Principles of mRNA control by human Pumilio proteins elucidated from global transcriptome stability, in vitro affinity profiling, and integrative data analysis

The human members of the PUF family of proteins, PUM1 and PUM2, are RNA-binding proteins that post-transcriptionally regulate gene expression through binding to a PUM recognition element (PRE) in the 3′ UTR of target mRNAs, promoting RNA decay. Recent RNA-seq experiments in PUM1/2 knockdown conditions have identified hundreds of known and new human PUM targets through measurement of changes in steady state RNA levels. However, steady-state RNA levels do not allow for measurement of changes in RNA stability between conditions and do not allow for the differentiation between the contributions of changes in transcription rates and changes in RNA decay. Here, we identify hundreds of human PUM1/2 targets that have changes in RNA stability following PUM1/2 knockdown. We separate the contributions of changes in transcription rate and RNA stability and find that human PUM proteins almost exclusively modulate RNA abundance through changing RNA stability and not transcription. In addition, we find that the sequence preferences for all possible 8mers are largely similar between PUM1 and PUM2, suggesting that PUM1 and PUM2 recognize similar targets. We identify an ideal PRE “rulebook” by determining key contextual features around PREs, including local AU content, location of a PRE within a 3′ UTR, clustering of PREs, and number of miRNA sites near a PRE, that help differentiate functional PREs from non-functional ones as measured by our decay dataset. Consistent with previously identified functional roles of mammalian PUMs, we find that human PUM1 and PUM2 modulate the decay of genes related to signaling cascades and neuronal function. Finally, we train machine learning models to predict functional regulation of RNA targets by the human PUM proteins and find that contextual features around PREs contribute meaningful information to our models. Bru-seq and BruChase-seq under PUM1 and PUM2 knockdown in HEK293 cells compared non targeting control with 4 replicates. SEQRS analysis of PUM1 and PUM2 homology domains sequence preference over 5 rounds of selection

PUF家族蛋白（PUF family of proteins）的人类成员PUM1与PUM2均为RNA结合蛋白，可通过结合靶mRNA的3'非翻译区（3′ UTR）中的PUM识别元件（PUM recognition element, PRE）促进RNA降解，进而介导基因表达的转录后调控。此前针对PUM1/2敲低模型开展的RNA测序（RNA-seq）实验，通过检测稳态RNA水平的变化，已鉴定出数百种已知及新发现的人类PUM靶基因。然而，稳态RNA水平既无法用于比较不同实验条件下的RNA稳定性差异，也无法区分转录速率变化与RNA降解变化对基因表达的相对贡献。本研究鉴定出数百种在PUM1/2敲低后出现RNA稳定性改变的人类PUM1/2靶基因，并拆分了转录速率变化与RNA稳定性变化对RNA丰度的影响，结果显示人类PUM蛋白几乎完全通过调控RNA稳定性而非转录速率来改变RNA丰度。此外，我们发现PUM1与PUM2对所有可能的8聚核苷酸序列的序列偏好性整体高度相似，提示二者识别的靶标具有较高的重合度。结合本研究的RNA降解数据集，我们通过解析PRE周围的关键上下文特征——包括局部AU碱基占比、PRE在3' UTR中的位置、PRE的簇集情况以及PRE附近的微小RNA（microRNA, miRNA）结合位点数量等，构建了可区分功能性与非功能性PRE的理想“规则手册”。与此前已报道的哺乳动物PUM蛋白的功能一致，我们发现人类PUM1与PUM2可调控参与信号级联反应与神经元功能的相关基因的RNA降解过程。最后，我们训练了机器学习模型以预测人类PUM蛋白对RNA靶标的功能性调控作用，结果表明PRE周围的上下文特征可为模型提供具有显著意义的预测信息。本研究在HEK293细胞中开展了PUM1/2敲低模型下的Bru-seq与BruChase-seq实验，以非靶向对照为参照，共设置4次生物学重复；同时针对PUM1与PUM2的同源结构域开展了5轮筛选的SEQRS分析，以解析其序列偏好性。