Region-Based Epigenetic Clock Design Improves RRBS-Based Age Prediction
Region-Based Epigenetic Clock Design Improves RRBS-Based Age Prediction
复制标题
基于区域的表观遗传时钟设计改进了基于 RRBS 的年龄预测
DOI：
10.1101/2023.01.13.524017
复制
发表时间：
2023
期刊：
影响因子：
--
通讯作者：
Simpson D
中科院分区：
文献类型：
--
作者：
Simpson D
最近的研究表明，表观遗传返老还童可以通过使用模拟卡路里限制的药物和重编程诱导返老还童等技术来实现。为了在体内有效地测试返老还童，小鼠模型是最安全的选择。然而，我们发现，最近为小鼠减少代表性亚硫酸氢盐测序（RRBS）数据开发的表观遗传时钟在应用于外部数据集时性能显着较差。我们发现，捕获的位点和年龄预测所需的关键CpG的覆盖范围在数据集之间差异很大，这可能导致RRBS时钟缺乏可转移性。为了缓解基于RRBS的年龄预测中的这些覆盖问题，我们提出了两种新的设计策略，它们使用大区域而不是单个CpG的平均甲基化，从而通过滑动窗口（例如5 kb）或基于密度的CpG聚类来定义区域。当应用于外部数据集时，我们观察到与已发表的基于个体CpG的技术相比，我们的区域血液时钟（RegBC）的相关性和误差有所改善。当应用于低覆盖率数据时，RegBC也更稳健，并且在经历卡路里限制的小鼠中检测到负的年龄加速。我们的RegBC提供了一个原则证明，即RRBS数据集的年龄预测可以通过考虑一个区域上的多个CpG来改善，这否定了目前阻碍基于单个CpG的方法的读取深度的缺乏。
Recent studies suggest that epigenetic rejuvenation can be achieved using drugs that mimic calorie restriction and techniques such as reprogramming‐induced rejuvenation. To effectively test rejuvenation in vivo, mouse models are the safest alternative. However, we have found that the recent epigenetic clocks developed for mouse reduced‐representation bisulphite sequencing (RRBS) data have significantly poor performance when applied to external datasets. We show that the sites captured and the coverage of key CpGs required for age prediction vary greatly between datasets, which likely contributes to the lack of transferability in RRBS clocks. To mitigate these coverage issues in RRBS‐based age prediction, we present two novel design strategies that use average methylation over large regions rather than individual CpGs, whereby regions are defined by sliding windows (e.g. 5 kb), or density‐based clustering of CpGs. We observe improved correlation and error in our regional blood clocks (RegBCs) compared to published individual‐CpG‐based techniques when applied to external datasets. The RegBCs are also more robust when applied to low coverage data and detect a negative age acceleration in mice undergoing calorie restriction. Our RegBCs offer a proof of principle that age prediction of RRBS datasets can be improved by accounting for multiple CpGs over a region, which negates the lack of read depth currently hindering individual‐CpG‐based approaches.