The most abundant mRNA post-transcriptional modification is N6-methyladenosine (m6A) that has broad roles in RNA biology1-5. In mammalian cells, the asymmetric distribution of m6A along mRNAs leaves relatively less methylation in the 5′ untranslated region (5′UTR) compared to other regions6,7. However, whether and how 5′UTR methylation is regulated is poorly understood. Despite the crucial role of the 5′UTR in translation initiation, very little is known whether m6A modification influences mRNA translation. Here we show that in response to heat shock stress, m6A is preferentially deposited to the 5′UTR of newly transcribed mRNAs. We found that the dynamic 5′UTR methylation is a result of stress-induced nuclear localization of YTHDF2, a well characterized m6A “reader”. Upon heat shock stress, the nuclear YTHDF2 preserves 5′UTR methylation of stress-induced transcripts by limiting the m6A “eraser” FTO from demethylation. Remarkably, the increased 5′UTR methylation in the form of m6A promotes cap-independent translation initiation, providing a mechanism for selective mRNA translation under heat shock stress. Using Hsp70 mRNA as an example, we demonstrate that a single site m6A modification in the 5′UTR enables translation initiation independent of the 5′ end m7G cap. The elucidation of the dynamic feature of 5′UTR methylation and its critical role in cap-independent translation not only expands the breadth of physiological roles of m6A, but also uncovers a previously unappreciated translational control mechanism in heat shock response.

Learning effective feature representations and similarity measures are crucial to the retrieval performance of a content-based image retrieval (CBIR) system. Despite extensive research efforts for decades, it remains one of the most challenging open problems that considerably hinders the successes of real-world CBIR systems. The key challenge has been attributed to the well-known "semantic gap" issue that exists between low-level image pixels captured by machines and high-level semantic concepts perceived by human. Among various techniques, machine learning has been actively investigated as a possible direction to bridge the semantic gap in the long term. Inspired by recent successes of deep learning techniques for computer vision and other applications, in this paper, we attempt to address an open problem: if deep learning is a hope for bridging the semantic gap in CBIR and how much improvements in CBIR tasks can be achieved by exploring the state-of-the-art deep learning techniques for learning feature representations and similarity measures. Specifically, we investigate a framework of deep learning with application to CBIR tasks with an extensive set of empirical studies by examining a state-of-the-art deep learning method (Convolutional Neural Networks) for CBIR tasks under varied settings. From our empirical studies, we find some encouraging results and summarize some important insights for future research.

Cells have evolved exquisite mechanisms to fine-tune the rate of protein synthesis in response to stress. Systemic mapping of start codon positions and precise measurement of the corresponding initiation rate would transform our understanding of translational control. Here we present quantitative translation initiation sequencing (QTI-seq), where the initiating ribosomes can be profiled in real time at single nucleotide resolution. The resultant initiation map not only delineates variations of start codon selection, but also highlights a dynamic range of initiation rates in response to nutrient starvation. The integrated data set provides unique insights into principles of alternative translation and mechanisms controlling different aspects of translation initiation. Using RiboTag mice, QTI-seq permits tissue-specific profiling of initiating ribosomes in vivo. Liver cell-specific ribosome profiling uncovers a robust translational reprogramming of the proteasome system in fasted mice. Our findings illuminate the prevalence and dynamic nature of translational regulation pivotal to physiological adaptation in vivo.

MicroRNAs (miRNAs) are RNA sequences of ;22 nucleotides that mediate post-transcriptional regulation of specific mRNAs. miRNA sequences are dispersed throughout the genome and are classified as intergenic (between genes) or intronic (embedded into a gene). Intergenic miRNAs are expressed by their own promoter, and until recently, it was supposed that intronic miRNAs are transcribed from their host gene. Here, we performed a genomic analysis of currently known intronic miRNA regions and observed that ;35% of intronic miRNAs have upstream regulatory elements consistent with promoter function. Among all intronic miRNAs, 30% have associated Pol II regulatory elements, including transcription start sites, CpG islands, expression sequence tags, and conserved transcription factor binding sites, while 5% contain RNA Pol III regulatory elements (A/B box sequences). We cloned intronic regions encompassing miRNAs and their upstream Pol II (miR-107, miR-126, miR-208b, miR548f-2, miR-569, and miR-590) or Pol III (miR-566 and miR-128-2) sequences into a promoterless plasmid, and confirmed that miRNA expression occurs independent of host gene transcription. For miR-128-2, a miRNA overexpressed in acute lymphoblastic leukemia, ChIP analysis suggests dual regulation by both intronic (Pol III) and host gene (Pol II) promoters. These data support complex regulation of intronic miRNA expression, and have relevance to disregulation in disease settings.

Cleavage stimulation factor 64 kDa (CstF64) is an essential premRNA 3′ processing factor and an important regulator of alternative polyadenylation (APA). Here we characterized CstF64-RNA interactions in vivo at the transcriptome level and investigated the role of CstF64 in global APA regulation through individual nucleotide resolution UV crosslinking and immunoprecipitation sequencing and direct RNA sequencing analyses. We observed highly specific CstF64-RNA interactions at poly(A) sites (PASs), and we provide evidence that such interactions are widely variable in affinity and may be differentially required for PAS recognition. Depletion of CstF64 by RNAi has a relatively small effect on the global APA profile, but codepletion of the CstF64 paralog CstF64τ leads to greater APA changes, most of which are characterized by the increased relative use of distal PASs. Finally, we found that CstF64 binds to thousands of dormant intronic PASs that are suppressed, at least in part, by U1 small nuclear ribonucleoproteins. Taken together, our findings provide insight into the mechanisms of PAS recognition and identify CstF64 as an important global regulator of APA.

The integrated stress response (ISR) facilitates cellular adaptation to stress conditions via the common target eIF2α. During ISR, the selective translation of stress-related mRNAs often relies on alternative mechanisms, such as leaky scanning or reinitiation, but the underlying mechanism remains incompletely understood. Here we report that, in response to amino acid starvation, the reinitiation of ATF4 is not only governed by the eIF2α signaling pathway, but is also subjected to regulation by mRNA methylation in the form of N-methyladenosine (mA). While depleting mA demethylases represses ATF4 reinitiation, knocking down mA methyltransferases promotes ATF4 translation. We demonstrate that mA in the 5' UTR controls ribosome scanning and subsequent start codon selection. Global profiling of initiating ribosomes reveals widespread alternative translation events influenced by dynamic mRNA methylation. Consistently, Fto transgenic mice manifest enhanced ATF4 expression, highlighting the critical role of mA in translational regulation of ISR at cellular and organismal levels.

No abstract

Tissue-specific alternative splicing is achieved through the coordinated assembly of RNA binding proteins at specific sites to enhance or silence splicing at nearby splice sites. We used high-throughput sequencing (RNA-Seq) to investigate the complete spectrum of alternative splicing events that are regulated by the epithelium-specific splicing regulatory proteins ESRP1 and ESRP2. We also combined this analysis with direct RNA sequencing (DRS) to reveal ESRP-mediated regulation of alternative polyadenylation. To define binding motifs that mediate direct regulation of splicing and polyadenylation by ESRP, SELEX-Seq analysis was performed, coupling traditional SELEX with high-throughput sequencing. Identification and scoring of high-affinity ESRP1 binding motifs within ESRP target genes allowed the generation of RNA maps that define the position-dependent activity of the ESRPs in regulating cassette exons and alternative 3= ends. These extensive analyses provide a comprehensive picture of the functions of the ESRPs in an epithelial posttranscriptional gene expression program. Nearly all human multiexon gene transcripts undergo alternative splicing, and most also undergo multiple alternative splicing events within the same transcript (29, 46). As a result, alternative splicing provides a mechanism that broadens gene expression through a nearly exponential expansion of the number of distinct gene products that can be produced from the rather modest number of ϳ20,000 human genes. The regulation of alternative splicing is mediated by RNA binding proteins (RBPs) that interact with exonic and intronic sequences and function as splicing enhancers or silencers, depending on the regulator and/or binding context (27). This regulation is combinatorial, with the splicing outcome being determined by the net activities of RBPs that bind within or near alternative exons (1,20). While many of the regulatory factors, such as the well-characterized hnRNP and SR families of proteins, are generally ubiquitously expressed, the number of known regulators with cell or context-specific expression is growing (6). Technological advances in the past several years have led to the identification of the genome-wide targets for several well-characterized splicing regulators, such as Nova1/2, the Fox family proteins, and PTBP1/PTBP2 (reviewed in reference 12). Most of the targets described to date consist of simple cassette exons, and these regulators can induce either splicing or skipping of different exons. A recurring theme has been that splicing regulators induce either exon splicing or skipping in a position-dependent manner. These observations have given rise to the concept of "RNA maps," whereby the binding position of the protein relative to a regulated alternative exon determines whether it promotes or represses splicing. These RNA maps along with other experimental and bioinformatics data suggest a broader "splicing code," wherein the collective identification of the expression levels and binding sites for all splicing regulators can p...