1 INTRODUCTION

The roots of using genomic data to search for signatures of relatively recent and/or ongoing selection trace back to the classic paper of Maynard Smith and Haigh (1974). They noted, as did Kojima and Schaffer (1967), that a linked neutral allele increases in frequency by hitchhiking along with the selected site, resulting in a region of significantly decreased background polymorphism around a recently selected site. Building on the basic premise of these papers, a very rich, and often highly technical, population genetics theory of selective sweeps has been developed (reviewed in Chapter 8 of Walsh & Lynch, 2018; henceforth denoted by WL). The length of the genomic region impacted by a favoured allele being swept to fixation is a function of the relative time scales of two competing processes: (a) recombination reshuffling the association between the selected and nearby neutral sites (scaling as roughly 1/c, the inverse of the recombination rate) and (b) selection driving the allele towards fixation (scaling as 1/s, the inverse of strength of selection on the allele). If the strength of selection is strong, a large genomic region is influenced by the sweep, as there is less time for recombination to occur before the favoured allele (and an unrecombined region around it) is fixed. Conversely, if the strength of recombination is strong relative to selection, then the impact of the sweep is small. Given this interplay between forces, it is not surprising that a number of measures of sweep impact scale as a function of s/c (WL Table 8.1).

This extensive body of theory has, in turn, led to the development of a very large, and often highly confusing, number of tests for selection using marker data (reviewed in WL Chapters 9, 10 and 12). A number of these test have been widely applied to human data and in the search for domestication genes (especially in plants). The high profile of much of this work leads to the obvious question of just how relevant these approaches might be for an animal breeder. Potential applications of these methods for the breeder fall loosely into three categories: (a) QTL mapping for traits under known artificial selection, (b) looking for signatures of selection at known QTL for a trait under selection and (c) trait‐independent searches for domestication, improvement or adaptation genes. These categories use different information and ask different questions. Categories (a) and (c) are best thought of as exploratory, either for finding QTL when a known trait is under selection (a) or searches of the genome for trait‐independent signatures of selection (c). Category (b) is conformational, showing that known QTLs for a candidate trait thought to be under recent selection do indeed show signatures of selection. Although this last category seems to be the most academic, and therefore usually the least practical, it has played an important role in some studies of the genetics of domestication.

We will consider these application categories in turn and will argue that Category (iii), and in particular, the searches for signatures of local adaptation, may be the most potent use of this approach by the breeder. The bottom line is that while marker‐based tests for selection can be useful, indeed important, tools, they must be used with caution. They are prone to a wide variety of false positives (e.g. the presence of population structure and/or a population expansion can yield false positives for many tests), and they are very likely to miss much of the actual selection. Finally, given that specific tests are designed for detecting very specific events (e.g. the signal from a completed versus that from an ongoing sweep), using the wrong class of tests for the question of interest is a fairly common mistake in the literature. A much more developed treatment of these issues can be found in WL Chapters 8–10 and 12, and it is highly recommended that any potential user of selection tests review this material before attempting to apply them.

中文翻译：

---

基因组选择特征和动物育种

1 介绍

使用基因组数据搜索相对近期和/或正在进行的选择特征的根源可以追溯到 Maynard Smith 和 Haigh ( 1974 )的经典论文。他们注意到，正如 Kojima 和 Schaffer ( 1967 )所做的那样，一个连锁的中性等位基因通过与选定的位点搭便车而增加了频率，导致最近选定的位点周围背景多态性显着降低的区域。在这些论文的基本前提的基础上，已经开发了一个非常丰富且通常具有高度技术性的选择性扫描种群遗传学理论（见 Walsh & Lynch，2018 年第 8 章综述） ; 此后用 WL 表示）。受偏爱等位基因影响的基因组区域的长度是两个竞争过程的相对时间尺度的函数：（a）重组重新调整所选中性位点和附近中性位点之间的关联（缩放为大约 1/c，重组率的倒数）和（b）选择驱动等位基因朝向固定（缩放为 1/s，等位基因选择强度的倒数）。如果选择的强度很大，则较大的基因组区域会受到扫描的影响，因为在固定有利等位基因（及其周围的未重组区域）之前发生重组的时间较短。相反，如果重组的强度相对于选择强，则扫描的影响就小。鉴于力之间的这种相互作用，

反过来，这一广泛的理论体系导致了大量使用标记数据进行选择的测试的发展，并且通常非常混乱（在 WL 第 9、10 和 12 章中进行了回顾）。其中许多测试已广泛应用于人类数据和寻找驯化基因（尤其是植物）。大部分工作的高调引出了一个明显的问题，即这些方法对动物饲养员的相关性有多大。这些方法对育种者的潜在应用大致分为三类：（a）已知人工选择下性状的 QTL 作图，（b）在已知 QTL 上寻找选择中的性状的选择特征和（c）性状独立搜索用于驯化、改良或适应基因。这些类别使用​​不同的信息并提出不同的问题。类别 (a) 和 (c) 最好被认为是探索性的，用于在选择已知性状时寻找 QTL (a) 或搜索基因组以寻找与性状无关的选择特征 (c)。类别 (b) 是构象，表明被认为是最近选择的候选性状的已知 QTL 确实显示了选择的特征。尽管最后一类似乎是最学术的，因此通常最不实用，但它在驯化遗传学的一些研究中发挥了重要作用。表明被认为处于最近选择之下的候选性状的已知 QTL 确实显示了选择的特征。尽管最后一类似乎是最学术的，因此通常最不实用，但它在驯化遗传学的一些研究中发挥了重要作用。表明被认为处于最近选择之下的候选性状的已知 QTL 确实显示了选择的特征。尽管最后一类似乎是最学术的，因此通常最不实用，但它在驯化遗传学的一些研究中发挥了重要作用。

我们将依次考虑这些应用类别，并将论证类别 (iii)，特别是搜索当地适应的特征，可能是育种者对这种方法最有效的使用。最重要的是，虽然基于标记的选择测试可能是有用的、确实很重要的工具，但必须谨慎使用。它们容易出现各种各样的误报（例如，种群结构的存在和/或种群扩张会在许多测试中产生误报），并且它们很可能会错过大部分实际选择。最后，鉴于特定测试是为检测非常特定的事件而设计的（例如，来自已完成扫描的信号与来自正在进行的扫描的信号），对感兴趣的问题使用错误的测试类别是文献中相当常见的错误。