Biology subjects

Yao, S.

Publications and source records attributed to Yao, S..

At least 55 records · Page 3Linked to original sources

bioRxiv2025

BRCA1-A and LIG4 complexes mediate ecDNA biogenesis and cancer drug resistance

Extrachromosomal circular DNA (ecDNA) are commonly produced within the nucleus to drive genome dynamics and heterogeneity, enabling cancer cell evolution and adaptation. However, the mechanisms underlying ecDNA biogenesis remain poorly understood. Here using genome-wide CRISPR screening in human cells, we identified the BRCA1-A and the LIG4 complexes mediate ecDNA production. Following DNA fragmentation, the upstream BRCA1-A complex protects DNA ends from excessive resection, promoting end-joining for circularization. Conversely, the MRN complex, which mediates end resection and thus antagonizes the BRCA1-A complex, suppresses ecDNA formation. Downstream, LIG4 conservatively catalyzes ecDNA production in Drosophila and mammals, with patient tumor ecDNA harboring junctions marked by LIG4 activity. Notably, disrupting LIG4 or BRCA1-A in cancer cells impairs ecDNA-mediated adaptation, hindering resistance to both chemotherapy and targeted therapies. Together, our study reveals the roles of the LIG4 and BRCA1-A complexes in ecDNA biogenesis, and uncovers new therapeutic targets to block ecDNA-mediated adaptation for cancer treatment.

cancer biology↗

bioRxiv2025

Technical and biological sources of noise confound multiplexed enhancer AAV screening

Cis-acting regulatory enhancer elements are valuable tools for gaining cell type-specific genetic access. Leveraging large chromatin accessibility atlases, putative enhancer sequences can be identified and deployed in adeno-associated virus (AAV) delivery platforms. However, a significant bottleneck in enhancer AAV discovery is charting their detailed expression patterns in vivo, a process that currently requires gold-standard one-by-one testing. Here we present a barcoded multiplex strategy for screening enhancer AAVs at cell type resolution using single cell RNA sequencing and taxonomy mapping. We executed a proof-of-concept study using a small pool of validated enhancer AAVs expressing in a variety of neuronal and non-neuronal cell types across the mouse brain. Unexpectedly, we encountered substantial technical and biological noise including chimeric packaging products, necessitating development of novel techniques to accurately deconvolve enhancer expression patterns. These results underscore the need for improved methods to mitigate noise and highlight the complexity of enhancer AAV biology in vivo.

neuroscience↗

bioRxiv2024

Modulating Cellular Deformability via 3D Dextran Hydrogel Cultivation to Regulate the Microcirculation of Mesenchymal Stem Cells in Murine Spleen and Liver

For mesenchymal stem cell (MSC) therapy to be effective, the vascular system may be used to deliver and steer the cells to the target tissue. However, the expanded MSCs in petri dishes are usually have a limited deformability and commonly excluded by the capillary networks when homing to the downstream organs via microcirculation. Here, we propose to utilize specially designed 3D dextran hydrogels and tuning the microscopic heterogeneity of hydrogel composition to make the administrated cells mechanically comply with the structure and mechanics of the capillary. The cell deformability in petri dishes, microcosmically homogeneous (HOM) and heterogeneous (HET) dextran hydrogels were investigated in vitro by measuring cell moduli through atomic force microscope (AFM), analyzing the expression of cytoskeletal protein via flow cytometry and fluorescent imaging. The in vitro experimental results demonstrate a progressive increase in cell deformability from 2D dishes, to HOM-hydrogel derived cells, and then to HET-hydrogel derived cells. The in vivo mouse experiment indicates the cells could deform accordingly and pass through easily with reduced resistance inside the mouse organs. It is suggested that the main destination of hMSC microcirculation could be selected between the spleen and liver of mice, by tuning cell mechanics that depends on the stimulus from microcosmically HOM or HET hydrogel, which lays a potential foundation for the mechanically modified MSC therapy for lesions in the organs.

biophysics↗

bioRxiv2024

Global analysis of protein and small-molecule substrates of ubiquitin-like proteins (UBLs)

AbstractUbiquitin-like proteins (UBLs) constitute a family of evolutionarily conserved proteins that share similarities with ubiquitin in 3D structures and modification mechanisms. For most UBLs including Small-Ubiquitin-like Modifiers (SUMO), their modification sites on substrate proteins cannot be identified using the mass spectrometry-based method that has been successful for identifying ubiquitination sites, unless a UBL protein is mutated accordingly. To identify UBL modification sites without having to mutate UBL, we have developed a dedicated search engine pLink-UBL on the basis of pLink, a software tool for identification of cross-linked peptide pairs. pLink-UBL exhibited superior precision, sensitivity, and speed than "make-do" search engines such as MaxQuant, pFind, and pLink. For example, compared to MaxQuant, pLink-UBL increased the number of identified SUMOylation sites by 50 [~] 300% from the same datasets. Additionally, we present a method for identifying small-molecule modifications of UBLs. This method involves antibody enrichment of a UBL C-terminal peptide following enrichment of a UBL protein, followed by LC-MS/MS analysis and a pFind 3 blind search to identify unexpected modifications. Using this method, we have discovered non-protein substrates of SUMO, of which spermidine is the major one for fission yeast SUMO Pmt3. Spermidine can be conjugated to the C-terminal carboxylate group of Pmt3 through its N1 or also likely, N8 amino group in the presence of SUMO E1, E2, and ATP. Pmt3-spermidine conjugation does not require E3 and can be reversed by SUMO isopeptidase Ulp1. SUMO-spermidine conjugation is present in mice and humans. Also, spermidine can be conjugated to ubiquitin in vitro by E1 and E2 in the presence of ATP. The above observations suggest that spermidine may be a common small molecule substrate of SUMO and possibly ubiquitin across eukaryotic species. HighlightsO_LIA specialized software pLink-UBL enables precise identification of UBL modification sites on protein substrates. C_LIO_LIpFind 3 blind search enables identification of unexpected small-molecule substrates of a UBL protein. C_LIO_LISpermidine is a small molecule substrate of fission yeast SUMO Pmt3 as well as mammalian SUMO proteins. C_LIO_LIThe C-terminal carboxyl group of Pmt3 can be attached to the N1 or likely also the N8 amino group of spermidine in the presence of E1, E2, and ATP, and can be detached by SUMO isopeptidase. C_LI

biochemistry↗

bioRxiv2024

Abnormal eNK cells contribute to endometrial fibrosis in intrauterine adhesions patients

Endometrial natural killer cells (eNK) at the proliferative phase are one of the major lymphocytes in the endometrium, but their phenotype and function remain elucidated. Here, we demonstrated that eNK cells had three subtypes and were different from peripheral blood NK (pbNK) and decidual NK (dNK) cells. They played an important role in maintaining the homeostasis of the endometrium through preventing endometrial stromal cells (ESCs) from the differentiation into myofibroblasts physiologically. However, in the fibrotic endometria of patients with intrauterine adhesions (IUA), eNK cells were significantly decreased and highly negatively correlated with an increase in myofibroblasts. In vitro experiments showed that eNK cells could inhibit the differentiation of ESCs into myofibroblasts and promote the dedifferentiation of myofibroblasts, in which the main effector molecules from eNK cells was Prostaglandin D2 (PGD2). PGD2 downregulated the expression of ZNF521 to decrease profibrotic protein synthesis. Furthermore, we confirmed the anti-endometrial fibrosis effect of eNK cells and its mechanism in an IUA-like murine model. These findings reveal an important role of eNK cells in endometrial homeostasis and provide potential therapeutic approaches for IUA patients.

immunology↗

bioRxiv2024

CIWARS: a web server for waterborne antibiotic resistance surveillance using longitudinal metagenomic data

The rise of antibiotic resistance (AR) is a major global health crisis, exacerbated by the overuse and misuse of antibiotics, leading to the rapid spread of antibiotic resistance genes (ARGs) in bacterial pathogens. This phenomenon poses significant threats to human and animal health, food security, and economic stability. Water bodies, particularly wastewater treatment plants (WWTPs), serve as critical reservoirs for ARGs, creating environments that favor the proliferation of resistant bacteria. Wastewater-based surveillance (WBS) has emerged as a cost-effective strategy for monitoring AR at the population level, providing real-time data to guide public health and policy decisions. Despite advancements in WBS, there are no comprehensive online analytical platforms for continuous environmental AR surveillance. This paper introduces CIWARS, a web server designed for AR analyses of longitudinal metagenomic data. CIWARS offers comprehensive ARG profiling, taxonomic annotation, and anomalous AR risk points detection. We demonstrate its capabilities through an interactive temporal data visualization, showcasing its potential for enhancing AR risk monitoring and guiding effective mitigation strategies. CIWARS is broadly applicable to longitudinal metagenomic data generated from any environment and aims to support global efforts in addressing the AR crisis by providing cyberinfrastructure for continuous AR surveillance. The web server is freely available at https://ciwars.cs.vt.edu/.

bioinformatics↗

bioRxiv2024

In vitro inhibition of the CFTR ion channel in the cervix thickens cervical mucus

Cervical mucus changes throughout the menstrual cycle in response to hormonal fluctuations, regulating access of sperm and pathogens to the reproductive tract. CFTR is an anion channel that plays a critical role in mediating epithelial mucus secretions. Primary endocervical cells obtained from rhesus macaques Macaca mulatta were cultured using conditional reprogramming and treated with vehicle controls or CFTR inhibitors. In order to measure changes in hydration and viscosity of secreted mucus, we adapted two airway mucus assays, airway surface liquid and particle-tracking microrheology, for our endocervical culture system. Endocervical cells treated with CFTR inhibitors demonstrated dehydrated, thicker mucus secretions compared to controls in both assay outputs. Our studies suggest that CFTR may be an important mediator of fertility changes and provide experimental evidence for the infertility phenotype seen in women with cystic fibrosis. Additionally, assays developed in these studies provide new endpoints for assessing cervical mucus changes in vitro. Summary SentenceInhibition of CFTR, a key epithelial ion channel in primary endocervical cells leads to dehydrated, thickened mucus production in vitro

cell biology↗

bioRxiv2024

Differential Impact of Multiple Sensory Deprivation on Spatial-coding Cells in Medial Entorhinal Cortex

Spatial navigation depends on anchoring internal spatial maps to external environments, guided by sensory cues such as visual and tactile information. The Medial Entorhinal Cortex (MEC) is crucial for integrating these sensory inputs during the formation of spatial maps. While the responsiveness of many spatially modulated cells to visual stimuli is well-established, the role of tactile sensation in spatial representation is less understood. Rodents primarily gather tactile information through their whiskers, which provide essential spatial and textural details via whisking movements, potentially vital for constructing accurate spatial maps. In our study, we employed miniature two-photon microscopy to monitor neural activity in the MEC of freely moving male mice subjected to visual and tactile deprivation. We found that after spatial maps were established under visual guidance, head direction and border cells were disrupted by both light deprivation and whisker trimming, whereas grid and spatial cells were primarily dependent on visual input. In complete darkness with enriched tactile cues, however, all spatially modulated cell types could anchor their tuning to environmental tactile cues, and whisker trimming markedly impaired grid and spatial cell representations. A subset of neurons remained stable under combined deprivation, possibly relying on boundary-related tactile inputs from direct environmental contact. Additionally, we identified certain MEC neurons whose activity correlated with whisker movements, suggesting a potential role in integrating tactile information into spatial representations. These findings demonstrate that the MEC flexibly integrates multiple sensory inputs to sustain spatial coding and highlight a previously underappreciated contribution of tactile information to spatial navigation.

neuroscience↗

bioRxiv2024

Enhancer AAVs for targeting spinal motor neurons and descending motor pathways in rodents and macaque

Experimental access to cell types within the mammalian spinal cord is severely limited by the availability of genetic tools. To enable access to lower motor neurons (LMNs) and LMN subtypes, we generated single cell multiome datasets from mouse and macaque spinal cords and discovered putative enhancers for each neuronal population. We cloned these enhancers into adeno-associated viral vectors (AAVs) driving a reporter fluorophore and functionally screened them in mouse. We extensively characterized the most promising candidate enhancers in rat and macaque and developed an optimized pan LMN enhancer virus. Additionally, we generated derivative viruses expressing iCre297T recombinase or ChR2-EYFP for labeling and functional studies, and we created a single vector with combined enhancer elements to achieve simultaneous labeling of layer 5 extratelencephalic projecting (ET) neurons and LMNs. This unprecedented LMN toolkit will enable future investigations of cell type function across species and potential therapeutic interventions for human neurodegenerative diseases.

neuroscience↗

bioRxiv2024

Single field evolution rule governs the dynamics of representational drift in mouse hippocampal dorsal CA1 region.

How the brain reconciles dynamism with stability to balance learning and reliable memory storage has not yet been fully understood. To address the critical question, we longitudinally recorded place cells in the hippocampal dorsal CA1 region over 7 to 56 days, utilizing multiple goal-oriented navigation paradigms across various environments. We found that over 80% of place cells displayed multiple fields, undergoing complex evolution events including field disappearance, formation, and retention. Place fields from the same neuron showed limited coordination ([~]5%), with a preference for synchronized changes. We further uncovered the single field evolution rule: the longer a field remains active, the more likely it is to continue being active; conversely, the longer a field remains inactive, the less likely it is to recover the future fate of a place field depends on its past activity. Mathematical modeling revealed that this rule sufficiently demonstrates the growing stability of the dCA1 spatial representation at the population level.

neuroscience↗

bioRxiv2024

CGPA: multi-context insights from the cancer gene prognosis atlas

Cancer transcriptomic data are used extensively to interrogate the prognostic value of targeted genes, yet basic scientists and clinicians have predominantly relied on univariable survival analysis for this purpose. This method often fails to capture the full prognostic potential and contextual relevance of the genes under study, inadvertently omitting a group of genes we term univariable missed-opportunity prognostic (UMOP) genes. Recognizing the complexity of revealing multifaceted prognostic implications, especially when extending the analysis to include various covariates and thresholds, we present the Cancer Gene Prognosis Atlas (CGPA). This platform greatly enhances gene-centric biomarker research across cancer types by offering an interactive and user-friendly interface for highly customized, in-depth prognostic analysis. CGPA notably supports data-driven exploration of gene pairs and gene-hallmark relationships, elucidating key composite biological mechanisms like synthetic lethality and immunosuppression. It further expands its capabilities to assess multi-gene panels using both public and user-provided data, facilitating a seamless mechanism-to-machine analysis. Additionally, CGPA features a designated portal for discovering prognostic gene modules using curated cancer immunotherapy data. Ultimately, CGPAs comprehensive, accessible tools allow cancer researchers, including those without statistical expertise, to precisely investigate the prognostic landscape of genes, customizing the model to fit specific research hypotheses and enhancing biomarker discovery and validation through a synergy of mechanistic and data-driven strategies. SignificanceCGPA is a streamlined, interactive platform for multi-context gene-centric prognostic analysis, simplifying biomarker discovery and validation in oncology for clinicians and basic scientists, and bridging a critical gap in translational cancer research.

bioinformatics↗

bioRxiv2024

Xist RNA binds select autosomal genes and depends on Repeat B to regulate their expression

Xist, a pivotal player in X chromosome inactivation (XCI), has long been perceived as a cis-acting long noncoding RNA that binds exclusively to the inactive X chromosome (Xi). However, Xists ability to diffuse under select circumstances has also been documented, leading us to suspect that Xist RNA may have targets and functions beyond the Xi. Here, using female mouse embryonic stem cells (ES) and mouse embryonic fibroblasts (MEF) as models, we demonstrate that Xist RNA indeed can localize beyond the Xi. However, its binding is limited to [~]100 genes in cells undergoing XCI (ES cells) and in post-XCI cells (MEFs). The target genes are diverse in function but are unified by their active chromatin status. Xist binds discretely to promoters of target genes in neighborhoods relatively depleted for Polycomb marks, contrasting with the broad, Polycomb-enriched domains reported for human XIST RNA. We find that Xist binding is associated with down-modulation of autosomal gene expression. However, unlike on the Xi, Xist binding does not lead to full silencing and also does not spread beyond the target gene. Over-expressing Xist in transgenic ES cells similarly leads to autosomal gene suppression, while deleting Xists Repeat B motif reduces autosomal binding and perturbs autosomal down-regulation. Furthermore, treating female ES cells with the Xist inhibitor, X1, leads to loss of autosomal suppression. Altogether, our findings reveal that Xist targets [~]100 genes beyond the Xi, identify Repeat B as a crucial domain for its in-trans function in mice, and indicate that autosomal targeting can be disrupted by a small molecule inhibitor.

genomics↗

bioRxiv2024

GPS-Net: discovering prognostic pathway modules based on network regularized kernel learning

The search for prognostic biomarkers capable of predicting patient outcomes, by analyzing gene expression in tissue samples and other molecular profiles, remains largely on single-gene-based or global-gene-search approaches. Gene-centric approaches, while foundational, fail to capture the higher-order dependencies that reflect the activities of co-regulated processes, pathway alterations, and regulatory networks, all of which are crucial in determining the patient outcomes in complex diseases like cancer. Here, we introduce GPS-Net, a computational framework that fills the gap in efficiently identifying prognostic modules by incorporating the holistic pathway structures and the network of gene interactions. By innovatively incorporating advanced multiple kernel learning techniques and network-based regularization, the proposed method not only enhances the accuracy of biomarker and pathway identification but also significantly reduces computational complexity, as demonstrated by extensive simulation studies. Applying GPS-Net, we identified key pathways that are predictive of patient outcomes in a cancer immunotherapy study. Overall, our approach provides a novel framework that renders genome-wide pathway-level prognostic analysis both feasible and scalable, synergizing both mechanism-driven and data-driven for precision genomics.

bioinformatics↗

bioRxiv2024

SmartImpute: A Targeted Imputation Framework for Single-cell Transcriptome Data

Single-cell RNA sequencing (scRNA-seq) has revolutionized our understanding of cellular heterogeneity and tissue transcriptomic complexity. However, the high frequency of dropout events in scRNA-seq data complicates downstream analyses such as cell type identification and trajectory inference. Existing imputation methods address the dropout problem but face limitations such as high computational cost and risk of over-imputation. We present SmartImpute, a novel computational framework designed for targeted imputation of scRNA-seq data. SmartImpute focuses on a predefined set of marker genes, enhancing the biological relevance and computational efficiency of the imputation process while minimizing the risk of model misspecification. Utilizing a modified Generative Adversarial Imputation Network architecture, SmartImpute accurately imputes the missing gene expression and distinguishes between true biological zeros and missing values, preventing overfitting and preserving biologically relevant zeros. To ensure reproducibility, we also provide a function based on the GPT4 model to create target gene panels depending on the tissue types and research context. Our results, based on scRNA-seq data from head and neck squamous cell carcinoma and human bone marrow, demonstrate that SmartImpute significantly enhances cell type annotation and clustering accuracy while reducing computational burden. Benchmarking against other imputation methods highlights SmartImputes superior performance in terms of both accuracy and efficiency. Overall, SmartImpute provides a lightweight, efficient, and biologically relevant solution for addressing dropout events in scRNA-seq data, facilitating deeper insights into cellular heterogeneity and disease progression. Furthermore, SmartImputes targeted approach can be extended to spatial omics data, which also contain many missing values.

bioinformatics↗

bioRxiv2024

Cellular and Molecular Changes During Aging in MEC: Unveiling the Role of Bglap3 Neurons in Cognitive Aging

Aging-correlated cognitive declines, including deficiencies in spatial orientation and memory, may reflect dysfunction in the hippocampus and medial entorhinal cortex (MEC). However, aging-related changes in MEC at the cellular and molecular levels remain unclear. In this study, we found fewer grid cells with reduced spatial stability in old mice. We compared gene expression profiles between young and old mice using 10x Genomics Visium technology. Among 1664 differentially expressed genes, we discovered Bglap3, a marker gene for subpopulation in MEC Layer III with decreased cell number with age. Silencing of Bglap3+ neurons in young mice impaired the spatial tuning of neurons in MEC and the spatial learning of a new platform location in water maze. These findings help us to understand the cellular and molecular changes in the MEC in healthy aging animals and the changes of Bglap3+ cells in old mice indicating a possible cause of aging-related MEC deficiency.

neuroscience↗

bioRxiv2024

Single-cell dissection of prognostic architecture and immunotherapy response in Helicobacter pylori infection associated gastric cancer

Most of the gastric cancer (GC) worldwide are ascribed to Helicobacter pylori (H. pylori) infections, which have a detrimental effect on the immunotherapys efficacy. Comprehensively dissecting the key cell players and molecular pathways associated with cancer immunotherapies is critical for developing novel therapeutic strategies against H. pylori infection associated GC. We performed a comprehensive single-cell transcriptome analysis of nine GC with current H. pylori infection (HpGC), three GC with previous H. pylori infection (ex-HpGC), six GC without H. pylori infection (non-HpGC), and six healthy controls (HC). We also investigated key cell players and molecular pathways associated with GC immunotherapy outcomes. We revealed the molecular heterogeneity of different cell components in GC including epithelium, immune cells, and cancer-associated fibroblasts (CAFs) at the single-cell level. The malignant epithelium of HpGC exhibited high expression level of inflammatory and epithelial-mesenchymal transition (EMT) signature, HpGC and ex-HpGC were enriched with VEGFA+ angiogenic tumor-associated macrophages (Angio-TAM) and IL11+ inflammatory CAF (iCAF), characterized by high expression levels of NECTIN2 and VEGFA/B. Additionally, we found significant correlations between the abundance of iCAF with Angio-TAM and TIGIT+ suppressive T cells, and iCAF interacted with Angio-TAM through the VEGF and ANGPTL angiogenic pathways. We also developed an immune signature and angiogenic signature and demonstrated that the iCAF abundance and angiogenic signature could predict poor immunotherapy outcomes in GC. We revealed the transcriptome characteristics and heterogeneity of various cellular constituents of HpGC and demonstrated that a synergistic combination of immunotherapy and anti-angiogenic targeted therapy may be an effective therapeutic modality for HpGC.

cancer biology↗

bioRxiv2024

Biomarker Identification by Proteomic Analysis of Vitreous Humor and Plasma in Diabetic Retinopathy

ImportanceIdentify detectable plasma and/or vitreous signals to potentially predict diabetic retinopathy (DR) progression for earlier disease intervention. ObjectiveTo determine the mediators and potential disease progression biomarkers of DR in vitreous humor (VH) and plasma samples using the SomaScan proteome profiling platform. DesignDifferential expression analysis was conducted on VH and plasma samples using the SomaScan Assay. SettingA non-interventional study conducted to collect and analyze VH and plasma samples from patients with diabetic retinopathy. ParticipantsSamples from DR (60 nonproliferative diabetic retinopathy/NPDR, 60 proliferative diabetic retinopathy/PDR) and 60 control patients were collected. Main outcomes and MeasuresDifferentially expressed proteins between disease and control groups were identified. Pathway enrichment analysis was conducted to identify significantly perturbed pathways in DR. Finally, a random forest model was used to identify predictive biomarkers of disease progression. ResultsSomaScan v3 is a pooled aptamer hybridization assay using 5080 SOMAmers to probe over 4100 proteoforms in VH and plasma samples from 3 groups (control, NPDR, and PDR). The most profound protein content change was observed in the VH samples of PDR patients, while minimal changes were measured in plasma samples, highlighting the regionality of PDR pathogenesis. Many key molecules and molecular pathways such as VEGF-A, erythropoietin, and inflammation-associated proteins implicated in DR were significantly affected in the VH of PDR patients. In addition to the classic pathways (hypoxia, immune response, mTORC1 signaling) known to be involved in PDR, novel signaling pathways, including HEME metabolism and adipogenesis, were identified in VH samples. Application of a machine learning algorithm identified a panel of plasma PDR predictive biomarkers and revealed SCARA5 as the top one based on the largest average Gini decrease in the model. ConclusionOur study identified profound alteration of protein expression and molecular pathways in the VH of PDR patients, supporting the key role of local pathogenic changes in DR progression compared to systemic factors. Although the systemic changes related to DR were small, a few disease progression predictive candidate biomarkers (SCARA5, PTK7, FAM3Band FAM3D) were identified, prompting further investigation. Key PointsQuestion: Are plasma/ vitreous humor (VH) proteins predictive of diabetic retinopathy (DR) progression? Findings: This study identifies substantial protein changes in the VH of proliferative diabetic retinopathy (PDR) patients, while early nonproliferative DR (NPDR) patients show minimal change. We identify multiple proteins linked to angiogenesis, inflammation, immune cells (microglia/macrophage/neutrophil), and leukostasis associated with PDR and reveal a potential plasma panel of disease progression (from NPDR to PDR) biomarkers (SCARA5, PTK7, FAM3B, FAM3D). Meaning: Identified disease progression predictive biomarkers permits potential development of prognostic tools to identify individuals most at risk for PDR progression and offering reduced disease burden by earlier intervention.

physiology↗

bioRxiv2024

Data-driven fine-grained region discovery in the mouse brain with transformers

Spatial transcriptomics offers unique opportunities to define the spatial organization of tissues and organs, such as the mouse brain. We address a key bottleneck in the analysis of organ-scale spatial transcriptomic data by establishing a workflow for self-supervised spatial domain detection that is scalable to multimillion cell datasets. This workflow uses a self-supervised framework for learning latent representations of tissue spatial domains or niches. We use a novel encoder-decoder architecture, which we named CellTransformer, to hierarchically learn higher-order tissue features from lower-level cellular and molecular statistical patterns. Coupling our representation learning workflow with minibatched GPU-accelerated clustering algorithms allows us to scale to multi-million cell MERFISH datasets where other methods cannot. CellTransformer is effective at integrating cells across tissue sections, identifying domains highly similar to ones in existing ontologies such as Allen Mouse Brain Common Coordinate Framework (CCF) while allowing discovery of hundreds of uncataloged areas with minimal loss of domain spatial coherence. CellTransformer domains recapitulate previous neuroanatomical studies of areas in the subiculum and superior colliculus, and characterize putatively uncataloged subregions in subcortical areas which currently lack subregion annotation. CellTransformer is also capable of domain discovery in whole-brain Slide-seqV2 datasets. Our workflows enable complex multi-animal analyses, achieving nearly perfect consistency of up to 100 spatial domains in a dataset of four individual mice with nine million cells across more than 200 tissue sections. CellTransformer advances the state of the art for spatial transcriptomics, by providing a performant solution for detection of fine-grained tissue domains from spatial transcriptomics data.

bioinformatics↗