Genomic prediction in a wild mammal

想象一下这样一个世界：科学家可以对一种动物或植物进行取样，并通过DNA分析来预测它的样子，它能活多久，它能有多少后代，以及它是否能在种群中胜过其他成员。虽然这个想法看起来很奇怪，但它已经成为一种可能，甚至对复杂生态系统中的野生种群也是如此。这项提议的目的是开发、测试和应用所谓的“基因组预测”方法，用于进化生态学。在过去的十年中，基因组学方法的显著进步，尤其是下一代测序，已经彻底改变了生物学研究的所有领域。现在，在任何生物体的基因组中，都可以在成千上万个可变位点上生成DNA图谱。许多这样的位点（被称为单核苷酸多态性，或SNPs）将位于或非常靠近导致表型变异的基因内。传统上，寻找这些基因，或数量性状位点（QTL），涉及到单独测试每个SNP，然后识别那些具有统计意义的。然而，这种方法是有问题的，因为它偏向于寻找对许多表型根本不存在的大影响的基因。如果像更常见的情况那样，有许多影响很小的基因，那么QTL就不会被检测到。在动植物育种中，通过同时考虑所有snp的表型效应，已经解决了这个问题。首先，使用已知表型的基因型样本的“训练群体”来估计每个SNP的效应大小。然后对第二个“测试”个体样本进行基因分型，基因分型用于预测表型；即进行基因组预测。这种方法为成功的现代人工选择项目奠定了基础，并将用于个性化医疗。然而，基因组预测从未应用于野生种群，尽管它有可能彻底改变进化生态遗传学。我们将在Hirta岛（苏格兰圣基尔达）的Soay羊的野生种群中测试和应用基因组预测；是世界上研究最深入的脊椎动物种群之一。自1985年以来，对村湾研究区出生的95%以上的动物进行了整个生命周期的监测，对7000多只羊的详细生活史（如出生日期、死亡日期、性别、双胞胎状态、形态测量、免疫测定、寄生虫负荷和终身健康）进行了描述。在整个发展过程中，许多特征已经被测量了无数次。此外，绵羊基因组已经测序，大多数Soay研究群体已经按照国际绵羊基因组学联盟发现的38K snp分型。使索伊羊成为测试基因组预测的理想系统的其他特点是：(1)不同的性状有很好的描述和非常不同的遗传结构。如。毛色和角型具有简单的遗传基础，而骨骼测量则具有更多的多基因（但仍然具有高度遗传性），并且（ii）连锁不平衡在基因组中延伸了很长的距离，因此芯片上的snp“标记”了大部分基因组。使用2010年之前出生的所有动物的“训练种群”，我们将估计单个snp的影响，然后使用这些估计来预测2010年之后出生的动物的表型。我们将把预测值与观测值进行比较；基因组预测首次在野生种群中进行了测试或应用。我们还将使用基因组预测来确定哪些特征对自然选择做出了进化反应。我们预测基因组预测将在我们的研究人群中实现，并且它将优于传统的基于系谱的方法来研究自然界的微观进化。

Imagine a world where a scientist could sample an animal or plant and, by DNA profiling, predict what it would look like, how long it would live, how many offspring it would have, and whether or not it would out-compete other members of its population. Although the idea seems fanciful, it has become a possibility, even for wild populations within complex ecological systems. The aim of this proposal is to develop, test and apply so called 'genomic prediction' methods for use in evolutionary ecology.In the last decade remarkable advances in genomics methods, most notably next-generation sequencing, have revolutionised all areas of biological research. It is now possible to generate DNA profiles at hundreds of thousands of variable sites across the genome, in any organism. Many of these sites (known as single nucleotide polymorphisms, or SNPs) will reside within, or very close to, genes that cause phenotypic variation. Traditionally, the search for these genes, or quantitative trait loci (QTL), has involved testing each SNP individually and then identifying those which are statistically significant. However, this approach is problematic, in that it is biased towards finding genes of large effect, which for many phenotypes simply do not exist. If, as is more common, there are many genes of small effect then QTL will remain undetected. In animal and plant breeding, the problem has been solved by considering the phenotypic effect of all SNPs simultaneously. First a 'training population' of genotyped samples with known phenotype are used to estimate effect sizes of each SNP. Then a second sample of 'test' individuals is genotyped, and the genotypes are used to predict phenotype; i.e. perform genomic prediction. This approach underpins successful modern artificial selection programmes and is set to be used in personalised medicine. However, genomic prediction has never been applied to wild populations, despite its potential to revolutionise evolutionary ecological genetics.We will test and apply genomic prediction in the feral population of Soay sheep on the island of Hirta (St Kilda, Scotland); one of the most intensively studied vertebrate populations in the world. Since 1985, over 95% of animals born in the Village Bay study area have been monitored over their entire lifetimes, such that detailed life histories (e.g. date of birth, date of death, sex, twin status, morphological measurements, immunological assays, parasite loads and lifetime fitness) are described for over 7000 sheep. Many traits have been measured numerous times across development. Furthermore, the sheep genome has been sequenced and most of the Soay study population has been typed at 38K SNPs discovered by the International Sheep Genomics Consortium. Additional features that make Soay sheep the ideal system for testing genomic prediction are: (i) different traits have well described and very different genetic architectures. eg. coat colour and horn type have a simple genetic basis while skeletal measurements are far more polygenic (but still highly heritable) and (ii) linkage disequilibrium extends for long distances in the genome, so that the SNPs on the chip 'tag' most of the genome. Using a 'training population' of all animals born until 2010 we will estimate the effects of individual SNPs, and then use these estimates to predict the phenotype of animals born after 2010. We will compare the predictions to observed values; the first time genomic prediction has been tested or applied in a wild population. We will also use genomic predictions to establish which traits have made an evolutionary response to natural selection.We predict that genomic prediction will be achievable in our study population and that it will outperform traditional pedigree-based approaches to studying micro-evolution in nature.