Biology subjects

Yu, D.

Publications and source records attributed to Yu, D..

At least 73 records · Page 4Linked to original sources

bioRxiv2024

Shox2 is necessary for normal thalamic spindle function

The cellular identity of thalamocortical neurons (TCNs), namely their firing properties, dictates brain-wide activity patterns, such as sleep spindles. Transcription factors are critical to the determination of cellular identity. Previously, we discovered that a subset of TCNs express the transcription factor, Shox2, and, in a global Shox2 KO, established that TCNs within the anterior nucleus of the thalamus rely on the expression of Shox2 to regulate key ion channels that are necessary to maintain their firing properties. From this, we hypothesized that Shox2 expression, through the regulation of firing properties of TCNs, is critical for the thalamocortical circuit to generate spindle oscillations. We exploited the somatosensory thalamocortical circuit to investigate this by creating a primary somatosensory thalamus (VB) Shox2 knockdown mouse model. We delivered Cre into the VB of P21 Shox2fl/fl mice using viral infection and compared in vitro, patch-clamp recordings from Shox2+ and Shox2 knockdown TCNs, finding that Shox2 expression is indeed critical to maintain burst and tonic firing properties of VB TCNs. Since Shox2 is important developmentally and firing from TCNs to cortex during development structures the circuit, we performed ultrasound-guided P3 injections at P3 to generate an early-stage, Shox2 VB knockdown, but found no changes in the layer four, barrel map (VB cortical target). Despite this, Shox2 knockdown mice exhibit reduced sleep-spindle EEG density. Further, key behaviors associated with spindles and proper VB thalamic function--memory consolidation and somatosensory perception--are significantly impaired. These results indicate that the impact on spindle function is likely due to cell autonomous changes to TCNs rather than circuit changes, confirming our hypothesis that Shox2 is necessary for normal thalamic spindle function and implicating a potential role for Shox2 in autism and schizophrenia pathologies.

neuroscience↗

bioRxiv2024

Lipid A phosphoethanolamine transferase-mediated site-selective modifications show association with colistin resistance phenotypes and fitness

Genes encoding lipid A modifying phosphoethanolamine transferases (PETs) are genetically diverse and can confer resistance to colistin and antimicrobial peptides. To better understand the functional diversity of PETs, we characterized three canonical mobile colistin resistance (mcr) alleles (mcr-1, -3, -9), one intrinsic pet (eptA), and two mcr-like genes (petB, petC). Using an isogenic expression system, we show that mcr-1 and mcr-3 are phenotypically similar by conferring colistin resistance with lower fitness costs. mcr-9, which is phylogenetically closely related to mcr-3, and eptA only provide fitness advantages in the presence of sub-inhibitory concentrations of colistin and significantly reduce fitness in media only. PET-B and PET-C were phenotypically distinct from bonafide PETs; neither conferred colistin resistance or caused considerable fitness cost in Escherichia coli. Strikingly, we found for the first time that different PETs selectively modify different phosphates of lipid A - MCR-1, MCR-3, and PET-C selectively modify the 4-phosphate, while MCR-9 and EptA modify the 1-phosphate. 4-phosphate modifications facilitated by MCR-1 and -3 are associated with high levels of colistin resistance and low toxicity. Our results suggest that PETs have a wide phenotypic diversity and that high level colistin resistance is associated with specific lipid A modification patterns that has been largely unexplored thus far. IMPORTANCERising levels of resistance to increasing numbers of antimicrobials has led to the revival of last resort antibiotic colistin. Unfortunately, resistance to colistin is also spreading in the form of mcr genes, making it essential to (i) improve identification of resistant bacteria to allow clinicians to prescribe effective drug regimens and (ii) develop new combination therapies effective at targeting resistant bacteria. Our results demonstrate that PETs, including MCR variants, are site-selective in E. coli, with site-selectivity correlating with the level of resistance and fitness costs conferred by certain PETs. Site-selectivity associated with a given PET may not only help predict colistin resistance phenotypes, but may also provide an avenue to (i) improved drug regimens and (ii) development of new combination therapies to better combat colistin resistant bacteria.

microbiology↗

bioRxiv2024

2'-O-Methyl-guanosine 3-base RNA fragments mediate essential natural TLR7/8 antagonism

Recognition of RNA fragments by Toll-like receptors (TLR) 7 and 8 is a key contributor to the initiation of a protective innate immune response against pathogens. A long-standing enigma is how degradation products of host RNAs, generated by the daily phagocytic clearance of billions of apoptotic cells, fail to activate TLR7 and TLR8 signalling1. Here, we report that select 2-O-methyl (2-Ome) guanosine RNA fragments as short as 3 bases, including those derived from host-RNAs, are potent TLR7 and TLR8 antagonists that reduce TLR7 sensing in vivo. Mechanistically, antagonistic fragments are directed towards a distinct binding site on these proteins by 5-end 2-Ome guanosine. Our results indicate that host-RNAs evade detection by TLR7/8 due to a pool of abundant host ribosomal 2-Ome-modified RNA fragments that naturally antagonize TLR7 and TLR8 sensing to avoid auto-immunity. Crucially, rare TLR7 and TLR8 mutations located at this antagonistic site decrease the inhibitory activity of 2-Ome guanosine RNA fragments and lead to auto-immunity in patients. Our findings also establish that select chemically synthesised 3-base oligonucleotides can harness the protective anti-inflammatory activity of this natural immune checkpoint for therapeutic targeting of TLR7-driven diseases. One Sentence SummaryShort 2-O-Methyl RNA fragments are natural TLR7/8 antagonists

immunology↗

bioRxiv2024

Hypergraph Cortical Cytoarchitectonic Parcellation with Multimodal Canine Brain Atlas

Brain atlases are vital tools in exploring the brain structure-function relationship. The burgeoning cross-species atlases have significantly accelerated our understanding of human brain development, evolution, function, and diseases. However, the existing coarse-grained macroscopic canine brain atlases greatly constrain their utility as an animal model for neurocognition research. Finer-grained brain atlas and partitions are crucial for decoding brain spatial heterogeneity and topology at different scales. Therefore, we conduct macroscopic and microscopic brain imaging to construct an interactive online dataset of multimodal canine brain atlas. Additionally, we develop a pioneering method for cortical cytoarchitectonic partitioning based on hypergraph learning. By integrating high-dimensional cytoarchitectonic features and spatial connections between cortical columns, the method leads to fine-grained partitioning patterns. This innovative approach aims to decode the biological heterogeneity of cortical microstructures, contributing to the structural annotation of canine atlas as well as public human brain atlases. The study not only offers valuable resources but also presents a novel zonation approach to investigate the cellular organization pattern and topology of the cortex.

neuroscience↗

bioRxiv2024

IL-2 enhances effector function but suppresses follicular localization of CD8+ T cells in chronic infection

Cytotoxic CD8+ T cells, essential in combating viral infections and cancer, become dysfunctional from prolonged antigen exposure. Precursors of exhausted T (TPEX) cells are pivotal in sustaining immune responses in chronic diseases and mediating immunotherapy efficacy. They also control viral infection within B-cell follicles, facilitated by CXCR5 expression. How cytokines regulate TPEX cell fate and follicular entry is not well understood. We reveal that IL-2 treatment enhances CD8+ T cell effector functions in chronic LCMV infection but hinders CXCR5+ TPEX cell formation and infection control within B-cell follicles. Mechanistically, IL-2 suppresses TPEX cell differentiation in a STAT5 and BLIMP1-dependent manner. Using an IL-2 fusion protein targeting CD122, we shifted the differentiation towards CX3CR1+ T cells with increased effector function. Clinical observations with low-dose IL-2 in autoimmune disease confirmed IL-2s inhibitory effect on CXCR5+ TPEX cells, underscoring IL-2s crucial regulatory role and therapeutic potential in modulating TPEX and effector T cell generation.

immunology↗

bioRxiv2024

Notch dimerization contributes to maintenance of intestinal homeostasis by a mechanism involving HDAC2.

A tri-protein complex containing NICD, RBPj and MAML1 binds DNA as monomer or, cooperatively, as dimers to regulate transcription. Mice expressing Notch dimerization-deficient (DD) alleles of Notch1 and Notch2 (NDD) are sensitized to environmental insults but otherwise develop and age normally. Transcriptomic analysis of colonic spheroids uncovered no evidence of dimer-dependent target gene miss-regulation, confirmed impaired stem cell maintenance in-vitro, and discovered an elevated signature of epithelial innate immune response to symbionts, the likely underlying cause for heightened sensitivity in NDD mice. TurboID followed by quantitative nano-spray MS/MS mass-spectrometry analyses in a human colon carcinoma cell line expressing either NOTCH2DD or NOTCH2 revealed an unbalanced interactome, with reduced interaction of NOTCH2DD with the transcription machinery but relatively preserved interaction with the HDAC2 interactome suggesting modulation via cooperativity. To ask if HDAC2 activity contributes to Notch loss-of-function phenotypes, we used the HDAC2 inhibitor Valproic acid (VPA) and discovered it could prevent the intestinal consequences of gamma secretase inhibitor (DBZ or DAPT) treatment in mice and spheroids, suggesting synergy between HDAC activity and pro-differentiation program in intestinal stem cells.

developmental biology↗

bioRxiv2024

Inferring gene regulatory networks by hypergraph variational autoencoder

In constructing Gene Regulatory Networks (GRNs), it is crucial to consider cellular heterogeneity and differential gene regulatory modules. However, traditional methods have predominantly focused on cellular heterogeneity, approaching the subject from a relatively narrow scope. We present HyperG-VAE, a Bayesian deep generative model that utilizes a hypergraph to model single-cell RNA sequencing (scRNA-seq) data. HyperG-VAE employs a cell encoder with a Structural Equation Model to address cellular heterogeneity and build GRNs, alongside a gene encoder using hypergraph self-attention to identify gene modules. Encoders are synergistically optimized by a decoder, enabling HyperG-VAE to excel in GRN inference, single-cell clustering, and data visualization, evidenced by benchmarks. Additionally, HyperG-VAE effectively reveals gene regulation patterns and shows robustness in varied downstream analyses, demonstrated using B cell development data in bone marrow. The interplay of encoders by the overlapping genes between predicted GRNs and gene modules is further validated by gene set enrichment analysis, underscoring that the gene encoder boosts the GRN inference. HyperG-VAE proves efficient in scRNA-seq data analysis and GRN inference.

bioinformatics↗

bioRxiv2024

Pathway-enhanced Transformer-based model for robust enumeration of cell types from the cell-free transcriptome

Analyzing cell types of origin of cell-free RNA can enhance the resolution of liquid biopsies, thereby deepening the understanding of molecular and cellular changes in development and disease processes. Existing deconvolution methods typically rely on meticulously curated gene expression profiles or employ deep neural network with vast and complex solution spaces that are difficult to interpret. These approaches overlook the synergistic and co-expression effects among genes in biological signaling pathways, compromising their generalizability and robustness. we developed Deconformer, a Transformer-based deconvolution model that integrates biological signaling pathways at the embedding stage, to address these issues. Compared to popular methods on multiple datasets, Deconformer demonstrates superior performance and robustness, and is capable of tracking the developmental process of the fetal and placenta. Additionally, pathway-level interpretability of Deconformer offers new insights into crosstalk, dependencies, and other interactions within cell-free RNA pathways, supporting further biological discoveries. We posit that Deconformer represents a significant advancement in the precise analysis of the cell-free transcriptome. It holds the promise of describing disease progression and severity with a new level of accuracy, focusing on the contributions of originating cell types and their pathway dependencies. This model has the potential to catalyze the development of non-invasive diagnostic tools and enhance our understanding of the underlying biology of diseases.

bioinformatics↗

bioRxiv2024

LXA4 promotes the browning of white adipose through miR-133a-3p/SIRT1 pathway

Lipoxin A4 (LXA4) promotes the browning of white adipose and energy consumption. The specific mechanism of which involved in white adipose browning is less clear. A high-fat diet (HFD) mouse model was constructed. Different groups of mice were treated with LXA4 accordingly. The body weight of mouse, subcutaneous and visceral fat, and food intake were recorded. The effect of LXA4 was examined by observing changes in pathology, serum insulin and lipid accumulation indices. The effects of LXA4/miR-133a-3p/Sirtuin1 on lipid droplet formation, fat browning-related genes, and the insulin receptor-AKT pathway in cells were examined after induction of adipocyte differentiation in 3T3-L1 precursors. At the cellular level, LXA4 promoted lipid droplet formation, expressions of fat browning genes and activation of the insulin receptor-AKT pathway in differentiated 3T3-L1 cells. MiR-133a-3p agomir partially offset the effects of LXA4. SIRT1 was a downstream target gene of miR-133a-3p, participating in the promotive effects of LXA4 on fat browning. LXA4 promotes white adipose browning and relieves insulin resistance through miR-133a-3p/SIRT1 pathway.

biochemistry↗

bioRxiv2024

Structure-based design of a soluble human cytomegalovirus glycoprotein B antigen stabilized in a prefusion-like conformation

Human cytomegalovirus (HCMV) glycoprotein B (gB) is a class III membrane fusion protein required for viral entry. HCMV vaccine candidates containing gB have demonstrated moderate clinical efficacy, but no HCMV vaccine has been FDA-approved. Here, we used structure-based design to identify and characterize amino acid substitutions that stabilize gB in its metastable prefusion conformation. One variant containing two engineered interprotomer disulfide bonds and two cavity-filling substitutions (gB-C7), displayed increased expression and thermostability. A 2.8 [A] resolution cryo-electron microscopy structure shows that gB-C7 adopts a prefusion-like conformation, revealing additional structural elements at the membrane-distal apex. Unlike previous observations for several class I viral fusion proteins, mice immunized with postfusion or prefusion-stabilized forms of soluble gB protein displayed similar neutralizing antibody titers, here specifically against an HCMV laboratory strain on fibroblasts. Collectively, these results identify initial strategies to stabilize class III viral fusion proteins and provide tools to probe gB-directed antibody responses. TeaserA structure-based design campaign leads to stabilization of the class III viral fusion protein from HCMV in a prefusion-like conformation.

molecular biology↗

bioRxiv2024

Primate-specific BTN3A2 protects against SARS-CoV-2 infection by interacting with and reducing ACE2

BackgroundCoronavirus disease 2019 (COVID-19) is an immune-related disorder caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). The complete pathogenesis of the virus remains to be determined. Unraveling the molecular mechanisms governing SARS-CoV-2 interactions with host cells is crucial for the formulation of effective prophylactic measures and the advancement of COVID-19 therapeutics. MethodsWe analyzed human lung single-cell RNA sequencing dataset to discern the association of butyrophilin subfamily 3 member A2 (BTN3A2) expression with COVID-19. The BTN3A2 gene edited cell lines and transgenic mice were infected by live SARS-CoV-2 in a biosafety level 3 (BSL-3) laboratory. Immunoprecipitation, flow cytometry, biolayer interferometry and competition ELISA assays were performed in BTN3A2 gene edited cells. We performed quantitative real-time PCR, histological and/or immunohistochemical analyses for tissue samples from mice with or without SARS-CoV-2 infection. FindingsThe BTN3A2 mRNA level was correlated with COVID-19 severity. BTN3A2 expression was predominantly identified in epithelial cells, elevated in pathological epithelial cells from COVID-19 patients and co-occurred with ACE2 expression in the same lung cell subtypes. BTN3A2 targeted the early stage of the viral life cycle by inhibiting SARS-CoV-2 attachment through interactions with the receptor-binding domain (RBD) of the Spike protein and ACE2. BTN3A2 inhibited ACE2-mediated SARS-CoV-2 infection by reducing ACE2 in vitro and in vivo. InterpretationThese results reveal a key role of BTN3A2 in the fight against COVID-19. Identifying potential monoclonal antibodies which mimic BTN3A2 may facilitate disruption of SARS-CoV-2 infection, providing a therapeutic avenue for COVID-19. FundingThis study was supported by the National Natural Science Foundation of China (32070569, U1902215, and 32371017), the CAS "Light of West China" Program, and Yunnan Province (202305AH340006). Research in contextO_ST_ABSEvidence before this studyC_ST_ABSOur understanding of the pathogenesis of COVID-19, especially key molecular events in the early stage of viral infection, remains incompletely albeit we witnessed many progresses. This knowledge gap hinders the finding for effective and specific antiviral agents against SARS-CoV-2. The entry of SARS-CoV-2 is mediated by the entry receptor angiotensin-converting enzyme 2 (ACE2) and is affected by host antiviral defenses. Targeting these universal host factors required for virus replication is the most promising approach for effective prevention and treatment of COVID-19. Added value of this studyOur study revealed that BTN3A2, a primate-specific gene, acts as a potent inhibitor of SARS-CoV-2 infection in vitro and in vivo. The up-regulation of BTN3A2 upon SARS-CoV-2 infection competed with the ACE2 receptor for binding to the Spike protein, subsequently reducing ACE2 expression and ACE2-mediated SARS-CoV-2 entry. Implications of all the available evidenceThese data highlighted that BTN3A2 as a novel host factor with protective effects against SARS-CoV-2 infection. The BTN3A2 holds considerable potential as a therapeutic drug for mitigating the impact of SARS-CoV-2 and its variants of concern (VOCs).

cell biology↗

bioRxiv2024

A new story of four Hexapoda classes: Protura as the sister to all other hexapods

Insects represent the most diverse animal group, yet previous phylogenetic analyses based on the morphological and molecular data have failed to agree on the evolutionary relationships of early insects and their six-legged relatives (together constituting the clade Hexapoda). In particular, the phylogenetic positions of the three early-diverging hexapod groups, the coneheads (Protura), springtails (Collembola), and two-pronged bristletails (Diplura), have been debated for over a century, with alternative topologies implying drastically different scenarios of the evolution of the insect body plan and hexapod terrestrialisation. We addressed this issue by sampling of all hexapod orders, and experimented with a broad range of across-site compositional heterogeneous models designed to tackle ancient divergences. Our analyses support Protura as the earliest-diverging hexapod lineage (Protura-sister) and Collembola as a sister group to the Diplura, a clade we refer to as Antennomusculata characterised by the shared possession of internal muscles in the antennal flagellum. The universally recognized Ellipura hypothesis is recovered under the site-homogenous LG model. Our cross-validation analysis shows that the CAT-GTR model that recovers Protura-sister fits significantly better than homogenous model. Furthermore, as a very unusual group, Protura as the first diverging lineage of hexapods is also supported by other lines of evidence, such as mitogenomics, comparative embryology, and sperm morphology. The backbone phylogeny of hexapods recovered in this study will facilitate the exploration of the underpinnings of hexapod terrestrialisation and mega-diversity.

evolutionary biology↗

bioRxiv2023

Nanotopography boosts cellular uptake by inducing macropinocytosis

Efficient cellular uptake of biomolecules, including genetic material, mRNA, proteins, and nanoparticles, requires novel approaches to overcome inherent cellular barriers. This study investigates how nanotopographical cues from nanoporous surfaces impact the uptake efficiency of diverse molecules by cells. The results demonstrate that cellular uptake efficiency increases significantly on nanoporous surfaces compared to flat surfaces. Notably, this process is found to be dependent on the size and morphology of the nanopores, reaching its peak efficacy with blind pores of 400 nm in size. Enhanced genetic transduction on nanoporous surfaces were observed for multiple vectors, including lentiviruses, baculoviruses, and mRNA molecules. The versatile nature of this approach allows co-transfection of cells with multiple mRNA vectors. Moreover, the nanoporous platform was used for efficient and fast manufacturing of CAR-T cells through lentiviral transduction. Furthermore, we pinpoint macropinocytosis as the predominant mechanism driving increased cellular uptake induced by the nanoporous surfaces. The method introduced here for enhancing genetic transduction of cells has applications in immunotherapy research, drug delivery, and cell engineering.

biophysics↗

bioRxiv2023

Thermo-optoplasmonic single-molecule sensing on optical microcavities

Whispering-gallery-mode (WGM) resonators are powerful instruments for single-molecule sensing in biological and biochemical investigations. WGM sensors leveraged by plasmonic nanostructures, known as optoplasmonic sensors, provide unprecedented sensitivity down to single atomic ions. In this article, we describe that the response of optoplasmonic sensors upon the attachment of single protein molecules strongly depends on the intensity of WGM. At low intensity, protein binding causes red shifts of WGM resonance wavelengths, known as the reactive sensing mechanism. By contrast, blue shifts are obtained at high intensities, which we explain as thermo-optoplasmonic (TOP) sensing, where molecules transform absorbed WGM radiation into heat. To support our conclusions, we experimentally investigated seven molecules and complexes; we observed blue shifts for dye molecules, amino acids and anomalous absorption of enzymes in the near-infrared spectral region. As an example of application, we propose a physical model of TOP sensing that can be used for the development of single-molecule absorption spectrometers. Significance StatementA notable contribution to the optical detection of single molecules has been brought about by using optical microcavities, specifically whispering-gallery-mode resonators that combine optical and plasmon resonances for the most sensitive single-molecule and ion detection. Adding absorption measurements to a single-molecule technique that operates in an aqueous solution would provide powerful new detection capabilities for sensing the properties of single molecules. We demonstrate here, for the first time, the detection of absorption cross-sections of protein molecules on optoplasmonic microcavities and validate the technique with seven different molecules and complexes. A proof-of-concept for single-molecule optoplasmonic absorption spectrometers is presented, based on a thermo-optoplasmonic sensing mechanism.

biophysics↗

bioRxiv2023

Successful spread of mcr-1-bearing IncX4 plasmids is associated with variant in replication protein of IncX4 plasmids

IncX4 plasmids are one of the most epidemiologically successful vehicles for mcr-1 spread. Here we found that the IncX4 plasmids carried two different replication proteins encoded by genes pir-1 and pir-2, respectively, but mcr-1 was only carried by IncX4 plasmid encoding pir-1. The copy number of pir-2 encoding plasmids (3.15{+/-}0.9 copies) are higher than that of pir-1 encoding plasmids (0.85{+/-}0.5 copies). When mcr-1 was cloned into IncX4 plasmid encoding pir-2, the higher copy number of these plasmids resulted in increased expression of mcr-1 and a greater fitness burden on their host cells. However, these plasmids exhibited a lower rate of invasion into the bacterial population compared to mcr-1 positive plasmids encoding the pir-1 gene. These findings collectively explain the absence of mcr-1 in all IncX4 plasmids encoding pir-2. Our results further confirmed that low-copy numbers are important for the spread of mcr-1 plasmid from the perspective of natural evolution.

microbiology↗

bioRxiv2023

MYC amplifies mitotic perturbations elicited by LXY18 to enable synthetic lethality

The MYC oncoprotein represents an intriguing target for cancer treatment, but its therapeutic potential has been hindered by the absence of specific pharmacological inhibitors. In this study, we demonstrate that the phenoxy quinoline compound LXY18 selectively targets and eliminates cells overexpressing MYC, leaving non-transformed cells unharmed. This synthetic lethality arises from an acute induction of multipolarity, resulting in a persistent arrest in early mitosis followed by cell death in mitosis or after mitotic slippage. Distinctively, LXY18s action contrasts with other antimitotic compounds, as they either fail to induce mitotic arrest or elicit mitotic arrest irrespective of MYC abundance. Furthermore, the MYC abundance in a panel of 98 tumor cell lines correlates with their sensitivity to LXY18. Collectively, our findings uncover LXY18 as an MYC- enabled mitotic blocker and open a new avenue to selectively target MYC-overexpressing tumor cells without affecting normal cells.

cancer biology↗

bioRxiv2023

Universal protection against SARS-CoV-2 viruses by multivalent mRNA vaccine in mice

The continual emergence of new severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) variants challenges available SARS-CoV-2 vaccines for adequate control of outbreaks. Currently, universal vaccines capable of obviating the need for exact strain matching between mRNA vaccines and circulating viruses are absent. In this study, we designed, manufactured, and evaluated a nucleoside-modified lipid nanoparticle mRNA vaccine, aimed for offering broad-spectrum protection against recent SARS-CoV-2 variants. Additionally, the protection efficiency of monovalent, bivalent, quadrivalent, and XBB.1.5 mRNA vaccines was compared with the proposed universal vaccine. The neutralizing antibody activity against wuhan-1, BA.4/5, XBB.1.5, B.1.1.529, BQ.1.1, EG.5.1 and JN.1 was assessed using enzyme-linked immunosorbent assay, rapid fiber-optic biolayer interferometry-based biosensor, and pseudovirus neutralization test. Our results reveal that the proposed multivalent vaccine affords comprehensive protection against previously circulating, current and previously unidentified SARS-CoV-2 strains.

immunology↗

bioRxiv2023

A 5' UTR Language Model for Decoding Untranslated Regions of mRNA and Function Predictions

The 5 UTR, a regulatory region at the beginning of an mRNA molecule, plays a crucial role in regulating the translation process and impacts the protein expression level. Language models have showcased their effectiveness in decoding the functions of protein and genome sequences. Here, we introduced a language model for 5 UTR, which we refer to as the UTR-LM. The UTR-LM is pre-trained on endogenous 5 UTRs from multiple species and is further augmented with supervised information including secondary structure and minimum free energy. We fine-tuned the UTR-LM in a variety of downstream tasks. The model outperformed the best-known benchmark by up to 42% for predicting the Mean Ribosome Loading, and by up to 60% for predicting the Translation Efficiency and the mRNA Expression Level. The model also applies to identifying unannotated Internal Ribosome Entry Sites within the untranslated region and improves the AUPR from 0.37 to 0.52 compared to the best baseline. Further, we designed a library of 211 novel 5 UTRs with high predicted values of translation efficiency and evaluated them via a wet-lab assay. Experiment results confirmed that our top designs achieved a 32.5% increase in protein production level relative to well-established 5 UTR optimized for therapeutics.

bioinformatics↗