Biology subjects

Li, S.

Publications and source records attributed to Li, S..

At least 73 records · Page 4Linked to original sources

bioRxiv2026

Reprogramming VHL with molecular glues enables selective degradation of caspase-2

Molecular glue degraders (MGDs) reprogram E3 ligases to eliminate neosubstrates, yet their application has largely been confined to CRBN. Here, we identify caspase-2 as a new neosubstrate for von Hippel-Lindau (VHL), expanding the scope of VHL-based MGDs. Guided by a focused VHL ligand library design, we employed TurboID-based proximity labeling to discover stereoisomeric compounds (dCASP2-1 and dCASP2-2) that selectively recruit caspase-2 to VHL and promote its ubiquitin-proteasome system-dependent degradation. Further structure-activity relationship (SAR) studies yielded dCASP2-3 and dCASP2-4, which enhanced degradation potency (by 622-fold relative to dCASP2-1) and abolished enantioselectivity. Mechanistic mapping localized the degrader-induced interface to a two-helix region of the caspase-2 CARD domain, with residues H33, P34, and D100 essential for VHL engagement. Degron-guided computational modeling of the VHL/MGD/caspase-2 ternary complex provided structural insight into neosubstrate recognition. Together, we report the development of VHL molecular glues that selectively and potently degrade caspase-2, offering chemical probes to interrogate its functions in apoptosis and stress responses, while broadening the substrate landscape of VHL-based MGDs.

biochemistry↗

bioRxiv2026

Auto-sumoylation of UBC9 coordinates meiotic prophase and protects ovarian reserves

The small ubiquitin-like modifier SUMO regulates key events of meiosis, including pairing and crossing over between homologous chromosomes. Auto-sumoylation of the SUMO E2-conjugating enzyme UBC9 at lysine 14 alters its substrate selectivity in vitro, but the role of this modification in vivo is unknown. Here, we show that UBC9-K14 auto-sumoylation helps coordinate meiotic prophase and is important for maintenance of the ovarian reserve. Ubc9K14R/K14R knock-in mice show a variety of defects in meiotic prophase I, including altered assembly of DNA strand-exchange complexes, delayed and defective homolog synapsis, and unstable crossover recombination complexes. In spermatocytes, these defects are associated with reduced efficiency of crossing over between the X and Y chromosomes. Oocytes from Ubc9K14R/K14R females show related but distinct defects in recombination and synapsis. Moreover, maintenance of the primordial follicle reserve is defective in Ubc9K14R/K14R females, and their fecundity is reduced. Finally, we identify the meiosis-specific homolog-axis protein SYCP3 as a direct target of UBC9 in vitro and show that SYCP3 modification is strongly stimulated by K14 auto-sumoylation. We infer that auto-sumoylation enables UBC9 to modify a subset of targets in vivo, helping to coordinate key events of meiotic prophase and enhance the survival of primordial follicles to maximize fecundity.

genetics↗

bioRxiv2026

scATrans: annotating single-cell differential expression as transcription- or stabilization-weighted using unspliced RNA

Single-cell differential expression (DE) reports changes in mature mRNA abundance, but the same fold-change can reflect faster synthesis or slower decay. Metabolic labeling resolves this ambiguity but is costly and cannot be applied retrospectively to the vast majority of published scRNA-seq. scATrans closes this gap using layers every standard pipeline already generates: from DE-selected genes, a reference-corrected unspliced residual annotates each change as transcription- or stabilization-weighted, with no additional experiment. Benchmarked against metabolic-labeling systems with independent kinetic ground truth, the residual separates the two mechanisms at matched mature abundance (ROC-AUC 0.68-0.74, full-length NASC-seq2 K562; 0.59-0.63, 3' scEU-seq RPE1); effect size scales with intron capture, not model complexity, and explicit kinetic fitting adds nothing over the static contrast. Per-gene, the residual recovers the classical bulk exon-intron contrast (EISA); what scATrans adds is the inference layer single-cell reanalysis actually needs--DE-defined membership, gene-structure correction, a capture-regime reliability pre-flight, induction-matched testing, and a permutation-calibrated program score--so that confident calls are reserved for where the data support them: gene programs, not single genes. Applied to standard 10x data with no labeling, scATrans recovers textbook post-transcriptional biology: a curated AU-rich-element program is called stabilization-weighted in LPS-stimulated PBMCs (confirmed by per-donor pseudobulk DE in an independent four-donor cohort), while a glucocorticoid-response program is called transcription-weighted in dexamethasone-treated A549 cells-- opposite mechanisms recovered from unlabeled counts. scATrans turns any spliced/unspliced-resolved DE table into a mechanism-typed one, retrospectively and at scale. Availability and implementationscATrans requires Python [≥]3.9, interoperates with the scverse ecosystem (AnnData, Scanpy), and is released under the Apache-2.0 license. Install with pip install scatrans or from Bioconda. Documentation and tutorials: https://scatrans.readthedocs.io. Analyses in this manuscript use software version 0.10.9.

bioinformatics↗

bioRxiv2026

Cell organization and disturbance-mapping using image activated cell-profiling (CODIAC)

Protein localization and abundance are tightly regulated to maintain cellular homeostasis, and their dysregulation is a hallmark of disease. However, methods that monitor the spatial localization changes of many proteins at once, and at scale, remain limited. Here we introduce CODIAC (Cell Organization and Disturbance-mapping using Image-Activated Cell-profiling), which couples image-enabled cell sorting (ICS) with improved Cas12a-assisted endogenous PCR tagging and machine-learning-guided gating to profile pools of fluorescently tagged proteins. Rather than isolating discrete cell populations, CODIAC reads the "phenotypic fingerprint" of each fluorescently tagged protein, defined by its distribution across a fixed set of image-defined sort bins. The method resolves localization-associated phenotypes, reproducibly profiles complex pools, generalizes to proteins absent from the training set, and detects shifts in protein localization and abundance following chemical perturbation. CODIAC thus extends ICS from outlier screening toward systematic profiling of the spatial proteome and its remodeling under perturbation.

cell biology↗

bioRxiv2026

Immune-metabolic PET/MRI uncovers microenvironmental reprogramming under combined immunotherapy and anti-angiogenic therapy

The limited efficacy of immune checkpoint inhibitor (ICI) therapy in triple-negative breast cancer (TNBC) highlights the need for combination strategies that enhance antitumor responses. Sorafenib, a multikinase inhibitor with anti-angiogenic and immunomodulatory activity, represents a rational partner for ICI-based combination therapy. However, therapeutic responses to such combinations are biologically complex and cannot be fully characterized by any single biomarker or imaging modality. Here, we evaluated ICI therapy combined with sorafenib in the aggressive and ICI-refractory orthotopic 4T1 TNBC model. Therapeutic responses were assessed using a unique longitudinal multimodal imaging framework integrating [Zr]Zr-DFO-anti-CD8 minibody and [{superscript 1}F]FDG PET, as well as perfluorocarbon (PFC)-based {superscript 1}F MRI and hyperpolarized {superscript 1}3C MRS, together with ex vivo analyses. Only the ICI-sorafenib combination suppressed tumor growth, whereas both monotherapies showed limited antitumor activity. Multimodal imaging, together with complementary ex vivo analyses, uncovered coordinated tumor microenvironment (TME) remodeling, including vascular normalization, elevated CD8 cell presence with modest enrichment in the tumor center, delayed increase in phagocyte-associated {superscript 1}F MRI signal coupled with reduced CD206 cell infiltration, and sustained metabolic activity. These findings support ICI-sorafenib combination therapy as a promising therapeutic strategy for TNBC. Therapeutic efficacy reflected coordinated vascular, immune, and metabolic remodeling. This multimodal imaging framework enables non-invasive longitudinal monitoring of these complementary TME changes, providing a comprehensive strategy for treatment assessment in immunotherapy-based combination therapies. One Sentence SummaryLongitudinal multimodal imaging identified a multidimensional TME response signature of effective ICI-sorafenib therapy in TNBC.

cancer biology↗

bioRxiv2026

Evolution-inspired multi-objective Bayesian optimization for protein engineering

Protein engineering requires efficient navigation of vast sequence spaces under limited evaluation budgets, especially when multiple properties must be optimized simultaneously. We developed Evolution-inspired Multi-Objective Bayesian Optimization (EvoMOBO), an active-learning framework that integrates path-dependent sequence generation, global competition among generated variants, and explicit multi-objective optimization. Benchmarking against state-of-the-art methods on complete steroid receptor DNA-binding domain and ParD3 antitoxin landscapes demonstrated robust target-region enrichment, Pareto-front advancement, and sequence diversity across two- and three-objective tasks. In the DBD landscape, simulation-derived geometric descriptors served as labels for both initialization and iterative updating, enriching variants with favorable measured activities without experimental labels. Building on this validation, we applied EvoMOBO to two enzyme-engineering tasks using simulation-derived mechanistic descriptors, with experiments reserved for final validation. For an old yellow enzyme (GkOYE), 16 of 26 tested variants outperformed the wild type, and the best increased non-native oxidative dehydrogenation conversion from 17.5% to 95%. For a formate oxidase (AoFOx), EvoMOBO identified aggregation-resistant variants, two of which nearly doubled diethyl phthalate degradation in a photoenzymatic cascade. Together, these results establish EvoMOBO as a modular framework for multi-objective protein engineering using experimental or mechanism-derived labels.

bioengineering↗

bioRxiv2026

Molecular and functional characterization of CYP6AN1 orthologs involved in capsaicin metabolism in two Helicoverpa species

The generalist Helicoverpa armigera and the specialist Helicoverpa assulta are closely related noctuid pests and are among the few insect herbivores capable of feeding on and damaging hot pepper fruits, which contain the defensive compound capsaicin. Cytochrome P450 monooxygenases (P450s) contribute to the metabolism of plant defensive compounds and can facilitate insect adaptation to chemically defended host plants. Here, we identified CYP6AN1 in H. assulta (HassCYP6AN1) and comparatively characterized the CYP6AN1 orthologs from H. armigera (HarmCYP6AN1) and H. assulta. RACE identified one full-length HarmCYP6AN1 transcript and three full-length HassCYP6AN1 transcript isoforms. The HassCYP6AN1 isoforms contained distinct 5' UTRs generated by alternative transcription initiation and splicing but shared an identical coding sequence. Sequence comparisons and phylogenetic analysis supported their assignment as orthologs. Constitutive CYP6AN1 expression was higher in the H. assulta midgut, whereas dietary capsaicin significantly induced CYP6AN1 expression in the H. armigera midgut. Recombinant CYP6AN1 proteins were co-expressed with NADPH-cytochrome P450 reductase in Escherichia coli, and their substrate-metabolizing activities were evaluated using HPLC-based depletion assays. Both orthologs metabolized capsaicin, but HarmCYP6AN1 exhibited an approximately 2.3-fold higher depletion activity than HassCYP6AN1 under the conditions tested. HarmCYP6AN1 also showed P450-content-dependent xanthotoxin depletion, whereas no detectable xanthotoxin metabolism was observed for HassCYP6AN1. These findings establish CYP6AN1 as a component of the capsaicin-metabolizing repertoire of both species and reveal substantial divergence between the orthologs in transcript organization, expression regulation, catalytic activity, and detectable substrate range.

zoology↗

bioRxiv2026

Pervasive Backdoor Vulnerabilities in Genomic Foundation Models

Genomic foundation models are increasingly used to interpret and design DNA sequences, yet their susceptibility to training-data manipulation remains poorly understood. Here we systematically evaluate backdoor poisoning across three model families, seven parameter scales ranging from 50 million to 7 billion, and 18 genomic classification tasks. We introduce two complementary 48-nucleotide triggers: a composition-matched synthetic sequence and a biologically grounded trigger derived from transposon terminal inverted repeats. Poisoning 5% of the training data induced high attack success rates across all tested models, with model-level median values ranging from 91.4% to 100%. Increasing parameter scale did not consistently improve resistance, whereas poisoning rate and trigger length had stronger effects on attack efficacy. Performance on unmodified sequences was generally preserved, with 79.4% of model - task - trigger configurations changing by no more than two percentage points, although larger task-specific losses occurred. We further developed a two-stage defense that combines single-nucleotide mutation-sensitivity screening with reference-database validation. Across 28 evaluated configurations, the method achieved 100% precision and a median recall of 92.95%, while localizing the trigger in nearly all detected poisoned sequences. These findings establish training-data poisoning as a pervasive and difficult-to-detect vulnerability in genomic foundation models and motivate stronger data-provenance controls, adversarial evaluation and post-training security auditing.

bioinformatics↗

bioRxiv2026

Catch-bond engineering surpasses high-affinity maturation for T cell receptor therapies against solid tumors

The sensitivity of T cell receptors (TCRs) has traditionally been attributed to their affinity, a principle that has long underpinned both T cell biology and TCR engineering. Currently, high-affinity maturation remains the predominant strategy employed to enhance TCR sensitivity; however, this approach has been associated with severe off-target toxicities in clinical settings. In this study, using the only FDA-approved TCR-T therapy as a clinical reference, we demonstrate that force-induced catch bonds, rather than static affinity, primarily determine TCR sensitivity and facilitate the development of more effective TCR engineering strategies. Among polar and charged amino acids, histidine was determined to be the most effective residue for pinpointing engineering hotspots. TCRs engineered to form catch bonds exhibited superior performance compared to the FDA-approved TCR subjected to high-affinity maturation, improving the efficacy of TCR-T cell therapies against solid tumors without eliciting off-target toxicity or alloreactivity. Mechanistically, a correlation was observed between TCR sensitivity and the strength of catch bonds, whereas no such relationship was found with affinity. The duration of T cell-tumor cell interactions, immunological synapse formation, and the intensity of subsequent signaling are governed by catch bonds rather than by affinity. Structural and computational investigations have demonstrated that force-induced reconfiguration of the ligand-receptor interface, along with the formation of a novel hydrogen bond network, augments the specificity of interactions between the TCR and antigenic peptides. Furthermore, bispecific T cell engagers utilizing the TCR scaffold also formed catch bonds with tumor antigens, indicating intrinsic mechanosensory properties of the extracellular domains of TCRs and suggesting a novel therapeutic modality based on catch-bond-engineered TCRs.

immunology↗

bioRxiv2026

Multiomic analysis of non-glaucomatous human trabecular meshwork cells.

Glaucoma is an irreversible blinding disease that affects millions of individuals worldwide. Elevated intraocular pressure (IOP), regulated by the trabecular meshwork (TM) in the anterior eye, is the only modifiable risk factor. Human TM cells can be cultured from donor eyes, providing a precious resource for studying factors that induce or prevent glaucoma. The goal of this study was to produce datasets that define the molecular profile of human non-glaucomatous TM cells. Using 18 human TM cell strains cultured from non-glaucomatous individuals deposited from seven laboratories in the USA and UK, this study used transcriptomic, proteomic, lipidomic, and metabolomic analyses to characterize the molecular content of TM cells. The data herein provides the most comprehensive multiomic analyses of human TM cells to date and will be a useful resource for researchers and clinicians in the TM and glaucoma fields.

cell biology↗

bioRxiv2026

Electron counting enables cryo-electron ptychography for near-atomic-resolution cryo-electron microscopy

Cryo-electron ptychography is an emerging technique for studying radiation-sensitive biological specimens, developed from four-dimensional transmission electron microscopy (4D-STEM). Although ptychography has achieved ultrahigh resolution beyond conventional transmission electron microscopy limits for radiation-resistant samples, its application to frozen hydrated biological specimens currently remains at sub-nanometer resolution. Here we overcome this limitation by implementing electron counting with a hybrid-pixel detector, establishing key technical foundations for near-atomic-resolution cryo-ptychography. This counting approach significantly improves weak signal detection in convergent-beam electron diffraction, enabling ptychographic reconstruction at doses below 1 e-/[A]{superscript 2}. Additionally, we found beam-induced motion is effectively eliminated within single scans, suggesting conventional cryoEMs dose-fractionation approach may need reevaluation. Demonstrating high contrast under both low-dose and tilted conditions, along with achieving 3.59 [A] resolution for the [~]700 kDa T20S proteasome, we validate cryo-electron ptychography as a viable general imaging modality that could complement or surpass conventional phase-contrast cryoEM methods.

biophysics↗

bioRxiv2026

Imaging-Based Age- and Sex-specific Cardiovascular and Skeletal Phenotypes in Fbn1C1041G/+ Mice

Marfan syndrome is a multisystem connective tissue disorder with age-dependent and sex-dependent phenotypic variability. This study characterized early cardiovascular, valvular, transmitral flow, and skeletal changes in Fbn1C1041G/+ mice. Male and female wild-type and Fbn1C1041G/+ mice were examined from 4 to 20 weeks of age. Transthoracic echocardiography was used to measure aortic root and ascending aortic diameters, anterior mitral leaflet length, and mitral inflow Doppler parameters. Lateral X-ray imaging was used to quantify kyphosis angle. Fbn1C1041G/+ mice of both sexes developed progressive aortic dilation, with earlier and broader aortic root involvement in males and delayed annular dilation in females. Anterior mitral leaflet elongation was detected from 8 weeks in males and at 20 weeks in females. In contrast, reduced E-wave velocity and E/A ratio were present from 4 weeks in both sexes, without changes in A-wave velocity. Kyphosis was detectable earlier in females than in males. These findings show that early Fbn1-related phenotypes follow organ-specific and sex-dependent temporal patterns, supporting age-resolved and sex-stratified assessment in preclinical Marfan syndrome studies.

pathology↗

bioRxiv2026

Hyaluronan and CD44 targeting reverses early matrix changes, proinflammatory signals and fibrosis in primary sclerosing cholangitis

Primary sclerosing cholangitis (PSC) is a rare, progressive liver disease characterized by biliary inflammation and bile duct strictures and no approved medical therapy. Despite its clinical severity, the pathological mechanisms underlying PSC remain poorly understood, largely due to early diagnostic challenges. Here we provide complementary evidence in human PSC samples, transcriptomic data, mouse models, and 3D cholangiocyte cultures that underscore the importance of hyaluronic acid (HA) and its cognate receptor CD44 in PSC pathogenesis. HA is a glycosaminoglycan abundant in the extracellular matrix in inflammatory disorders, yet its role in PSC has not been well characterized. We demonstrate that in early-stage PSC, cholangiocytes aberrantly produce high molecular weight HA that accumulates in the peribiliary matrix, increasing local tissue stiffness. This mechanical signal is transduced by a CD44/Integrin {beta}1 receptor complex in cholangiocytes, driving cell proliferation, YAP mechanosignaling, pro-inflammatory cytokine production with a transition to a ductular reactive phenotype. CD44 knockdown in cholangiocyte cell lines and mouse models significantly lowered stiffness, and attenuated inflammation. Together, these findings reveal a mechano-inflammatory axis in which HA-driven matrix stiffening perpetuates biliary inflammation and disease progression, identifying HA targeting and CD44 as promising therapeutic strategies. One Sentence SummaryHyaluronan and CD44 mediate matrix changes and progressive fibrosis in primary sclerosing cholangitis.

pathology↗

bioRxiv2026

Large-scale functional coupling and computational modelling reveal frontotemporal hotspots in distributed network connectivity during working memory recognition, encoding, and retrieval

Working memory depends on coordinated activity across distributed brain networks, but how these interactions support recognition, encoding, and retrieval remains unclear. Here, we analyzed intracranial local field potentials from 1,652 sites in 36 epilepsy patients performing a naturalistic card-matching task. Phase-amplitude and phase-phase coupling revealed widespread condition-dependent interactions, with frontal and temporal regions emerging as recurrent network hubs. Event-related analyses showed reproducible within-trial coupling dynamics across participants. Directed phase transfer entropy further identified frequency-specific information flow, including frontal regions acting as a theta-band source and an alpha-band sink. Finally, a time-delay-embedded hidden Markov model identified task-locked latent states with condition-dependent occupancy trajectories and distinct coherence profiles. Together, these results suggest that working memory is implemented through dynamic, directionally organized large-scale networks, in which frontotemporal hubs coordinate flexible transitions among distributed neural states rather than operating in isolation.

neuroscience↗

bioRxiv2026

A germline shortcut in protein language model retrieval of adaptive immune receptors

Protein language models (PLMs) are increasingly used to represent adaptive immune receptors; however, their advantages over classical alignment-based methods remain unclear. We benchmarked the ESM2 family, ESM-C, antibody-, and TCR-specific PLMs against sequence-alignment methods across four B-cell and T-cell receptor databases using CDR3, clonotype, paratope, and full variable domain representations. At the CDR3 level, alignments consistently out-performed all PLMs, except SCEPTR, a contrastively pre-trained TCR model that additionally encodes the germline V gene. BLOSUM62 and Levenshtein distance metrics exceeded all ESM2 variants by 7-12 percentage points in top-1 retrieval accuracy, and domain-specific PLMs did not close this gap. At the full variable domain level, the alignments and PLMs converged. Alignment performance peaked at the paratope level, whereas PLMs benefited primarily from the addition of conserved framework information in the full-length sequences. The one exception was heavily-mutated HIV-1 antibodies, where PLMs outperformed alignments. These results, together with germline reversion analyses, indicate that full-length PLM performance takes advantage of a germline shortcut rather than improved extraction of antigen-specific information from CDRs. We further showed that conventional random train-test splits inflate retrieval accuracy by 15-28 percentage points owing to clonal leakage. Together, these findings define the strengths and limitations of frozen zero-shot PLM embeddings for immune receptor retrieval and establish clone-aware benchmarking as a practical standard of comparison. Key PointsO_LIClassical sequence alignment consistently outperformed frozen protein language models (PLMs) for CDR3 immune-receptor retrieval across the four datasets. C_LIO_LIProtein language models reached parity only in the full variable domain by exploiting the conserved germline-derived sequence, thus revealing a germline shortcut. C_LIO_LIFrozen PLMs exceeded alignment only for highly mutated HIV-1 antibodies. C_LIO_LIOne PLM matching alignment at CDR3, SCEPTR, encoded germline V gene information. C_LIO_LIRandom train-test splitting inflated accuracy by 15-28 percentage points; clone-aware splitting should be the standard. C_LI Bibliographical NoteDaron M. Standley is Professor of Genome Informatics at the Research Institute for Microbial Diseases, Osaka University. His research developed computational methods for immune repertoire analysis and antibody discovery.

bioinformatics↗

bioRxiv2026

Accumulation of Lipid Droplets in Microglia following Neonatal Brain Hypoxia-Ischemia

BackgroundHypoxic-ischemic encephalopathy (HIE) is a major cause of neonatal mortality and neurodevelopmental impairments. Following brain hypoxia-ischemia (HI), microglia face substantial metabolic stress; and upon phagocytosis, they become overloaded with lipids derived from engulfed dead neurons and myelin debris. It is unclear how microglia respond to and process the lipid cargo, and whether lipid accumulation may affect microglia function following neonatal HI. MethodsThe postnatal day 10 mice were subjected to HI using the Vannucci model. Lipid droplets (LD) were assessed by histology and immunofluorescent staining. Single-nucleus RNA sequencing (snRNA-seq) was performed using brain tissue from HI-injured and sham-operated mice at 72 hours after HI. LD-accumulating microglia (LDAM) were identified by a specific LD marker gene perilipin 2 (Plin2). Differential gene expression was analyzed between Plin2-positive and Plin2-negative microglia after HI. Human HIE brain sections were also examined for LD accumulation. The dynamic changes of PLIN2-expressing microglia and infiltrating monocyte-derived macrophages (MDM) at 24 hours, 72 hours and 7 days after HI were compared using flow cytometry. In addition, mouse BV2 microglia were subjected to oxygen-glucose deprivation (OGD) to study phagocytosis and cytokine expression. ResultsLipid droplets accumulated primarily in microglia after HI in neonatal mice and in human HIE brain. LD were not found in astrocytes or neurons. Plin2-expressing LDAM emerged as new microglia clusters after HI. Compared with microglia without LD, LDAM showed a distinct transcriptional profile with upregulation of genes linked to microglial activation, enhanced cholesterol and lipid processing, and a shift towards phagocytic and pro-inflammatory state. Blocking LD biogenesis reduced elevated phagocytosis and IL-1{beta} expression in BV2 cells following OGD. ConclusionOur study revealed that microglia accumulate lipid droplets as part of their metabolic responses to HI in the neonatal brain. Microglial lipid droplet formation is associated with a pro-inflammatory phenotype at early stage after HI, and increased phagocytosis in vitro. The lipid metabolic changes may regulate microglial function and influence HI outcomes.

neuroscience↗

bioRxiv2026

Deformability screening identifies NUDT5 as a mediator of cellular mechanobiology

How cells deform, sense, and respond to mechanical cues drives physiological and disease processes ranging from development to cancer metastasis; however, unbiased approaches to identify mechanical mediators are lacking. We screened 1280 compounds to identify modulators of cancer cell deformability using a cellular filtration assay and identified 92 compounds that significantly reduced deformability of ovarian cancer cells; top hits also reduced migration and invasion. Connectivity mapping of the top 21 compounds identified NUDT5 (Nudix hydrolase 5) as a predicted mechanical mediator; transcriptomic analyses implicated NUDT5 in mechanobiology and metabolic processes. We confirmed that NUDT5 mediates intracellular ATP and cellular mechanical behaviors, including morphology and deformability. In ovarian cancer, increased NUDT5 levels were associated with higher tumor stage and worse patient survival; NUDT5 inhibition reduced migration and colony formation in vitro and peritoneal tumor burden in mice. These findings establish deformability-based screening as a platform for discovering mechanical mediators and identify NUDT5 as a therapeutic target in ovarian cancer. TeaserScreening cells based on deformability provides an unbiased approach to identify NUDT5 as a mediator of cell mechanics

cell biology↗

bioRxiv2026

A multi-agent workflow converts CAR-T patient evidence into experimentally testable hypotheses

Clinical studies of chimeric antigen receptor (CAR)-T therapy generate diverse molecular and clinical evidence that remains fragmented across publications, public repositories and patient-derived datasets, limiting systematic therapeutic discovery. Here we present BioPathfinder, an evidence-guided multi-agent workflow for closed-loop biomedical discovery. BioPathfinder constructs a provenance-aware knowledge resource linking publications with patient single-cell RNA sequencing (scRNA-seq) datasets and clinical metadata, and uses role-specialized large language model agents to generate, review and prioritize diverse, falsifiable and dataset-aware mechanistic hypotheses for computational and experimental validation. Applied to a curated corpus of CAR-T-treated patient studies and matched scRNA-seq datasets, BioPathfinder identified candidate mechanisms underlying CAR-T persistence, dysfunction and therapeutic resistance. The workflow prioritized the hypothesis that genes associated with an NK-like transition programme could be targeted to reduce CAR-T exhaustion and improve persistence. Analysis of patient scRNA-seq datasets showed enrichment of this programme in exhausted post-infusion CAR-T cells. Virtual perturbation prioritized transition-associated receptor genes, including KLRC1, KLRD1 and KLRG1, and expert review selected KLRC1, encoding NKG2A, for experimental validation. In vitro and in vivo chronic-stimulation models showed that NKG2A marked activated, exhaustion-associated CD8 CAR-T cells, whereas NKG2A blockade enhanced antitumour activity and persistence-associated functional readouts in vivo. BioPathfinder establishes a generalizable framework that transforms fragmented clinical single-cell evidence into experimentally validated therapeutic hypotheses, providing a scalable strategy for AI-guided biomedical discovery. HIGHLIGHTSO_LIBioPathfinder integrates fragmented CAR-T clinical studies, patient scRNA-seq datasets and metadata into a provenance-aware evidence resource for AI-guided hypothesis discovery. C_LIO_LIA multi-agent workflow generates, critiques and prioritizes diverse, falsifiable, dataset-aware mechanisms underlying CAR-T persistence, dysfunction and therapeutic resistance. C_LIO_LIBioPathfinder prioritizes KLRC1/NKG2A as a therapeutic target, and experimental validation demonstrates that NKG2A blockade enhances CAR-T persistence and antitumour function. C_LI

bioengineering↗