• 《中国科学引文数据库(CSCD)》来源期刊
  • 中国科技期刊引证报告(核心版)期刊
  • 《中文核心期刊要目总览》核心期刊
  • RCCSE中国核心学术期刊

大豆GmGST7基因耐酸铝功能研究

胡康, 金晓雨, 张雪, 王金玉, 程艳波, 连腾祥, 年海, 马启彬

胡康, 金晓雨, 张雪, 等. 大豆GmGST7基因耐酸铝功能研究[J]. 华南农业大学学报, 2023, 44(5): 769-779. DOI: 10.7671/j.issn.1001-411X.202307063
引用本文: 胡康, 金晓雨, 张雪, 等. 大豆GmGST7基因耐酸铝功能研究[J]. 华南农业大学学报, 2023, 44(5): 769-779. DOI: 10.7671/j.issn.1001-411X.202307063
HU Kang, JIN Xiaoyu, ZHANG Xue, et al. Study on the tolerant function of soybean GmGST7 gene to acidic aluminum stress[J]. Journal of South China Agricultural University, 2023, 44(5): 769-779. DOI: 10.7671/j.issn.1001-411X.202307063
Citation: HU Kang, JIN Xiaoyu, ZHANG Xue, et al. Study on the tolerant function of soybean GmGST7 gene to acidic aluminum stress[J]. Journal of South China Agricultural University, 2023, 44(5): 769-779. DOI: 10.7671/j.issn.1001-411X.202307063

大豆GmGST7基因耐酸铝功能研究

基金项目: 2022—2023年乡村振兴战略专项省级种业振兴行动项目(2022-NPY-00-021);中乌联合实验室合作项目(SQ2018YFE010044);广东省科技创新战略项目(2020B122060062);国家自然科学基金(31771816,31971965);转基因生物新品种培育(2016ZX08004002-007);国家大豆产业技术体系岗位专家专项(CARS-04-PS11)
详细信息
    作者简介:

    胡 康,硕士研究生,主要从事大豆耐酸铝基因功能研究,E-mail: 1264936727@qq.com

    通讯作者:

    马启彬,教授,博士,主要从事大豆遗传育种研究,E-mail: maqibin@scau.edu.cn

  • 中图分类号: S565.1

Study on the tolerant function of soybean GmGST7 gene to acidic aluminum stress

  • 摘要:
    目的 

    耐酸铝基因GsMYB7过表达转化大豆品种‘华春6号’后,从转基因株系的表达谱中获得目标基因GmGST7,该基因受酸铝胁迫诱导上调,且位于GsMYB7基因下游,进一步分析其耐酸铝功能,以期提高大豆酸铝耐受能力。

    方法 

    采用生物信息学方法分析GmGST7基因的碱基序列、蛋白结构域和构建系统进化树。通过烟草叶片瞬时转化法完成亚细胞定位。通过RT-qPCR分析该基因组织表达特异性。设计0、25、50、75 和 100 µmol/L 5个AlCl3浓度梯度,研究GmGST7对酸铝胁迫的响应。在50 µmol/L AlCl3处理下,设计0、4、8、12、16、24、36、48和72 h共9个时间梯度,对GmGST7的表达模式进行分析。过表达GmGST7基因遗传转化拟南芥,鉴定阳性植株,并对转基因株系进行耐酸铝表型验证、氧化水平测定、耐酸铝标志基因及下游基因的表达分析。

    结果 

    GmGST7基因位于大豆第7号染色体,序列全长为1 128 bp。该基因含有2个外显子和1个内含子,2个外显子分别编码GST高度保守的N端和不保守的C端;GmGST7基因编码226个氨基酸,编码的蛋白为大豆GST蛋白的tau类家族成员,定位于细胞质和细胞核中;GmGST7基因在大豆根、茎、叶、花和幼荚中均有表达,且在根中的表达量最高;GmGST7基因在50 µmol/L AlCl3处理24 h时表达最高;AlCl3处理后,野生型拟南芥相对根伸长显著低于转基因株系的,野生型拟南芥氧化水平高于转基因株系的,耐酸铝标志基因和下游基因的表达量在转基因株系中较高。

    结论 

    GmGST7基因属于大豆GST tau类家族成员,在细胞核和细胞质中行使功能,呈组成型表达模式,且在大豆根中表达最高,对酸铝胁迫响应显著;GmGST7过表达通过激活酸铝胁迫标志基因及其下游基因的表达提高拟南芥的酸铝耐受能力。

    Abstract:
    Objective 

    The GmGST7 gene was obtained from the gene expression profile of the GsMYB7 overexpressed lines of soybean ‘Huachun 6’ which was tolerant to acidic aluminum stress. GmGST7 lied downstream of the GsMYB7 gene, and was up-regulated by acidic aluminum stress. Its function of acidic aluminum resistance was further investigated to enhance the tolerance to aluminum stress in soybean.

    Method 

    The bioinformatics of the GmGST7 gene was analyzed using the base sequence, protein domain and phylogenetic tree. Subcellular localization of GmGST7 protein was accomplished by transient transformation in tobacco leaves. The tissue expression specificity of the GmGST7 gene was analyzed by RT-qPCR. Five AlCl3 concentration gradients of 0, 25, 50, 75 and 100 µmol/L were designed to study the response of GmGST7 to aluminum stress. Under the treatment of 50 µmol/L AlCl3, nine time gradients of 0, 4, 8, 12, 16, 24, 36, 48 and 72 h were designed to investigate the expression patterns of GmGST7. Arabidopsis (Col-0) was transformed by overexpression of GmGST7, positive plants were identified by molecular technology. The phenotype identification of Arabidopsis tolerant to acidic aluminum stress were performed with the oxidation level determination, the expression analysis of the genes response to aluminum stress and downstream genes of GmGST7.

    Result 

    The full-length sequence of GmGST7 located on chromosome 7 of soybean was 1 128 bp. The GmGST7 gene contained two exons and one intron which encodes a highly conserved N domain and a unconserved C domain of GST, respectively. GmGST7 encoded 226 amino acids. The GmGST7 protein was a tau member of the GST family in soybean, and localized in the nuclear and cytoplasm. GmGST7 was expressed in soybean root, stem, leaf, flower and young pod, and rich in root. The GmGST7 gene was up-regulated by AlCl3 with the highest relative expression under 50 µmol/L AlCl3 for 24 h. The relative root elongation of wild type was significantly lower than that of the transgenic lines, the oxidation level was higher, and the expression levels of acidic aluminum stress response genes and downstream genes were higher.

    Conclusion 

    The GmGST7 gene is a tau member of the GST family in soybean, locates in the nucleus and cytoplasm. The GmGST7 gene holds a constitutive expression pattern, and is rich in soybean root. GmGST7 is significantly up-regulated by acidic aluminum stress. Overexpression of GmGST7 enhances the tolerance to aluminum stress in Arabidopsis by activating the expression of the marker genes response to acidic aluminum stress and its downstream genes.

  • 奶牛体尺能够反映其生长发育状况、繁殖能力、产奶能力及潜在经济价值等重要信息[1-3]。长期以来,我国奶牛体尺测量主要采用传统的接触式测量,由经验丰富的工作人员利用卡尺、皮尺、卷尺、测杖等工具测量或者通过人眼估算,存在复杂繁琐、测量效率低、易受环境和测量人员主观因素影响等问题[4]

    随着机器视觉技术的快速发展,家禽家畜的体尺测量有更多的解决方案,主要可分为基于二维图像的体尺测量和基于三维点云的体尺测量2种方式[4]。在基于二维图像的体尺测量方面,Tasdemir等[5]拍摄荷斯坦奶牛顶部和侧面RGB图像,采用直接线性变化法标定相机,利用图像分析软件基于点云计算奶牛体尺,该方法测得奶牛体高、臀高、体长、臀宽准确率分别为97.72%、98.00%、97.89%、95.25%。刘同海等[6]在饮水器正上方采集120日龄猪只的体尺图像,将猪体目标从二值图像中分割,实时提取体尺测点,该方法测得猪只体长的平均相对误差为0.92%。薛广顺等[7]利用Logitech Webcam Pro 9000摄像机采集牛体图像,采用贝叶斯决策皮肤检测法提取牛体,基于尺度不变特征变换(Scale-invariant feature transform,SIFT)算法实现特征点匹配,利用双目相机成像模型结合相机标定参数获取牛体三维点云。张晨光等[8]采用双目视觉技术进行奶牛图像采集,提取奶牛背部轮廓与侧面轮廓,通过SIFT算法,结合相机标定参数获取特征点空间坐标,最终利用欧式距离计算奶牛体高、体长和体宽。在基于三维点云的体尺测量方面,赵建敏等[9]使用双目相机采集体尺图像,利用YOLOv5检测牛体及牛头、躯干、牛蹄、牛尻、腿关节等特征部位,基于改进的Canny边缘检测算法及曲线拟合算法提取牛体局部边缘轮廓,通过U弦长得到体尺测量特征点,最终完成体高、体长、体斜长等体尺测量,该方法平均体尺测量误差为2.4%。Huang等[10]利用IFM O3D303相机获取秦川牛侧面点云数据,采用KD-Network训练牛体轮廓提取模型,通过快速点特征直方图获得特征表面中心完成体尺测量,该方法最大测量误差为2.0%。初梦苑等[11]使用Kinect DK深度相机拍摄奶牛侧面与俯视图像,基于视频错帧补全奶牛侧面点云,采用迭代最近点算法配准奶牛侧面点云与俯视点云,基于几何特征测量奶牛体直长、肩宽、臀宽等7种体尺参数。Yang等[12]使用智能手机环绕奶牛拍摄图像,通过运动结构法构造三维点云模型,根据形态特征自动标记体尺测量点,该方法体尺测量平均相对误差小于4.67%。基于三维点云的体尺测量方法提供了精确的测量,但计算量大、测量过程复杂、测量成本高,并且需要特定的环境条件,在规模化养殖场实现其快速准确体尺测量尚需进一步验证。相比之下,使用单个深度相机进行体尺测量具有计算负担低、测量设备部署简单和设备成本低等优点,更适合家禽家畜体尺参数的快速测量。

    测量关键点的快速准确检测是奶牛体尺快速准确测量的前提。Yin等[13]采用对猪只身体部位平均分割的方法定位测量关键点,并将点云姿态标准化,实现了体尺的自动测量。Hu等[14]通过PointNet++对不同姿势的猪只点云进行分割,根据几何特征定位猪体测量关键点,该方法最大体尺测量误差为5.26%。陆明洲等[15]利用图像处理技术获取山羊体尺关键点,测量最大相对误差为5.5%。本研究基于YOLOv8n-Pose快速准确获取测量关键点,利用双目立体视觉模型获取深度信息,完成奶牛体尺的快速准确测量。

    奶牛双目图像数据采集自陕西省咸阳市杨陵区官村奶牛养殖场与杨凌科元克隆股份有限公司奶牛养殖场,采集时间为2022年11月至2023年4月,包括奶牛侧面双目图像及视频数据,采用ZED 2i相机采集奶牛侧面图像。图像数据的图像分辨率为2 208像素×1 242像素,用于构建目标检测及关键点检测数据集;双目视频数据的视频分辨率为4 416像素×1 242像素,帧率为15 帧/s,用于最终体尺测量。图1为数据采集及数据集构建示意图,图1a为数据采集平台示意图,ZED 2i双目立体相机固定于相机支架上,距离地面约1.1 m,将相机支架置于奶牛侧面2~3 m处,拍摄奶牛侧面图像或采集视频数据,在采集过程中避免阳光直射相机,减少阳光直射对立体匹配的影响;图1b为通过采集平台采集单幅奶牛侧面图像,共采集1 895幅图像,按8∶2划分训练集与验证集,构建奶牛目标检测及关键点检测数据集;图1c为视频数据采集及体尺测量数据集构建示意图,将图1a所示的采集平台置于奶牛挤奶通道外2.5 m处采集奶牛在挤奶通道的侧面视频,并将采集的视频进行分帧,以2帧/s构建体尺测量数据集。

    图  1  数据采集与数据集构建示意图
    Figure  1.  Data acquisition and data set construction diagram

    本研究涉及奶牛体斜长、体高、臀高和尻长4项体尺测量。图2为奶牛体尺测量标准的示意图,奶牛体斜长近似为坐骨端(C)到肩端(D)的直线距离,体高为鬐甲部最高点(A)到前蹄与地面交点(E1)的直线距离,臀高为臀部最高点(B)到后蹄与地面交点(E2)的直线距离,尻长为坐骨端(C)到尻尖(F)的直线距离。在构建的目标检测及关键点检测数据集中标注鬐甲部、尻尖、坐骨端、肩端、蹄部等8个测量关键点。

    图  2  奶牛直线体尺测量标准
    A:鬐甲部最高点,B:臀部最高点,C:坐骨端,D:肩端,E1、E2:前、后蹄与地面的交点,F:尻尖;G:眼睛
    Figure  2.  Standard for linear body size measurement of dairy cow
    A: The highest point of wither; B: The highest point of hip; C: Ischial tuberosity; D: Point of shoulder; E1, E2: Contact points of the front and rear hooves with the ground; F: Hip; G: Eye

    奶牛体尺真实数据的准确获取是试验测试的前提,但由于手动测量体尺依赖于人工经验,存在较大的偶然误差与粗大误差,Hu等[14]采用PointNet++分割奶牛点云后获取奶牛体尺,与手动测量相比,体长、体高的平均偏差仅为2.34%,说明点云数据测量与手动测量相差少。Tran等[16]和Deris等[17]证实了ZED相机在点云三维重建与测量方面具有高精度的优势。因此,本研究采用ZED 2i相机采集奶牛侧面点云数据并依据手动测量的方法获取体尺的真实值。

    小孔成像模型是常用的相机模型,双目相机基于小孔成像模型,利用2个相机从不同角度同时拍摄,通过标定参数进行畸变矫正、极限矫正,根据拍摄点在左右图像中成像位置的差异,确定点在空间中的位置。根据文献[18],设计双目相机成像模型(图3),OlOr分别是双目立体相机左右相机的光心,olor分别是左右相机光轴与成像平面的交点,假设两成像平面宽度为W,成像平面与相机之间的距离为焦距fP是三维空间中的一个点,plpr是点P在左右相机成像平面上的像点,xlxrplprx坐标,zP到左右相机所在直线的距离,根据相似三角形可得公式(1):

    图  3  双目相机成像模型
    Ol:左相机光心;Or:右相机光心;ol:左相机光轴与成像平面交点;or:右相机光轴与成像平面交点;W:成像平面宽度;f:相机焦距;P:三维空间中某点;plpr为点P在左右相机成像平面上的像点;xlxrplprx坐标;zP到左右相机的距离;b:左右相机基线距离
    Figure  3.  Binocular camera imaging model
    Ol: Optical center of the left camera; Or: Optical center of the right camera; ol: Intersection point of the left camera’s optical axis with the imaging plane; or: Intersection point of the right camera’s optical axis with the imaging plane; W: Width of the imaging plane; f: Camera focal length; P: A point in three-dimensional space; pl and pr: Image points of P on the imaging planes of the left and right cameras; xl and xr: x-coordinates of pl and pr; z: Distance from P to the cameras; b: Baseline distance between the left and right cameras
    $$ \dfrac{{z - f}}{z} = \dfrac{{b - \left( {{x_{\mathrm{l}}} - \dfrac{1}{2}W} \right) - \left( {\dfrac{1}{2}W - {x_{\mathrm{r}}}} \right)}}{b}, $$ (1)

    式中,b为左右相机的基线距离。

    xlxr为视差(d),根据公式(1),则点P的深度为:

    $$ z = {{bf} \mathord{\left/ {\vphantom {{bf} d}} \right. } d}。 $$ (2)

    根据视差值(d),结合双目立体相机标定参数基线(b)和焦距(f)推导每个像素点在三维空间中的深度,进而推导出三维空间坐标。

    立体匹配是从一对图像中推断每个像素的深度信息。传统的立体匹配算法是将立体匹配问题转化为在左右两幅图像中寻找最相似的像素点的过程。相比传统立体匹配算法,基于深度学习的立体匹配方法具有更高的精度和更好的鲁棒性,能够处理复杂的场景[19]。因此,本研究采用CREStereo算法[20]对奶牛双目图像进行立体匹配获取视差值。图4为CREStereo算法立体匹配推理过程,推理网络采用堆叠级联架构,在保证鲁棒性的基础下,保留高分辨率输入的细节,利用循环更新模块(Recurrent update module,RUM)处理实际应用中的非理想矫正情况,尽可能地缓解下采样过程中区域的特征退化。RUM采用循环架构,逐步更新和细化深度图,通过多次迭代整合上下文信息,提高视差计算的整体精度。

    图  4  CREStereo立体匹配
    Figure  4.  CREStereo stereo matching

    本文基于YOLOv8n-Pose估计网络、SimAM注意力机制[21]及CoordConv卷积[22]改进网络,提出适用于奶牛关键点检测的姿态估计模型。其中YOLOv8n-Pose网络是基于YOLOv8及YOLO-Pose估计网络而提出的多任务联合学习网络,能够同时实现目标分类及姿态估计任务。图5为改进的YOLOv8n-Pose估计网络。

    图  5  改进的YOLOv8n-Pose估计网络
    Figure  5.  Improved YOLOv8n-Pose estimation network

    SimAM注意力模块是Yang等[21]基于神经科学理论,受人脑注意力机制的启发,提出的一种3D注意力模块,SimAM注意力机制通过构建一个优化能量函数来挖掘神经元的重要性以计算注意力权重,公式(3)为能量函数(et):

    $$\begin{split} {e_t}\left( {{w_t},{b_t},y,{x_i}} \right) = & \dfrac{1}{{M - 1}} \displaystyle\sum \limits_{i = 1}^{M - 1} {\left[ { - 1 - \left( {{w_t}{x_i} + {b_t}} \right)} \right]^2} +\\ & {\left[ { - 1 - \left( {{w_t}t + {b_t}} \right)} \right]^2} + \lambda {w_t}^2\text{,} \end{split} $$ (3)

    式中,t为输入特征的目标神经元;λ为正则化系数;i为空间维度索引号;xi为其他神经元;M为通道上所有神经元的个数;y为标签值,用于确定该神经元是否为重要神经元;wt为权重;bt为偏置。根据公式(3),推导最小能量函数如公式(4):

    $$ e_t^* = \dfrac{{4\left( { {{\hat \sigma ^2}} + \lambda } \right)}}{{{{\left( {t - \hat u} \right)}^2} + 2{{\hat \sigma }^2} + 2\lambda }}, $$ (4)

    式中,û为输入特征t的均值,${\hat \sigma ^2}$为输入特征t的方差。

    由公式(4)可知,能量越低,特征相关的神经元(t)与周围神经元区别越大,重要程度越高,则神经元重要性为1/et*

    本研究增加SimAM,位置如图5 Backbone部分所示,SimAM注意力模块嵌入在C2f与SPPF之间,以观察整个Backbone中的重要神经元,增强重要特征,抑制其他神经元,从而聚焦于更有利于奶牛关键点检测的特征表示。

    CoordConv卷积是由Liu等[22]开发的卷积方法,旨在提高神经网络对坐标信息的感知能力。图6为CoordConv结构,在传统的卷积神经网络中,通过1个拼接操作在通道维度上引入2个额外的通道,这2个通道分别包含x坐标(j)和y坐标(i)的信息。CoordConv显著提升了模型对空间布局的理解能力,使得网络能够更有效地处理具有明确空间关系的任务。在本研究中,参考PP-YOLO[23]网络,如图5 Head部分所示,在YOLOv8网络的Head部分引入2层卷积核大小为1×1的CoordConv卷积层,旨在增强网络对奶牛空间坐标的感知,从而提升目标检测及关键点检测的性能。

    图  6  CoordConv结构
    h:输入特征的高度;w:输入特征的宽度;c:输入特征图通道数;i:特征的x坐标信息;j:特征的y坐标信息;h′:输出特征的高度;w′:输出特征的宽度;c′:输出特征图通道数
    Figure  6.  CoordConv structure
    h: Height of input features; w: Width of input features; c: Number of channels in input feature map; i: x-coordinate information of features; j: y-coordinate information of features; h′: Height of output features; w′: Width of output features; c′: Number of channels in output feature map

    本研究中,奶牛直线体尺自动测量算法分为以下5个步骤:

    1)利用双目相机拍摄奶牛侧面图像,并进行立体校正;

    2)采用CREStereo算法对奶牛双目图像立体匹配,并计算左目图像深度信息;

    3)采用改进YOLOv8n-Pose检测奶牛左目图像中的关键点,并分别计算每头奶牛目标检测及关键点检测置信度,优先选择关键点检测置信度高的奶牛作为最优体尺测量目标;

    4)结合双目相机参数与左目图像深度信息,将奶牛关键点映射至三维空间中;

    5)完成奶牛体斜长、体高、臀高和尻长指标的测量。

    为计算奶牛体尺,需将关键点像素坐标映射到相机坐标系或世界坐标系下,采用相机坐标系作为三维空间坐标系,以减少因相机标定不准确及坐标变换对精度的影响。假设由步骤3)改进的YOLOv8n-Pose模型进行奶牛关键点检测后,奶牛鬐甲部最高点像素坐标为(uA, vA),臀部为(uB, vB),坐骨端为(uC, vC),肩端为(uD, vD),奶牛前、后蹄与地面交点为(uE1, vE1)和(uE2, vE2),尻尖为(uF, vF)。根据公式(2)得到对应的相机坐标为公式(5):

    $$ \left\{ {\begin{array}{*{20}{c}} {{X'_{\mathrm{C}}} = \left( {u - {c_x}} \right) \times {Z'_{\mathrm{C}}}/{f_x}} \\ {{Y'_{\mathrm{C}}} = \left( {v - {c_y}} \right) \times {Z'_{\mathrm{C}}}/{f_y}} \end{array}} \right., $$ (5)

    式中,(u, v)为图像坐标系的像素坐标,(${X'_{\mathrm{C}}} $, ${Y'_{\mathrm{C}}} $, ${Z'_{\mathrm{C}}} $)为像素坐标对应相机坐标系的坐标,fx为相机在水平方向的焦距,fy为相机在垂直方向的焦距。

    关键点像素坐标经公式(5),可映射到三维坐标系中,根据欧氏距离计算奶牛体斜长如公式(6):

    $$ {l}_{体斜长}=\sqrt{{\left({X}_{\rm{C}}-{X}_{\rm{D}}\right)}^{2}+{\left({Y}_{\rm{C}}-{Y}_{\rm{D}}\right)}^{2}+{\left({Z}_{\rm{C}}-{Z}_{\rm{D}}\right)}^{2}} \text{,} $$ (6)

    式中,(XC, YC, ZC)为坐骨端坐标,(XD, YD, ZD)为肩端坐标。

    奶牛体高计算方法为鬐甲部最高点到地面距离,具体计算方式如公式(7):

    $$ {h}_{体高}=\sqrt{{\left({X}_{\rm{A}}-{X}_{{\rm{E}}1}\right)}^{2}+{\left({Y}_{\rm{A}}-{Y}_{{\rm{E}}1}\right)}^{2}+{\left({Z}_{\rm{A}}-{Z}_{{\rm{E}}1}\right)}^{2}} \text{,} $$ (7)

    式中,(XA, YA, ZA)为鬐甲部最高点坐标,(XE1, YE1, ZE1)为蹄部与地面交点坐标。

    臀高为奶牛尻尖到地面的垂直距离,具体计算方式如公式(8):

    $$ {h}_{臀高}=\sqrt{{\left({X}_{\rm{B}}-{X}_{{\rm{E}}2}\right)}^{2}+{\left({Y}_{\rm{B}}-{Y}_{{\rm{E}}2}\right)}^{2}+{\left({Z}_{\rm{B}}-{Z}_{{\rm{E}}2}\right)}^{2}} \text{,} $$ (8)

    式中,(XB, YB, ZB)为臀部最高点坐标,(XE2, YE2, ZE2)为蹄部与地面交点坐标。

    尻长为坐骨端与尻尖的直线距离,具体计算方式如公式(9):

    $$ {l}_{尻长}=\sqrt{{\left({X}_{\rm{C}}-{X}_{\rm{F}}\right)}^{2}+{\left({Y}_{\rm{C}}-{Y}_{\rm{F}}\right)}^{2}+{\left({Z}_{\rm{C}}-{Z}_{\rm{F}}\right)}^{2}} \text{,} $$ (9)

    式中,(XF, YF, ZF)为尻尖坐标。

    为探究2~3 m不同距离的双目立体匹配结果,选用5头奶牛在距离2~3 m处进行立体匹配,图7为不同距离下立体匹配结果,表1为不同距离的体尺手动点云测量与人工测量的相对误差。可见,在2~3 m深度范围内CREStereo方法能正确地从ZED 2i拍摄的奶牛双目图像中估算深度信息,满足奶牛体尺测量要求,但在实际测量中为避免距离过近导致的信息缺失和距离过远造成的图像质量下降,应保证奶牛位于在双目图像中央完整且占据图像主体的距离。

    图  7  不同距离立体匹配结果
    Figure  7.  Results of stereo matching at different distances
    表  1  不同距离下手动点云测量与人工测量奶牛体尺的相对误差
    Table  1.  Relative errors of manual point clouds measurement compared to manual measurement of dairy cows body size at different distances %
    距离/m
    Distance
    体高
    Body height
    体斜长
    Body length
    臀高
    Hip height
    尻长
    Rump length
    2.00 3.01 2.87 2.96 5.00
    2.25 2.85 2.81 2.98 4.52
    2.50 2.77 2.74 2.67 3.26
    2.75 2.62 2.62 2.55 2.73
    3.00 2.81 2.75 2.67 2.66
    下载: 导出CSV 
    | 显示表格

    本文提出的改进的YOLOv8n-Pose模型在GPU计算服务器训练的硬件配置为Intel(R) Xeon(R) Platinum 8255C CPU @ 2.50 GHz 12核、RTX 3090 24 GB、内存为43 GB;软件环境为Ubuntu18.04、Python3.8、PyTorch1.8.1、Cuda11.1;网络训练初始学习率为0.001,周期学习率为0.02,采用Adam优化器。

    为了验证SimAM注意力机制和CoordConv卷积在奶牛关键点检测上的作用,以YOLOv8n-Pose模型为基础,设计消融试验验证改进效果,图8为消融试验结果。在YOLOv8n-Pose中分别添加SimAM机制和CoordConv后,重叠度(Intersection over union,IoU)阈值为50%的平均精确度均值(mAP50)和IoU阈值为50%~95%的平均精确度均值(mAP50~95)均有所提高,最终网络较原网络目标检测mAP50~95提高了1.1个百分点(图8a),关键点检测mAP50~95提高了4.5个百分点(图8b)。

    图  8  消融试验测试YOLOv8n-Pose的改进效果
    mAP50:重叠度(Intersection over union,IoU)阈值为50%的平均精确度均值;mAP50~95:IoU为50%~95%的平均精确度均值
    Figure  8.  Improvement effect of YOLOv8n-Pose tested by ablation experiment
    mAP50: Mean average precision at 50% intersection over union (IoU) threshold; mAP50−95: Mean average precision across IoU threshold from 50% to 95%

    表2为消融试验模型参数量与计算量的对比结果,从表2可以看出,最终改进模型的参数量仅增加了0.34%,而运算量降低了1.18%。

    表  2  通过消融试验比较模型改进前后的参数量与计算量
    Table  2.  Parameters and operations comparisons before and after model improvements through ablation tests
    模型
    Model
    层数
    Layers
    参数量/M
    Parameters
    运算量/G
    Operations
    YOLOv8n-Pose1872.988.5
    YOLOv8n-Pose+SimAM1893.018.6
    YOLOv8n-Pose+CoordConv1953.018.5
    YOLOv8n-Pose+SimAM+CoordConv1972.998.4
    下载: 导出CSV 
    | 显示表格

    为检验改进后模型的关键点检测性能,利用测试集验证改进的YOLOv8n-Pose模型,图9为模型在不同复杂情景下的奶牛关键点检测结果。如图9a为奶牛侧面存在遮挡时的检测结果,图9b为不同奶牛品种的检测结果,图9c为多目标奶牛检测结果;改进后的模型在多种复杂因素影响下均可较好地完成奶牛关键点检测任务。

    图  9  改进后的YOLOv8n-Pose在不同情景下的检测结果
    Figure  9.  Detection results of the improved YOLOv8n-Pose in different scenarios

    为客观地衡量改进的YOLOv8n-Pose性能,与Hourglass[24]、HRNet[25]、SimCC[26]和RTMpose[27]等其他优秀算法比较,各算法使用相同数据集和软硬件环境,表3为对比结果。本文提出的改进YOLOv8n-Pose的精度最高,且模型参数量和浮点运算量均表现优秀。

    表  3  不同算法关键点检测结果对比
    Table  3.  Comparison of key point detection results by different algorithms
    方法
    Method
    网络骨干
    Backbone
    图像尺寸/像素
    Image size
    参数量/M
    Parameters
    计算量/G
    Operations
    平均精确度均值/%
    Mean average precision
    平均召回率/%
    Average recall
    模型大小/MB
    Model size
    Hourglass Hourglass-52 384 94.85 64.48 85.9 88.2 362.0
    SimCC HRNet-w32 384 × 288 28.54 17.33 89.3 91.4 110.0
    HRNet HRNet-w48 384 × 288 63.60 35.48 89.5 92.2 244.0
    RTMpose-m CSPNeXt-P5 256 × 192 13.25 1.90 92.5 94.7 101.0
    改进YOLOv8n-Pose
    Improved YOLOv8n-Pose
    CSPDarknet-P5 640 2.99 8.40 94.3 98.2 6.2
    下载: 导出CSV 
    | 显示表格

    在拍摄的15头奶牛的双目视频中分别选取5帧站立视频,计算体尺平均值作为测量值。图10为体尺测量最优目标判断结果,图10a为单头奶牛且环境较暗,图10b为两头奶牛,图10c多头奶牛且相互遮挡,图10d为多头奶牛且被栏杆遮挡。

    图  10  奶牛体尺测量最优目标选择
    Figure  10.  Optimal target selection for body size measurement of dairy cows

    表4为15头奶牛的体尺测量结果,测量值为本文提出的奶牛体尺测量方法的平均测量结果,真实值为通过点云数据手动平均测量结果。

    表  4  15头奶牛的体尺测量结果
    Table  4.  Body size measurement results of 15 dairy cows
    编号
    Code
    体斜长/cm Body length体高/cm Body height臀高/cm Hip height尻长/cm Rump length
    测量值
    Measured
    真实值
    True
    测量值
    Measured
    真实值
    True
    测量值
    Measured
    真实值
    True
    测量值
    Measured
    真实值
    True
    1169.56165.91147.29151.92157.31159.1945.4547.78
    2172.35170.61135.24135.14146.82141.4654.0651.19
    3167.93171.47152.99154.52157.03158.4458.8456.92
    4163.25160.63146.40145.99148.92148.1646.4649.56
    5162.10165.91140.63143.59150.46152.3550.0554.77
    6176.91171.69135.27144.74155.02148.1851.4854.12
    7170.31172.68150.45147.24146.01148.9254.2255.69
    8175.99178.13154.96159.60143.14153.8356.2358.59
    9178.34171.91146.60146.08148.84151.0957.6956.73
    10171.47175.43138.61152.57141.76154.1059.1657.35
    11174.62174.09151.92153.80152.05150.6061.9361.52
    12169.25172.57156.13153.88138.83152.7456.0453.23
    13175.21172.34151.94156.04154.99152.9051.7151.95
    14179.81184.20158.55159.31160.85158.0952.7957.39
    15168.69167.72145.67148.28148.95150.3755.2453.38
    下载: 导出CSV 
    | 显示表格

    利用箱线图分析各项体尺的相对误差,图11为体尺测量相对误差箱线图,其中体斜长相对误差最小,臀高相对误差最大。体斜长的最大相对误差为3.74%,平均相对误差为1.84%,平均绝对误差为3.2 cm;体高的最大相对误差为6.54%,平均相对误差为2.34%,平均绝对误差为3.5 cm;臀高的最大相对误差为9.11%,平均相对误差为2.98%,平均绝对误差为4.5 cm;尻长的最大相对误差为8.62%,平均相对误差为4.19%,平均绝对误差为2.3 cm。分析误差来源,主要如下:1)在奶牛关键点检测数据集中,部分奶牛轮廓不明显或标注不规范,导致奶牛关键点检测存在一定偏差;2)双目相机在实际使用中存在一定的畸变,奶牛深度信息计算的偏差导致体尺测量偏差。

    图  11  体尺测量相对误差箱线图
    IQR:四分位距;25%~75%:数据的中间50%,从数据的第1四分位数到第3四分位数的范围;1.5IQR范围:从第1四分位数减去1.5倍IQR到第3四分位数加上1.5倍IQR,超出此范围的点为异常值
    Figure  11.  Box diagram of relative error of body size measurement
    IQR: Interquartile range; 25%−75%: The middle 50% of the data, which ranges from the first quartile to the third quartile; 1.5IQR range: Extending from the first quartile minus 1.5 times IQR to the third quartile plus 1.5 times IQR, points outside this range are considered as outliers

    基于YOLOv8n-Pose模型,通过引入SimAM注意力机制和CoordConv卷积,提出了改进的YOLOv8n-Pose奶牛关键点检测模型,模型检测精度为94.30%,参数量为2.99 M,计算量为8.40 G,视频平均检测速度为55.6帧/s;相比于原网络,有着更好的精度,更小的计算量,更适合奶牛养殖场实际的部署与应用。本研究融合CREStereo立体匹配与改进的YOLOv8n-Pose,提出一种奶牛体尺测量方法,利用CREStereo对奶牛双目图像立体匹配计算视差值,利用改进的YOLOv8n-Pose检测左目图像中奶牛关键点,结合双目相机标定参数将奶牛关键点像素坐标转换至空间坐标,通过计算关键点间的欧氏距离完成奶牛体尺自动测量,体尺测量最大平均相对误差为4.19%。

  • 图  1   GmGST7蛋白结构域

    Figure  1.   Protein domain of GmGST7

    图  2   GmGST7蛋白序列对比(A)和进化树(B)分析

    红色横线代表GST-N,蓝色横线代表GST-C,红色点标注为目标基因GmGST7

    Figure  2.   Sequence comparison (A) and evolutionary tree (B) analysis of GmGST7 protein

    The red horizontal line represents GST-N, the blue horizontal line represents GST-C, and the red dot is labeled as the target gene GmGST7

    图  3   GmGST7互作蛋白预测

    Figure  3.   Interaction protein prediction of GmGST7

    图  4   GmGST7基因的组织表达模式分析

    “**”表示该组织与幼荚在P < 0.01水平差异显著(t检验)

    Figure  4.   Analysis of tissue expression pattern of the GmGST7 gene

    “**” indicates that the tissue differed from the young pod at P < 0.01 level (t test)

    图  5   GmGST7基因酸铝胁迫响应

    各图中,“**”表示处理与对照(CK)在P < 0.01水平差异显著(t检验)

    Figure  5.   Response of the GmGST7 gene to acidic aluminum stress

    “**” indicates that the treatment differed from the control (CK) at P < 0.01 level in each figure (t test)

    图  6   GmGST7基因克隆及载体构建

    A:GmGST7基因cDNA序列克隆,根据引物位置克隆大小为742 bp;B:连亚细胞定位载体后,根据载体引物位置PCR产物大小为1 416 bp;C:连接过表达载体后,根据载体引物位置PCR产物大小为870 bp

    Figure  6.   Cloning and vector construction of the GmGST7 gene

    A: Cloning of GmGST7 gene cDNA sequence, according to the location of the primer, cloning size is 742 bp; B: After connecting the subcellular localization vector, according to the location of the carrier primer, the size of the PCR product is 1 416 bp; C: After connecting the overexpressed vector, according to the location of the vector primer, the size of the PCR product was 870 bp

    图  7   GmGST7蛋白亚细胞定位分析

    Figure  7.   Subcellular localization analysis of GmGST7 protein

    图  8   GmGST7转基因拟南芥阳性苗除草剂喷洒鉴定

    阳性苗正常生长,假阳性苗则枯萎

    Figure  8.   Herbicide spraying identification of GmGST7 transgenic Arabidopsis positive seedlings

    Positive seedlings grow normally, false positive seedlings wilt

    图  9   GmGST7转基因拟南芥阳性苗DNA鉴定

    Figure  9.   DNA identification of GmGST7 transgenic Arabidopsis seedlings

    图  10   转基因拟南芥T3代株系中GmGST7基因表达量

    “**”表示株系与WT在0.01水平差异显著(t检验)

    Figure  10.   Expression of GmGST7 gene in T3 transgenic Arabidopsis thaliana

    “**” indicates that the line differed from the WT at 0.01 level (t test)

    图  11   过表达GmGST7拟南芥耐酸铝表型

    Figure  11.   Overexpressed GmGST7 Arabidopsis aluminium-resistant phenotype

    表  1   引物序列

    Table  1   Primer sequence

    引物名称
    Primer name
    引物序列(5′→3′)
    Primer sequence
    pTF101-GmGST7-F gagaacacgggggactctagaATGGCTGCTAATCAGGAAGATGTG
    pTF101-GmGST7-R cgatcggggaaattcgagctcTTTTGAAGCAGAAAGACTTTCATGG
    Super1300-GmGST7-F acgggggactcttgaccatggCTATGGCTGCTAATCAGGAAGATGTG
    Super1300-GmGST7-R aagttcttctcctttactagtTTTTGAAGCAGAAAGACTTTCATGG
    GmGST7-F GTCCTGATTCCCGGCTCAAT
    GmGST7-R AACTCACAAATGAGAGACCAGT
    RT-GmGST7-F TCAACCACCCTGTTGTCAAAC
    RT-GmGST7-R AAGACTTTCATGGCAGGCTTTGT
    下载: 导出CSV

    表  2   荧光定量引物序列

    Table  2   qPCR primer sequence

    基因
    Gene
    引物序列(5′→3′)
    Primer sequence
    AtALMT F: TCCCATGGGTAAAGACAAAG
    R: ATAGTCTGCTTTCTGCCAAA
    AtMATE F: CATTCGAATCCATCGAGATT
    R: CGAATGTTGCACTCTGTTTT
    AtALS3 F: AGCTTCGAGATGACATCAAA
    R: ACGGTTTTGCAGCTATCTAA
    AtWAK1 F: TGGCCGCTGATATTACAAAT
    R: CAGATTGGCTACTGGTTAGT
    At1G78660 F: CAGGTTTGAGTGTATCGGTG
    R: CATCTGATTCTTCTGCCCAA
    At1G78670 F: TCCTCTGAGATGTGGAGATT
    R: TAGTTGAGGTTTGGATCAGC
    At1G78680 F: AAAATGGTGGATTTTGCAGG
    R: ATAGGCTGACGTTCAAAGTT
    At4G33090 F: TGGATCAGTTCAAAGGTGAG
    R: GACTATGTCGAGATCGATGG
    下载: 导出CSV
  • [1]

    VON UEXKÜLL H R, MUTERT E. Global extent, development and economic impact of acid soils[J]. Plant and Soil, 1995, 171(1): 1-15. doi: 10.1007/BF00009558

    [2]

    KINRAIDE T B, PARKER D R. Cation amelioration of aluminum toxicity in wheat[J]. Plant Physiology, 1987, 83(3): 546-551. doi: 10.1104/pp.83.3.546

    [3]

    KINRAIDE T B. Assessing the rhizotoxicity of the aluminate ion, Al(OH)4[J]. Plant Physiology, 1990, 93(4): 1620-1625. doi: 10.1104/pp.93.4.1620

    [4]

    DELHAIZE E, RYAN P R, RANDALL P J. Aluminum tolerance in wheat (Triticum aestivum L.) II: Aluminum-stimulated excretion of malic acid from root apices)[J]. Plant Physiology, 1993, 103(3): 695-702. doi: 10.1104/pp.103.3.695

    [5]

    SIVAGURU M, BALUSKA F, VOLKMANN D, et al. Impacts of aluminum on the cytoskeleton of the maize root apex. short-term effects on the distal part of the transition zone[J]. Plant Physiology, 1999, 119(3): 1073-1082. doi: 10.1104/pp.119.3.1073

    [6]

    KOLLMEIER M, FELLE H H, HORST W J. Genotypical differences in aluminum resistance of maize are expressed in the distal part of the transition zone. is reduced basipetal auxin flow involved in inhibition of root elongation by aluminum?[J]. Plant Physiology, 2000, 122(3): 945-956. doi: 10.1104/pp.122.3.945

    [7]

    VAZQUEZ M D, POSCHENRIEDER C, et al. Change in apoplastic aluminum during the initial growth response to aluminum by roots of a tolerant maize variety[J]. Plant Physiology, 1999, 119(2): 435-444. doi: 10.1104/pp.119.2.435

    [8]

    SILVA I R, SMYTH T J, MOXLEY D F, et al. Aluminum accumulation at nuclei of cells in the root tip. Fluorescence detection using lumogallion and confocal laser scanning microscopy[J]. Plant Physiology, 2000, 123(2): 543-552. doi: 10.1104/pp.123.2.543

    [9]

    TAYLOR G J, MCDONALD-STEPHENS J L, HUNTER D B, et al. Direct measurement of aluminum uptake and distribution in single cells of Chara corallina[J]. Plant Physiology, 2000, 123(3): 987-996. doi: 10.1104/pp.123.3.987

    [10]

    FRANTZIOS G, GALATIS B, APOSTOLAKOS P. Aluminium causes variable responses in actin filament cytoskeleton of the root tip cells of Triticum turgidum[J]. Protoplasma, 2005, 225(3/4): 129-140.

    [11]

    ČIAMPOROVÁ M. Morphological and structural responses of plant roots to aluminium at organ, tissue, and cellular levels[J]. Biologia Plantarum, 2002, 45(2): 161-171. doi: 10.1023/A:1015159601881

    [12]

    JONES D L, KOCHIAN L V. Aluminum inhibition of the inositol 1, 4, 5-trisphosphate signal transduction pathway in wheat roots: A role in aluminum toxicity?[J]. The Plant Cell, 1995: 1913-1922.

    [13]

    BARCELÓ J, POSCHENRIEDER C. Fast root growth responses, root exudates, and internal detoxification as clues to the mechanisms of aluminium toxicity and resistance: A review[J]. Environmental and Experimental Botany, 2002, 48(1): 75-92. doi: 10.1016/S0098-8472(02)00013-8

    [14]

    KOCHIAN L V, PIÑEROS M A, HOEKENGA O A. The physiology, genetics and molecular biology of plant aluminum resistance and toxicity[J]. Plant and Soil, 2005, 274(1/2): 175-195. doi: 10.1007/s11104-004-1158-7

    [15]

    WANG H J. GsMYB7 encoding a R2R3-type MYB transcription factor enhances the tolerance to aluminum stress in soybean (Glycine max L.) [J]. BMC Genomics, 2022. 23(1): 529. doi: 10.1186/s12864-022-08744-w.

    [16]

    SHEEHAN D, MEADE G, FOLEY V M, et al. Structure, function and evolution of glutathione transferases: Implications for classification of non-mammalian members of an ancient enzyme superfamily[J]. The Biochemical Journal, 2001, 360(Pt1): 1-16.

    [17]

    SAADAT M, MOHABATKAR H. Polymorphisms of glutathione S-transferases M1 and T1 do not account for inter individual differences for smoking behavior[J]. Pharmacology Biochemistry and Behavior, 2004, 77(4): 793-795. doi: 10.1016/j.pbb.2004.02.003

    [18]

    SASAN M, BABAK S, HASSAN M. A new member of tau-class glutathione S-transferase from barley leaves[J]. Excli Journal, 2009, 8: 190-194.

    [19]

    DROOG F, HOOYKAAS P, VAN DER ZAAL B J. 2, 4-dichlorophenoxyacetic acid and related chlorinated compounds inhibit two auxin-regulated type-III tobacco glutathione S-transferases[J]. Plant Physiology, 1995, 107(4): 1139-1146. doi: 10.1104/pp.107.4.1139

    [20]

    ÖZTETIK E. A tale of plant glutathione S-transferases: Since 1970[J]. The Botanical Review, 2008, 74(3): 419-437. doi: 10.1007/s12229-008-9013-9

    [21]

    LIGHT G G, MAHAN J R, ROXAS V P, et al. Transgenic cotton (Gossypium hirsutum L.) seedlings expressing a tobacco glutathione S-transferase fail to provide improved stress tolerance[J]. Planta, 2005, 222(2): 346-354. doi: 10.1007/s00425-005-1531-7

    [22]

    DIXON D P, CUMMINS I, COLE D J, et al. Glutathione-mediated detoxification systems in plants[J]. Current Opinion in Plant Biology, 1998, 1(3): 258-266. doi: 10.1016/S1369-5266(98)80114-3

    [23]

    NUTRICATI E, MICELI A, BLANDO F, et al. Characterization of two Arabidopsis thaliana glutathione S-transferases[J]. Plant Cell Reports, 2006, 25(9): 997-1005. doi: 10.1007/s00299-006-0146-1

    [24]

    PANTELIDES I S, TJAMOS S E, PAPLOMATAS E J. Ethylene perception via ETR1 is required in Arabidopsis infection by Verticillium dahliae[J]. Molecular Plant Pathology, 2010, 11(2): 191-202. doi: 10.1111/j.1364-3703.2009.00592.x

    [25]

    BELA K, HORVÁTH E, GALLÉ Á, et al. Plant glutathione peroxidases: Emerging role of the antioxidant enzymes in plant development and stress responses[J]. Journal of Plant Physiology, 2015, 176: 192-201. doi: 10.1016/j.jplph.2014.12.014

    [26]

    BIELACH A, HRTYAN M, TOGNETTI V. Plants under stress: Involvement of auxin and cytokinin[J]. International Journal of Molecular Sciences, 2017, 18(7): 1427. doi: 10.3390/ijms18071427.

    [27]

    LIU J P, LUO X Y, SHAFF J, et al. A promoter-swap strategy between the AtALMT and AtMATE genes increased Arabidopsis aluminum resistance and improved carbon-use efficiency for aluminum resistance[J]. The Plant Journal, 2012, 71(2): 327-337. doi: 10.1111/j.1365-313X.2012.04994.x

    [28]

    MANGEON A, PARDAL R, MENEZES-SALGUEIRO A D, et al. AtGRP3 is implicated in root size and aluminum response pathways in Arabidopsis[J]. PLoS One, 2016, 11(3): e0150583. doi: 10.1371/journal.pone.0150583.

    [29]

    MINERVA B. Five decades with glutathione and the GSTome[J]. Journal of Biological Chemistry, 2012, 287(9): 6072-6083. doi: 10.1074/jbc.X112.342675

    [30]

    LABROU N E, PAPAGEORGIOU A C, PAVLI O, et al. Plant GSTome: Structure and functional role in xenome network and plant stress response[J]. Current Opinion in Biotechnology, 2015, 32: 186-194. doi: 10.1016/j.copbio.2014.12.024

    [31]

    JHA B, SHARMA A, MISHRA A. Expression of SbGSTU (tau class glutathione S-transferase) gene isolated from Salicornia brachiata in tobacco for salt tolerance[J]. Molecular Biology Reports, 2011, 38(7): 4823-4832. doi: 10.1007/s11033-010-0625-x

    [32]

    TIWARI V, PATEL M K, CHATURVEDI A K, et al. Functional characterization of the tau class glutathione-S-transferases gene (SbGSTU) promoter of Salicornia brachiata under salinity and osmotic stress[J]. PLoS One, 2016, 11(2): e0148494. doi: 10.1371/journal.pone.0148494.

    [33]

    XU J, ZHENG A Q, XING X J, et al. Transgenic Arabidopsis plants expressing grape glutathione S-transferase gene (VvGSTF13) show enhanced tolerance to abiotic stress[J]. Biochemistry (Moscow), 2018, 83(6): 755-765. doi: 10.1134/S0006297918060135

    [34]

    SRIVASTAVA D, VERMA G, CHAUHAN A S, et al. Rice (Oryza sativa L.) tau class glutathione S-transferase (OsGSTU30) overexpression in Arabidopsis thaliana modulates a regulatory network leading to heavy metal and drought stress tolerance[J]. Metallomics, 2019, 11(2): 375-389. doi: 10.1039/C8MT00204E

    [35]

    KAMPRANIS S C, DAMIANOVA R, ATALLAH M, et al. A novel plant glutathione S-transferase/peroxidase suppresses bax lethality in yeast[J]. Journal of Biological Chemistry, 2000, 275(38): 29207-29216. doi: 10.1074/jbc.M002359200

    [36]

    LOYALL L, UCHIDA K, BRAUN S, et al. Glutathione and a UV light-induced glutathione S-transferase are involved in signaling to chalcone synthase in cell cultures[J]. The Plant Cell, 2000, 12(10): 1939-1950.

    [37]

    MUELLER L A, GOODMAN C D, SILADY R A, et al. AN9, a petunia glutathione S-transferase required for anthocyanin sequestration, is a flavonoid-binding protein[J]. Plant Physiology, 2000, 123(4): 1561-1570. doi: 10.1104/pp.123.4.1561

    [38]

    WEISERBS K F, JACOBSON J S, BEGG M D, et al. A cross-sectional study of polycyclic aromatic hydrocarbon-DNA adducts and polymorphism of glutathione S-transferases among heavy smokers by race/ethnicity[J]. Biomarkers, 2003, 8(2): 142-155. doi: 10.1080/1354750031000086269

    [39]

    GARCERÁ A, BARRETO L, PIEDRAFITA L, et al. Saccharomyces cerevisiae cells have three Omega class glutathione S-transferases acting as 1-cys thiol transferases[J]. The Biochemical Journal, 2006, 398(2): 187-196. doi: 10.1042/BJ20060034

    [40]

    FEDERICI L, MASULLI M, GIANNI S, et al. A conserved hydrogen-bond network stabilizes the structure of Beta class glutathione S-transferases[J]. Biochemical and Biophysical Research Communications, 2009, 382(3): 525-529. doi: 10.1016/j.bbrc.2009.03.052

    [41]

    LALLEMENT P A, RORET T, TSAN P, et al. Insights into ascorbate regeneration in plants: Investigating the redox and structural properties of dehydroascorbate reductases from Populus trichocarpa[J]. Biochemical Journal, 2016, 473(6): 717-731. doi: 10.1042/BJ20151147

    [42]

    GONZALEZ D, FRAICHARD S, GRASSEIN P, et al. Characterization of a Drosophila glutathione transferase involved in isothiocyanate detoxification[J]. Insect Biochemistry and Molecular Biology, 2018, 95: 33-43. doi: 10.1016/j.ibmb.2018.03.004

    [43]

    TAJC S G, TOLBERT B S, BASAVAPPA R, et al. Direct determination of thiol pKa by isothermal titration microcalorimetry[J]. Journal of the American Chemical Society, 2004, 126(34): 10508-10509. doi: 10.1021/ja047929u

    [44]

    ATKINSON H J, BABBITT P C. Glutathione transferases are structural and functional outliers in the thioredoxin fold[J]. Biochemistry, 2009, 48(46): 11108-11116. doi: 10.1021/bi901180v

    [45]

    OHNO S. Patterns in genome evolution[J]. Current Opinion in Genetics & Development, 1993, 3(6): 911-914.

    [46]

    FORCE A, LYNCH M, PICKETT F B, et al. Preservation of duplicate genes by complementary, degenerative mutations[J]. Genetics, 1999, 151(4): 1531-1545. doi: 10.1093/genetics/151.4.1531

    [47]

    TANAKA K M, TAKAHASI K R, TAKANO-SHIMIZU T. Enhanced fixation and preservation of a newly arisen duplicate gene by masking deleterious loss-of-function mutations[J]. Genetics Research, 2009, 91(4): 267-280. doi: 10.1017/S0016672309000196

    [48] 马立功, 孟庆林, 张匀华, 等. 向日葵谷胱甘肽-S-转移酶基因的克隆及抗病功能研究[J]. 中国油料作物学报, 2015, 37(5): 635-643. doi: 10.7505/j.issn.1007-9084.2015.05.007
    [49] 韩少怀, 李佳佳, 张璟曜, 等. 大豆GmGST12基因的克隆及表达分析[J]. 大豆科学, 2015, 34(5): 782-788. doi: 10.11861/j.issn.1000-9841.2015.05.0782
    [50] 李永生, 方永丰, 李玥, 等. 玉米逆境响应基因ZmGST23克隆和表达分析[J]. 农业生物技术学报, 2016, 24(5): 667-677.
    [51] 孙兰兰, 麻荣慧, 薛飞, 等. 玉米GST31基因的克隆与表达分析[J]. 作物学报, 2023, 49(10): 2717-2726.
图(11)  /  表(2)
计量
  • 文章访问数:  109
  • HTML全文浏览量:  14
  • PDF下载量:  15
  • 被引次数: 0
出版历程
  • 收稿日期:  2023-09-10
  • 网络出版日期:  2023-11-12
  • 发布日期:  2023-09-18
  • 刊出日期:  2023-09-09

目录

/

返回文章
返回