Smart control of crop diseases: how can we best combine fungicides and plant resistance genes?

真菌病原菌三孢发酵壳针孢菌（Zymoseptoria triacetum，Zt）引起三孢壳针孢菌斑点病（Septoria triacetum blotch，STB），这是欧洲小麦最具破坏性的病害，也是全球小麦生产的最大限制之一。这种疾病在英国特别严重，因为气候条件有利。由于Zt能够迅速进化对杀菌剂的抗性并适应抗病小麦品种和环境条件，因此控制STB变得越来越困难。面对病原体臭名昭著的适应能力，没有任何单一的控制措施是持久的，因此，两种关键的控制方法-杀真菌剂和小麦抗病基因-需要以一种不仅优化短期控制效果，而且优化其长期可持续性的方式相结合。这个跨学科项目将利用大规模田间试验与新型高通量表型分析技术、生物信息学分析、最先进的机器学习和数学建模的强大组合，为实现这一目标做出重大贡献。目标1：揭示小麦定量STB抗性的新遗传基础。由于目前表型分析方法的局限性，对STB的定量抗性的遗传基础在很大程度上仍未被探索：它们使用对疾病严重程度的不够准确的视觉评估，使用有限数量的病原体菌株进行人工接种，以及在小麦生长季节进行单次测量。这一目标将通过使用我们最近开发的精确、高通量数字表型分析方法表征该领域的自然STB流行病发展来实现。选择的小麦群体是多亲本高级世代互交（MAGIC）群体，其结合了高遗传多样性和丰富的重组，代表了用于鉴定负责抗病性的新的数量性状基因座（QTL）的强大资源。新的抗病性的数字表型以前没有与小麦MAGIC群体一起使用。目标2：实现STB流行发展的准确和可靠的预测。该项目设计的预测模型将使用先进的机器学习方法来结合联合收割机三种类型的大型数据集：精确的疾病测量，小麦基因组数据和气象数据。基因组数据由MAGIC群体中分离的> 13，000个单核苷酸多态性（SNP）组成。我们将首先使用传统的QTL定位技术来确定与流行发展/抗病性相关的性状相关的SNPs。接下来，我们将使用基于惩罚线性回归的线性机器学习方法，这些方法能够结合这三种类型的数据。此外，我们将采用更多的计算密集型算法的基础上捕获非线性依赖的决策树。最强大的预测因子将被确定，以构建模型，提供足够的准确性，同时最大限度地减少成本的数据acquisition.Objective 3：优化组合使用杀菌剂和抗病基因在小麦。这将通过将目标1和2的成果纳入最先进的数学建模框架来实现。我们将整合目标1中获得的定量抗病性知识与目标2中设计的STB预测模型以及杀真菌剂剂量-反应数据集到流行病学/进化建模框架中。一个多目标优化算法将用于优化杀菌剂处理方案和抗病小麦品种在一个单一的生长季节的短期内的选择。我们将比较他们的优化结果进行了较长时间的连续生长季节的数量。

The fungal pathogen Zymoseptoria tritici (Zt) causes septoria tritici blotch (STB), the most damaging disease of wheat in Europe and one of the largest constraints on wheat production globally. The disease is especially serious in the UK because of conducive climatic conditions. It is becoming increasingly difficult to control STB, because Zt is capable of rapidly evolving resistance to fungicides and adapting to disease-resistant wheat varieties and environmental conditions. No single control measure is durable in the face of the pathogen's notorious adaptive capacity, hence the two key control methods - fungicides and disease resistance genes in wheat - need to be combined in a manner that optimizes not only control efficacy in the short term, but also their sustainability in the longer term. This interdisciplinary project will make a major contribution to this goal using a powerful combination of large-scale field experimentation with novel high-throughput phenotyping techniques, bioinformatic analyses, state-of-the-art machine learning and mathematical modelling.Objective 1: Reveal the novel genetic bases of quantitative STB resistance in wheat. Genetic basis of quantitative resistance to STB remains largely unexplored because of the limitations in the current phenotyping methods: they use insufficiently accurate visual assessments of disease severity, artificial inoculation with limited numbers of pathogen strains, and single measurements during the wheat growing season. The objective will be achieved by characterizing natural STB epidemic development in the field using precision, high-throughput digital phenotyping approaches that we recently developed. The wheat population of choice is the Multiparent Advanced Generation Inter-Cross (MAGIC) population, which combines a high genetic diversity with abundant recombination, representing a powerful resource for identifying new quantitative trait loci (QTL) responsible for disease resistance. New digital phenotyping of disease resistance has not been previously used in conjunction with wheat MAGIC populations.Objective 2: Achieve accurate and robust predictions of STB epidemic development. The predictive models devised in the project will use advanced machine learning approaches to combine large datasets of three types: precision disease measurements, wheat genome data and meteorological data. Genomic data consists of >13,000 single nucleotide polymorphisms (SNPs) segregating in MAGIC population. We will first use conventional QTL mapping techniques to identify the SNPs associated with the traits related to epidemic development/disease resistance. Next, we will use linear machine learning approaches based on penalized linear regression that are capable of combining the three types of data. Furthermore, we will employ more computationally intensive algorithms based on decision-trees capturing nonlinear dependencies. Most powerful predictors will be identified to construct models that provide sufficient accuracy, while minimizing the costs of data acquisition.Objective 3: Optimize the combined use of fungicides and disease resistance genes in wheat. This will be achieved by incorporating the outcomes of Objectives 1 and 2 into the state-of-the-art mathematical modelling framework. We will integrate the knowledge on quantitative disease resistance acquired in Objective 1 with predictive models of STB devised in Objective 2 together with fungicide dose-response datasets into an epidemiological/evolutionary modelling framework. A multi-objective optimization algorithm will be used to optimize choices of fungicide treatment programmes and disease-resistant wheat cultivars over a short term of a single growing season. We will compare them with the outcomes of optimization conducted over a longer term of a number of consecutive growing seasons.