Biology subjects
Publications and source records attributed to Deng, Y..
SummaryAncestryHub is a web version of local ancestry analysis software. It contains four built-in software and a set of built-in population reference panels. AncestryHub will significantly reduce the efforts and requirements for the users on their personal computational skills and computational hardware environment for ancestry analysis. Availabilityhttps://www.ancestryhub.4dgenome.com Contact: lima@4dgenome.com
HTT1a was identified in human and mouse Huntingtons disease brain as the pathogenic exon 1 mRNA generated from aberrant splicing between exon 1 and 2 of HTT that contributes to aggregate formation and neuronal dysfunction.1 Detection of the huntingtin exon 1 protein (HTT1a) has been accomplished with fluorescence-based reporter assays (Meso Scale Discovery, Homogeneous Time Resolved Fluorescence) and immunoprecipitation assays in Huntingtons disease knock-in mice but direct detection in homogenates by gel electrophoresis and western blot assay has been lacking. Subcellular fractions prepared from mouse and human Huntingtons disease brain were separated by gel electrophoresis and probed by western blot with neo-epitope monoclonal antibodies 1B12 and 11G2 directed to the C-terminal eight residues of HTT1a. In caudate putamen of an allelic series of 6 month old Huntingtons disease knock-in mice (Q50, Q80, Q111, Q140 and Q175) HTT1a migration was inversely correlated with CAG repeat length and appeared as a SDS soluble high molecular mass smear in Q111, Q140 and Q175 mice but weakly in Q80 and not in WT mice or Q50 indicating a CAG repeat size threshold for detecting HTT1a. HTT1a immunoreactivity diminished if 1B12 and 11G2 antibodies were preincubated with an eight amino acid peptide containing the C-terminus of HTT1a but not with unrelated peptide sequence. Migration of HTT1a and its high molecular mass smear changed with age in caudate putamen of Q111, Q175 and YAC128 mice. Treating Q111 mice with siRNA to MSH3, a modifier of CAG repeat expansion, significantly reduced levels of the high molecular mass smear indicating that the effects of curbing CAG repeat expansion were quantifiable. A prominent 56-60 kDa doublet detected by 1B12 and 11G2 antibodies in lysates from human Huntingtons disease brain was not blocked by preincubation with C-terminal HTT1a blocking peptide and also appeared in brains of Parkinsons disease patients. 1B12 and 11G2 antibodies did not immunoprecipitate HTT proteins from either Huntingtons disease mouse or human brain lysates using conditions that pulled down full length HTT with anti-HTT antibody 2B7. Altogether these data show that 11G2 and 1B12 antibodies can be used in western blot assays to track and quantify immunoreactive HTT1a levels, solubility, and subcellular localization in Huntingtons disease mouse brain. Abbreviated SummarySapp et al., report that pathogenic exon 1 protein HTT1a is detected in brain of mouse models of Huntingtons disease by direct western blot assay using monoclonal antibodies 11G2 and 1B12. Lowering MSH3 mRNA in the caudate putamen to prevent CAG repeat expansion reduced levels of HTT1a.
Large-scale characterization of gut viral genomes provides strain-resolved insights into host-microbe interactions. However, existing viral genomes are mainly derived from Western populations, limiting our understanding of global gut viral diversity and functional variations necessary for personalized medicine and addressing regional health disparities. Here, we introduce the Chinese Gut Viral Reference (CGVR) set, consisting of 120,568 viral genomes from 3,234 deeply sequenced fecal samples collected nationwide, covering 72,751 viral species-level clusters (VSCs), nearly 90% of which are absent from current databases. Analysis of single nucleotide variations (SNVs) in 233 globally prevalent VSCs revealed that 18.9% showed significant genetic stratification between Chinese and non-Chinese populations, potentially linked to bacterial infection susceptibility. The predicted bacterial hosts of population-stratified viruses exhibit distinct genetic components associated with health-related functions, including multidrug resistance. Additionally, viral strain diversity at the SNV level correlated with human phenotypic traits, such as age and gastrointestinal issues like constipation. Our analysis also indicates that the human gut bacteriome is specifically shaped by the virome, which mediates associations with human phenotypic traits. This foundational resource underscores the unique genetic makeup of the gut virome across populations and emphasizes the importance of recognizing gut viral genetic heterogeneity for deeper insights into regional health implications.
Traditional directed evolution is limited by labor-intensive iterative steps and low-throughput selection and screening. To address these challenges, we developed a growth-coupled continuous directed evolution (GCCDE) approach, enabling automated and efficient enzyme engineering. By linking enzyme activity to bacterial growth and utilizing the MutaT7 system, GCCDE combines in vivo mutagenesis and high-throughput selection of superior enzyme variants in a single process. To validate this approach, we evolved the thermostable enzyme CelB to enhance {beta}-galactosidase activity at lower temperatures while maintaining thermal stability. CelB activity was coupled to the growth of E. coli, allowing variants with improved activity to metabolize lactose more efficiently and promote faster bacterial growth in a minimal medium. Using a continuous culture system, we achieved automated mutagenesis and real-time selection of over 109 variants per culture. Integrating in vitro and in vivo mutagenesis further increased genetic diversity, yielding CelB variants with significantly enhanced low-temperature activity compared to the wild type while preserving thermostability. DNA sequencing identified key mutations likely responsible for improved substrate binding and catalytic turnover. This GCCDE approach is broadly applicable for optimizing diverse enzymes, demonstrating the potential of automated continuous evolution for industrial and research applications. IMPORTANCEEnzyme engineering aims to develop enzymes with improved or novel traits, but traditional methods are slow and require repetitive manual steps. This study presents a faster, automated protein engineering approach. We utilized an in vivo mutagenesis technique, MutaT7 tools, to induce mutations in living bacteria and established a direct link between enzyme activity and bacterial growth. A continuous culture setup was used to enable growth-coupled high-throughput selection of better-performing variants. Bacteria with improved enzymes grew faster, selecting superior variants without manual intervention. Using this method, we engineered CelB with better performance at lower temperatures while maintaining high-temperature stability. The approach is adaptable to many enzymes. It offers a faster and more efficient solution for enzyme engineering. This system enables high-throughput mutagenesis and selection simultaneously, showing the power of automated continuous evolution to advance enzyme engineering.
BackgroundAtypical teratoid rhabdoid tumor (ATRT) is the most common malignant brain tumor in infants, and more than 60% of children with ATRT die from their tumor. ATRT is associated with mutational inactivation/deletion of SMARCB1, a member of the SWI/SNF chromatin remodeling complex, suggesting that epigenetic events play a critical role in tumor development and progression. Moreover, disruption of SWI/SNF allows unopposed activity of epigenetic repressors, which contribute to tumorigenicity. We therefore explored the role of the CoREST repressor complex in ATRT. MethodsWe evaluated the effects of the bifunctional LSD1/HDAC1/2 small molecule CoREST inhibitor, corin, on ATRT tumor cell growth, apoptosis, differentiation, gene expression and chromatin accessibility. ResultsWe found that corin inhibited the growth of ATRT cells regardless of their epigenetic subgroup, and was associated with increased tumor cell apoptosis and differentiation. ATAC-seq showed increases in chromatin accessibility in corin-treated ATRT cells, with changes seen at genes associated with neuronal differentiation and synaptic function. RNA-seq confirmed increased expression of neuronal differentiation genes and decreased DNA replication/cell cycle-associated genes in ATRT cells treated with corin. Corin suppressed orthotopic ATRT tumor growth, leading to significant extension of lifespan. In addition, increased histone acetylation (H3K9ac, H3K27ac) and methylation (H3K4Me1) was seen in corin-treated ATRT orthotopic xenografts, consistent with on-target pharmacodynamics. ConclusionThe CoREST inhibitor, corin, suppresses tumor growth, induces differentiation, and promotes apoptosis in ATRT, leading to significantly increased survival of mice bearing ATRT orthotopic xenografts. Our results suggest a potential application of CoREST complex inhibitors in patients with ATRT. Key PointsO_LICoREST complex inhibition by corin leads to decreased cell growth and increased apoptosis in ATRT C_LIO_LICorin promotes chromatin accessibility and neuronal differentiation in ATRT C_LIO_LICorin inhibits tumor growth and extends lifespan in ATRT animal models C_LI Importance of the StudyLoss of function of SMARCB1 is a hallmark of ATRT which leads to dysfunction of the mammalian SWI/SNF complex and an inability to counteract epigenetic repressor complexes. The CoREST complex functions as a chromatin remodeling complex that represses neuronal differentiation genes during development. Inhibition of the CoREST complex by corin in ATRT leads to decreased tumor cell growth, induction of apoptosis and increased survival of mice bearing ATRT orthotopic xenografts. These changes are associated with increased chromatin accessibility and expression of genes associated with neuronal differentiation. Corin therefore reverses the primary block of differentiation that maintains a stem cell state in ATRT, which contributes to tumorigenesis. These studies significantly improve our understanding of how to therapeutically address the underlying epigenetic drivers of ATRT and support further development of corin for ATRT.
Salmonella is one of the most important bacterial pathogens in the world, causing an estimated 120 million infections each year. There is considerable diversity, with [~]2,600 serovars, but it remains unclear how the genomes have evolved as the species and subspecies of Salmonella differentiate. Here, we have reconstructed the ancient orthologous chromosomes of each major Salmonella lineage and traced their evolutionary process. In total, 911 rearrangement events were identified, with 64% of events occurring in a locus-specific way. Using RNA sequencing and multi-strain association analysis, we demonstrate that genetic rearrangements have a significant effect on gene expression across Salmonella lineages. Moreover, we perform chromosome conformation capture (3C) sequencing analysis, which demonstrates large variations for the organization of ter chromosomal interaction domains among Salmonella lineages. In conclusion, our work delineates the evolutionary trajectory of Salmonella chromosomes, and demonstrates the influence of rearrangements on gene expression profiles and chromosomal conformation. ImportanceThis study reconstructed the ancient orthologous chromosomes of the Salmonella genus, species and well-recognized subspecies, by which the trajectory of gene flow caused by genetic rearrangements was delineated chronologically. The rearrangements show apparent hotspot distribution property and are enriched with genes important for bacterial fitness. Correlation analysis further disclosed the general influence of the rearrangements on gene expression and the organization and conformation of chromosome interaction domains. The results provide new insights on the evolution of Salmonella chromosomes, especially the genetic rearrangements and their epigenetic consequences.
Tumor-host interaction plays a critical role in malignant tumor-induced organ wasting across multiple species. Despite known regulation of regional wasting of individual peripheral organs by tumors, whether and how tumors utilize critical host catabolic hormone(s) to simultaneously induce systemic host wasting, however, is largely unknown. Using the conserved yki3SA-tumor model in Drosophila, we discovered that tumors increase the production of adipokinetic hormone (Akh), a glucagon-like catabolic hormone, to cause systemic host wasting, including muscle dysfunction, lipid loss, hyperglycemia, and ovary atrophy. We next integrated RNAi screening and Gal4-LexA dual expression system to identify that yki3SA-gut tumors secrete Pvf1 to remotely activate its receptor Pvr in Akh-producing cells (APCs), ultimately promoting Akh production. The underlying molecular mechanisms involved the Pvf1-Pvr axis that triggers Mmp2-dependent ECM remodeling of APCs and enhances innervation from the excitatory cholinergic neurons. Interestingly, we also confirmed the similar mechanisms governing tumor-induced glucagon release and organ wasting in mammals. Blockade of either glucagon or PDGFR (homolog of Pvr) action efficiently ameliorated organ wasting in the presence of malignant tumors. Therefore, our results demonstrate that tumors remotely promote neural-associated Akh/glucagon production via Pvf1-Pvr axis to cause systemic host wasting.
Liquid chromatography-mass spectrometry (LC-MS) based proteomics, particularly in the bottom-up approach, relies on the digestion of proteins into peptides for subsequent separation and analysis. The most prevalent method for identifying peptides from data-dependent acquisition (DDA) mass spectrometry data is database search. Traditional tools typically focus on identifying a single peptide per tandem mass spectrum (MS2), often neglecting the frequent occurrence of peptide co-fragmentations leading to chimeric spectra. Here, we introduce MSFragger-DDA+, a novel database search algorithm that enhances peptide identification by detecting co-fragmented peptides with high sensitivity and speed. Utilizing MSFraggers fragment ion indexing algorithm, MSFragger-DDA+ performs a comprehensive search within the full isolation window for each MS2, followed by robust feature detection, filtering, and rescoring procedures to refine search results. Evaluation against established tools across diverse datasets demonstrated that, integrated within the FragPipe computational platform, MSFragger-DDA+ significantly increases identification sensitivity while maintaining stringent false discovery rate (FDR) control. It is also uniquely suited for wide-window acquisition (WWA) data. MSFragger-DDA+ provides an efficient and accurate solution for peptide identification, enhancing the detection of low-abundance co-fragmented peptides. Coupled with the FragPipe platform, MSFragger-DDA+ enables more comprehensive and accurate analysis of proteomics data.
Infections with antimicrobial resistant pathogens, such as Pseudomonas aeruginosa, are a frequent occurrence in healthcare settings. Human P. aeruginosa infections are predominantly caused by a small number of sequence types (ST), such as ST235, ST111, and ST175. Although ST111 is recognized as one of the most prevalent high-risk P. aeruginosa clones worldwide and frequently exhibits multidrug-resistant or extensively drug-resistant phenotypes, the basis for this dominance remains unclear. In this study, we used a genome-wide transposon insertion library screen to discover that the competitive advantage of ST111 strains over certain non-ST111 strains is through production of R pyocins. We confirmed this finding by showing that competitive dominance was lost by ST111 mutants with R pyocin gene deletions. Further investigation showed that sensitivity to ST111 R pyocin (specifically R5 pyocin) is caused by deficiency in the O-antigen ligase waaL, which leaves lipopolysaccharide (LPS) bereft of O antigen, enabling pyocins to bind the LPS core. In contrast, sensitivity of waaL mutants to R1 or R2 pyocins depended on additional genomic changes. In addition, we found the PA14 mutants in lipopolysaccharide biosynthesis (waaL, wbpL, wbpM) that cause high susceptibility to R pyocins also exhibit poor swimming motility. Analysis of 5,135 typed P. aeruginosa strains revealed that several international, high-risk sequence types (including ST235, ST111, and ST175) are enriched for R5 pyocin production, indicating a correlation between these phenotypes and suggesting a novel approach for evaluating risk from emerging prevalent P. aeruginosa strains. Overall, our study sheds light on the mechanisms underlying the dominance of ST111 strains and highlighting the role of waaL in extending spectrum of R pyocin susceptibility.
It was recently reported that a severe ancient bottleneck occurred around 900 thousand years ago in the ancestry of African populations, while this signal is absent in non-African populations. Here, we present evidence to show that this finding is likely a statistical artifact.
A promising drug target, SETDB1, is a dual Kme reader and methyltransferase, which has been implicated in cancer and neurodegenerative disease progression. To help understand the role of the triple Tudor domain (3TD) of SETDB1, its Kme reader, we first identified a low micromolar small molecule ligand, UNC6535, which occupies simultaneously both the TD2 and TD3 reader binding sites. Further optimization led to the discovery of UNC10013, the first covalent 3TD ligand targeting Cys385 of SETDB1. UNC10013 is potent with a kinact/KI of 1.0 x 106 M-1s-1 and demonstrated proteome-wide selectivity. In cells, negative allosteric modulation of SETDB1-mediated Akt methylation was observed after treatment with UNC10013. Therefore, UNC10013 is a potent, selective and cell-active covalent ligand for the 3TD of SETDB1, demonstrating negative allosteric modulator properties and making it a promising tool to study the biological role of SETDB1 in disease progression. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=81 SRC="FIGDIR/small/615363v1_ufig1.gif" ALT="Figure 1"> View larger version (31K): org.highwire.dtl.DTLVardef@6db8e2org.highwire.dtl.DTLVardef@1852ee3org.highwire.dtl.DTLVardef@c95bb6org.highwire.dtl.DTLVardef@1a43951_HPS_FORMAT_FIGEXP M_FIG C_FIG
Typical antibiotic susceptibility testing (AST) of microbial samples is performed in homogeneous cultures in batch environments, which does not account for the highly heterogeneous and dynamic nature of antibiotic responses. The most common mutation found in P. aeruginosa lineages evolved during chronic infections in the human lung, a loss of function of repressor MexZ, increases basal levels of multidrug efflux MexXY, but does not increase resistance by traditional minimal inhibitory concentration (MIC) assays. Here, we use single cell microfluidics to show that P. aeruginosa response to aminoglycosides is highly heterogeneous, with only a subpopulation of cells surviving exposure. In contrast, strains carrying mexZ mutations bypass the lengthy process of MexXY activation, increasing survival to sudden drug exposures and conferring a fitness advantage in fluctuating environments. Building on the data we present here, we propose a simple "Response Dynamics" assay to quantify the rate of population-level recovery to drug exposures across strains. We used this assay to profile a representative panel of 49 P. aeruginosa strains from diverse environments, showing that the presence of mexZ mutations correlates with faster population recovery from exposures to aminoglycosides, and thus confers an advantage to cells exposed to a sudden, large dose of antibiotic. We propose that the Response Dynamics assay can be used alongside MIC assays for profiling of antibiotic sensitivity to better predict clinical outcomes from in vitro sensitivity/resistance profiles. SignificanceCommon mutations affecting the regulation of antibiotic resistance in bacterial pathogens often do not increase resistance by traditional measures. However, antibiotic resistance is typically measured in stable cultures, without accounting for fluctuations in drug concentration. Here, we show that P. aeruginosa response to aminoglycosides is highly heterogeneous, and that the most common mutation found in clinical isolates improves resistance by increasing single-cell survival to drug exposures. Therefore, the success of antibiotic treatments depends on the dynamics of drug delivery and the heterogeneous activation of microbial responses. We then develop an assay to measure resistance in a dynamic context, which captures this overlooked aspect of antibiotic resistance and can be used alongside traditional measures in the profiling of clinical isolates.
The phenotypic and functional states of a cell are modulated by a complex interactive molecular hierarchy of multiple omics layers, involving the genome, epigenome, transcriptome, proteome, and metabolome. Spatial omics approaches have enabled the capture of information from different molecular layers directly in the tissue context. However, current technologies are limited to map one to two modalities at the same time, providing an incomplete representation of cellular identity. Such data is inadequate to fully understand complex biological systems and their underlying regulatory mechanisms. Here we present spatial-Mux-seq, a multi-modal spatial technology that allows simultaneous profiling of five different modalities, including genome-wide profiles of two histone modifications and open chromatin, whole transcriptome, and a panel of proteins at tissue scale and cellular level in a spatially resolved manner. We applied this technology to generate multi-modal tissue maps in mouse embryos and mouse brains, which discriminated more cell types and states than unimodal data. We investigated the spatiotemporal relationship between histone modifications, chromatin accessibility, gene and protein expression in neuron differentiation revealing the relationship between tissue organization, function, and gene regulatory networks. We were able to identify a radial glia spatial niche and revealed spatially changing gradient of epigenetic signals in this region. Moreover, we revealed previously unappreciated involvement of repressive histone marks in the mouse hippocampus. Collectively, the spatial multi-omics approach heralds a new era for characterizing tissue and cellular heterogeneity that single modality studies alone could not reveal.
While a subset of patients with metastatic melanoma achieves durable responses to immune checkpoint blockade (ICB) therapies, the majority ultimately exhibit either innate or acquired resistance to these treatments. However, the molecular mechanisms underlying resistance to ICB therapies remain elusive and are warranted to elucidate. Here, we comprehensively investigated the tumor and tumor immune microenvironment (TIME) of paired pre- and post-treatment tumor specimens from metastatic melanoma patients who were primary or secondary resistance to anti-CTLA-4 and/or anti-PD-1/PD-L1 therapies. Differentially expressed gene (DEG) analysis and single-sample gene set enrichment analysis (ssGSEA) with transcriptomic data identified cell cycle and c-MYC signaling as pathway-based resistance signatures. And weighted gene co-expression network analysis (WGCNA) revealed the activation of a cross-resistance meta-program involving key signaling pathways related to tumor progression in ICB resistant melanoma. Moreover, spatially-resolved, image-based immune monitoring analysis by using NanoStrings digital spatial profiling (DSP) and Cyclic Immunofluorescence (CyCIF) showed infiltration of suppressive immune cells in the tumor microenvironment of melanoma with resistance to ICB therapies. Our study reveals the molecular mechanisms underlying resistance to ICB therapies in patients with metastatic melanoma by conducting such integrated analyses of multi-dimensional data, and provides rationale for salvage therapies that will potentially overcome resistance to ICB therapies. Statement of translational relevanceThis study paves the way for the creation of innovative therapeutic strategies, aimed at subverting resistance to immune checkpoint blockade (ICB) therapies in metastatic melanoma patients. By unraveling the specific molecular mechanisms underlying resistance, scientists can design effective alternative treatments that target pathways such as pathways associated with cell cycle dysregulation and c-MYC signaling. Furthermore, through the application of advanced immune monitoring techniques such as NanoString Digital Spatial Profiling (DSP) and Cyclic Immunofluorescence (CyCIF), this study has significantly enriched our understanding of the tumor microenvironment. This enhanced characterization facilitates the discovery of potential biomarkers that may forecast a patients response to ICB treatment. Ultimately, these advancements could potentially refine patient outcomes and foster the development of more personalized cancer treatments in the future.
Symmetry breaking is a fundamental process that underlies key cellular behaviors such as cell polarity and migration, but the mechanism - how a uniform plasma membrane spontaneously transitions to an asymmetric state - is still unknown. Here in this study, using a combination of Dictyostelium amoeba, multiple mammalian leukocytes, and human cancer cells and 3D organoid systems, we monitored the localization dynamics of RasGTP and PIP5K, dissected the effects of eliminating, conditionally increasing, or optogenetically manipulating membrane PIP5K levels, screened for key regulators of Ras activation, and tracked single-molecules of PIP5K. Our data converge on a core biochemical circuit involving mutually inhibitory interactions between PIP5K and RasGTP that is both necessary and sufficient for symmetry breaking. The process is initiated by stochastic, localized Ras activation coupled with a decrease in the lifetime of the association of PIP5K with the membrane. A resulting localized reduction in PI(4,5)P2 facilitates the recruitment of a RasGEF to the corresponding domain, amplifying RasGTP production through a positive feedback loop. Dissociated PIP5K relocates to other membrane regions, where it suppresses Ras activation. These events separate the membrane into distinct active and inactive zones, even when receptor inputs or cytoskeletal activities are absent. This same core biochemical circuit controls the spatial organization of downstream PI3K/Akt/Rac signaling and actin/actomyosin activities, which generate the localized protrusions to define polarity and migration mode. While many models have been forwarded to explain polarity or symmetry breaking, the PIP5K-Ras mutually inhibitory bistable circuit we present here is the first universal molecular mechanism to explain the initiation of asymmetry as well as subsequent polarity and migration.
The fibroblast (FB), cardiomyocyte (CM), and vascular endothelial cell (Vas_EC) are the three major cell types in the heart, yet their relationships during development are largely unexplored. To address this gap, we employed RNA staining of the FB marker gene Col1a1 together with the CM marker gene Actn2 and the Vas_EC marker gene Cdh5 at different stages. This approach enabled us to discern the anatomical pattern of cardiac FBs and identify approximately one EC and four CMs directly interacting with each FB. Molecularly, through the analysis of single-cell mRNA sequencing (scRNA-seq) data, we unveiled collagen as the top signaling molecule derived from FBs influencing CM and Vas_EC development. Subsequently, we used a Pdgfra-CreER controlled diphtheria toxin A (DTA) system to ablate the FBs at different stages. We found that the ablation of FBs disrupted myocardium and vasculature development and led to embryonic heart defects. Using scRNA-seq, we further profiled the ablated hearts and identified molecular defects in their ventricular CMs and Vas_ECs compared to control hearts. Moreover, we identified a reduction of collagen in the ablated hearts and predicted collagen as the major signaling pathway regulating the differentially expressed genes in the ablated ventricular CMs. Finally, we performed both short-term and long-term fibroblast ablation at the neonatal stage. We found that short-term ablation caused a reduction in collagen and Vas_EC density, while long-term ablation may induce compensatory collagen expression without causing heart function reduction. In summary, our study has identified the function of fibroblasts in regulating myocardium and vasculature development and implicated an important role for the collagen pathway in this process.
Activity cliffs, which refer to pairs of molecules that are structurally similar but show significant differences in their potency, can lead to model representation collapse and make the model challenging to distinguish them. Our research indicates that as molecular similarity increases, graph-based methods struggle to capture these nuances, whereas imagebased approaches effectively retain the distinctions. Thus, we developed MaskMol, a knowledge-guided molecular image selfsupervised learning framework. MaskMol accurately learns the representation of molecular images by considering multiple levels of molecular knowledge, such as atoms, bonds, and substructures. By utilizing pixel masking tasks, MaskMol extracts fine-grained information from molecular images, overcoming the limitations of existing deep learning models in identifying subtle structural changes. Experimental results demonstrate MaskMols high accuracy and transferability in activity cliff estimation and compound potency prediction across 20 different macromolecular targets, outperforming 25 state-of-the-art deep learning and machine learning approaches. Visualization analyses reveal MaskMols high biological interpretability in identifying activity cliff-relevant molecular substructures. Notably, through MaskMol, we identified candidate EP4 inhibitors that could be used to treat tumors. This study not only raises awareness about activity cliffs but also introduces a novel method for molecular image representation learning and virtual screening, advancing drug discovery and providing new insights into structure-activity relationships (SAR). Code is available at https://github.com/ZhixiangCheng/MaskMol.
The Mediator complex regulates various aspects of hematopoietic development, but whether composition of the Mediator complex undergoes dynamic changes for diversifying transcription and functional outputs is unknown. Here, we found that MED26, a subunit in the core Mediator complex, played a distinctive role in facilitating transcription pausing essential for erythroid development. While most Mediator subunits drastically decreased during this process, MED26 remained relatively abundant. Intriguingly, in the early stages, more than half of MED26 occupancy sites did not co-localize with MED1, a representative Mediator subunit, suggesting these subunits exert context-dependent gene regulation. We revealed that MED26-enriched loci were associated with RNA polymerase pausing. MED26 manifested a markedly preferential recruitment of pausing-related factors, leading to an increase in Pol pausing critical for genome-wide transcription repression during erythropoiesis. Moreover, MED26 exhibited pronounced condensate-forming capability, which was necessary for its function in promoting erythropoiesis and recruiting pausing-related factors. Collectively, this study provides mechanistic insights into the functional coordination of distinct Mediator subunits during development and highlights the switch of transcription condensates towards a MED26 enriched form, which modulates transcription pausing to facilitate transcription repression and erythroid development.