Position-dependent Codon Usage Bias in the Human Transcriptome

All life depends on the reliable translation of RNA to protein according to complex interactions between translation machinery and RNA sequence features. While ribosomal occupancy and codon frequencies vary across coding regions, well-established metrics for computing coding potential of RNA do not capture such positional dependence. Here, we investigate position-dependent codon usage bias (PDCUB), which dynamically accounts for the position of protein-coding signals embedded within coding regions. We demonstrate the existence of PDCUB in the human transcriptome, and show that it can be used to predict translation-initiating codons with greater accuracy than other models. We further show that observed PDCUB is not accounted for by other common metrics, including position-dependent GC content, consensus sequences, and the presence of signal peptides in the translation product. More importantly, PDCUB defines a spectrum of translational efficiency supported by ribosomal occupancy and tRNA adaptation index (tAI). High PDCUB scores correspond to a tAI-defined translational ramp and low ribosomal occupancy, while low PDCUB scores exhibit a translational valley and the highest ribosomal occupancy. Finally, we examine the relationship between PDCUB intensity and functional enrichment. We find that transcripts with start codons showing the highest PDCUB are enriched for functions relating to the regulation of synaptic signaling and plasticity, as well as skeletal, heart, and nervous-system development. Furthermore, transcripts with high PDCUB are depleted for functions related to immune response and detection of chemical stimulus. These findings lay important groundwork for advances in our understanding of the regulation of translation, the calculation of coding potential, and the classification of RNA transcripts.

m6Am-seq reveals the dynamic m6Am methylation in the human transcriptome

N6,2′-O-dimethyladenosine (m6Am), a terminal modification adjacent to the mRNA cap, is a newly discovered reversible RNA modification. Yet, a specific and sensitive tool to directly map transcriptome-wide m6Am is lacking. Here, we report m6Am-seq, based on selective in vitro demethylation and RNA immunoprecipitation. m6Am-seq directly distinguishes m6Am and 5′-UTR N6-methyladenosine (m6A) and enables the identification of m6Am at single-base resolution and 5′-UTR m6A in the human transcriptome. Using m6Am-seq, we also find that m6Am and 5′-UTR m6A respond dynamically to stimuli, and identify key functional methylation sites that may facilitate cellular stress response. Collectively, m6Am-seq reveals the high-confidence m6Am and 5′-UTR m6A methylome and provides a robust tool for functional studies of the two epitranscriptomic marks.

Detecting and Profiling Endogenous RNA G-Quadruplexes in the Human Transcriptome

G-quadruplexes are the non-canonical nucleic acid structures that are preferentially formed in G-rich regions. This structure has been shown to be associated with many biological functions. Regardless of the broad efforts on DNA G-quadruplexes, we still have limited knowledge on RNA G-quadruplexes, especially in a transcriptome-wide manner. Herein, by integrating the DMS-seq and the bioinformatics pipeline, we profiled and depicted the RNA G-quadruplexes in the human transcriptome. The genes that contain RNA G-quadruplexes in their specific regions are significantly related to immune pathways and the COVID-19-related gene sets. Bioinformatics analysis reveals the potential regulatory functions of G-quadruplexes on miRNA targeting at the scale of the whole transcriptome. In addition, the G-quadruplexes are depleted in the putative, not the real, PAS-strong poly(A) sites, which may weaken the possibility of such sites being the real cleaved sites. In brief, our study provides insight into the potential function of RNA G-quadruplexes in post-transcription.

Inflammation but not Glycemic Control is Associated with Neurocognitive Decline After Cardiac Surgery

Background: Whether perioperative glycemic control or markers of inflammation is associated with neurocognitive decline (NCD) after cardiac surgery was examined. Methods: Thirty patients undergoing cardiac surgery utilizing cardiopulmonary bypass (CPB) were screened for NCD preoperatively and on post-operative day four (POD4). Serum cytokine levels were measured and human transcriptome analysis was performed on blood samples. Neurocognitive data are presented as a change from baseline to POD4 in a score standardized with respect to age and gender. Results: A decline in neurocognitive function was identified in 73% (22/30) of patients on POD4. Patients with postoperative leukocytosis (WBC ≥ 10.5) had more NCD when compared to their baseline function (p=0.03). Patients with elevated IL-8 levels at 6 hours postoperatively had a significant decline in NCD at POD4 (p=0.04). Surprisingly, TNF-α, IL-1β, IL-2, or IL-6 levels were not associated with NCD (p>0.3 for all). There was no difference in neurocognitive function between patients with elevated HbA1c levels preoperatively (p=0.973) or elevated fasting blood glucose levels the morning of surgery (>126mg/dL, p=0.910), or a higher maximum blood glucose levels during CPB (>180mg/dL, p=0.252), or higher average glucose levels during CPB (>160mg/dL, p=0.639). Human transcriptome analysis demonstrated unique and differential patterns of gene expression in patients depending on the presence of DM and NCD. Conclusions: Perioperative glycemic control does not have an effect on NCD soon after cardiac surgery. Postoperative leukocytosis and elevated IL-8 levels are associated with neurocognitive decline. The profile of gene expression was altered in patients with NCD with or without diabetes.

Unbiased screen of human transcriptome reveals an unexpected role of 3ʹUTRs in translation initiation

Although most eukaryotic mRNAs require a 5ʹ-cap for translation initiation, some can also be translated through a poorly studied cap-independent pathway. Here we developed a circRNA-based system and unbiasedly identified more than 10,000 sequences in the human transcriptome that contain Cap-independent Translation Initiators (CiTIs). Surprisingly, most of the identified CiTIs are located in 3ʹUTRs, which mainly promote translation initiation in mRNAs bearing highly structured 5ʹUTR. Mechanistically, CiTI recruits several translation initiation factors including eIF3 and DHX29, which in turn unwind 5ʹUTR structures and facilitate ribosome scanning. Functionally, we showed that the translation of HIF1A mRNA, an endogenous DHX29 target, is antagonistically regulated by its 5ʹUTR structure and a new 3ʹ-CiTI in response to hypoxia. Therefore, deletion of 3ʹ-CiTI suppresses cell growth in hypoxia and tumor progression in vivo. Collectively, our study uncovers a new regulatory mode for translation where the 3ʹUTR actively participate in the translation initiation.

Multilayer modelling of the human transcriptome and biological mechanisms of complex diseases and traits

Here, we performed a comprehensive intra-tissue and inter-tissue multilayer network analysis of the human transcriptome. We generated an atlas of communities in gene co-expression networks in 49 tissues (GTEx v8), evaluated their tissue specificity, and investigated their methodological implications. UMAP embeddings of gene expression from the communities (representing nearly 18% of all genes) robustly identified biologically-meaningful clusters. Notably, new gene expression data can be embedded into our algorithmically derived models to accelerate discoveries in high-dimensional molecular datasets and downstream diagnostic or prognostic applications. We demonstrate the generalisability of our approach through systematic testing in external genomic and transcriptomic datasets. Methodologically, prioritisation of the communities in a transcriptome-wide association study of the biomarker C-reactive protein (CRP) in 361,194 individuals in the UK Biobank identified genetically-determined expression changes associated with CRP and led to considerably improved performance. Furthermore, a deep learning framework applied to the communities in nearly 11,000 tumors profiled by The Cancer Genome Atlas across 33 different cancer types learned biologically-meaningful latent spaces, representing metastasis (p < 2.2 × 10−16) and stemness (p < 2.2 × 10−16). Our study provides a rich genomic resource to catalyse research into inter-tissue regulatory mechanisms, and their downstream consequences on human disease.

A Computational and Biochemical Study of -1 Ribosomal Frameshifting in Human mRNAs

Abstract−1 programmed ribosomal frameshifting (−1 PRF) is a translational recoding mechanism used by many viral and cellular mRNAs. −1 PRF occurs at a heptanucleotide slippery sequence and is stimulated by a downstream RNA structure, most often in the form of a pseudoknot. The utilization of −1 PRF to produce proteins encoded by the −1 reading frame is wide-spread in RNA viruses, but relatively rare in cellular mRNAs. In human, only three such cases of −1 PRF events have been reported, all involving retroviral-like genes and protein products. To evaluate the extent of −1 PRF utilization in the human transcriptome, we have developed a computational scheme for identifying putative pseudoknot-dependent −1 PRF events and applied the method to a collection of 43,191 human mRNAs in the NCBI RefSeq database. In addition to the three reported cases, our study identified more than two dozen putative −1 PRF cases. The genes involved in these cases are genuine cellular genes without a viral origin. Moreover, in more than half of these cases, the frameshift site locates far upstream (>250 nt) from the stop codon of the 0 reading frame, which is nonviral-like. Using dual luciferase assays in HEK293T cells, we confirmed that the −1 PRF signals in the mRNAs of CDK5R2 and SEMA6C are functional in inducing efficient frameshifting. Our findings have significant implications in expanding the repertoire of the −1 PRF phenomenon and the protein-coding capacity of the human transcriptome.