Again, three independent boxes were generated containing 250 alkane molecules and one xenon atom. The boxes were equilibrated for 30 ns and the subsequent three consecutive production runs lasted 60 ns. Since there is no significant effect of electrostatic interaction for these systems, as for ILs, slowing down the dynamics, we ran the simulations at 300 K.
The Gromacs built-in software utilities were used to calculate radial distribution functions (RDFs) and meansquared displacements (MSDs). The diffusion coefficient was then obtained by linear fitting of the MSD in an appropriate The Journal of Physical Chemistry B pubs.acs.org/JPCB Article range: for the cation and anion diffusion coefficients, the MSD was found to be linear normally up to 50 ns; for the single Xe atom, we limited the fitting to the first 2 ns. The first 0.2 ns were excluded from the linear fitting procedure.
■ RESULTS AND DISCUSSION NMR Diffusion. NMR diffusion experiments 42 are based on the measurement of the signal decay after applying a train of field gradient pulses (PFG) of duration δ and increasing intensity g along the z direction. The signal decay intensity I(q, t d ) measured at a fixed time t d can be related to the meansquared displacement (MSD), ⟨z 2 ⟩, as follows where q = (δγg)/2π and γ is the magnetogyric ratio of the observed nucleus. In the case of diffusing species whose motion is described by the Langevin 43 equation (hence Fickian diffusion), the MSD scales linearly with the observation time t d according to eq 2 obtained for the case of application of field gradients along the z direction only with D being the particle self-diffusion coefficient. This relation properly describes not only the free diffusion motion of liquid samples but also all of the diffusion processes that, even in the presence of barriers or obstacles, are described by a Gaussian distribution of displacement probabilities. 44,45 This condition occurs whenever the observation time t d and the mean diffusion distance ⟨z⟩ = (MSD) 1/2 traveled by the molecules during t d become much larger than the characteristic lengthscales λ associated with the obstacles. 46 Different from free diffusion, in complex heterogeneous systems, whenever ⟨z⟩ ∼ λ the molecule feels the effects of the obstacles and the MSD is related to the elapsed time t d through a more general equation where D′ is a generalized diffusion coefficient (whose units are α-dependent) and the parameter α ≠ 1 is defined as the anomalous diffusion exponent. The motion regime may be defined as non-Fickian and, depending on the α value, as anomalous subdiffusive (0 < α < 1) or anomalous superdiffusive (α > 1). Only a few systems deviate from this equation, such as molecular crystals, 20,47 where geometrical constraints to the motion produce a non-Gaussian distribution of displacements. Figure 1A shows the normalized experimental signal decay I(q, t d ) plotted on a semilogarithmic scale vs q 2 for Xe@ C 10 C 1 imCl with observation time t d in the range 0.5−2 s. The slopes of the linear fits provide the MSD values for each observation time. The log−log plot of xenon MSD vs t d is also reported in Figure 1B. It is important to note that a log−log plot based on eq 3 provides the experimental α values as the All of the data were obtained at 305 K. The experimental data are estimated to be accurate between ±3 and ±6%.
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article slope and D′ as the intercept of the linear regression. An immediate indication of the motion regime is obtained. The analysis of xenon motion was performed for Xe@[C 10 C 1 im]Cl sample and the scaling exponent α was found to be 1.05 ± 0.02, providing evidence of Fickian diffusion for the gas in an IL environment. This indicates that the diffusing Xe atoms undergo unrestricted diffusion and do not experience diffusion barriers or obstacles of length-scale comparable with ⟨z⟩ (12−24 μm) accessible by PGSE experiments during the observation time (0.5−2 s). The diffusion coefficients can be calculated according to eq 2 and they are reported in Table 1 for Xe@[C n C 1 im]Cl and Xe@[C n C 1 im][PF 6 ]. In parallel, the diffusion coefficients of the IL cation and hexafluorophosphate anion were also measured using 1 H and 19 F NMR. A summary of the experimental diffusion coefficients is reported in Table 1. Table 1 also collects the T 1 data, which will be discussed in the next section.
Xenon diffusion in ILs is many orders of magnitude smaller than that of free xenon gas 48 (5.3 × 10 −6 m 2 /s) and about 1 order of magnitude smaller than xenon dissolved in water 49 (2.2 × 10 −9 m 2 /s) or alkanes, thereby indicating that xenon dynamics is influenced by the peculiar structural features of the IL systems. Furthermore, xenon diffusion is about 2 orders of magnitude faster than the diffusion of both the IL's cation and anion. Thus, the dynamics of the observed species follows the general trend D(Xe) ≫ D(cat) ≥ D(an) for all of the samples, despite the different anion and chain lengths. This finding also pointed out that the diffusivity of Xe is scarcely influenced by the different viscosities of the alkylimidazolium ILs as a function of the length of the alkyl chain. 129 Xe NMR Relaxation. Spin−lattice relaxation 50 designated by the time constant T 1 is sensitive to the magnetic intra/intermolecular interactions as well as to their time dependence arising from molecular tumbling in solution. Among the several relaxation mechanisms, spin−rotation interaction is responsible for xenon relaxation in the gas phase, while 129 Xe− 1 H dipole−dipole coupling is the predominant mechanism accounting for xenon relaxation in solution. 51 129 Xe T 1 values, reported in Table 1, are obtained by fitting the experimental data with a monoexponential function, suggesting that xenon gas in all ILs, as well as in alkane samples, experiences a single average environment. Errors are estimated as the average absolute deviation over three independent runs. They are between ±1 and ±5% for the diffusion coefficients of the ions and about ±10% for the diffusion coefficient of xenon. b T = 300 K. c D(alkane).
The Journal of Physical Chemistry B pubs.acs.org/JPCB Article Different dynamic behavior of xenon is observed in the two sets of IL and alkane samples; the results, reported in Table 1 and Figure 2B). (iv) For all samples, the measured Xe diffusivity shows an opposite trend with respect to the cation/ anion of the ionic liquid on passing from small to large n values (i.e., with progressively longer alkyl chains): while the components of the ILs diffuse at a slower rate with larger n, Xe diffusivity grows correspondingly. The latter finding is unexpected and counterintuitive, indicating that the two motional regimes are decoupled. Actually, the available literature data on the viscosity of the ILs under investigation, taken from the database published by Yu et al., 52 clearly indicate that the viscosity increases with the increasing length of the alkyl chains for both the PF 6 − and Cl − series at the same conditions of T and P. The D values of Table 1 related to both cations and anions of the examined ILs decrease with increasing viscosity, whereas the corresponding D(Xe) and T 1 relaxation values are not affected by the solvent viscosity in the same way. The transport of Xe atoms in the ILs seems to be related to the extension of the nonpolar domain, as indicated by the extension of n, and basically independent of the motion of the anion−cation components of the ILs.
To better understand the structural features responsible for the differences experimentally observed in Xe@IL motion, we performed classical MD simulations as described below (see the Supporting Information for additional details). Since the electrostatic interaction in nonpolarizable force fields is known to significantly slow down the dynamics and the sampling of the phase space, 53 we used a higher temperature compared to experiments for Xe@IL systems, T = 400 K. The results of The Journal of Physical Chemistry B pubs.acs.org/JPCB Article these simulations, therefore, only have a qualitative meaning but nonetheless, as we will see, they provide essential insights into the interpretation of the results. MD Simulations. The results of classical MD simulations are reported in Table 2 and Figure 2D.
First, the results for the two simulated systems Xe@alkanes compare very well with the experiments. We note that, being noncharged systems, nonpolarizable force fields are expected to perform rather well, almost at a quantitative level; see Tables 1 and 2. Concerning the ionic systems, in Figure S11, we see an analogous trend to the experimental data of the diffusion coefficients of cations and anions as the chain length is increased; moreover, the cation of the chloride salt has a slower diffusion than the cation of the hexafluorophosphate salt for the same alkyl chain length, both in experiments and simulations, confirming the reliability of the simulations of the Xe@IL systems at least from a qualitative point of view.
For Xe@[C n C 1 im][PF 6 ], D(Xe) varies relatively little from the system C 2 (3.6 × 10 −10 m 2 /s) to the system C 10 (4.1 × 10 −10 m 2 /s), though it appears to have a minimum variation for the C 6 salt. In contrast, for the chloride salt, the xenon diffusion is much more strongly dependent on the chain length, as observed experimentally. Moreover, the diffusion of xenon in the hexafluorophosphate salt is faster than in the chloride salt for short chains, while the two diffusion coefficients tend to become closer for longer chains. It is also worth to mention that the Xe dynamics described by the MD simulations appears to be well described by a linear dependence of the MSD with time; see Figures S8−S10 in the Supporting Information.
It is possible to interpret these data by considering the structural features of the two systems, as obtained from MD simulations and previously validated by a comparison of experimental and calculated xenon chemical shifts. 40 It is well known that in ILs, the ionic parts are, on average, arranged in a continuum polar network separated by the hydrophobic domains. 54 The Journal of Physical Chemistry B pubs.acs.org/JPCB Article the ionic liquids: these are the terminal methyl carbons of the alkyl chain in Figure 3; the imidazolium ring carbon in position 2 of the ring (see Scheme 1), labeled CR in Figure 4; and the anion in Figure 5. The RDFs clearly show that xenon is preferentially solvated by the alkyl chains rather than by the ionic moieties of the IL (see also ref 57), as indicated by the first strong peak in the RDFs in Figure 3; moreover, for the chloride salts, such solvation appears stronger than with the corresponding hexafluorophosphate salts. From the RDFs in Figure 4, it is clear that the interaction of Xe with the imidazolium ring is very weak in all cases. However, the hydrophobic anion [PF 6 − ] can penetrate into the hydrophobic alkyl domains more easily than the hard and hydrophilic Cl − . This is evident from Figure 5 where the peak in the RDF of Xe with the P atom in the [PF 6 ] salt is significantly higher in intensity than the analogous one with chloride. Therefore, the nanosegregation is stronger and more defined in the chloride salt compared to the hexafluorophosphate salt. As the chain length increases, the hydrophobic domains become larger but also more connected in the chloride salt.
Finally, the RDF between the center of mass of the anion and the terminal methyl group of the chain, see Figure 6, shows a clear peak in the probability of finding the hexafluorophosphate at contact distance with the terminal methyl group even for the C10 systems, while such probability is strongly reduced for the chloride salt. This confirms that PF 6 − can, to some extent, penetrate the hydrophobic domains where xenon is preferentially solvated, in agreement with ref 40 (see also Figure S12 in the Supporting Information).
This means that for the chloride salt, increasing the alkyl chain length produces a significant change in the environment felt by xenon, that is, the growing hydrophobic domain that become more and more segregated from the polar network of ions; in contrast, for the hexafluorophosphate salt, such a change is to some extent mitigated by the fact that the anions are more easily dispersed within the hydrophobic domain and such a domain is, in fact, more loosely defined than for the chloride salt. Therefore, the smoother the change in the environment as the chain length is increased, the weaker the dependence of the diffusion coefficient, as observed in Figure and of Xe relaxation time T 1 support the picture of an interconnected network of hydrophobic domains where xenon preferentially resides and diffuses. Their dependence on the chain length reflects the extent of the variation of the structural features of such domains in the studied ILs, see Figure 7. The values of D(Xe) and T 1 in the chloride and hexafluorophosphate salts are quite different for short chains, while they tend to become closer for longer alkyl chains. These results are also in agreement with previous data on 129 Xe chemical shift in the same ILs. 57,59 δ( 129 Xe) was found to be highly dependent on chain length although the nature of the anion can invert the slope of the variation: in chloride-based ILs, δ( 129 Xe) decreases with increasing alkyl side-chain lengths, meanwhile for the series based on [PF 6 ] ion, xenon chemical shift increases. In both cases, δ( 129 Xe) to converge to a common value for long-chain ILs as the hydrophobic domains, preferentially hosting the Xe atom, becomes more relevant.
■ CONCLUSIONS
Combined NMR diffusion-relaxation experimental data and computational MD simulation of 129 Xe gas in two representative classes of IL systems provided relevant information on the The Journal of Physical Chemistry B pubs.acs.org/JPCB Article structure−dynamics relationship. The measured diffusivity for Xe@[C 10 C 1 im]Cl exhibits a linear relation with the observation time (Fickian diffusion). This indicates that in the IL nanostructure, at least for this system with relatively long alkyl chains, there are no diffusion barriers, and xenon atoms diffuse in a more rigid homogeneous medium. This picture is confirmed by the results of the MD simulations that show an interconnected network of alkyl domains. Nevertheless, the alkyl chain length and type of anion, and hence the detailed structure of the nanosegregated domains, influence the gas diffusion coefficient and spin−lattice relaxation. This can be particularly appreciated by a comparison of the dynamics in simple alkanes: here, the Xe diffusion, alkane diffusion, and 129 Xe T 1 decrease with increasing chain length because of the increasing viscosity. For xenon, the opposite trend is observed in ILs, that is, an increase of the diffusive motion and T 1 with increasing chain length, while the cations and anions still exhibit the expected trend with viscosity.
These results improve significantly the understanding of noble gases' motion in innovative materials such as RTILs, thus facilitating their use for cost-efficient Xe recycling and recovery as well as other conceivable industrial applications. The Journal of Physical Chemistry B pubs.acs.org/JPCB Article | 5,674.2 | 2020-07-02T00:00:00.000 | [
"Chemistry",
"Physics"
] |
PySMILESUtils – Enabling deep learning with the SMILES chemical language
Recent years have seen a large interest in using the Simplified Molecular Input Line Entry System (SMILES) chemical language as input for deep learning architectures solving chemical tasks. Many successful applications have been demonstrated within de novo molecular design, quantitative structure-activity relationship modelling, forward reaction prediction and single-step retrosynthetic planning as examples. PySMILESUtils aims to enable these tasks by providing readyto-use and adaptable Python classes for tokenization, augmentation, dataset, and dataloader creation. Classes for handling datasets larger than memory and speeding up training by minimizing padding are also provided. The framework subclasses PyTorch dataset and dataloaders but should be adaptable for other deep learning frameworks. The project is open-sourced with a permissive license and made available at GitHub: https://github.com/MolecularAI/pysmilesutils
Introduction
Machine learning and deep learning have seen a research boom in the latest years. Deep learning in particular has significantly improved some long standing problems in machine learning, such as image recognition, speech recognition, and natural language processing (NLP). Deep learning can be understood as automatic feature extraction, rather than engineering features for shallow neural networks or other machine learning algorithms. In deep learning, the first layers extract and create useful features for the subsequent layers and final predictions. As such, deep learning can use formats that are more "raw" and yet reach similar or better performance than approaches that use engineered approaches. On the downside, deep learning needs enough data to learn the complex, often nonlinear, transformations necessary to convert the raw format into useful features for predictive modelling.
In chemistry, molecules have previously been featurized with descriptors and fingerprints [1], but recently other formats such as images [2], strings [3] and graphs [4] have been used as input to deep neural networks which solve chemically related tasks such as property prediction. Moreover, the graphs and strings can also be sampled autoregressively and thus used to predict molecular structures [5], [6]. This particular generative or molecular read-out capability has enabled a wide range of advanced algorithms ranging from autoencoders [7] to molecular transformers for e.g. reaction informatics [8], representation learning [9] and molecular optimizations [10].
The Simplified Molecular Input Line Entry System (SMILES) format is a single-line molecular notation, which has been widely used for handling molecules in a convenient way. The SMILES strings can easily be organized in common spreadsheets and sent by e-mail. However, in combination with NLP network architectures, it provides an easy and convenient way to do deep learning on molecules. Techniques such as data augmentation [3] can improve performance and bring the fidelity of the molecular creation close to 100% [11].
PySMILESUtils strives to make the use of SMILES for deep learning applications even easier by providing reusable and flexible objects for turning SMILES into tensors for deep learning, and sampled tensors back to SMILES again, as illustrated in Figure 1. SMILES strings can be tokenized in different ways, the most simple isto encode each character as a token. However, single atoms are often encoded on their own, so that e.g. chlorine "Cl" and bromine "Br" have their own token and can easily be distinguished from carbon "C" and boron "B". The tokenization class supports different tokenization schemes and can use regular expression patterns or explicit token patterns to search for tokens in the strings in the dataset. The token patterns are used to analyze the dataset and build the vocabulary that contains the translation table between tokens and integer indexes or one-hot encoding. For multicharacter tokens with potential overlap, ambiguity can exists, as example if both "ccc", "cc" and "c" token patterns exits. In that instance the order in the token list is important, as the tokens are extracted in order. Additionally. a class for data augmentation of the SMILES is also provided, as are PyTorch [12] compatible datasets and dataloaders. [3], or by randomizing the string during SMILES creation [13].
Figure 2 PySMILESUtils consists of reusable and customizable elements that work together to create mini-batches for training of artificial neural networks. Datasets provides a consistent interface that can be customized for different data types. The SMILES augmenter and tokenizer augments and converts SMILES strings into tensors, respectively. The DataLoader chooses samples from datasets and collates them into minibatch tensors ready for training.
PySMILESUtils has a range of different classes and sub-classes that serve different purposes during the creation of minibatches ( Figure 2), and are similar to and compatible with the framework available within PyTorch, but has been customized to enable handling of the data in SMILES format.
Tokenizers
The tokenizers are central for working with text-based formats such as SMILES strings. The tokenizer in PySMILESUtils consists of an abstract class with methods that are considered common to tokenizers. A provided subclass, the SMILESAtomTokenizer, uses regular expression token patterns for tokenization. Regular token strings can be provided as well. Given token patterns and potential tokens, the dataset is analyzed and the found tokens are used to build the vocabulary that will be used to encode and decode the SMILES strings into tensors of indices ( ). Optionally one-hot encoded tensors can be created. Code Box 1 shows an example of how to work with the SMILESAtomTokenizer using the default regular expression patterns to create a vocabulary from the SMILES in the dataset. The tokens created are chemical symbols, where e.g. Chlorine is not split into "C" and "l", but rather kept as "Cl". After fitting the vocabulary it's easy to convert a given SMILES string into a PyTorch tensor of indices that could go into an embedding layer of a deep learning model. Page 4 of 8
Code Box 1: Example of working with the SMILESAtomTokenizer. A vocabulary is created upon instantiation from the provided SMILES using the default regular expression token pattern, and the tokenizer is then called to tokenize a single SMILES.
import pandas as pd from pysmilesutils.tokenize import SMILESAtomTokenizer data = pd.read_csv("data/pande_dataset.csv") #Example dataset tokenizer = SMILESAtomTokenizer(smiles=list(data.reactants + data.products)) print(tokenizer.vocabulary)
SMILES augmentation
SMILES augmentation is provided via the SMILESAugmenter object. After instantiation and configuration, the object itself can be called with a SMILES string or list of SMILES strings and will return a list of augmented SMILES strings. By default, the object will use atom order permutation [3], but full randomization [13] is also available. The augmenter object can be deactivated by setting the .active property to False, and it will then pass through the SMILES strings unaltered. This makes it possible to switch off the augmentation if needed, for example during model evaluation or inference.
Datasets
A couple of different datasets are provided with the PySMILESutils package. An example of a simple dataset is the SMILESDatasets, which uses one or more lists or list-like objects of SMILES strings, that it returns tuples from when indexed. Furthermore, the object has a property .sorted_indices, which contains the indexes of the samples sorted by the lengths of the elements of the first list, which is necessary for the efficiency of the BucketBatchSampler (see section Combining the objects …). The usage of SMILESdatasets is illustrated in Code Box 3.
An example of a more advanced dataset is the MultiDataset, demonstrated at the bottom of Code Box 3. Here the MultiDataset is used to step through a list of data, which can then be used one after another. This is useful if there is a need to switch dataset in between epochs, one example could be to switch dataset because offline data augmentation such as Levenshtein augmentation [14] has been used. Subclassing the MultiDataset would then be necessary to load the data off the disk. An example of a MultiDataset loading from disk is available in the PickledMultiDataset, which uses a list of filenames of pickle files to load them one by one.
DataLoaders and variants
If a single epoch of the prepared dataset can't fit in memory, the BlockDataloader can provide efficient mitigation. The BlockDataloader loads data in chunks or blocks into memory from e.g. an HDF file that is too large for the machine memory. The chunck size can be chosen to fit the resource restraints, i.e. the amount of memory. The minibatches are then sampled randomly from the current in-memory Page 6 of 8 chunk. The dataloader discards the current block from memory and loads the next chunk when needed. If the chunks are sufficiently large, the random minibatches will probably be different between each epoch, while at the same time loading from disk into memory is efficient as the read is consecutive. The approach breaks with a full stochastic sampling of the dataset, but in practice we have not observed a difference in training and test performance if the dataset is pre-shuffled and the blocks are kept sufficiently large, as was done with the pre-processed graph tensors and action tensors in GraphInvent [15].
Combining the objects for an efficient whole As shown in Figure 2, the individual elements must be combined to enable the creation of mini-batches for training. However, the objects are flexible and can be combined in several different ways. To investigate the most efficient way we constructed a small transformer model [16] and trained it with several different ways of organizing the code and training. First, we put the tokenization and collating of the mini-batches into the training loop itself. Secondly, we put the tokenization into the dataset and padded it with a customized collate function in the dataloader. Alternatively the dataset could produce padding of a fixed width. Lastly we added both the tokenization, padding and collating into a custom collate function used by the dataloader. The timings for training an epoch of test data (USPTO-50K [17] has 40.000 training samples) are shown in Table 1. It is evident that the slowest option is to put the tokenization into the training loop itself or using the pre-tokenized dataset. Customizing the collate function is nearly twice as fast, presumably because this enables the dataloader to work on one or several workers in parallel to the GPU training. The bottom half of Table 1 shows the speedup that can be obtained by bundling together SMILES of similar length in the mini-batches. Sorting the samples by length is the optimal case for speed-up, but would likely be detrimental to training as the mini-batches would be the same in each epoch. The BucketBatchSampler object balances the need for similar size batches with the need for random batches, by dividing the data into a number of "buckets" based on the length of the SMILES strings. Mini-batches are then drawn randomly from each bucket, ensuring different mini-batches of similar length SMILES for each epoch. Padding is thus minimized and it is evident from Table 1 that the training performance in terms of speed is close to the optimally obtainable with the sorted lengths. The full example code for the different code organizations can be found as a %delimited examples_training.py script in the examples directory in the code package.
Comparable frameworks and projects
Several other projects and frameworks contain code for working with SMILES based deep learning for chemistry. One such project is SMILES-enumeration (https://github.com/EBjerrum/SMILESenumeration) [3]. It is a more or less monolithic class for both tokenization and data augmentation aimed at providing tensors for Keras [18] training. It only supports single-character based tokenization, one-hot encoding and fixed-length padding. It has been superseded by MolVecGen, which also contains SMILES based tensor generation as well as objects for creating chemception images [2] or tensors based on RDKit fingerprints [1] (https://github.com/EBjerrum/molvecgen). The wellestablished DeepChem [19] project has a SmilesToSeq class amongst the featurizers (https://deepchem.readthedocs.io/en/latest/api_reference/featurizers.html#smilestoseq), which supports character-based tokenization and fixed width padding, where it uses indices-based encodings. Augmentation utilities seem not to be available and would have to be coded independently.
OpenChem [20] is another project which can be used for SMILES based deep learning (https://github.com/Mariewelt/OpenChem/tree/master/openchem/data). The project seemingly aims to both provide models and various dataset classes, where one supports SMILES tokenization and augmentation. The code is adapted from the SMILES enumeration project mentioned above.
In comparison with the other frameworks, PySMILESUtils aims to only handle the SMILES conversions and datamodel in a flexible and extendible way and is fully model agnostic, so that it can be adapted to the needs of the task and deep learning algorithm. Many other projects aimed for single models or algorithms, also contain code for tokenization and handling, but are not aiming to be libraries or frameworks. As examples can be mentioned Reinvent [21] and Molecular Transformer [8], for de novo design and synthesis prediction. A full review of all projects using SMILES based deep learning is however out of scope of this application note.
Conclusion
PySMILESUtils is a framework for working with SMILES based deep learning architectures in PyTorch. It has classes for efficient vocabulary and tokenization, data augmentation, as well as optimized data loading for pre-augmented datasets, out-of-memory datasets. Significant speedups can be achieved by smart sampling of mini-batches based on SMILES length. The framework should provide a good and extensible starting point for building SMILES based deep learning models. By releasing it as opensource we hope to lower the barrier in for using SMILES based methods in deep learning for the molecular data science community and allow for greater collaboration and consistency across projects
Conflicts of Interests
The authors declare no conflicts of interests. | 3,094 | 2021-06-29T00:00:00.000 | [
"Chemistry",
"Computer Science"
] |
Validity of Project Model Science Learning Tools Assisted by Augmented Reality to Improve Students' Literacy and Creative Thinking Abilities
: The development of project model science learning tools assisted by augmented reality has been carried out to improve students' literacy and creative thinking abilities. This development research aims to produce a valid product and describe the results of the validation of the learning device so that it is suitable for use in the learning process. This type of research is research and development with 4D model design which consists of the definition, design, development and dissemination stages. Validation of learning tools was carried out by three competent expert validators at Mataram University. The development of augmented reality-assisted project model science learning tools developed consists of a syllabus, lesson plans, teaching materials, LKPD, augmented reality media as well as scientific literacy test instruments and creative thinking ability test instruments. The results of the research show that the project model science learning tools assisted by augmented reality have very valid and reliable criteria so that they are suitable for use in the science learning process in schools to improve students' scientific literacy and creative thinking abilities.
Introduction
The world development of the 21st Century is marked by the use of information and communication technology in various life activities which will of course also have an impact on the world of education.Education has a very important role in producing quality Human Resources (HR) to be able to keep up with developments and competition in the current era of globalization.So to deal with this, adaptation or adjustments need to be made in the educational process (Prihatmojo et al., 2019).Dewi et al. (2022) stated that several skills needed to face the developments that occur include creativity, entrepreneurship, literacy, communication, problem solving, critical thinking and working together.Thus, students need to be prepared to have good competencies including the ability to think logically, critically and creatively, being able to communicate and collaborate as well as the ability to master technology and scientific literacy.
Scientific literacy is a person's ability to apply their scientific knowledge in identifying a problem and drawing conclusions based on scientific evidence in order to make decisions regarding human activities towards nature (Hadiprayitno et al., 2020).The facts from the PISA survey results from 2000 to 2018 show that Indonesia is one of the countries with a relatively low scientific literacy ranking.This shows that there is a gap in the applied science learning (Narut & Supardi, 2019).Apart from scientific literacy skills, the 4C skills that are important to develop are creativity, including the ability to think creatively.Creative thinking is a thinking process to get a new relationship from various things that is done through receiving, remembering and analyzing activities so that the results can be used to solve problems (Ananda, 2019).It is very important to hone creative thinking skills because it can train and familiarize students with solving problems in various ways that suit their thinking (Trimawati et al., 2020).The 2015 Global Creativity Index (GCI) data shows that Indonesia's level of creativity is still low and is ranked 115th out of 139 countries, which indirectly shows that the ability to think creatively is also still relatively low (Yulaikah et al., 2022).
Based on observations and interviews conducted at one of the MTs in West Lombok, information was obtained that the science learning process implemented had not directed students to train their thinking skills and solve problems scientifically.Learning resources and media in science learning are still very limited and have not been integrated with the use of technology.The problem that is the main focus is the limitations of teachers in preparing learning tools, especially those oriented towards growing and improving scientific literacy and students' creative thinking abilities.Therefore, innovation or new ideas are needed through the development of learning tools to serve as a reference for implementing active, creative and fun learning using a model that can support and facilitate the development of students' scientific literacy and creative thinking abilities.The learning model used is the project based learning model.
Project based learning is a learning model that produces a project, where in the project students will create a product and are given the freedom to determine the product that will be created and presented.This PjBL model can help students train critical and creative thinking skills to produce quality products (Elisabet et al., 2019).In using learning models, it is very good if combined with the use of learning media (Ulfa et al., 2022).This is closely related to the use of information and communication technology (ICT) which is integrated in learning.ICT-based learning is very important to apply, especially to science learning (Muzaki et al., 2022).One technology-based media that can be used in science learning is Augmented Reality (AR).AR is a technology that combines two-dimensional or three-dimensional virtual objects into a real environment which is then displayed or projected in real time (Setyawan et al., 2019).The benefit is that it can help students to understand objects in science learning more realistically with flexible time and impressive experience, thereby fostering students' interest and motivation to learn and can train their thinking abilities (Sudarmayana et al., 2021;Vari, 2022).Based on the description above, it is necessary to develop science learning tools with a project model assisted by augmented reality to improve students' literacy and creative thinking abilities.This research aims to produce a scientific learning device with a project model assisted by augmented reality that is valid to be applied in the learning process.In line with research by Putra et al., (2018) which states that the learning tools developed need to be validated to ensure their quality.A learning device is declared suitable for use if the validation results show a high validity and reliability category.This is a reference for researchers to carry out validation tests of project model science learning devices assisted by augmented reality to improve students' scientific literacy and creative thinking abilities before implementing them in the learning process at school.
Method
This type of research is research and development.
The research design used is the 4D model.Thiagarajan in (Sugiyono, 2019) revealed that 4D model research consists of four steps, namely Define, Design, Development and Dissemination.Define is an activity to analyze and define a problem, weakness or condition which is the basis for the importance of carrying out development.Design is the stage of making plans in the form of designs related to the product that has been determined.Development is the activity of creating and developing a design into a product and carrying out product validation tests so that results are obtained in accordance with previously determined specifications.Dissemination is the activity of spreading products that have been tested so that they can be used by other people.To facilitate the research process, a research and development flow has been arranged with stages or procedures as in Figure 1.
Figure 1. 4D Research Procedures
The type of data obtained in this research is qualitative and quantitative data.Qualitative data is obtained from expert validators' suggestions and input on learning tools that have been developed to serve as a basis and reference for making revisions or improvements.Meanwhile, quantitative data was obtained from validation results by expert validators in the form of learning device validation questionnaire scores on a scale of 1 to 4. The formula used to calculate the validity of the learning device products that have been developed is as stated in Equation 1.
% validation =
Validator score Maximum score × 100% (1) The percentage data on validity results that have been obtained can then be matched according to the validity criteria as stated in Table 1.Apart from validity testing, reliability testing of learning device products is also carried out.In this research, the calculation of the reliability of learning tools uses the Borich method, known as Percentage Agreement (PA), which is the percentage of agreement between assessors which is a percentage of agreement between the first assessor and the second assessor using the formula stated in equation 2.
With A being a larger assessor's score and B being a smaller assessing score.The larger score (A) is always subtracted from the smaller score (B).Learning tools are said to be reliable if the percentage agreement value is more or equal to 75% (Makhrus et al., 2020).
Result and Discussion
This research is a type of research and development with a 4D model design.The 4D development model consists of 4 stages, namely definition, design, development and dissemination (Rajagukguk et al., 2021).This research is limited to the development stage, namely product validity testing.This research aims to produce a product in the form of a project model science learning device assisted by augmented reality.
Define is an activity to analyze and define a problem, weakness or condition which is the basis for the importance of carrying out development (Gogahu & Prasetyo, 2020).At this stage there are several activities carried out including beginning-to-end analysis, student analysis, task analysis, concept analysis and specification of learning objectives (Amali et al., 2019).The results of the initial and final analysis show that science learning has not directed students to train their thinking skills and solve problems scientifically, science learning media and resources are still very limited, there is no integration of technology in science learning and the available learning tools are incomplete and lacking adequate.Furthermore, the results of the student analysis show that the level of literacy and creative thinking abilities of students is still relatively low because they are rarely trained and have not yet become accustomed to it.Students become more motivated when learning is integrated with the use of technology.The students involved in this research were class VIII MTs students in West Lombok.Task analysis contains a collection of procedures for determining the content in a learning plan which includes Core Competencies (known with KI) and Basic Competencies (known with KD) in accordance with the applicable 2013 curriculum (Purwasi & Fitriyana, 2020).The basic competencies used as a reference in the development of this learning tool are KD 3.8 (Understanding substance pressure and its application in everyday life, including blood pressure, osmosis and capillarity of transport tissue in plants) and 4.8 (Presenting the results of testing the application of the concept of substance pressure in everyday life).Concept analysis is carried out by identifying main concepts and systematically arranging the main concepts to be taught.Detailing and systematic preparation of the material was carried out to form a concept map, which in this case is material about pressure of substances and its application in daily life for class VIII SMP/MTs.Specification of learning objectives, namely the formulation of learning objectives in more detail based on KI and KD which are adjusted to the main indicators of learning objectives (Kristianto & Rahayu, 2020), namely measuring students' scientific literacy and creative thinking abilities.
Design is the stage of making plans in the form of designs related to products that have been determined (Hafidh & Lena, 2023).Next, an initial draft or prototype of the augmented reality-assisted project model learning device was prepared, which was developed including syllabus, lesson plans, worksheet, teaching materials, augmented reality media and scientific literacy test instruments and students' creative thinking abilities.The syllabus created in this research refers to Minister of Education and Culture Regulation No. 22 of 2016 and was developed in the form of learning activities in accordance with the stages or steps of the project model to improve students' scientific literacy and creative thinking abilities.The Learning Implementation Plan (RPP) is made in accordance with the project model syntax which contains school identity, KI, KD, IPK (known with Competency Achievement Indicators), instructional objectives, materials, models, methods, learning media, learning resources, learning and assessment steps to improve students' scientific literacy and creative thinking abilities.Teaching materials are prepared based on the 2013 SMP/MTs Curriculum on Substance Pressure and its Application in Daily Life.The content in the teaching materials is designed to be integrated with augmented reality media.Student Worksheets (LKPD) are designed in the form of activities with project model stages to improve students' scientific literacy and creative thinking abilities.Augmented reality media is Android-based media designed to display science learning objects in three dimensions.This media was designed using the Unity, Vuforia and Blender applications.The AR media developed is an application that can be installed on Android versions 7 to 12.The test instrument was created to measure students' scientific literacy and creative thinking abilities.The test instruments developed are in accordance with indicators of scientific literacy and creative thinking abilities.At this stage a validation sheet is also designed which will be used by the validator as a reference to provide an assessment of the product being developed.
Development is the activity of creating and developing a design into a product and carrying out product validation tests so that results are obtained in accordance with previously determined specifications (F.Fitri & Ardipal, 2021).Validated products include syllabus, lesson plans, student worksheets, teaching materials, augmented reality media and scientific literacy test instruments and students' creative thinking abilities.Validation was carried out by expert validators, namely three lecturers from the Master of Science Education study program, Mataram University.The data on the validity and reliability of learning devices can be seen in Table 2.The results of expert validation are one of the criteria that serve as a reference that the learning device products developed can be used or applied in the learning process in schools.Based on the data in Table 2, it shows that each product consisting of syllabus, lesson plans, teaching materials, student worksheet, augmented reality media, scientific literacy test instruments and creative thinking skills has very valid categories or criteria.This is in line with the statement (Fitri et al., 2020) explain that learning tools that already have very valid criteria can be applied or tested in schools after making revisions based on input and suggestions from expert validators.
The syllabus developed refers to KD 3.8 and 4.8 material on pressure of substances and its application in everyday life.The average assessment result of the syllabus by expert validators is 96.15% with a very valid category and the average value of percentage agreement (PA) is 96.40% with reliable category.The syllabus developed already contains the complete components that must be included in the syllabus (Yulianingsih et al., 2022).
The RPP developed consists of three meetings and is structured based on the syntax or learning steps of the project model.The average assessment results of the RPP by expert validators were 95.24% in the very valid category and the average percentage agreement (PA) value was 94.70% in the reliable category.The improvements made were based on input and suggestions from the validator, namely explaining the apperception section more clearly and in detail and refining the learning objectives that will be measured during the research.(Wati, 2021) states that lesson plans that have a valid category can be applied in learning.
The teaching materials developed contain material about pressure of substances and its application in daily life for class VIII SMP/MTs.This teaching material is integrated with augmented reality (AR) media which is also being developed, which includes markers that can be scanned using the AR application.The average assessment results of teaching materials by expert validators were 94.17% in the very valid category and the average percentage agreement (PA) value was 94.40% in the reliable category.Improvements made include adjusting the colors and including the source of the image and adding numbering to the writing of the equation.In line with the results of other research which shows that teaching materials assisted by augmented reality have valid criteria as a learning resource for computer systems informatics (Alimka et al., 2024).
The developed LKPD is prepared based on the project model.The initial part of the LKPD contains basic questions related to the material topic, sections for preparing project activity schedules, project design plans, project creation procedures (tools, materials and project creation steps), as well as tables of progress, obstacles and solutions in the project creation process.The average assessment results of the LKPD by expert validators were 93.75% in the very valid category and the average percentage agreement (PA) value was 94.60% in the reliable category.Improvements made include adding images to make them more attractive and complemented by the source.This is in line with research by (Murni & Yasin, 2021) which states that LKPD based on project models is valid, practical and effective for use in Water Cycle Science material.
The augmented reality media developed is an application that can be installed on Android versions 7 to 12 with the available menus, namely developer info, application info, AR camera menu and quiz.The average assessment result of AR media by expert validators was 93.23% in the very valid category and the average percentage agreement (PA) value was 93.60% in the reliable category.The improvements made were adjusting the image display.In line with Hidayat (2024) research which states that science learning media regarding the solar system using augmented reality applications can increase students' learning motivation.Apart from that, Vari (2022) stated that the use of augmented reality in science learning can train 21st century thinking skills depending on the learning activities taking place.
The test instrument developed consists of 8 questions with details of 4 questions to measure scientific literacy abilities and 4 questions to measure creative thinking abilities.These questions are prepared based on indicators of scientific literacy and creative thinking abilities.The average assessment results of scientific literacy and creative thinking ability test instruments by expert validators were 91.67% in the very valid category and the average percentage agreement (PA) value was 94.40% in the reliable category.Improvements made include checking and readjusting the question script with the indicators.In line with research of Putri (2020) which explains that scientific literacy test instruments in science learning that have been validated and have valid and reliable criteria can be used by teachers to measure the level of students' scientific literacy abilities.The research results of (Trimawati et al., 2020) stated that science assessment instruments in project-based learning which have very valid and reliable criteria require the implementation of learning in an efficient time to improve students' creative thinking abilities.
Dissemination is the activity of spreading products that have been tested so that they can be used by other people (Insani & Rossa, 2024).At this stage, research products will be disseminated in the form of science learning tools with augmented reality-assisted project models that have been developed.Dissemination activities will be carried out after limited-scale and widescale trial activities to obtain data on the product's practicality and effectiveness.Next, the final product that has been improved is given to the school where the research was carried out and one other school.
Conclusion
The learning tools developed in this research consist of a syllabus, RPP, LKPD, teaching materials, augmented reality learning media, scientific literacy test instruments and creative thinking abilities.Expert validators have provided a validation assessment of the learning tools to ensure their quality before being implemented in learning activities.Based on the validation data obtained, the science learning tools developed using a project model assisted by augmented reality to increase students' scientific literacy and creative thinking abilities have very valid and reliable criteria so they are suitable for use in the science learning process in schools (SMP/MTs).
Table 2 .
Validity and Reliability Test Results of Learning Devices | 4,447.8 | 2024-08-31T00:00:00.000 | [
"Education",
"Computer Science"
] |
Symmetry, Integrability and Geometry: Methods and Applications Boundary Liouville Theory: Hamiltonian Description and Quantization ⋆
The paper is devoted to the Hamiltonian treatment of classical and quantum properties of Liouville field theory on a timelike strip in 2d Minkowski space. We give a complete description of classical solutions regular in the interior of the strip and obeying constant conformally invariant conditions on both boundaries. Depending on the values of the two boundary parameters these solutions may have different monodromy properties and are related to bound or scattering states. By Bohr-Sommerfeld quantization we find the quasiclassical discrete energy spectrum for the bound states in agreement with the corresponding limit of spectral data obtained previously by conformal bootstrap methods in Euclidean space. The full quantum version of the special vertex operator $e^{-\phi}$ in terms of free field exponentials is constructed in the hyperbolic sector.
Introduction
In connection with D-brane dynamics in string theory there has been a renewed interest in Liouville theory with boundaries. Within the boundary state formalism of conformal field theory the complete set of boundary states representing Dirichlet conditions (ZZ branes) [1] and generalized Neumann conditions (FZZT branes) [2,3] has been constructed, including an intriguing relation between ZZ and FZZT branes [4].
While FZZT branes naturally arise in the quantization of the Liouville field with certain classical boundary conditions, the set of ZZ branes, counted by two integer numbers, so far has not found a complete classical counterpart. It seems to us an open question whether or not all ZZ branes can be understood as the quantization of a classical set up. Thus motivated we are searching for a complete treatment of the boundary Liouville theory relying only on Minkowski space Hamiltonian methods.
A lot of work in this direction has been performed already in the early eighties by Gervais and Neveu [5,6,7,8], see also [9]. They restricted themselves to solutions with elliptic monodromy representing bound states of the Liouville field. Obviously such a restriction is not justified if one wants to make contact with the more recent results mentioned above. They also get a quantization of the parameters characterizing the FZZT branes [7,8], which is not confirmed by the more recent investigations [2,3].
Our paper will be a first step in a complete Minkowski space Hamiltonian treatment of Liouville field theory on a strip with independent FZZT type boundary conditions on both sides, which for a special choice of the boundary and monodromy parameters become of ZZ type. Concerning the classical field theory, parts of our treatment will be close to the analysis of coadjoint orbits of the Virasoro algebra in [10]. As a new result we consider the assignment of these orbits to certain boundary conditions. We also give a unified treatment of all monodromies, i.e. bound states and scattering states, derive the related symplectic structure, discuss a free field parametrization and perform first steps to the quantization.
Classical description
Let us consider the Liouville equation on the strip σ ∈ (0, π), τ ∈ R 1 , where σ and τ are space and time coordinates, respectively. Introducing the chiral coordinates x = τ + σ,x = τ − σ and the exponential field V = e −ϕ , equation (2.1) can be written as The conformal transformations of the strip are parameterized by functions ξ(x), which satisfy the conditions ξ ′ (x) > 0, ξ(x + 2π) = ξ(x) + 2π. (2.3) Note that the function ξ is the same for the chiral and the anti-chiral coordinates x → ξ(x), x → ξ(x). This group usually is denoted by Diff + (S 1 ), since it is a covering group of the group of orientation preserving diffeomorphisms of the circle Diff + (S 1 ). The space of solutions of equation (2.2) is invariant under the transformations , ξ(x)), (2.4) which is the basic symmetry of the Liouville model, and a theory on the strip has to be specified by boundary conditions invariant with respect to (2.4). The invariance of the Dirichlet conditions is obvious, but note that the corresponding Liouville field becomes singular ϕ → +∞ at σ = 0 and σ = π. Taking into account that ξ ′ (x) = ξ ′ (x) at the boundaries, another set of invariant boundary conditions can be written in the Neumann form with constant boundary parameters l and r. We study the Minkowskian case and our aim is to develop the operator approach similarly to the periodic case [11,12,13,14,15]. In this section we describe the conformally invariant classes of Liouville fields on the strip and give their Hamiltonian analysis; preparing, thereby, the systems for quantization.
The energy-momentum tensor of Liouville theory is chiral ∂xT = 0 = ∂ xT . The linear combinations T +T and T −T correspond to the energy density E and the energy flow P, respectively, The Neumann conditions (2.6) provide vanishing energy flow at the boundaries, which leads to and The Dirichlet condition (2.5) allows ambiguities for the boundary behaviour of T andT . In this case we introduce the conditions (2.8) as additional to (2.5), which means that we assume regularity of T andT at the boundaries and require vanishing energy flow there. Then, due to (2.7) and (2.2), the field V , in both cases (2.5) and (2.6), can be represented by Here ψ(x), χ(x) are linearly independent solutions of Hill's equation with the unit Wronskian 11) and the coefficients a, b, c, d form a SL(2, R) matrix: ad − bc = 1. With the notations M e = ± cos πθ sin πθ − sin πθ cos πθ , θ ∈ (0, 1), (2.12) which are called hyperbolic, parabolic and elliptic monodromies, respectively. The matrices A, M and the freedom related to the transformations Ψ → SΨ can be specified by the boundary conditions. First we consider the case (2.5).
Dirichlet condition
The functions ψ 2 (τ ), χ 2 (τ ) and ψ(τ )χ(τ ) are linearly independent and the boundary condition V | σ=0 = 0 with (2.9) leads to a = 0 = d = b + c. Then, we find (2.13) which corresponds to b = 1 = −c. Note that the behaviour of (2.13) near to the boundary σ ∼ ǫ is given by V = 2mǫ + O(ǫ 3 ), the other choice c = 1 = −b corresponds to negative V near to σ = 0 and has to be neglected. Applying the boundary condition V | σ=π = 0 to (2.13) and using the monodromy property we obtain Ψ(x + 2π) = ±Ψ(x). Expanding V now near to the right boundary σ ∼ π − ǫ, we get V = ∓2mǫ + O(ǫ 3 ). Thus, the allowed monodromy is Ψ(x + 2π) = −Ψ(x) only. After fixing the matrices A and M we have to specify the class of functions ψ and χ, which ensures the positivity of V in the whole bulk σ ∈ (0, π). Representing (ψ, χ) in polar coordinates, due to the unit Wronskian condition, the radial coordinate is fixed in terms of the angle variable ξ(x)/2 resulting in and by (2.13) the V -field becomes The obtained monodromy of Ψ(x) leads to ξ(x+2π) = ξ(x)+2π(2n+1) with arbitrary integer n, but (2.15) is positive in the whole strip σ ∈ (0, π) for n = 0 only. Thus, ξ(x) turns out to be just a function parameterizing a diffeomorphism according to (2.3 and for ξ(x) = x it is constant T = − 1 4 . The corresponding Liouville field is time-independent and it is associated with the vacuum of the system. The vacuum solution is invariant under the SL(2, R) subgroup of conformal transformations generated by the vector fields ∂ x , cos x∂ x and sin x∂ x and this symmetry is a particular case of (2.16) for ξ(x) = x. Therefore, the solutions of the Liouville equation with Dirichlet boundary conditions form the conformal orbit of the vacuum solution (2.18). The energy functional on this orbit is bounded below and the minimal value is achieved for the vacuum configuration [10].
To get the Hamiltonian description we first specify the boundary behaviour of Liouville fields. Due to (2.10), (2.11), (2.13), near to the boundaries V is given by and we find The signs + and − for ∂ σ ϕ correspond to the right (σ = π) and left (σ = 0) boundaries and the argument of T is τ + π and τ , respectively. The Liouville equation (2.1) is equivalent to the Hamilton equations obtained from the canonical action with the Hamiltonian given by the energy functional (2.19). Note that ∂ 2 σσ ϕ can not be integrated into a boundary term due to the singularities (2.21).
The canonical 2-form related to (2.22) is well defined on the class of singular functions (2.20) and using the parameterization (2.13)-(2.14), we find this 2-form in terms of the ξ-field (see Appendix A) It is degenerated, but has to be reduced on the space Diff + (S 1 )/ SL(2, R), where it becomes symplectic.
One can of course also study the case with Dirichlet conditions on one and Neumann conditions on the other boundary of the strip, say Starting again with (2.13), the boundary condition at σ = π, together with the unit Wronskian forces the trace of the monodromy matrix to be equal to 2r. Thus for |r| < 1 one has elliptic and for r > 1 hyperbolic monodromy (r < −1 is excluded by arguments of the next section).
Then an analysis similar to the one above gives for −1 ≤ r < 1 with r = cos πθ. The related energy momentum tensor is The result for r > 1 is obtained by the replacement θ = ip with real p.
Generalized Neumann conditions
To analyze the Neumann conditions (2.6) we first construct the fields corresponding to constant energy-momentum tensor T (x) = T 0 and then obtain others by conformal transformations. It appears that this construction covers all regular Liouville fields on the strip. For T 0 = p 2 /4 > 0 convenient solutions of Hill's equation are The related monodromy is hyperbolic. To get it in the normal form (2.12) one has to switch to corresponding exponentials. The form chosen allows a smooth limit for ψ and χ to the parabolic and elliptic cases below. The corresponding V -field (2.9), which obeys the Neumann conditions (2.6), reads and τ 0 is an arbitrary constant. The positivity of the V -field (2.25) imposes restrictions on the parameters of the theory. One can show that if l < −1 or r < −1, then the positivity of (2.25) fails for any p > 0; but if l ≥ −1 and r ≥ −1, then (2.25) is positive in the whole bulk σ ∈ (0, π) for all p > 0.
This case corresponds to the parabolic monodromy. The positivity of (2.27) requires l ≥ −1, r ≥ −1 and l + r < 0. Note that (2.27) is also obtained from (2.25) in the limit p → 0, if l + r < 0. Among these parabolic solutions there are two time-independent solutions which correspond to the degenerated cases of (2.27) for l = −1, r = 1 and l = 1, r = −1, respectively.
Other restrictions on the parameters come from the analysis of the equation which defines an ellipse on the (l, r)-plane. The ellipse is centered at the origin, its half axis with length √ 1 ± cos πθ are situated on the lines r ± l = 0. It is tangential to the lines l = −1 and r = −1 at the points B and C with the coordinates B = (−1, cos πθ) and C = (cos πθ, −1) . The curve BC in Fig. 1 indicates the corresponding arc of the ellipse. The positivity of V -field (2.28) requires (l, r) ∈ Ω ABC , where Ω ABC is the 'triangle' bounded by the lines l = −1, r = −1 and the arc BC.
Having so far discussed which values of l and r are allowed for given θ, we now turn the question around. We start with (l, r) somewhere in the triangle ADE, then all θ obeying Thus, we have all three monodromies if (l, r) is inside the triangle ADE, and only hyperbolic monodromy if (l, r) is outside of it. In the first case (2.28) is a continuation of (2.25) from positive to negative T 0 and for T 0 = 0 it coincides with (2.27). The conformal orbits, generated out of these fields with constant T (x) = T 0 can be written as where p = 2 √ T 0 and ξ(x) is the group parameter on the orbits. The parameter τ 0 is absorbed by the zero mode of ξ(x).
The energy-momentum tensor for (2.32) is given by (2.33) Here S ξ (x) is the Schwarz derivative (2.17), which defines the inhomogeneous part of the transformed T (x). Note that there are other orbits, either with T 0 < − 1 4 , or those, which do not contain orbits with constant T (x) [10]. Using the classification of Liouville fields by coadjoint orbits, one can show that the orbits (2.32) with T 0 ≥ − 1 4 cover all regular Liouville fields on the strip. One can also prove that the energy functional (2.19) is bounded below just on these orbits only [10]. Thus, boundary Liouville theory selects the class of fields with bounded energy functional and, therefore, its quantum theory should provide highest weight representations of the Virasoro algebra.
The Hamiltonian approach based on the action (2.22) is applicable for the Neumann conditions as well. Indeed, the boundary conditions (2.6) are equivalent to and its variation yields ∂ σ (δϕ) − (∂ σ ϕ)δϕ| σ=0,π = 0, which cancels the boundary term for the variation of (2.22). The Legendre transformation of (2.22) and the integration of ∂ 2 σσ ϕ into the boundary terms leads to the action [2] Note that for l = −1 or/and r = −1 the V -field vanishes at the boundary for τ = τ 0 and one can not pass to (2.34), due to the singularities of e ϕ . The canonical 2-form (2.23) can be calculated in the variables (T 0 , ξ) similarly to (2.24) and we obtain (see Appendix A) In the context of Liouville theory this symplectic form was discussed in [16], where it was obtained as a generalization of the symplectic form on the co-adjoint orbits of the 2d conformal group. Note that the form of (2.35) does not depend on the boundary parameters l and r. This dependence implicitly is encoded in the domain of T 0 : if (l, r) is inside the triangle ADE, then T 0 ≥ −θ 2 * , (2.30); and T 0 > 0 if (l, r) is outside the triangle. The symplectic form (2.35) provides the following Poisson brackets where λ(x, y) = ξ(x) − ξ(y) − πǫ(x − y) and ǫ(x) is the stair-step function: ǫ(x) = 2n + 1, for x ∈ (2πn, 2πn + 2π), which is related to the periodic δ-function by ǫ ′ (x) = 2δ(x). From (2.36) we obtain which define the conformal transformations for the fields ξ(x) and T (x). Using the Fourier mode expansion for ξ(x) ξ(x) = x + n∈Z ξ n e −inx and the first equation of (2.36), we find that 2πξ 0 is the canonical conjugated to T 0 For T 0 < 0 the variable α = 2 √ −T 0 ξ 0 is cyclic (α ∼ α + 2π), since the exponentials in (2.32) become oscillating. By (2.37), α is canonical conjugated to πθ, where θ = 2 √ −T 0 . This allows a remarkable first conclusion concerning quantization. Semi-classical Bohr-Sommerfeld quantization of θ yields θ n = − n/π + θ * (l, r), which implies a quantization of T 0 with integer n as long as (T 0 ) n < 0. As shown in Appendix B, this spectrum with the identification = 2πb 2 and the trivial shifts n 2 → n(n + 1), (T 0 ) n → (T 0 ) n + 1/4 agrees with the quasiclassical expansion of the spectrum derived in [3] by highly different methods.
After this short aside we start preparing for the full quantization of our system. The variables (ξ, p) are not suitable for this purpose due to the complicated form of the Poisson brackets (2.36). A natural approach in this direction is a free-field parameterization with a perspective of canonical quantization.
For T 0 = p 2 /4 > 0, free-field variables can be introduced similarly to the periodic case [15] φ Here the x-independent part given by the last term is chosen for further convenience, for u(l, r; p) see (2.26). The field φ(x) obviously has the monodromy φ(x + 2π) = φ(x) + πp, which allows the mode expansion The integration of (2.39) yields where A p (x) is the integral of the equation A ′ p (x) = 2 sinh πpe 2φ(x) with the monodromy property A p (x + 2π) = e 2πp A p (x) and it can be written as The free-field form of (2.35) Note that these brackets and (2.40) lead to (2.36). The energy-momentum (2.33) takes also a free-field form with a linear improvement term The field Φ = φ(x) + φ(x) is the full free-field on the strip. It satisfies for all allowed values of l and r the standard Neumann boundary conditions ∂ σ Φ| σ=0 = 0 = ∂ σ Φ| σ=π and has the following mode expansion a n n e −inτ cos nσ, Since p > 0, A p (x) and A p (x) vanish for τ → −∞. Therefore Φ(τ, σ) is the in-field for the Liouville field: ϕ(τ, σ) → Φ(τ, σ), for τ → −∞.
The chiral out-field is introduced similarly to (2.39) replacing p by −p and its mode expansion can be written as The relation between in and out fields defines a canonical map between the modes (p, q; a n ) and (q, −p,ã n ). Quantum mechanically this map is given by the S-matrix and finding its closed form is one of the basic open problems of Liouville theory.
Canonical quantization
In this section we consider canonical quantization applying the technique developed for the periodic case [11,12,13,14,15]. Our discussion has some overlap with [9]. But in contrast to their parametrization in terms of two related free fields we use only one parametrizing free field. We mainly treat the hyperbolic case. The quantum theory of other sectors can be obtained by analytical continuation in the zero mode p, choosing appropriate values of the boundary parameters (l, r).
and have a standard realization in the Hilbert space L 2 (R + ) ⊗ F, where L 2 (R + ) corresponds to the momentum representation of the zero modes with p > 0 and F stands for the Fock space of the non-zero modes a n . We use the same notations for classical and corresponding normal ordered quantum expressions, which, in general, have to be deformed in order to preserve the symmetries of the theory. The guiding principle for the construction of quantum operators are the conformal symmetry and infinite dimensional translation symmetry generated by φ ′ (x). A semi-direct product of these symmetry groups is provided by the Poisson bracket (2.43), which quantum mechanically admits a deformation of the central term. This implies a deformation of the coefficient in front of the linear term in the energy-momentum tensor (2.42) The related Virasoro generators satisfy the standard commutation relations with the central charge c = 1 + 12πη 2 / . The deformation parameter η is fixed by conformal properties of free-field exponentials. Using the decomposition a free-field exponential is introduced in a standard normal ordered form Requiring unit conformal weight of e 2φ(x) , one finds η = 1 + b 2 , with 2πb 2 = .
Our aim is to construct the vertex operator corresponding to the Liouville exponential (2.44). Building blocks for this construction are the chiral operators
2)
A p (x) = The operators ψ(x) and A p (x) are obviously hermitian and the p-dependent shift of φ 0 in (3.3) is motivated by hermiticity of χ(x) (see (C.9)). The unit conformal weight of e 2φ(x) provides zero conformal weight of A p (x) and, therefore the conformal weights of the operators ψ and χ are the same, like in the classical case. Exchange relations of these operators and their classical counterparts are derived in Appendix C. It is important to note that these relations for the ψ and χ fields are the same Based on (2.44), we are looking for the vertex operator V in the form with p-dependent coefficients B p , C p and D p . The phase factor e −i( /8) provides hermiticity of the first term of V -operator, which corresponds to the in-field exponential. The last term describes the out-field exponential, respectively. To fix B p , C p and D p we use the conditions of locality and hermiticity The analysis of these equations can be done effectively with the help of exchange relations between the ψ and χ operators. There are two kind of exchange relations. The first exchanges the ordering of the arguments x and y χ(x)ψ(y) = e i( /4)ǫ(x−y) sinh (πp + i /2) sinh πp ψ(y)χ(x) − i sin( /2) e πpǫ(x−y) sinh πp χ(y)ψ(x) , (3.8) and another the ordering of χ and ψ fields Applying these relations to (3.7) we obtain a set of equations for the functions B p , C p , D p . They relate the values of these coefficients with shifted arguments and we have found the following solution of these equations (see Appendix D) The parameters m b , l b and r b arise in the solution as p independent constants. Comparing these expressions with their classical analogs (2.45), we find a naturally interpretation of m b and (l b , r b ) as a renormalized mass and renormalized boundary parameters, respectively. To cover parabolic and elliptic monodromies, one has to investigate analytical properties (in the variables p, l b , r b ) of the vertex operator V . Work in this direction is in progress.
Conclusions
For Minkowski space Liouville theory on the strip we have performed a complete analysis of classical solutions regular in the bulk of the strip. These solutions, falling into conformal coadjoint orbits of the energy-momentum tensor [10], can be parameterized by the constant energy density T 0 of the lowest energy solution in the orbit and an element ξ(x) of the conformal group of the strip.
Depending on the parameters l and r, describing the conformally invariant generalized Neumann boundary conditions (FZZT branes) on the left and right boundary of the strip, the solutions have elliptic, parabolic or hyperbolic monodromies. Avoiding singularities in the bulk requires l, r ≥ −1. Solutions with elliptic monodromy correspond to bound states, those with hyperbolic monodromy to scattering states. For l + r > 0 all positive values of T 0 are allowed, the monodromy is then always hyperbolic. For l + r < 0 negative energies above a threshold depending on l, r and elliptic monodromy are allowed as well as all positive energies and hyperbolic monodromy. The peculiarities of zero energy and parabolic monodromy have been touched, too.
For l or r = −1 and certain related T 0 the Liouville field develops a controlled singularity on the boundaries, just realizing a Dirichlet condition (ZZ brane).
For the Hamilton description of the system the Poisson brackets and the canonical two form has been expressed in terms of the variables T 0 and ξ(x). To prepare the system for quantization an alternative description in terms of a free field has been given, similar to the corresponding construction for the Liouville field theory on a cylinder [11,12,13,14,15].
We could get a first estimate of quantum effects by discussing semi-classical Bohr-Sommerfeld quantization. The bound state energy levels become quantized and the spectrum agrees with the corresponding quasiclassical limit of the spectrum gained in [3] by conformal bootstrap techniques in Euclidean space.
Finally we have constructed the quantum version of the degenerated exponential of the Liouville field e −ϕ . The quantum deformation of the weights in its representation in terms of free field exponentials has been fixed by requiring locality and hermiticity.
There is an obvious schedule for further investigations. From the free field representation of e −ϕ in the hyperbolic sector one can read off the reflection amplitude. Its poles should give information on the full quantum bound state spectrum. With the quantum e −ϕ at hand one can construct generic correlation functions following the technique used for the periodic case [17]. We also hope to fully explore the limiting ZZ case within the canonical quantization.
A Calculation of 2-forms
A.1 Dirichlet condition Equation (2.15) leads to the following parametrization of the canonical coordinates The canonical form (2.23) can be represented in the form ω = ω 0 +ω 0 + ω 1 , where while ω 1 turns to a boundary term, since it is represented as an integral from a derivative by σ. The 2-form ω 1 vanishes due to the monodromy properties of ξ. Using the doubling trick as in (2.19), we rewrite the sum ω 0 +ω 0 into (2.24).
A.2 Neumann conditions
The general solution (2.32) can be written in the standard Liouville form Applying the same technique as before, we express the canonical form (2.23) in terms of parameterizing F andF fields with the boundary term Then, using (A.1) and the monodromy properties of ξ-field we get (2.35).
B Comparison of quasiclassical quantization with the corresponding limit of the conformal bootstrap spectrum First we write our formula (2.30) for θ * (l, r) in a form more suitable for the comparison with [3]. Denoting l = l 1 , r = l 2 and defining ϑ j in (0, π) for |l j | < 1 by This brings (2.38) in the form The dictionary to compare our normalizations of the Liouville field, the mass and boundary parameters ϕ, m, l j with that of [3] (φ T , µ, ρ j ) is According to [3], the state space of Liouville theory on the strip is the direct sum of highest weight (∆ β = β(Q − β), Q = 1/b + b) representations of the Virasoro algebra. There is a continuum contribution β ∈ Q/2 + iR + , and depending on the boundary parameters a discrete contribution [3] characterized by where n,n are non-negative integers and σ ± = i(s 2 ± s 1 ) with The evaluation of (B.3) in the quasiclassical limit b → 0 expressed in our boundary parameters l j (for |l j | < 1) gives Inserting this into (B.2) one first notices that for small enough b the optionn = 0 is switched off. On top of this, in this limit only the choices σ + and arccos l j ∈ (0, π) obey the inequality in (B.2). Altogether this leads to After the identifications = 2πb 2 and (T 0 ) n = b 2 ∆ n this agrees with (B.1) up to the trivial shift by −1/4 and the replacement n(n + 1) → n 2 , valid for large n and common for the quasiclassical approximation. The continuous spectrum in [3] corresponds to the our solutions with hyperbolic monodromy.
Although not touching the issue of quantization for the Dirichlet case in this paper, we nevertheless can add already one interesting observation concerning the spectrum of T 0 . From Subsection 2.1 we know that classically there is only one value for T 0 allowed. It is T 0 = −1/4, if on both sides of the strip Dirichlet conditions are imposed, and T 0 = −(arccos r) 2 /(4π 2 ) for Dirichlet on the left and generalized Neumann with parameter r on the right. With the just derived translation rule T 0 = b 2 ∆ − 1/4 this corresponds to the conformal dimensions of highest weight states of the contributing Verma modules ∆ = 0 and ∆ = s 2 + 1/(4b 2 ), respectively. This agrees in leading order of b with the full quantum result via conformal bootstrap reported in [1,18] for the (1, 1) ZZ brane. Note that s j defined according to [3] in our equation (B.4) differs by a factor 1/2 from s in [1,18].
C Exchange relations C.1 Poisson brackets algebra of chiral fields
In this appendix we use the method applied in [19]. The chiral field ψ(x) = e −φ(x) is the classical analog of the operator (3.2) and the canonical Poisson brackets (2.41) are equivalent to which quantum mechanically becomes (3.5). The operator (3.3) corresponds to the field A p (x) = Due to the stair-step character of the ǫ-function the following identity holds and since cosh πp ± sinh πp = e ±πp , we find Inserting (C.4) into (C.2) and using that the function 2φ(y + z) + πpǫ(x − y − z) is periodic in z, we can shift the integration domain in the last term and obtain To find a closed form of the Poisson brackets in terms of the A p -field, we use the identity which follows from (C.4). The contributions of the last two terms of (C.7) in the integral (C.6) cancel each other and provide the result The calculation of the Poisson brackets between χ-fields is now straightforward and we end up with which indicates that the fields ψ and χ are related canonically. The derivation of the exchange relation between the ψ and χ operators (see (3.4)) is now straightforward and we obtain ψ(x)χ(y) = e i( /4)ǫ(x−y) sinh (πp − i /2) sinh πp χ(y)ψ(x)+i sin( /2) e −πpǫ(x−y) sinh πp ψ(y)χ(x) . (C.11)
C.2 Operator algebra
The exchange relation (3.8) is derived in a similar way and (3.9) follows from (C.11) and (3.8) by simple algebraic manipulations. The next step is the exchange relation between the A p -operators, which is obtained in the same manner and in a symmetrized form it reads A p (x)A p (y)e −i( /2)ǫ(x−y) − A p (y)A p (x)e i( /2)ǫ(x−y) = i sin( /2) e (πp+i )ǫ(x−y) sinh(πp + i ) This finally provides (3.6).
D Locality and Hermiticity of V -operator
The locality condition (3.7) is equivalent to the symmetry of the product V (σ, −σ) V (σ ′ , −σ ′ ) under σ ↔ σ ′ . Let us collect the terms with a given power N of the χ-field. The number N changes from 0 to 4. There is only one term with N = 0 C σ,σ ′ = e −i /4 ψ(−σ)ψ(σ)ψ(−σ ′ )ψ(σ ′ ), which is symmetric due to (3.5). The case N = 4 is similar because of (3.6). For the terms with N = 1 we use the exchange relation (3.9), moving the χ-field in each term to the right hand side. Replacing then χ by ψA p , we find the following structure with p dependent coefficients Λ (1) p , . . . , Λ p . The symmetry of (D.1) requires Λ in the form Thus, with X p = 2L and Y p = 2R, where L and R are arbitrary complex numbers, we find B p = Le −πp + R sinh πp sinh(πp − i /2) , C p = Le πp + R sinh πp sinh(πp + i /2) .
The hermiticity condition (3.7) puts restrictions on the parameters L and R. Making use of the exchange relations (C.11) and (3.8), one finds a relation between B p and C p and their complex conjugates, reducing the freedom of two complex parameters to two real ones. With an additional free real parameter from D p we finally obtain with real l b , r b and m b (3.10) and (3.11). Due to the symmetry between the ψ and χ fields, the case N = 3 gives the same result as N = 1.
The analysis of the case N = 2 can be done similarly, but now with the known B p , C p and for D p we end up with (3.12). | 7,625.8 | 2007-01-01T00:00:00.000 | [
"Physics"
] |
Polarization of Human Macrophages by Interleukin-4 Does Not Require ATP-Citrate Lyase
Macrophages exposed to the Th2 cytokines interleukin (IL) IL-4 and IL-13 exhibit a distinct transcriptional response, commonly referred to as M2 polarization. Recently, IL-4-induced polarization of murine bone marrow-derived macrophages (BMDMs) has been linked to acetyl-CoA levels through the activity of the cytosolic acetyl-CoA-generating enzyme ATP-citrate lyase (ACLY). Here, we studied how ACLY regulated IL-4-stimulated gene expression in human monocyte-derived macrophages (MDMs). Although multiple ACLY inhibitors attenuated IL-4-induced target gene expression, this effect could not be recapitulated by silencing ACLY expression. Furthermore, ACLY inhibition failed to alter cellular acetyl-CoA levels and histone acetylation. We generated ACLY knockout human THP-1 macrophages using CRISPR/Cas9 technology. While these cells exhibited reduced histone acetylation levels, IL-4-induced gene expression remained intact. Strikingly, ACLY inhibitors still suppressed induction of target genes by IL-4 in ACLY knockout cells, suggesting off-target effects of these drugs. Our findings suggest that ACLY may not be the major regulator of nucleocytoplasmic acetyl-CoA and IL-4-induced polarization in human macrophages. Furthermore, caution should be warranted in interpreting the impact of pharmacological inhibition of ACLY on gene expression.
INTRODUCTION
Macrophages respond to changes in their environment, such as bacterial or viral infection, hormones, cytokines, or nutrients, with remodeling their transcriptome. Consequently, they alter their phenotype, a response known as macrophage polarization (1). Historically, macrophage polarization was described in a dichotomous fashion with a pro-inflammatory response to bacterial lipopolysaccharide in combination with a Th1 cytokine interferon-γ (M1 response) as opposed to an anti-inflammatory response elicited by Th2 cytokines interleukin-4 (IL-4) or IL-13 (M2 response) (2)(3)(4).
M1 and M2 responses are remarkably different not only in the signaling pathways involved, but also how they engage variable metabolic pathways (5). While metabolism primarily serves to provide energy and substrates to support macrophage functional responses, e.g., phagocytosis, several metabolites directly affect transcription through epigenetic mechanisms (6). Acetyl-CoA is a metabolite with a distinct role in epigenetic and transcriptional regulation through its widespread use as a substrate for acetylation of histones and other proteins, including transcription factors (7). Although several reactions provide acetyl-CoA for histone acetylation, ATP-citrate lyase (ACLY) is considered to predominantly contribute to nuclear pool of acetyl-CoA (8). ACLY role in epigenetic control was initially described for cancer cells (9), where ACLY critically supports de novo lipogenesis and thus, cell proliferation (10,11). Further studies reported epigenetic regulation through ACLY in adipocytes (9,12) or myocytes (13).
Recently, ACLY was shown to regulate transcriptional responses to IL-4 in murine bone marrow-derived macrophages (BMDMs) (14). IL-4 triggered the Akt-mediated serine phosphorylation of ACLY, which is supposed to increase ACLY enzymatic activity (15,16). Accordingly, pharmacological inhibition of Akt or ACLY prevented induction of a subset of IL-4-responsive mRNAs, which was associated with reduced histone acetylation at promoters of ACLY-sensitive genes (14). How ACLY regulates the response of human macrophages to IL-4 is unknown.
Since we previously noticed considerable differences in metabolic requirements of human vs. murine macrophages toward IL-4-induced polarization (17), we questioned the role of ACLY in regulating human monocyte-derived macrophage (MDM) responses to IL-4. Our data suggest that ACLY has little impact on transcriptional regulation of IL-4-responsive genes. Surprisingly, we observed a widespread inhibition of IL-4-induced target gene expression by pharmacological ACLY inhibitors, which persisted in ACLY knockout THP-1 macrophages, suggesting off-target effects of these substances.
Acetyl-CoA Determination
Cells were rapidly washed with saline, and metabolism was quenched by putting the dishes into liquid nitrogen. Cells were scraped into methanol:water (5:3) mix on ice/dry ice followed by addition of cold chloroform (cat. no. 4432.1, Carl Roth), and vortexing for 10 min at 4 • C. Aqueous phase was separated, evaporated, and re-suspended in 10% methanol (cat. no. 32213, Sigma-Aldrich). Acetyl-CoA concentration was determined with a Sciex QTrap5500 mass spectrometer operating in multiple reaction monitoring mode in positive electrospray ionization mode. Chromatographic separation was performed on an Agilent 1290 Infinity LC system (Agilent) using an Acquity HSS T3 column. The mobile phase consisted of (A) water, 10 mM ammonium formate (cat. no. 70221, Sigma-Aldrich), 0.01% ammonia, and (B) methanol, 10 mM ammonium formate, 0.01% ammonia. Elution of analytes was carried out under gradient conditions at a flow rate of 0.3 mL/min going from 2% B to 70% B in 5 min, increasing to 95% B in 1 min, hold 95% B for 0.5 min, and equilibrate at 2% B for 2.5 min. Calibration curve was performed with an authentic standard. All samples and dilutions of the standards were spiked with heavy isotope labeled internal standard containing 13 C2-acetyl-CoA (cat. no. 658650, Sigma-Aldrich). Acetyl-CoA concentration was determined by reference to the standard. Analyst 1.6.2 and MultiQuant 3.0 (both Sciex), were used for data acquisition and analysis, respectively. Acetyl-CoA amounts in the sample were normalized to sample DNA concentration measured following incubation with Höchst 33342 fluorescent DNA dye (cat. no. B2261, Sigma-Aldrich) on a Tecan fluorescence plate reader.
CRISPR/Cas 9 Knockout of ACLY in THP-1 Cells pLentiCRISPRv2 (Addgene: cat. no. #52961) vectors harboring different sgRNAs designed using the benchling software package ( Table 1) were obtained by target-specific oligonucleotide annealing using the GoldenGate protocol (18). Cell-free lentiviral supernatants were produced by co-transfection of pLentiCRISPRv2 vectors, gag/pol helper plasmid, and envelope plasmid encoding the glycoprotein of vesicular stomatitis virus into HEK293T cells using the JetPrime transfection reagent (Polyplus Transfection). Seventy-two hours post transfection viral supernatants were harvested and sterile filtered. THP-1 cells were infected with lentiviruses, and transduced cells expressing EGFP were sorted using FACS Aria cell sorter, followed by dilution in cell culture media to obtain single-cell suspensions. Resulting single cell-derived colonies were analyzed for ACLY knockout using Western Blotting.
Statistical Analysis
Data are presented as means ± S.E. of at least three independent experiments. Data were analyzed by one-way analysis of variance (ANOVA) with Bonferroni post-hoc means comparison using GraphPad Prism. Differences were considered statistically significant at p < 0.05.
Ethics
Investigations were conducted in accordance with the ethical standards and according to the Declaration of Helsinki and to the national and international guidelines and have been approved by the authors' institutional review board. The ethics committee of Goethe-University waived the necessity of written informed consent when using the buffy coats from anonymized blood donors.
RESULTS
To investigate the role of ACLY in IL-4-stimulated human macrophage polarization we initially analyzed mRNA expression of arachidonate 15-lipoxygenase (ALOX15) in MDMs in the presence of pharmacological ACLY inhibitors BMS 303141 (19), SB 204990 (20), MEDICA16 (21), and hydroxycitrate (22). Inhibitor concentrations were previously described in the literature. ALOX15 was chosen because of its possible expression sensitivity to metabolic perturbations through regulation by the central metabolic sensor AMP-activated protein kinase (23). MDMs were pre-incubated with ACLY inhibitors for 1 h followed by 24 h-treatment with IL-4 in the presence of inhibitors. As shown in Figures 1A-D, all ACLY inhibitors concentration-dependently suppressed IL-4-induced ALOX15 mRNA expression. Similarly, ACLY inhibitors prevented ALOX15 mRNA induction in MDMs treated with the Th2 cytokine IL-13 ( Figure 1E).
A previous study showed that ACLY was necessary for IL-4induced expression of a subset of IL-4 target genes in murine macrophages (14). To assess the ubiquitous nature of ACLY in affecting IL-4-stimulated gene expression in MDMs we analyzed mRNA expression of several well described IL-4 targets, e.g., CCL17, F13A1, CCL18, and MRC1 (CD206) (24). Whereas, IL-4stimulated CCL17 and F13A1 mRNA expression was uniformly inhibited by ACLY inhibitors, this was not the case for CCL18 and MRC1 (Figures 1F-I). Apparently, the sensitivity toward ACLY inhibition is IL-4 target gene-specific. Remarkably, analyzing the list of top 50 IL-4 responsive genes sensitive to ACLY inhibition in murine BMDMs (14), we found that only 5 genes (CCL17, Frontiers in Immunology | www.frontiersin.org CAMK2A, PHF19, ALDH1A2, ITGAX) were induced by IL-4 more than 1.5-fold in MDMs (25). This highlights the differences between the transcriptional responses toward IL-4 in human and murine systems. Except for CCL17, none of these genes showed more than two-fold induction by IL-4 or ACLY inhibitor sensitivity in quantitative PCR analyses (data not shown).
Next, we assessed the effect of silencing ACLY mRNA expression on IL-4-induced MDM polarization. For this, we treated MDMs for 96 h with control or ACLY siRNAs prior to 24 h-treatment with IL-4. Surprisingly, a knockdown of ACLY failed to reproduce the effect of ACLY inhibitors on IL-4-stimulated gene expression (Figures 2A-C), despite of a 90% decrease of ACLY mRNA (Figure 2D), an 80% reduced ACLY protein expression (Figure 2E), and a 60% decrease in ACLY enzymatic activity ( Figure 2F). We also did not observe any significant changes of IL-4-induced target gene mRNA expression upon overexpression of ACLY in MDMs (Supplementary Figure 1).
Since the impact of ACLY on gene expression is linked to reduced acetylation of histone proteins in macrophages (14) and other cells (9, 13), we analyzed acetylation of lysine 27 and 14 on histone H3. Acetylation of H3K14 and H3K27 was previously shown to respond to alterations of ACLY activity (12,26). A knockdown of ACLY failed to alter acetylation of H3K14 and H3K27 ( Figure 2G). These data suggest that ACLY silencing in primary human MDMs does not recapitulate the effect of pharmacologic ACLY inhibition on IL-4-induced gene expression.
Considering these discrepancies, we decided to investigate the impact of ACLY inhibitors on human macrophage metabolism in more detail. Surprisingly, we did not find any differences in the levels of acetyl-CoA in cells treated with BMS 303141 or SB 204990 for 24 h (Figure 3A). Accordingly, treating MDMs with ACLY inhibitors for 24 h did not affect histone H3 acetylation at Lys14 or Lys27 ( Figure 3B). Next, we exploited treatments known to increase nucleocytosolic levels of acetyl-CoA and histone acetylation. In these experiments, MDMs were preincubated with acetate (9), inhibitor of acetyl-CoA carboxylase TOFA (27), or octanoate (28), for 1 h prior to treatments with SB 204990, hydroxycitrate, and IL-4. As Figures 3C-F show, acetate, TOFA, and octanoate failed to reverse inhibition of IL-4-stimulated ALOX15 and CCL17 expression by SB 204990 or hydroxycitrate. Alternatively, we aimed to block the transport of the ACLY substrate citrate from mitochondria to the cytosol by pre-treating MDMs with different concentrations of pharmacological inhibitors of the mitochondrial citrate carrier SLC25A1 1,2,3-benzene-tricarboxylic acid (BTA) (29) and 4chloro-3-{[(3-nitrophenyl)amino]sulfonyl}benzoic acid (CTPi) (30) for 1 h prior to 24 h-treatment with IL-4. Neither BTA, nor CTPi influenced IL-4-stimulated gene expression (Figures 3G,H). These results indicate that the impact of ACLY inhibitors on IL-4-induced gene expression may be unrelated to regulation of nucleocytosolic acetyl-CoA.
We then investigated the influence of ACLY inhibitors on initial steps of IL-4-induced signal transduction by preincubating MDMs with inhibitors for 1 h followed by 0.5 htreatment with IL-4. ACLY inhibitors did not affect IL-4triggered tyrosine phosphorylation of STAT6 with the exception of BMS 303141 and hydroxycitrate ( Figure 3I). We noticed 25-30% inhibition of STAT3 tyrosine phosphorylation in ACLY inhibitor-exposed, IL-4-stimulated MDMs (Figure 3I). Whereas, STAT3 phosphorylation at Ser727 and acetylation at Lys685 were reported to affect STAT3 transcriptional activity, neither IL-4, nor ACLY inhibitors influenced these post-translational modifications in our system (data not shown).
Murine BMDMs responded to IL-4-stimulation with an Aktdependent increase of ACLY phosphorylation at Ser454, which exhibited a delayed kinetics as compared with IL-4-induced Akt phosphorylation (14). We followed kinetics of Akt and ACLY phosphorylation in MDMs stimulated with IL-4 for different times. In our hands, Akt phosphorylation at Ser473 was transiently increased in IL-4-stimulated human MDMs ( Figure 4A). However, we noticed very modest differences in ACLY phosphorylation during the time course of IL-4 treatment ( Figure 4A). ACLY is known to be partially present in the nuclear fraction of different cancer cell lines (9). We also observed ACLY in the nuclear fraction of human MDMs, although most of the enzyme was located in the cytosol. IL-4 did not cause any alteration of ACLY localization (Figure 4B). We failed to detect ACLY phosphorylation in the nuclear fraction in response to IL-4. Finally, we tracked histone acetylation in MDMs stimulated with IL-4 for different times. Whereas, lysine 27 of histone H3 exhibited no significant changes in acetylation, K14 acetylation increased within 0.5 h of IL-4 treatment and remained elevated up to 24 h ( Figure 4C). Collectively, these observations indicate that in human MDMs ACLY does not seem to be regulated by phosphorylation or nuclear transport. This is in contrast to results obtained in the murine system. An ACLY knockdown in human MDMs left substantial amounts of residual ACLY activity. Therefore, we proceeded to create a knockout of ACLY in human myeloid THP-1 cells, which upon differentiation resemble human MDMs. Although they have a defect in IL-4 receptor signaling through the absence of a common gamma receptor chain (31), THP-1 cells retain transcriptional responses to IL-4 stimulation and are used to investigate IL-4-induced human macrophage polarization (32). Figure 5A shows the complete absence of ACLY protein in ACLY knockout THP-1 cells. As expected, ACLY-deficient THP-1 cells exhibited delayed growth, likely through deficiencies in de novo lipogenesis (Figure 5B). In contrast to primary macrophages, ACLY knockout THP-1 macrophages showed reduced levels of histone H3 acetylation on lysines 9, 14, 23, and 27 ( Figure 5C). However, ablation of ACLY did not prevent the ability of THP-1 to respond to 24 h IL-4-stimulation with increased gene expression (Figures 5D,E). This was also reflected by intact phosphorylation of STAT6 after 0.5 h IL-4-treatment in ACLY knockout cells (Figure 5F). Most strikingly, pre-incubating ACLY knockout THP-1 cells with ACLY inhibitors for 1 h still suppressed IL-4-induced mRNA expression of CCL13 and F13A1 (Figures 5G,H). These data strongly indicate that ACLY inhibitors suppress IL-4-induced gene expression independently of ACLY through off-target effects.
DISCUSSION
ACLY is thought to link metabolism and epigenetic control of transcription through its provision of acetyl-CoA for nuclear histone acetylation (8). Strikingly, our study suggests that ACLY has little, if any, influence on IL-4-induced transcriptional responses and histone acetylation in human MDMs. Our data thus contrast with observations in murine BMDMs, where ACLY was shown to significantly contribute to the induction of at least a subset of the IL-4-sensitive transcriptome by increasing histone acetylation (14). Moreover, we show that several commonly used pharmacological ACLY inhibitors influence IL-4-induced gene expression even in the absence of ACLY, strongly pointing to off-target effects of these drugs. This warrants caution in interpretation of the previous study (14) as well as other reports relying on use of ACLY inhibitors.
Which differences between our and murine BMDM system may explain the observed discrepancies? Obviously, species difference can be a major factor. Studies from our and other groups show that murine and human macrophages differ not only in how their transcriptome is altered in response to IL-4 (25,33), but also how their metabolism controls IL-4-dependent gene expression (17). We now observe that initial steps in signal transduction downstream of the IL-4 receptor differ between human and murine systems. Thus, human MDMs do not show Akt-dependent ACLY phosphorylation in response to IL-4. Akt activation in our system is only transient in contrast to prolonged activation in BMDMs (14). Whether this relates to differences in insulin receptor substrate 2 engagement by the IL-4 receptor, which is thought to be responsible for Akt activation by IL-4 (31), remains to be investigated. Another difference is sustained proliferation in response to macrophage colony-stimulating factor and IL-4 (34), which was noticed in BMDMs (14). Since high ACLY activity is a pre-requisite for ongoing proliferation of cells through its contribution to de novo lipogenesis (10), proliferating macrophages may be characterized by increased ACLY activity, exerting greater influence on nuclear acetyl-CoA and, consequently, histone acetylation. In contrast, our experimental setup employed fully differentiated macrophages, which do not proliferate. Negligible de novo lipogenesis was evidenced by the failure to incorporate C13-carbon from C13labeled glucose to cellular palmitate (data not shown). Thus, the role of ACLY in metabolism and epigenetic regulation of terminally differentiated human macrophages remains unclear and warrants further research. Of note, ACLY may have greater impact during human macrophage differentiation, since this process is characterized by a temporary rise of de novo lipogenesis (35). Another proposed function of ACLY in macrophages, based on observations in U937 cell line, is the provision of substrates for increased synthesis of bioactive mediators, such as prostaglandin E 2 , nitric oxide, or reactive oxygen species, in response to proinflammatory stimuli (36), which remains to be validated in primary cells.
Based on our studies, a critical unresolved question is, which enzymatic system provides nuclear acetyl-CoA for histone acetylation in human MDMs. Whereas, our data show that ACLY definitely contributes to histone acetylation in the acute myeloid leukemia cell line THP-1, we obtained no evidence for such a behavior in human MDMs. Several alternative sources of nuclear acetyl-CoA are described. The major source is conversion of acetate to acetyl-CoA by the action of nucleocytosolic acyl-CoA synthetase short-chain family member 2 (ACSS2). ACSS2 contributes to histone acetylation in cancer cells, especially under hypoxia or glucose deprivation (37,38), and can also participate in recycling of acetate released from histones by the action of histone deacetylases (39). Acetyl-CoA can also be synthesized in the nucleus through nuclear translocation of the pyruvate dehydrogenase complex (40,41). An alternative cytosolic acetyl-CoA generation system, involving coordinated actions of mitochondrial succinyl-CoA:3-ketoacid-CoA transferase and cytosolic acetoacetyl-CoA synthetase and acetyl-CoA acyltransferase activities was described for pancreatic beta-cells (42). Finally, activity of carnitine acetyl-CoA transferase and other uncharacterized pathways were proposed to link mitochondrial and nucleocytosolic acetyl-CoA (28,43,44). Which of these pathways contributes to nucleocytosolic acetyl-CoA in human MDMs remains the topic of current investigation.
Our results also highlight the notorious proneness of pharmacological inhibitors to off-target effects, which is in our case particularly remarkable, since we used structurally dissimilar substances (hydroxylated citrate, tertramethylated long chain dicarboxylic fatty acid, tricyclic aromatic sulfonamide, and a dychlorphenylhexyl-substituted hydroxylated derivative of tetrahydrofuranacetic acid). Off-target effects of ACLY inhibitors on IL-4-stimulated gene transcription could only be revealed using ACLY knockout cell line. Our findings also point to the limitation of studying human primary cells, since ACLY knockdown leaves substantial residual activity left whereas pharmacological inhibitors are unsuitable due to off-target activities. Whether induced pluripotent stem cell-derived macrophages, where creation of knockout models is possible (45), will be a better suitable model for studying metabolic regulation of macrophage polarization as compared to acute myeloid leukemia THP-1 cells, should be revealed by future investigation.
AUTHOR CONTRIBUTIONS
DN designed the study, performed the experiments, and wrote the manuscript. SZ and IF contributed to acetyl-CoA measurements. FS, NK, and DF contributed to CRISPR/Cas9 THP-1 knockout cell line creation. BB contributed to study design and edited the final manuscript. All authors contributed to manuscript revision, read, and approved the submitted version. | 4,316.2 | 2018-12-04T00:00:00.000 | [
"Biology"
] |
Characterization of hydroxypropyl-beta-cyclodextrins used in the treatment of Niemann-Pick Disease type C1
2-Hydroxypropyl-beta-cyclodextrin (HPβCD) has gained recent attention as a potential therapeutic intervention in the treatment of the rare autosomal-recessive, neurodegenerative lysosomal storage disorder Niemann-Pick Disease Type C1 (NPC1). Notably, HPβCD formulations are not comprised of a single molecular species, but instead are complex mixtures of species with differing degrees of hydroxypropylation of the cyclodextrin ring. The degree of substitution is a critical aspect of the complex mixture as it influences binding to other molecules and thus could potentially modulate biological effects. VTS-270 (Kleptose HPB) and Trappsol® Cyclo™ are HPβCD products under investigation as novel treatments for NPC1. The purpose of the present work is to compare these two different products; analyses were based on ion distribution and abundance profiles using mass spectrometry methodology as a means for assessing key molecular distinctions between products. The method incorporated electrospray ionization and analysis with a linear low-field ion mobility quadrupole time-of-flight instrument. We observed that the number of hydroxypropyl groups (the degrees of substitution) are substantially different between the two products and greater in Trappsol Cyclo than in VTS-270. The principal ions of both samples are ammonium adducts. Isotope clusters for each of the major ions show doubly charged homodimers of the ammonium adducts. In addition, both products show doubly charged homodimers from adduction of both a proton and ammonium. Doubly charged heterodimers are also present, but are more intense in Trappsol Cyclo than in VTS-270. Based on the analytical differences observed between VTS-270 and Trappsol Cyclo with respect to the degree of substitution, the composition and fingerprint of the complex mixture, and the impurity profiles, these products cannot be considered to be the same; the potential biological and clinical implications of these differences are not presently known.
Introduction
Cyclodextrins are cyclic oligosaccharides that, for many years, have been used to modulate the composition of cholesterol and other lipids in biological membranes and have also been used as pharmaceutical excipients in the formulation of hydrophobic drugs [1,2]. More recently, 2-Hydroxypropyl-β-cyclodextrins (HPβCDs) have gained attention as a potential therapeutic intervention for Niemann-Pick Disease Type C1 (NPC1), an autosomal-recessive, rare, fatally progressive, neurodegenerative lysosomal storage disease characterized by endo-lysosomal accumulation of cholesterol and other lipids [3][4][5][6][7]. Currently, there are no approved pharmacological drugs for the treatment of NPC1 in the United States, and options available to patients are limited to supportive therapies and use of miglustat, which is off-label in the United States; miglustat is an iminosugar that reduces glycosphingolipid production through the inhibition of the enzyme glucosylceramide synthase [8]. Management of the disease is mainly aimed at symptomatic relief [9,10], leaving unmet medical needs for NPC1 patients. However, progress has been made recently in the development of novel therapeutics for NPC1 [4,11,12]. In the United States and European Union, two HPβCD products, VTS-270 (Vtesse, Inc., Gaithersburg, MD) and Trappsol 1 Cyclo™ (CTD Holdings, Inc., Alachua, FL) have received orphan drug designations. VTS-270 uses Kleptose 1 HPB (Roquette Pharma, France) as the active ingredient, which has shown highly encouraging results in a phase 1/2a clinical trial (NCT01747135) and is currently being studied in a global pivotal phase 2b/3 clinical trial (NCT02534844) as a treatment for the neurological manifestations of NPC1 [7,13]. Trappsol Cyclo is a second HPβCD and is the subject of a recent investigational new drug filing. Given the use of these HPβCD products in NPC1 clinical investigations, it is important to understand if they are chemically different and clinically equivalent. Not only has there been extensive interest in HPβCD materials for NPC1, but also for a number of other disorders including atherosclerosis, Alzheimer's disease, Parkinson's disease, and Huntington's disease [14].
Unlike most FDA-approved drugs, which are characterized as single chemical or molecular species, HPβCDs are complex mixtures of different species, and variations in production may lead to differences in composition. HPβCDs are synthesized by condensation between β-cyclodextrin (βCD) and propylene oxide [15]. There are 21 hydroxyl groups on βCD, all of which are potential sites for the condensation reaction (Fig 1) [15]. Following condensation, the number of substituted hydroxypropyl groups on the cyclodextrin ring varies with synthesis conditions and leads to a distribution in the degrees of substitution on HPβCD [15]. Importantly, the condensation reactions do not lead to a single molecular HPβCD species but rather to a complex mixture of HPβCD species with differing degrees of hydroxypropylation. The average degree of substitution (DS) is a critical characteristic of the complex mixture, as it influences the ability of HPβCDs to bind other molecules [16] and affects the degree of aqueous solubility [1], which may influence biological activity. Thus, in the setting where these materials are used as therapeutic agents, the need to accurately characterize the complex mixture of HPβCD species becomes necessary in order to attain consistent clinical safety and efficacy responses. While quality standards for HPβCDs used as pharmaceutical excipients are based on United States Pharmacopeia (USP) specifications [17], these characterization criteria were not developed for active therapeutic agents that directly treat disease and are considered to be unsatisfactorily broad and insufficiently specific for such applications. For potential therapeutic applications, it is important to characterize HPβCD mixtures to aid in the future determination of which specific components may be therapeutically active and which may be toxic. Herein, we report an approach to characterize HPβCDs using ion mobility mass spectrometry to determine both the DS and describe specific ions arising from interactions between HPβCDs that yield homo-and heterodimeric ions of both protonated and ammonium adducts; we term these characteristic interaction ions a "molecular fingerprint." Using this method, VTS-270 and Trappsol Cyclo were analyzed and characterized, and substantial chemical differences were observed.
Materials and methods Materials
All reagents were used as supplied unless otherwise noted. Water was purified using a Hydroservices system (Levittown, PA) and then polished using a Millipore Simplicity UV purification unit (Billerica, MA); HPLC grade acetonitrile (ACN) was obtained from Burdick and Jackson (Morris Plains, NJ).
VTS-270 (Vtesse, Inc., Gaithersburg, MD) was supplied in 14 separate synthetic batches for comparison. Two separate synthetic batches of Trappsol Cyclo (CTD Holdings, Inc., Alachua, FL) were used for comparison. HPβCD solutions were prepared as a 1 mg/mL stock in water and infused at either a 200-fold dilution in 1:1 water:ACN or a 400-fold dilution in water:80% (v/v) ACN. Based on the chemical structure and stock solution, a 200-fold dilution represents a concentration of approximately 5 μM.
Measurement approach
An Agilent Model 6560 Ion Mobility Quadrupole Time-of-Flight Mass Spectrometer (Agilent Technologies, Santa Clara, CA) was used for all measurements. The instrument has been Characterization of HP-betaCDs for Niemann-Pick Disease type C1 described previously [18]. In brief, electrospray-generated ions were formed by direct infusion of HPβCD solutions using a nano-electrospray ionization (ESI) source with a flow rate set at 600 nL/min, V cap potential of 1500 V, and sheath gas flow at 5.0 L/min at 150˚C. With this instrument, ions are sampled into the drift tube through a series of ion funnels, the last of which serves as a trapping ion funnel to deliver ions at 80-msec intervals to the ion mobility drift tube. RF amplitudes in the trapping ion funnel were adjusted to optimize ion intensities. The drift tube was operated with a N 2 bath gas pressure at 3.94 Torr and 27˚C as well as 1450 V potential across the tube. Upon exiting the drift tube, ions passed through another series of optics and were mass analyzed by a high-resolution Q-TOF mass spectrometer. Ion signals were collected for 3-minute intervals prior to analysis and measurements were made in duplicate each day for a total of 3 days over 3 weeks (ie, 12 replicate measurements). All spectra collected were analyzed using Agilent Mass Hunter software (version 7.01) for visualization and manually interpreted and annotated. Mass assignments were based upon accurate mass measurement. Ion collision cross sections were determined using software supplied in this same package.
Ion collision cross section determination
Under conditions of low drift fields in an appropriate pressure regimen, as defined by Mason and McDaniel [19] and approximating both an ion and the buffer gas as hard spheres, the collision cross section, O, can be calculated in units of Å 2 from the expression: where k B is the Boltzmann constant, z and e are the charge state and electronic charge, respectively, N Ã is the gas number density in the drift tube, m i is the ion mass, and m B is the mass of the buffer gas. Widespread usage has allowed the spherical assumptions to be extended to what are clearly non-spherical ions and to non-spherical, polarizable drift gases. K 0 is determined from measurements of apparent ion drift times that are corrected for passage time in the instrument outside of the drift tube region and transformed to STP conditions.
Results
All batches of VTS-270 and both Trappsol Cyclo batches showed consistent behavior; data are shown for a representative single batch for each product.
Ion mobility-mass spectrometry profiles of VTS-270 and Trappsol Cyclo
Initial experiments compared the ion mobility profile of the two HPβCD products. [20]. Ions with the same m/z values but different charge states have different drift times, with higher charge state ions having shorter drift times [20]. These components of the ion mobility response give rise to the several "families" of ion responses corresponding to ions with related structures and different charge states seen in the two panels of Fig 2. Considering that the two materials are infused under the same preparation conditions and at the same concentration, it is apparent that there is a much greater level of non-specific chemical "noise" associated with Trappsol Cyclo compared with VTS-270, particularly in the lower m/z regions over drift times up to approximately 45 msec. In addition, the Trappsol Cyclo heat map shows the presence of
Mass spectra of VTS-270 and Trappsol Cyclo
Additional differences between VTS-270 and Trappsol Cyclo can be seen in Fig 3, which presents the mass spectra extracted from Region 1 denoted on Fig 2. These mass spectra exhibit resolutions of 20,000 or greater and peak assignments are consistent with 10 ppm or better mass accuracy when compared with expected values. This region of the heat maps covers the mass range 1200 to 1700 Da and drift times between 25 and 50 msec. Fig 3 represents a "compression" of all the ion intensities in this drift region and therefore combines the intensities of multiply charged ions of the same m/z, namely, monomeric singly charged and doubly charged ions formed from two HPβCD molecules as either two molecules of a single DS (homodimers) or two different degrees of substitution (heterodimers). The most intense ions in the spectra are those formed by NH 4 + adduction to neutral HPβCD species with smaller contributions from protonated adducts. The spectra show a distribution of ion intensities corresponding to the degrees of substitution of the two materials studied, and demonstrate clear differences between them. The numbers above the most intense ions of each spectrum show the DS of the ammonium adduct for that species (eg, nominal mass 1384 corresponds to the ammonium adduct of a DS = 4). Close examination of the isotope clusters for each of the major ions shows both doubly same preparation conditions and at the same concentration of HPβCD. Trappsol Cyclo exhibits greater chemical heterogeneity than VTS-270 as observed by the more complex overall heat map and the m/z singly charged trend line (long oval-shaped area), the region labeled "higher order complexes," and the region corresponding to triply charged dimers.
https://doi.org/10.1371/journal.pone.0175478.g002 Fig 5A. Similar fractional intensity changes as a function of DS are seen for both materials, but the relative fractional intensities of these dimers appear to be greater for Trappsol Cyclo than for VTS-270. Fig 5B shows the fraction of total ions represented by heterodimers as a function of DS for both materials. The difference between the two is more apparent in this case as the degree of heterodimer formation is seen to be much greater for Trappsol Cyclo than for VTS-270, particularly at higher DS. While these data were collected at 5 μM solution conditions, it is important to note that these observable differences were preserved under conditions of a 400-fold dilution in 80% ACN.
Ion collision cross sections
Collision cross section (Ω) areas of the major ions of HPβCD were calculated using the VTS-270 and Trappsol Cyclo data as described and presented in Tables 1-6. These novel data were generated and included to add to the growing publicly available characterization of the ion collision cross sections measurements of various molecular species. While there are differences between VTS-270 and Trappsol Cyclo in the various ions present as described above, the ion collision cross sections for each of the ions present in both materials were similar, as expected. Tables 1-6 show values of O for a series of different ions with varying DS observed in the m/z region between 1200 and 1620 Da. As seen in the mass spectra in Figs 3 and 4, these ions consist of a mixture of adducts of protons as singly and doubly charged species. As also seen in the spectra, not all ions of each adduct type are present in both VTS-270 and Trappsol Cyclo; however, when both species are present, the values calculated for their collision cross sections are in close agreement as noted in Tables 1-6.
Discussion
Using electrospray coupled with ion mobility mass spectrometry, we demonstrated substantial differences between HPβCDs from two different sources currently being administered to NPC1 subjects. Specifically, both the DS and spectral fingerprints can be used to distinguish between HPβCDs of different sources; we found clear differences between two HPβCD products. Trappsol Cyclo was found to have a higher DS compared with VTS-270, increased levels of dimeric ions, and additional differences in ion mobility profiles. Furthermore, Trappsol Cyclo had greater nonspecific chemical noise with higher order complexes of HPβCDs compared with VTS-270. The differences observed may be a consequence of the higher average DS of Trappsol Cyclo, giving rise to a greater propensity to form dimeric ions compared with VTS-270, which has a lower DS. For data acquisition, the instrument was operated under conditions designed to facilitate the observation of species existing in solution (ie, high temperature was not implemented). Accordingly, our observation of dimeric ions is most likely a reflection of species existing in solution prior to ion formation.
Dimer intensities appear to increase as a function of the DS, but absolute signal intensities arising from their presence is reduced by the decreasing intensity of the higher DS forms. In the case of Trappsol Cyclo, the overall higher DS leads to an overall relative increase in dimer signals. Considering the possibilities for hydrogen bonded non-covalent interactions that arise from greater levels of hydroxyl-propyl substitution, the existence of more extensive dimer interactions with Trappsol Cyclo rather than with VTS-270 is not surprising. The robustness of these interactions is apparently high because they were preserved during a 400-fold dilution and 80% ACN. Since these materials are normally administered at concentrations approximately 100-fold higher than used in these studies, the presence of dimers in solution appears to be likely. The observation of the strong heterodimers in Trappsol Cyclo suggests a different level of intermolecular interactions in Trappsol Cyclo than in VTS-270.
To our knowledge, there have not been any other studies directly comparing VTS-270 with Trappsol Cyclo regarding ion composition profiles or using the methods described herein to obtain and differentiate molecular fingerprints for the two materials. However, in a recently reported study that examined a series of α-, β-and γ-cyclodextrins for potential use in NPC, the average DS was noted to be 7.0 for Trappsol Cyclo and 4.3 for VTS-270, which was comparable with our findings, and the binding constants for unesterified cholesterol were found to be 3,250 ± 80 M -1 for Trappsol Cyclo and 4,400 ± 66 M -1 for VTS-270 [6]. Further studies are needed to examine potential differences in biological and therapeutic effects of Trappsol Cyclo and VTS-270. The data presented here strongly suggest that biological and potential therapeutic equivalence should not be assumed, as the activities of individual molecular species are not yet well understood or described. In the preparation of HPβCDs, the proportion of contaminants has been shown to be greater with HPβCDs with higher DS [15], and this holds true in comparing Trappsol Cyclo with VTS-270. The differences in the DS, dimer ion intensity, chemical noise, and contaminants indicate that VTS-270 and Trappsol Cyclo are not chemically equivalent and therefore may not be biochemically equivalent or lead to comparable formulations from a clinical development perspective. Therefore, when considering biological or clinical safety and efficacy data, it should not be assumed that effects observed with VTS-270 would also occur with Trappsol Cyclo and vice versa. Each product must be examined and evaluated independently and comparatively to fully understand the potential clinical similarities and differences with regard to safety and efficacy.
Conclusions
These studies demonstrate the significant complexity that exists within the composition of HPβCDs. Based on ion mobility and mass spectral data, two HPβCD materials (VTS-270 and Trappsol Cyclo) were found to have substantial mass spectral differences and thus should not be considered to be chemically equivalent. Further research is needed to determine the impact these differences may have biologically and/or clinically in investigations of NPC. | 4,030.6 | 2017-04-17T00:00:00.000 | [
"Chemistry"
] |
Inhibition of Vascular Endothelial Growth Factor (VEGF)-induced Endothelial Cell Proliferation by a Peptide Corresponding to the Exon 7-Encoded Domain of VEGF165 *
Vascular endothelial growth factor (VEGF) is a potent mitogen for endothelial cells (EC) in vitro and a major regulator of angiogenesis in vivo. VEGF121 and VEGF165 are the most abundant of the five known VEGF isoforms. The structural difference between these two is the presence in VEGF165 of 44 amino acids encoded by exon 7 lacking in VEGF121. It was previously shown that VEGF165 and VEGF121 both bind to KDR/Flk-1 and Flt-1 but that VEGF165 binds in addition to a novel receptor (Soker, S., Fidder, H., Neufeld, G., and Klagsbrun, M. (1996)J. Biol. Chem. 271, 5761–5767). The binding of VEGF165 to this VEGF165-specific receptor (VEGF165R) is mediated by the exon 7-encoded domain. To investigate the biological role of this domain further, a glutathioneS-transferase fusion protein corresponding to the VEGF165 exon 7-encoded domain was prepared. The fusion protein inhibited binding of125I-VEGF165 to VEGF165R on human umbilical vein-derived EC (HUVEC) and MDA-MB-231 tumor cells. The fusion protein also inhibited significantly125I-VEGF165 binding to KDR/Flk-1 on HUVEC but not on porcine EC which express KDR/Flk-1 alone. VEGF165had a 2-fold higher mitogenic activity for HUVEC than did VEGF121. The exon 7 fusion protein inhibited VEGF165-induced HUVEC proliferation by 60% to about the level stimulated by VEGF121. Unexpectedly, the fusion protein also inhibited HUVEC proliferation in response to VEGF121. Deletion analysis revealed that a core inhibitory domain exists within the C-terminal 23-amino acid portion of the exon 7-encoded domain and that a cysteine residue at position 22 in exon 7 is critical for inhibition. It was concluded that the exon 7-encoded domain of VEGF165 enhances its mitogenic activity for HUVEC by interacting with VEGF165R and modulating KDR/Flk-1-mediated mitogenicity indirectly and that exon 7-derived peptides may be useful VEGF antagonists in angiogenesis-associated diseases.
Angiogenesis, the process in which new blood vessels sprout from pre-existing vessels, normally occurs during reproduction, embryonic development, and wound repair. On the other hand, pathological processes such as tumor progression may lead to aberrant angiogenesis (reviewed in Refs. [1][2][3][4]. The discovery that tumor growth is angiogenesis-dependent has led to the identification of a number of angiogenesis-promoting factors such as basic (bFGF) 1 and acidic fibroblast growth factor, vascular endothelial growth factor (VEGF), tumor necrosis factor-␣, transforming growth factor-, platelet-derived endothelial cell growth factor, and interleukin-8 (reviewed in Refs. 2, 4, and 5). Concomitant with the discovery of positive regulators of angiogenesis, inhibitors of angiogenesis have been identified including thrombospondin-1, interferon-␥, thalidomide, AGM-1470, the 16-kDa fragment of prolactin, cartilage-derived inhibitor, angiostatin, and endostatin (reviewed in Refs. 2, 4, and 5).
There is mounting evidence that VEGF may be a major regulator of angiogenesis (reviewed in Refs. 6 -8). VEGF was initially purified from the conditioned media of folliculostellate cells (9) and from a variety of tumor cell lines (10,11). VEGF was found to be identical to vascular permeability factor, a regulator of blood vessel permeability that was purified from the conditioned medium of U937 cells at the same time (12). VEGF is a specific mitogen for endothelial cells (EC) in vitro and a potent angiogenic factor in vivo. The expression of VEGF is up-regulated in tissues undergoing vascularization during embryogenesis and the female reproductive cycle (13,14). High levels of VEGF are expressed in various types of tumors, but not in normal tissue, in response to tumor-induced hypoxia (15)(16)(17)(18). Treatment of tumors with monoclonal antibodies directed against VEGF resulted in a dramatic reduction in tumor mass due to the suppression of tumor angiogenesis (19).
VEGF exists in five different isoforms that are produced by alternative splicing from a single gene containing eight exons (6, 20 -22). Human VEGF isoforms consist of monomers of 121, 145, 165, 189, and 206 amino acids, each capable of making an active homodimer (22,23). The VEGF 121 and VEGF 165 isoforms are the most abundant. VEGF 121 is the only VEGF isoform that does not bind to heparin and is totally secreted into the culture medium. VEGF 165 is functionally different than VEGF 121 in that it binds to heparin and cell surface heparan sulfate proteoglycans (HSPGs) and is only partially released into the culture medium (24,25). The remaining isoforms are entirely associated with cell surface and extracellular matrix HSPGs (24,25).
VEGF receptor tyrosine kinases, KDR/Flk-1 and/or Flt-1, are expressed by EC and by several types of non-EC such as NIH 3T3, Balb/c 3T3, human melanoma, and HeLa cells (26 -30). It appears that VEGF activities such as mitogenicity, chemotaxis, and induction of morphological changes are mediated by KDR/ Flk-1 but not Flt-1, even though both receptors undergo phosphorylation upon binding of VEGF (31)(32)(33)(34). Recently, we have characterized a new VEGF receptor which is expressed on EC and various tumor-derived cell lines such as breast cancerderived MDA-MB-231 (231) cells (35). Although both VEGF 121 and VEGF 165 bind to KDR/Flk-1 and Flt-1, only VEGF 165 binds to the new receptor. Thus, this is an isoform-specific receptor and has been named as the VEGF 165 receptor (VEGF 165 R). VEGF 165 R has a molecular mass of approximately 130 kDa, and it binds VEGF 165 with a K d of about 2 ϫ 10 Ϫ10 M, compared with approximately 5 ϫ 10 Ϫ12 M for KDR/Flk-1. In structurefunction analysis, it was shown directly that VEGF 165 binds to VEGF 165 R via its exon 7-encoded domain which is absent in VEGF 121 (35).
VEGF 165 is a more potent mitogen for EC than is VEGF 121 (36). One possible explanation is that the interaction of VEGF 165 with VEGF 165 R enhances KDR/Flk-1-mediated VEGF 165 bioactivity. To address this hypothesis, a glutathione S-transferase (GST) fusion protein containing a peptide corresponding to the 44 amino acids encoded by exon 7 (amino acids 116 -159 of VEGF 165 ) was prepared. The GST-exon 7 fusion protein inhibited the binding of 125 I-VEGF 165 to receptors on human umbilical cord vein-derived EC (HUVEC) and on 231 cells. The inhibitory activity was localized to the C-terminal portion of the exon 7-encoded domain. Furthermore, the fusion protein inhibited VEGF-induced proliferation of HUVEC. These results suggest that the exon 7-encoded domain contributes to the enhanced VEGF 165 mitogenic activity for HUVEC and that exon 7-derived peptides are potential antagonists of VEGF mitogenic activity for EC.
EXPERIMENTAL PROCEDURES
Materials-Human recombinant VEGF 165 and VEGF 121 were produced in Sf-21 insect cells infected with recombinant baculovirus encoding human VEGF 165 or VEGF 121 as described previously (35,37). VEGF 165 was purified from the conditioned medium of infected Sf-21 cells by heparin affinity chromatography, and VEGF 121 was purified by anion exchange chromatography. Basic FGF was kindly provided by Dr. Judith Abraham (Scios, Sunnyvale, CA). Cell culture media were purchased from Life Technologies, Inc. 125 I-Sodium was purchased from NEN Life Science Products. Disuccinimidyl suberate and IODO-BEADS were purchased from Pierce. Glutathione-agarose, NAP-5 columns, and pGEX-2TK plasmid were purchased from Pharmacia Biotech Inc. TSK-heparin columns were purchased from TosoHaas (Tokyo, Japan). Molecular weight marker was purchased from Amersham Corp. IL). Porcine intestinal mucosal-derived heparin was purchased from Sigma.
Cell Culture-Human umbilical vein endothelial cells (HUVEC) were obtained from the American Type Culture Collection (ATCC) (Rockville, MD) and grown on gelatin-coated dishes in M-199 medium containing 20% fetal calf serum (FCS) and a mixture of glutamine, penicillin, and streptomycin (GPS). Basic FGF (1 ng/ml) was added to the culture medium every other day. Porcine endothelial cells (PAE), parental and transfected to express KDR/Flk-1 (PAE-KDR), were kindly provided by Dr. Lena Claesson-Welsh and grown in F12 medium containing 10% FCS and GPS as described (32). MDA-MB-231 (231) cells were obtained from ATCC and grown in Dulbecco's modified Eagle's medium containing 10% FCS and GPS.
Endothelial Cell Proliferation Assay-HUVEC were seeded in gelatin-coated 96-well dishes at 4,000 cells/200 l/well in M-199 containing 5% FCS and GPS. After 24 h, VEGF isoforms and VEGF exon 7-GST fusion proteins were added to the wells at the same time. The cells were incubated for 72 h, and [ 3 H]thymidine (1 Ci/ml) was added for 10 -12 h. The medium was aspirated, and the cells were trypsinized and harvested by an automatic cell harvester (TOMTEC) and loaded onto Filtermats (Wallac). The Filtermats were scanned and cpm were deter-mined by a MicroBeta counter (Wallac). The results represent the average of samples assayed in triplicate, and the standard deviations were determined. All experiments were repeated at least three times and similar results were obtained.
Radioiodination of VEGF-The radioiodination of VEGF 165 and VEGF 121 was carried out using IODO-BEADS according to the manufacturer's instructions. Briefly, one IODO-BEAD was rinsed with 100 l of 0.1 M sodium phosphate, pH 7.2, dried, and incubated with 125 Isodium (0.2 mCi/g protein) in 100 l of 0.1 M sodium phosphate, pH 7.2, for 5 min at room temperature. VEGF (1-3 g) was added to the reaction mixture, and after 5 min the reaction was stopped by removing the bead. The solution containing 125 I-VEGF was adjusted to 2 mg/ml gelatin and purified by size exclusion chromatography using a NAP-5 column that was pre-equilibrated with PBS containing 2 mg/ml gelatin. Aliquots of the iodinated proteins were frozen on dry ice and stored at Ϫ80°C. The specific activity ranged from 40,000 to 100,000 cpm/ng protein.
Binding and Cross-linking of 125 I-VEGF-Binding and cross-linking experiments using 125 I-VEGF 165 and 125 I-VEGF 121 were performed as described previously (29,35). VEGF binding was quantified by measuring the cell-associated radioactivity in a ␥-counter (Beckman, Gamma 5500). The counts represent the average of three wells. All experiments were repeated at least three times, and similar results were obtained. 125 I-VEGF cross-linked complexes were resolved by 6% SDS-PAGE, and the gels were exposed to a phosphor screen and scanned after 24 h by a PhosphorImager (Molecular Dynamics). Subsequently, the gels were exposed to x-ray film.
Preparation of GST-VEGF Exon 7 and 8 Fusion Proteins-Different segments of exons 7 and 8 of VEGF were amplified by the polymerase chain reaction from human VEGF cDNA using the following primers: exon 7 ϩ 8 (Ex 7ϩ8), CGGGATCCCCCTGTGGGCCTTGCTC and GG-AATTCTTACCGCCTCGGCTTGTC; exon 7 (Ex 7), CGGGATCCCCCT-GTGGGCCTTGCTC and GGAATTCTTAACATCTGCAAGTACGTT; exon 7 with residues 1-10 deleted (Ex 7d-(1-10)), CGGGATCCCATTT-GTTTGTACAAGAT and GGAATTCTTAACATCTGCAAGTACGTT; exon 7 with residues 1-21 deleted (Ex 7d-(1-21)), CGGGATCCTGTT-CCTGCAAAAACACAG and GGAATTCTTAACATCTGCAAGTACGTT; exon 7 with residues 1-22 deleted (Ex 7d-(1-22)), CGGGATCCTGCA-AAAACACAG and GGAATTCTTAACATCTGCAAGTACGTT, and exon 7 with residues 30 -44 deleted (Ex 7d-(30 -44)), CGGGATCCCCCTGT-GGGCCTTGCTC and GGAATTCTAGTCTGTGTTTTTGCA. The amplified products were digested with BamHI and EcoRI restriction enzymes and cloned into the vector pGEX-2TK (Pharmacia Biotech Inc.) encoding GST (38) to yield the plasmid p2TK-exon 7ϩ8 and its derivatives. Escherichia coli (DH5␣) were transformed with p2TK-exon 7ϩ8 and derivatives to produce GST fusion proteins (see Fig. 5B for sequences). Bacterial lysates were subsequently separated by a glutathione-agarose affinity chromatography (38). Samples eluted from glutathione-agarose were analyzed by 15% SDS-PAGE and silver staining. GST fusion proteins were further purified on a TSK-heparin column as described previously (35). 165 and VEGF 121 differ in their ability to interact with VEGF receptors expressed on HUVEC (35,39). VEGF 121 binds to KDR/Flk-1 to form a 240-kDa labeled complex (Fig. 1, lane 2), whereas VEGF 165 , in addition to forming this size complex, also forms a lower molecular mass complex of 165-175 kDa (Fig. 1, lane 1). This isoform-specific receptor has been named the VEGF 165 receptor (VEGF 165 R). These differential receptor binding properties suggest that VEGF 165 and VEGF 121 might also have differential mitogenic activities. Accordingly, the ability of the two VEGF isoforms to stimulate HUVEC proliferation was tested. VEGF 165 was a more potent mitogen for HUVEC than was VEGF 121 (Fig. 2). VEGF 165 stimulated half-maximal DNA synthesis at 1 ng/ml and maximal stimulation at 4 ng/ml resulting in an 8-fold increase over control. On the other hand, 2 ng/ml VEGF 121 were required for half-maximal stimulation and 20 ng/ml for maximal stimulation resulting in a 4-fold increase in HUVEC proliferation over control. Thus, twice as much VEGF 121 compared with VEGF 165 was needed to attain half-maximal stimulation, and VEGF 121 -induced proliferation was saturated at about one-half the level induced by VEGF 165 . Taken together, these results suggest that there might be a correlation between the enhanced mitogenic activity of VEGF 165 for EC compared with VEGF 121 and the ability of VEGF 165 to bind to an additional receptor (VEGF 165 R) on HUVEC.
Differential Receptor Binding and Mitogenic Activities of VEGF 165 and VEGF 121 for HUVEC-VEGF
A (35). This finding suggested that an excess of exon 7-encoded peptide might inhibit VEGF 165 binding to VEGF 165 R. GST fusion proteins containing a peptide encoded by VEGF exon 7 or by VEGF exons 7 and 8 were prepared. The 6 amino acids encoded by exon 8 which is C-terminal to exon 7 were included to facilitate the preparation of the fusion protein but did not affect the results in any way (data not shown). The exon 7 fusion protein binds directly to VEGF 165 R on 231 cells (35). It also binds directly to VEGF 165 R on HUVEC but not to KDR/FLK-1 on HUVEC (Fig. 1, lane 3). The ability of the GST-VEGF 165 exons 7-and 8-encoded peptide (GST-Ex 7ϩ8) to compete with 125 I-VEGF 165 binding to HUVEC, which express both KDR/Flk-1 and VEGF 165 R, to PAE-KDR cells which express only KDR/ Flk-1 (32), and to 231 cells which express only VEGF 165 R (35) was tested (Fig. 3). Increasing concentrations of GST-Ex 7ϩ8 markedly inhibited the binding of 125 I-VEGF 165 to HUVEC by about 85-95% (Fig. 3A) and to 231 cells by 97-98% (Fig. 3B). However, the fusion protein did not inhibit the binding of 125 I-VEGF 165 to PAE-KDR cells which do not express any VEGF 165 R (Fig. 3C). GST protein alone even at concentrations of 20 g/ml had no significant effect on the binding of 125 I-VEGF 165 to any of the cell types. Taken together, these binding studies suggested that GST-Ex 7ϩ8 competes for 125 I-VEGF 165 binding by interacting directly with VEGF 165 R but not with KDR.
These binding experiments were extended to analyze the effects of GST-Ex 7ϩ8 on 125 I-VEGF 165 binding to the individual VEGF receptor species by cross-linking (Fig. 4). Crosslinking of 125 I-VEGF 165 to 231 cells resulted in the formation of labeled complexes with VEGF 165 R (Fig. 4, lane 3). The formation of these complexes was markedly inhibited in the presence of 15 g/ml GST-Ex 7ϩ8 (Fig. 4, lane 4). 125 I-VEGF 165 crosslinking to HUVEC resulted in the formation of labeled complexes of higher molecular mass with KDR/Flk-1 and lower molecular mass complexes with VEGF 165 R (35) (Fig. 4, lane 1). GST-Ex 7ϩ8 markedly inhibited the formation of the 165-175-kDa labeled complexes containing VEGF 165 R (Fig. 4, lane 2). Unexpectedly, GST-Ex 7ϩ8 also inhibited markedly the formation of the 240-kDa labeled complex in HUVEC containing KDR/Flk-1 (Fig. 4, lane 2). On the other hand, the fusion protein did not inhibit cross-linking of 125 I-VEGF 165 to KDR/ Flk-1 on the PAE/KDR cells (not shown). Taken together, since (i) VEGF 165 binds to KDR/Flk-1 via the amino acids encoded by exon 4 (40), (ii)) VEGF 165 binds to VEGF 165 R via the amino acids encoded by exon 7, and (iii) GST-Ex 7ϩ8 binds to VEGF 165 R but not to KDR ( Fig. 1 and Fig. 3), these results suggested that by interfering directly with the binding of 125 I-VEGF 165 to VEGF 165 R, GST-Ex 7ϩ8 also inhibits indirectly the binding of 125 I-VEGF 165 to KDR/Flk-1.
Localization of the Core Inhibitory Region within the Exon 7-encoded Domain-The GST-Ex 7 fusion protein encompasses the entire 44 amino acid exon 7-encoded domain. To determine whether a core inhibitory region exists, deletions were made at the N and C termini of exon 7, and the effects on 125 I-VEGF 165 binding to HUVEC were measured (Fig. 5). In these experiments a fusion protein containing the exon 7-encoded domain plus the cysteine residue at position 1 of exon 8 was used as the parental construct. The cysteine residue of exon 8 was included to keep the number of cysteine residues in the VEGF portion of the fusion protein even. The GST-Ex 7 fusion protein inhibited 125 I-VEGF 165 binding to HUVEC by 80% at 2 g/ml fusion protein (Fig. 5). Inhibition of 125 I-VEGF 165 binding to HUVEC and 231 cells was comparable to that of GST-Ex 7ϩ8 (data not shown). Deletion of the first 10 (GST-Ex 7d-(1-10)) or 21 (GST-Ex 7d-(1-21)) N-terminal amino acids did not reduce the inhibitory activity of the fusion proteins. Actually, 1 g/ml of GST-Ex 7d-(1-21) had a greater inhibition activity than the same concentration of GST-Ex 7 suggesting that there may be a region within exon 7 amino acids 1-21 that interferes with the inhibitory activity. On the other hand, deletion of the cysteine residue at position 22 in exon 7 (GST-Ex 7 d-(1-22)) resulted in a complete loss of inhibitory activity. Deletion of the 15 C-terminal amino acids (GST-Ex 7 d- (30 -44)) also resulted in a complete loss of inhibitory activity (Fig. 5). These results indicated that the inhibitory core is found within amino acids 22-44 of exon 7. Moreover, it seems that the cysteine residue at position 22 in exon 7, which is Cys 137 in VEGF, is crucial for maintaining a specific structure required for the inhibition.
GST-Ex 7ϩ8 Inhibits VEGF 165 -induced Proliferation of HUVEC-The inhibition of VEGF 165 binding to KDR/Flk-1 by the GST-Ex 7ϩ8 fusion protein as shown in Fig. 4 suggested that it might also be an inhibitor of VEGF 165 mitogenicity since KDR/Flk-1 mediates VEGF mitogenic activity (32). Addition of 1-5 ng/ml VEGF 165 to HUVEC resulted in a 5.5-fold increase in the proliferation rate, peaking at 2.5 ng/ml (Fig. 6). When 15 g/ml GST-Ex 7ϩ8 was added in addition to VEGF 165 , HUVEC proliferation was reduced by about 60%. GST protein prepared in a similar way did not inhibit HUVEC proliferation even at 25 g/ml indicating that the inhibitory effect was due solely to the presence of the exon 7ϩ8-encoded domain within the fusion protein. It was concluded that exon 7ϩ8 peptide-mediated inhibition of VEGF 165 binding to VEGF receptors on HUVEC correlates with the inhibition of HUVEC proliferation. 1 and 2) and MDA-MB-231 cells (lanes 3 and 4) in 6-cm dishes. The binding was carried out in the presence (lanes 2 and 4) or the absence (lanes 1 and 3) of 15 g/ml GST-Ex 7ϩ8. Heparin (1 g/ml) was added to each dish. At the end of a 2-h incubation, 125 I-VEGF 165 was chemically crosslinked to the cell surface. The cells were lysed, and proteins were resolved by 6% SDS-PAGE. The gel was dried and exposed to x-ray film.
FIG. 5. Localization of a core inhibitory region within exon 7.
GST-Ex 7 fusion proteins containing full-length exon 7-encoded domain or truncations at the N-terminal and C-terminal ends were prepared as described under "Experimental Procedures." A, 125 I-VEGF 165 (5 ng/ml) was bound to subconfluent HUVEC cultures, as described in Fig. 3, in the presence of increasing concentrations of the GST fusion proteins. At the end of a 2-h incubation, the cells were washed and lysed, and the cell-associated radioactivity was determined with a ␥ counter. The counts obtained are expressed as percentage of the counts obtained in the presence of PBS without fusion protein. B, the amino acid sequences of VEGF exon 7 derivatives. These derivatives were prepared to contain the first cysteine residue of exon 8 at their C termini to keep an even number of cysteine residues.
FIG. 6. GST-Ex 7؉8 fusion protein inhibits VEGF 165 -stimulated HUVEC proliferation. HUVEC were cultured in 96-well dishes (5,000 cell/well) as in Fig. 2. Increasing concentrations of VEGF 165 (open circles), together with 15 g/ml GST-Ex 7ϩ8 (closed circles) or 25 g/ml GST (squares), were added to the medium, and the cells were incubated for 4 more days. DNA synthesis was measured in HUVEC as described in Fig. 2. The results represent the average counts of three wells, and the standard deviations were determined. duced mitogenicity (Fig. 7). GST-Ex 7ϩ8, at 15 g/ml, also inhibited VEGF 121 -mediated HUVEC proliferation, by about 2-fold. This was an unexpected result considering that VEGF 121 does not contain exon 7. To understand better the nature of the VEGF 121 inhibition, the effect of GST-Ex 7ϩ8 on the binding of 125 I-VEGF 121 to VEGF receptors was analyzed by cross-linking studies. Cross-linking of 125 I-VEGF 121 to HUVEC resulted in the formation of 240-kDa labeled complexes (Fig. 8, lane 1), which have been shown to contain VEGF 121 and KDR/Flk-1 (35,39). Formation of these complexes was significantly inhibited by GST-Ex 7ϩ8 at 15 g/ml (Fig. 8, lane 2). It was concluded that GST-Ex 7ϩ8 inhibits VEGF 121 -induced mitogenicity possibly by inhibiting its binding to KDR/Flk-1.
DISCUSSION
The most abundant of the VEGF isoforms are VEGF 165 and VEGF 121 . An important question in terms of understanding VEGF biology is whether these isoforms differ in their biochemical and biological properties. To date, it has been demonstrated that VEGF 165 , but not VEGF 121 , binds to cell-surface HSPG (23)(24)(25) and that VEGF 165 is a more potent EC mitogen than is VEGF 121 (36) (Fig. 2). In addition, we recently characterized a novel 130-kDa VEGF receptor found on the surface of HUVEC and tumor cells that is specific in that it binds VEGF 165 but not VEGF 121 (35). VEGF 165 binds to this receptor, termed VEGF 165 R, via the 44 amino acids encoded by exon 7, the exon which is present in VEGF 165 but not VEGF 121 . In contrast KDR/Flk-1 and Flt-1 bind both VEGF 165 and VEGF 121 and do so via the VEGF exons 4 and 3, respectively (40). Our goal in the present study was to determine whether exon 7 modulated VEGF 165 activity, in particular mitogenicity for HUVEC, and by what mechanism. To do so, we developed a strategy of inhibiting the binding of VEGF 165 to VEGF 165 R using a GST fusion protein containing the exon 7-encoded domain and examining any subsequent effects on HUVEC proliferation. Cross-linking experiments demonstrated, as expected, that the exon 7 fusion protein could bind to VEGF 165 R but not to KDR/Flk-1. The exon 7 fusion protein was found to be a potent inhibitor of 125 I-VEGF 165 binding to 231 cells which express VEGF 165 R alone, by 98%, and to HUVEC which express both KDR/Flk-1 and VEGF 165 R, by 85-95%. It did not, however, inhibit at all the binding of 125 I-VEGF 165 to PAE-KDR cells which express KDR/Flk-1 but not VEGF 165 R. GST protein alone did not inhibit binding to any of the cell types demonstrating that the inhibition was due solely to the exon 7 portion of the fusion protein. Cross-linking analysis, which demonstrated the formation of specific 125 I-VEGF 165 ⅐receptor complexes, confirmed that GST-Ex 7ϩ8 markedly inhibited the binding of 125 I-VEGF 165 to VEGF 165 R on HUVEC and 231 cells. Taken together, these results indicate that the exon 7 fusion protein interacts directly with VEGF 165 R and can act as a competitive inhibitor of binding of 125 I-VEGF 165 to this receptor.
The GST-Ex 7ϩ8 fusion protein inhibited VEGF 165 -induced proliferation of HUVEC by about 60%, to a level equivalent to that induced by VEGF 121 suggesting that activation of the KDR/Flk-1 tyrosine kinase receptor was somehow being adversely affected. Indeed, cross-linking analysis showed that the fusion protein not only inhibited cross-linking of 125 I-VEGF 165 to VEGF 165 R but to KDR/Flk-1 as well. This result was unexpected since our cross-linking studies show that the exon 7 fusion protein does not bind directly to KDR/Flk-1 consistent with the previous demonstration that VEGF 165 interacts with KDR/Flk-1 via its exon 4-encoded domain (40). Thus it appears that the binding of 125 I-VEGF 165 to VEGF 165 R via the exon 7-encoded domain modulates indirectly the interaction of the growth factor with KDR/Flk-1. A possible mechanism for this inhibitory effect of GST-Ex 7ϩ8 on HUVEC proliferation is that KDR/Flk-1 and VEGF 165 R are co-localized in close proximity on the cell surface. In this model, a VEGF 165 dimer interacts simultaneously with KDR/Flk-1 via the exon 4 domain and with VEGF 165 R via the exon 7 domain, generating a three-component complex. The GST-Ex 7ϩ8 fusion protein by competing directly with the binding of VEGF 165 to VEGF 165 R impairs indirectly the ability of VEGF 165 to bind to the signaling receptor, KDR/Flk-1. Thus, an efficient binding of VEGF 165 to KDR/Flk-1 might be dependent in part on successful interaction with VEGF 165 R. An alternative possibility is that the exon 7-encoded domain contains a heparin-binding domain (35) and that an excess of GST-Ex 7ϩ8 prevents VEGF 165 from binding to cell-surface HSPGs that are required for efficient binding of VEGF 165 to its receptors (29).
Surprisingly, GST-Ex 7ϩ8 also inhibited the mitogenic activity of VEGF 121 for HUVEC, by about 50%, even though VEGF 121 does not bind to VEGF 165 R (35). A possible explanation is that VEGF 165 R and KDR/Flk-1 are in proximity on the cell surface and that excess GST-Ex 7ϩ8 bound to VEGF 165 R g/ml GST-Ex 7ϩ8. Heparin (1 g/ml) was added to each dish. At the end of a 2-h incubation, 125 I-VEGF 121 was chemically cross-linked to the cell surface. The cells were lysed, and proteins were resolved by 6% SDS-PAGE. The gel was dried and exposed to x-ray film. sterically inhibits access of VEGF 121 to KDR/Flk-1. Cross-linking analysis did indeed show diminished binding of 125 I-VEGF 121 to KDR/Flk-1 in the presence of GST-Ex 7ϩ8 which does not bind directly to KDR/Flk-1, suggesting an indirect effect of the fusion protein on the binding of VEGF 121 to KDR/Flk-1.
GST-Ex 7ϩ8 also inhibits VEGF 165 binding to 231 breast cancer cells, which express VEGF 165 R and not KDR/Flk-1. However, VEGF is not mitogenic for these cells and at present we do not know the consequence of inhibiting VEGF 165 binding to these tumor cells.
The coordinate binding of VEGF 165 to a higher and to a lower affinity receptor (KDR/Flk-1 and VEGF 165 R, respectively) on HUVEC (35) and the inhibitory effects of GST-Ex 7ϩ8 fusion protein on the binding of VEGF 165 to these two receptors suggest that there is a dual receptor system at work in mediating VEGF 165 activity. Several other growth factors have been shown to bind to high and low affinity receptors. Transforming growth factor- generates a complex with three receptors; two of them, receptors I and II, are the signaling receptors, whereas transforming growth factor- receptor III/betaglycan is a lower affinity accessory binding molecule (41). The low affinity receptor for the nerve growth factor family, p75, is part of a complex with the signaling TRK receptors (42). A different type of dual receptor recognition is the binding of bFGF to cell-surface HSPGs and to its signaling receptors (43,44). It has been suggested that the binding of bFGF to its low affinity receptors (HSPGs) may induce conformational changes in bFGF so that the HSPG-bound bFGF could be efficiently presented to its high affinity, signaling receptors (43,44). Thus, the binding of VEGF 165 to both VEGF 165 R and KDR/Flk-1 appears to be part of a general mechanism wherein two different types of receptors are used to modulate growth factor activity.
Receptor binding studies were used to identify an inhibitory core within the 44 amino acids encoded by exon 7. Deletions were made in both the N-terminal and C-terminal regions of exon 7, and the inhibitory activity was localized to the 23amino acid C-terminal portion of exon 7 (amino acids . Of these 23 amino acids, 5 are cysteine residues. The high proportion of cysteine residues suggests that this domain has a defined three-dimensional structure required for efficient binding to VEGF 165 R. The cysteine residue at position 22 of the exon 7 domain is critical for inhibitory activity, probably for maintenance of a necessary three-dimensional structure. A study that examined the role of cysteine residues at different positions in VEGF 165 showed that a substitution of Cys 146 , which lies within the core inhibitory domain of exon 7 (at position 31 in exon 7), by a serine residue resulted in a 60% reduction in VEGF 165 permeability activity and a total loss of EC mitogenicity (45). The Cys 146 mutation had no effect on the dimerization of VEGF (45). Thus, it appears that this cysteine residue is not involved in the formation of interdisulfide bonds between two VEGF monomers but might rather involve intradisulfide bonding within the monomer. These results support our hypothesis that a three-dimensional structure stabilized by cysteine residues exists in the C-terminal half of exon 7 that contributes to VEGF 165 biological activity, such as interaction with VEGF 165 R. Interestingly, a fusion protein corresponding to a deletion of the N-terminal 21 amino acid residues encoded by exon 7 was a more potent inhibitor than the intact exon 7-encoded peptide. It may be that the N-terminal portion interferes in part with the interaction of the C-terminal portion with VEGF 165 R and therefore a deletion of the N-terminal portion results in enhanced binding to VEGF 165 R and yields a better competitor of VEGF 165 .
Since the identification of VEGF as a major angiogenesis factor and contributor to tumor pathology, numerous attempts had been made to design specific VEGF antagonists. These antagonists include anti-VEGF antibodies (19) and soluble KDR/Flk-1 and Flt-1 ectodomains (46 -48). We now add to this group the peptide encoded by exon 7 of VEGF and possibly a smaller core inhibitory peptide. Since the exon 7-encoded peptide inhibits both VEGF 165 -and VEGF 121 -induced mitogenicity for HUVEC, it and its derivatives may be useful as general VEGF inhibitors. The VEGF exon 7-encoded domain is an example of a portion of an EC mitogen being an EC inhibitor. Previously, it has been shown that fragments of SPARC (secreted protein, acidic and rich in cysteine) inhibit EC proliferation while the intact SPARC maintains angiogenic activity (49). Several other EC inhibitors are fragments of larger proteins, which in themselves are devoid of inhibitory activity. These include the 16-kDa fragment of prolactin (50), fragments of laminin (51), plasmin-cleaved fragments of fibronectin (52), angiostatin which is a fragment of plasminogen (53), and endostatin which is a fragment of collagen XVIII (54). Thus, it seems that there are numerous examples of EC inhibitors being generated from larger proteins. Our identification of the NEGF exon 7-encoded domain as an EC antagonist is based on the analysis of VEGF and VEGF receptor structure-function relationships. In the future, further analysis of the exon 7 domain might be useful for the design of small pharmacological peptides that would serve as VEGF antagonists in angiogenesis-related diseases. | 7,243.6 | 1997-12-12T00:00:00.000 | [
"Biology",
"Medicine"
] |