Research Radar — 2026-05-28
Methods & AI
Computational
Scoring Gene Importance by Interpreting Single-Cell Foundation Models
Nature Biotechnology Published 2026-05-27 research article DOI: 10.1038/s41587-026-02845-2
single-cell foundation model gene importance interpretability transformer scGPT scFoundation computational method
Summary: Introduces a framework for scoring gene importance directly from single-cell foundation models without requiring task-specific fine-tuning. Single-cell foundation models — large transformer models pre-trained on millions of single-cell transcriptomes — have shown remarkable ability to capture gene-gene relationships and cellular states, but extracting biologically interpretable gene-level importance scores from these black-box models has been challenging. This study develops an attribution-based method that interrogates the internal representations of pre-trained single-cell foundation models (including scGPT, scFoundation, and Geneformer) to rank genes by their contribution to specific cellular phenotypes or conditions. The approach leverages the models' learned attention patterns and embedding geometries to identify genes that are most informative for distinguishing cell states, predicting perturbation responses, or stratifying disease conditions — all without additional training. Benchmarking against traditional differential expression and CRISPR-based gene essentiality screens demonstrates that foundation model-derived gene importance scores often capture regulatory relationships that are missed by conventional methods, particularly for lowly-expressed transcription factors and signaling molecules whose effects propagate through network-level changes rather than direct expression differences. The framework is demonstrated across diverse applications including disease-associated gene discovery, perturbation response prediction, and cell-type-specific regulatory network inference.
Why it matters: Single-cell foundation models represent a massive investment in pre-training, but their practical utility for biologists has been limited by the difficulty of extracting interpretable insights. This framework transforms these models from black-box embeddings into gene prioritization tools that can guide experimental follow-up — effectively turning foundation models into hypothesis generators. The ability to score gene importance without task-specific fine-tuning makes this approach immediately applicable to any single-cell dataset, democratizing access to foundation model-powered analysis.
Why for Yiru: Gene prioritization in TME contexts — identifying which genes drive immune evasion, therapy resistance, or T cell dysfunction — is a central challenge in computational oncology. A method that leverages pre-trained single-cell foundation models to identify regulatory genes missed by differential expression could surface novel immunotherapy targets, particularly transcription factors and signaling molecules whose effects on the TME are mediated through network rewiring rather than expression changes alone.
Generalizable Mutation-Effect Prediction Across Adaptive Immune Recognition via Unified Multimodal Framework
Nature Machine Intelligence Published 2026-05-27 research article DOI: 10.1038/s42256-026-01245-5
mutation effect prediction immune recognition T cell receptor B cell receptor MHC multimodal deep learning immunoinformatics protein language model
Summary: Presents a unified multimodal deep learning framework that predicts how mutations — both in pathogens and in self-proteins — affect recognition by the adaptive immune system across T cell, B cell, and MHC presentation modalities. Predicting whether a mutation will create, destroy, or alter an immune epitope is critical for vaccine design, cancer neoantigen identification, and understanding autoimmune disease — but existing tools are fragmented across modalities (separate models for MHC binding, TCR recognition, and B cell epitope prediction). This framework unifies these tasks by representing both the mutating protein and the immune receptor/MHC molecule in a shared multimodal embedding space learned from large-scale immunopeptidomics, TCR repertoire, and antibody-antigen interaction data. The model leverages protein language model embeddings to capture the structural and evolutionary context of mutations, then projects these into an immune recognition space that predicts binding, presentation, and immunogenicity across modalities. A key innovation is the use of contrastive learning to align representations of peptides, MHC alleles, TCRs, and BCRs in a way that generalizes to unseen mutations, alleles, and even entirely new pathogens. The framework achieves state-of-the-art performance on multiple benchmarks including SARS-CoV-2 variant immune escape prediction, cancer neoantigen prioritization, and autoimmune epitope identification.
Why it matters: Immune evasion through mutation is a central challenge in infectious disease and cancer — variants escape existing immunity, and tumours acquire mutations that avoid T cell recognition. A unified framework that predicts immune consequences of mutations across modalities would accelerate vaccine updates for viral variants, improve neoantigen selection for personalized cancer vaccines, and identify autoimmune triggers. The multimodal approach is biologically appropriate because real immune recognition involves the coordinated action of MHC presentation, T cell recognition, and antibody binding — yet computational tools have historically treated these as separate problems.
Why for Yiru: Neoantigen prediction is a key application for TME research — identifying which tumour mutations generate T cell-recognizable epitopes is essential for personalized immunotherapy and for understanding immune editing during tumour evolution. A unified framework that simultaneously evaluates MHC presentation and TCR recognition could improve neoantigen prioritization by filtering out mutations that bind MHC but fail to elicit T cell responses, reducing the false-positive rate that has limited neoantigen-based vaccines.
TADShop: Systematic Benchmarking and Identification of Topologically Associating Domains
Nature Methods Published 2026-05-27 research article DOI: 10.1038/s41592-026-03112-y
topologically associating domains TAD Hi-C 3D genome chromatin organization benchmarking computational method
Summary: Presents TADShop, a comprehensive benchmarking framework that systematically evaluates 28 TAD-calling algorithms against multiple metrics including reproducibility, robustness to sequencing depth, boundary precision, and biological relevance. Topologically associating domains (TADs) are megabase-scale chromatin interaction domains that constrain enhancer-promoter contacts and are fundamental units of 3D genome organization. Despite their importance, the field has lacked consensus on which computational method to use — different algorithms applied to the same Hi-C data can yield dramatically different TAD sets, undermining reproducibility and biological interpretation. TADShop addresses this by running all major algorithms on standardized benchmark datasets with known biological features, evaluating each method against ground-truth criteria including conservation across cell types, enrichment of CTCF/cohesin at boundaries, and concordance with orthogonal measures of chromatin organization (e.g., replication timing, histone modifications). The analysis reveals that no single algorithm dominates across all metrics, but identifies specific algorithms that are optimal for different use cases — e.g., some excel at detecting robust, conserved TADs while others are better at capturing cell-type-specific or dynamic TAD boundaries. The framework also provides guidelines for parameter selection and sequencing depth requirements.
Why it matters: TADs are fundamental organizers of gene regulation — disruption of TAD boundaries can lead to enhancer adoption and oncogene activation, as demonstrated in medulloblastoma and T-cell acute lymphoblastic leukemia. Yet the field's inability to agree on TAD-calling methods has limited the translation of 3D genome biology into clinical applications. TADShop provides an objective basis for method selection and establishes community standards that will improve the reproducibility of 3D genome studies, analogous to how benchmarks like CASP transformed protein structure prediction.
Why for Yiru: 3D genome organization is increasingly recognized as relevant to cancer — structural variants that disrupt TAD boundaries can create oncogenic enhancer-promoter interactions, and epigenetic therapies may act in part by restoring normal TAD architecture. Standardized TAD calling is a prerequisite for integrating 3D genome features into TME multi-omic analyses, particularly for understanding how chromatin organization differs between tumour and immune cells and how it changes during immune cell activation.
EnzymeTuning Improves Enzyme-Constrained Metabolic Modeling and Proteome Abundance Prediction Through Deep Learning
Nature Communications Published 2026-05-27 research article DOI: 10.1038/s41467-026-54432-8
metabolic modeling enzyme constraint deep learning proteome prediction genome-scale model GEM computational method
Summary: Introduces EnzymeTuning, a deep learning framework that enhances genome-scale metabolic models (GEMs) by accurately predicting enzyme abundance constraints from transcriptomic data, improving flux predictions and proteome allocation estimates. Traditional GEMs predict metabolic fluxes using stoichiometric constraints but lack information about enzyme capacity — how much of each enzyme the cell actually produces — leading to unrealistic flux distributions where biomass production and metabolic reactions can proceed at implausible rates. Enzyme-constrained models (ecGEMs) address this by incorporating proteomic constraints, but their accuracy depends on having measured enzyme abundances, which are rarely available. EnzymeTuning solves this by training a deep learning model to predict enzyme abundances from readily available transcriptomic data, learning the complex nonlinear relationships between mRNA levels and protein abundance — including translational regulation, protein degradation rates, and post-translational modifications that decouple transcript and protein levels. The predicted enzyme abundances are integrated into ecGEMs as constraints, dramatically improving flux predictions compared to both unconstrained GEMs and ecGEMs using naive transcript-to-protein conversions. The framework is validated on multiple organisms and conditions, demonstrating improved prediction of growth rates, metabolic exchange fluxes, and proteome allocation.
Why it matters: Genome-scale metabolic models are widely used in metabolic engineering, drug targeting, and understanding disease metabolism, yet their predictive power has been limited by the assumption that enzyme availability does not constrain flux. By making enzyme-constrained modeling practical with only transcriptomic data — no proteomics required — EnzymeTuning democratizes ecGEMs and enables their application to the thousands of existing transcriptomic datasets. This is analogous to how AlphaFold democratized protein structure prediction by eliminating the need for experimental structure determination.
Why for Yiru: Metabolic modeling of the TME is a growing area — understanding how tumour, immune, and stromal cells compete for nutrients and how metabolic interventions (dietary, pharmacological) reshape this competition. Enzyme-constrained models that accurately predict metabolic fluxes from the abundant single-cell transcriptomic data in the TME could identify cell-type-specific metabolic vulnerabilities and predict which metabolic interventions will most effectively shift the balance toward anti-tumour immunity.
SOFisher: Reinforcement Learning-Guided Experiment Designs for Spatial Omics
Nature Communications Published 2026-05-25 research article DOI: 10.1038/s41467-026-54325-w
spatial omics reinforcement learning experimental design tissue sampling region of interest computational method active learning
Summary: Presents SOFisher, a reinforcement learning framework that optimally selects regions of interest (ROIs) for spatial omics experiments — determining which tissue areas to profile to maximize information gain given a fixed budget of measurements. Spatial omics technologies (spatial transcriptomics, spatial proteomics, imaging mass cytometry) generate rich molecular maps of tissues but are expensive and time-consuming, typically profiling only a fraction of a tissue section. Selecting which regions to profile is typically done ad hoc — based on morphology, coarse markers, or uniform sampling — which often misses rare but biologically important regions. SOFisher frames ROI selection as a sequential decision problem: an RL agent learns to navigate tissue images, selecting sampling locations that maximize the diversity of molecular states captured while balancing exploration of unknown regions and exploitation of known interesting areas. The agent is guided by a reward function that incorporates molecular diversity, spatial coverage, morphological features from H&E images, and prior biological knowledge about the tissue context. Demonstrations on spatial transcriptomic datasets from tumour samples show that SOFisher-selected ROIs capture significantly more cell types, more rare cell states, and more spatial niches than human-selected or uniformly sampled ROIs, for the same number of measurements. The framework is compatible with multiple spatial omics platforms and can incorporate platform-specific constraints such as minimum ROI size or instrument field of view.
Why it matters: Spatial omics is transforming tissue biology but the cost per measurement remains high, creating a tension between coverage and depth. Intelligent ROI selection that maximizes information per measurement dollar can dramatically increase the effective throughput of spatial omics experiments — essentially getting more biology for the same budget. This is especially critical for clinical specimens where tissue is limited and cannot be re-sampled.
Why for Yiru: Spatial omics of the TME is central to our understanding of immune-tumour interactions, but current studies are limited by the number of ROIs that can be profiled. SOFisher could guide TME spatial profiling to ensure that rare but critical regions — immune exclusion zones, TLS-adjacent areas, invasive fronts — are systematically captured rather than missed through random sampling. This is directly relevant to TME spatial studies where the goal is to understand the full diversity of spatial niches.
Biomedical discoveries
Biomedicine
αKG-Mediated Carnitine Synthesis Drives DNA Repair via Histone Acetylation
Nature Published 2026-05-27 research article DOI: 10.1038/s41586-026-10678-2
α-ketoglutarate αKG carnitine DNA repair histone acetylation metabolism epigenetics cancer
Summary: Discovers that α-ketoglutarate (αKG), a central TCA cycle metabolite, promotes DNA repair by fueling carnitine synthesis, which in turn drives histone acetylation at DNA damage sites to facilitate repair factor recruitment. αKG is well known as a co-substrate for αKG-dependent dioxygenases — including TET DNA demethylases and JmjC histone demethylases — that regulate the epigenome. This study reveals an unexpected parallel pathway: αKG is diverted into carnitine biosynthesis through the enzyme trimethyllysine dioxygenase (TMLD), and the resulting carnitine is used by carnitine acetyltransferase (CrAT) to generate acetyl-CoA in the nucleus. This locally produced acetyl-CoA serves as the acetyl donor for histone acetyltransferases (HATs) that acetylate histones at DNA double-strand break sites, creating an open chromatin conformation that enables recruitment of DNA repair factors including 53BP1 and BRCA1. The pathway is particularly important in rapidly proliferating cells — including cancer cells — where nuclear acetyl-CoA pools are limiting. Disruption of the αKG→carnitine→acetyl-CoA axis impairs DNA repair, sensitizes cells to DNA-damaging agents, and reduces tumour growth in xenograft models. The study establishes a direct metabolic route from TCA cycle activity to nuclear histone acetylation that is spatially and functionally coupled to DNA repair, distinct from the canonical role of αKG in dioxygenase reactions.
Why it matters: The connection between metabolism and DNA repair has been recognized — cells need energy and building blocks to repair DNA — but the specific molecular conduits linking metabolic state to repair capacity have been unclear. This study identifies a concrete pathway: αKG→carnitine→nuclear acetyl-CoA→histone acetylation→DNA repair. This has immediate implications for cancer therapy, suggesting that metabolic interventions (αKG supplementation, carnitine modulation) could enhance or impair DNA repair in cancer cells, with consequences for genomic stability and response to DNA-damaging chemotherapies and PARP inhibitors.
Why for Yiru: Metabolic crosstalk in the TME — where nutrient competition between tumour and immune cells shapes both cancer progression and anti-tumour immunity — is central to TME biology. If αKG availability influences DNA repair capacity in tumour cells, then metabolic competition in the TME that limits αKG could create DNA repair-deficient tumour subpopulations that are specifically vulnerable to PARP inhibitors or immunotherapy (through increased mutational burden). Understanding how TME metabolic gradients affect the αKG-carnitine-DNA repair axis could identify metabolic combination strategies.
A PI(3,5)P2/CHMP4B Axis on Lysosomes Is Essential for Microautophagic Degradation of STING
Nature Communications Published 2026-05-27 research article DOI: 10.1038/s41467-026-54418-6
STING microautophagy lysosome PI(3,5)P2 CHMP4B innate immunity cGAS-STING protein degradation
Summary: Identifies a phosphoinositide-regulated microautophagy pathway on lysosomes that directly degrades STING, establishing a new mechanism for terminating innate immune signaling. The cGAS-STING pathway detects cytosolic DNA and triggers type I interferon responses essential for antiviral immunity and antitumour T cell priming — but STING must be tightly controlled because chronic activation causes autoinflammation and interferonopathies. While STING degradation via the lysosome after trafficking from the ER through the Golgi is known (macroautophagy-independent), this study reveals a distinct microautophagy mechanism: STING molecules on the lysosomal surface are directly engulfed by lysosomal membrane invaginations in a process requiring the endosomal sorting complex required for transport (ESCRT)-III component CHMP4B and the lysosome-specific phosphoinositide PI(3,5)P2. The lipid PI(3,5)P2, produced by the PIKfyve kinase complex on lysosomes, recruits CHMP4B to initiate membrane deformation and scission of STING-containing vesicles into the lysosomal lumen for degradation. Genetic or pharmacological disruption of PI(3,5)P2 synthesis or CHMP4B function results in STING accumulation on lysosomes and prolonged interferon signaling, leading to enhanced antitumour immunity in mouse models but also increased systemic inflammation. The pathway is distinct from STING degradation via ER-to-lysosome trafficking and from canonical macroautophagy, representing a dedicated microautophagic mechanism for regulating this critical immune sensor.
Why it matters: STING agonists are a major class of cancer immunotherapeutics in clinical development, designed to activate innate immunity in the TME. Understanding how STING is naturally terminated — and how to modulate this termination — could enhance STING agonist efficacy or enable more controlled activation. Conversely, blocking STING degradation could treat viral infections or boost vaccine responses. The identification of a specific, druggable lipid-protein axis (PI(3,5)P2/CHMP4B) controlling STING levels on lysosomes opens a new avenue for pharmacological STING modulation independent of cGAS activation.
Why for Yiru: STING signaling in the TME is a double-edged sword — chronic STING activation can promote T cell priming and antitumour immunity, but sustained interferon signaling can also drive immune suppression and T cell exhaustion. Understanding the mechanisms that tune STING degradation — particularly in different TME cell types (tumour cells, dendritic cells, macrophages, T cells) — could inform strategies to selectively enhance STING signaling in antigen-presenting cells while limiting it in tumour cells or exhausted T cells.
Macrophage ALDH2 Drives Immunotherapy Resistance by Silencing CXCL9 Through Metabolic-Epigenetic Crosstalk
Nature Communications Published 2026-05-21 research article DOI: 10.1038/s41467-026-54156-9
ALDH2 macrophage immunotherapy resistance CXCL9 metabolic-epigenetic T cell recruitment acetaldehyde tumour microenvironment
Summary: Demonstrates that the metabolic enzyme aldehyde dehydrogenase 2 (ALDH2) in tumour-associated macrophages (TAMs) drives immunotherapy resistance by silencing the T cell-recruiting chemokine CXCL9 through a metabolic-epigenetic mechanism. ALDH2 is best known for its role in alcohol metabolism — detoxifying acetaldehyde generated from ethanol oxidation — but is broadly expressed and participates in the metabolism of endogenous aldehydes produced during oxidative stress and lipid peroxidation. This study shows that in TAMs, ALDH2 activity generates acetate as a byproduct of aldehyde detoxification, and this acetate is used by acetyl-CoA synthetase to produce acetyl-CoA that fuels histone acetylation at the CXCL9 locus. Paradoxically, increased histone acetylation at CXCL9 in ALDH2-high TAMs is associated with gene silencing rather than activation — the acetylation promotes binding of a transcriptional repressor complex that suppresses CXCL9 expression. The resulting loss of CXCL9 secretion reduces CD8+ T cell recruitment into tumours, creating an immune-excluded TME that is resistant to anti-PD-1 therapy. Pharmacological inhibition of ALDH2 with the clinically approved drug disulfiram restores CXCL9 expression, increases T cell tumour infiltration, and sensitizes tumours to checkpoint blockade in multiple mouse models. Analysis of human tumour samples confirms that high TAM ALDH2 expression correlates with low CXCL9, reduced CD8+ T cell infiltration, and poor response to immunotherapy.
Why it matters: Immunotherapy resistance remains the major clinical challenge in oncology — most patients do not respond, and understanding why has been a central research focus. This study identifies a metabolic enzyme in TAMs as a driver of resistance through a mechanism that is therapeutically targetable with an existing FDA-approved drug (disulfiram). The finding that ALDH2-mediated acetate production epigenetically silences CXCL9 adds a new dimension to TAM biology — metabolic activity in macrophages is not just about energy and biosynthesis but actively shapes the chemokine landscape that determines T cell infiltration. The repurposing potential of disulfiram for immunotherapy sensitization is immediately testable in clinical trials.
Why for Yiru: TAM-focused immunotherapy strategies are a major area of TME research. This study provides a concrete metabolic-epigenetic mechanism by which TAMs create immune exclusion — directly connecting a specific enzyme activity to a specific chemokine to T cell infiltration. The ALDH2-CXCL9 axis could be interrogated in spatial transcriptomic data of the TME by examining co-localization of ALDH2-expressing macrophages, CXCL9 gradients, and CD8+ T cell density across tumour regions. This is precisely the kind of spatial-metabolic-immune axis that computational TME analysis can reveal.
Iron Overload in the Tumor Microenvironment Induces CD8+ T Cell Ferroptosis and Dysfunction
Nature Communications Published 2026-05-22 research article DOI: 10.1038/s41467-026-54218-0
iron overload ferroptosis CD8+ T cell tumour microenvironment immunotherapy lipid peroxidation T cell dysfunction
Summary: Reveals that iron accumulation in the tumour microenvironment selectively induces ferroptosis in tumour-infiltrating CD8+ T cells, representing a previously unrecognized mechanism of immune dysfunction in cancer. Iron is essential for cellular proliferation and metabolism — including T cell activation and effector function — but excess iron catalyzes Fenton reactions that generate lipid reactive oxygen species, triggering the non-apoptotic cell death pathway ferroptosis. The tumour microenvironment is known to be iron-rich due to haemorrhage, tumour cell necrosis releasing intracellular iron, and dysregulated iron metabolism in cancer cells (which often upregulate iron import and downregulate iron export to support proliferation). This study shows that CD8+ T cells infiltrating iron-rich tumour regions accumulate labile iron through the transferrin receptor CD71, making them selectively vulnerable to ferroptosis while regulatory T cells (which express lower CD71) are protected. Iron-driven CD8+ T cell ferroptosis depletes the functional effector T cell pool, impairing anti-tumour immunity and reducing immunotherapy efficacy. Iron chelation with deferoxamine, or genetic ablation of CD71 specifically in CD8+ T cells, protects tumour-infiltrating T cells from ferroptosis, enhances their effector function, and improves tumour control in combination with anti-PD-1 therapy. Analysis of human tumour specimens confirms that iron deposition correlates with CD8+ T cell ferroptosis markers and poor prognosis across multiple cancer types.
Why it matters: T cell dysfunction in the TME has been attributed to chronic antigen stimulation (exhaustion), immunosuppressive cytokines, nutrient deprivation, and inhibitory receptor signaling — but the contribution of metal ion toxicity has been largely overlooked. This study identifies iron-driven ferroptosis as a major cause of CD8+ T cell loss in tumours, distinct from classical exhaustion, and demonstrates that iron chelation can protect T cells and enhance immunotherapy. This opens a new therapeutic axis — targeting iron homeostasis in the TME to preserve T cell function — that is immediately actionable with existing iron chelators.
Why for Yiru: Ferroptosis has been primarily studied as a tumour-cell death mechanism inducible by therapy, but the concept that iron selectively kills anti-tumour T cells in the TME inverts this paradigm — the same iron-rich environment that might kill some tumour cells could simultaneously be destroying the immune cells needed for durable responses. This dual-edged nature of TME iron highlights the importance of cell-type-specific analysis in TME computational studies — bulk iron signatures may mask opposing effects on tumour vs. immune cells that can only be resolved at single-cell resolution.
Targeting Arginine Metabolism Reverses Bone Immunosuppressive Microenvironment and Metastasis in ARID1A-Deficient Triple-Negative Breast Cancer
Nature Communications Published 2026-05-26 research article DOI: 10.1038/s41467-026-54387-w
arginine metabolism bone metastasis triple-negative breast cancer ARID1A immunosuppressive microenvironment arginase myeloid-derived suppressor cell MDSC
Summary: Demonstrates that ARID1A-deficient triple-negative breast cancer (TNBC) remodels arginine metabolism in the bone microenvironment to create an immunosuppressive niche that promotes metastatic colonization, and that targeting this metabolic adaptation reverses immune exclusion and reduces bone metastasis. ARID1A is a subunit of the SWI/SNF chromatin remodeling complex and is among the most frequently mutated genes in cancer, including TNBC. Bone is a common site of breast cancer metastasis, and bone metastases are largely incurable, in part because the bone microenvironment is naturally immunosuppressive. This study shows that ARID1A loss in TNBC cells upregulates arginase-1 expression through epigenetic derepression, causing the tumour cells to consume arginine from the bone microenvironment. Arginine depletion suppresses T cell proliferation and effector function while simultaneously promoting the expansion of polymorphonuclear myeloid-derived suppressor cells (PMN-MDSCs), which further deplete arginine and produce immunosuppressive factors. The resulting profoundly immunosuppressive bone niche permits metastatic colonization and growth. Pharmacological inhibition of arginase or dietary arginine supplementation restores T cell function in the bone microenvironment, reduces MDSC accumulation, and significantly decreases bone metastasis burden in mouse models. Combination with immune checkpoint blockade shows enhanced efficacy, suggesting that targeting arginine metabolism could sensitize bone metastases to immunotherapy.
Why it matters: Bone metastases are a major source of cancer morbidity and are notoriously resistant to immunotherapy — the bone microenvironment has been considered an immune-privileged site, but the molecular mechanisms underlying this privilege have been unclear. This study identifies a specific metabolic mechanism — tumour cell arginine consumption driving T cell starvation — that is targetable. The finding that ARID1A loss, one of the most common cancer mutations, drives this process through epigenetic regulation of arginase suggests that ARID1A-mutant cancers may be selectively vulnerable to arginine-targeted therapies, providing a biomarker-stratified therapeutic approach.
Why for Yiru: Organ-specific TMEs — the unique immune and metabolic environments of different metastatic sites — are an emerging frontier. The bone TME is particularly understudied from a computational perspective because bone tissue is technically challenging for single-cell and spatial omics. This study provides a concrete metabolic pathway (ARID1A→arginase→arginine depletion→T cell suppression) that could be interrogated in existing transcriptomic and metabolomic data from bone metastases to identify patients who would benefit from arginine-targeted therapy.
Cross-disciplinary watchlist
Other Fields
β-Arrestin Condensates Regulate G-Protein-Coupled Receptor Function
Nature Published 2026-05-27 research article DOI: 10.1038/s41586-026-10682-6
β-arrestin GPCR biomolecular condensate phase separation signal transduction endocytosis cell signaling
Summary: Discovers that β-arrestins — the universal regulators of G protein-coupled receptor (GPCR) signaling — form biomolecular condensates upon receptor activation, and that this phase separation is essential for their canonical functions in receptor desensitization, endocytosis, and signaling. GPCRs are the largest family of drug targets, and β-arrestins are the key proteins that terminate G protein signaling (desensitization), recruit the endocytic machinery to internalize activated receptors, and serve as scaffolds for G protein-independent signaling cascades including MAP kinase activation. Despite decades of study, how a relatively small number of β-arrestin molecules can coordinate these diverse functions at hundreds of different GPCRs has been unclear. This study shows that activated GPCRs — through their phosphorylated C-terminal tails — nucleate the formation of β-arrestin condensates that concentrate the endocytic adaptor AP-2, clathrin, and signaling kinases (ERK, JNK) to create localized reaction hubs. The multivalent interactions between β-arrestin, receptor phospho-tail, and phosphoinositides in the plasma membrane drive liquid-liquid phase separation, forming micron-sized puncta visible by live-cell imaging. Mutations that disrupt β-arrestin condensation — without affecting receptor binding — impair receptor internalization and G protein-independent signaling, demonstrating that condensate formation is functionally essential. The condensates are dynamic, dissolving as receptors are internalized, and their stability is tuned by the phosphorylation pattern ('barcode') on the receptor tail, providing a mechanistic basis for how different GPCRs produce different arrestin-mediated outcomes.
Why it matters: GPCRs are the target of approximately one-third of all FDA-approved drugs, and β-arrestin-biased ligands — drugs that activate arrestin signaling while blocking G protein signaling, or vice versa — are a major frontier in pharmacology. The discovery that arrestin function depends on phase separation adds a new dimension to drug development: ligands may differ not just in which pathways they activate but in the physical properties (size, stability, dynamics) of the condensates they nucleate. This could explain the complex, often nonlinear pharmacology of GPCR drugs and provides new design principles for biased agonism. More broadly, the concept that signal transduction uses phase separation is expanding beyond transcription to membrane-proximal signaling.
Why for Yiru: GPCRs are increasingly recognized as important modulators of immune cell function in the TME — chemokine receptors (a major GPCR subfamily) direct immune cell trafficking, and specific GPCRs regulate T cell activation, macrophage polarization, and tumour cell survival. Understanding that GPCR-arrestin signaling occurs in phase-separated condensates rather than through simple binary interactions could reframe how we think about chemokine gradients and GPCR drug action in the TME — drug effects may depend on whether they promote or dissolve arrestin condensates at specific subcellular locations.
Universal Transcriptomic Hallmarks of Mammalian Ageing and Mortality
Nature Published 2026-05-27 research article DOI: 10.1038/s41586-026-10674-6
ageing transcriptomics mortality gene expression biological age longevity cross-species RNA-seq
Summary: Identifies a universal set of transcriptomic changes that occur across mammalian tissues with ageing and that predict both chronological age and time to death — representing a conserved gene-expression signature of the ageing process. The study analyzes RNA-seq data from over 40 tissues across multiple mammalian species (mouse, rat, human, and several long-lived species including naked mole-rat and bat), spanning different lifespans and ageing trajectories, to identify gene expression changes that are consistently associated with ageing independent of species, tissue, or environment. The core ageing signature includes upregulation of inflammatory and stress-response genes (consistent with inflammageing) and downregulation of genes involved in mitochondrial function, protein homeostasis, and DNA repair. A key finding is that this signature predicts not just chronological age but also remaining lifespan — individuals whose transcriptomes appear 'older' than their chronological age have significantly higher near-term mortality risk, establishing these gene expression changes as functional biomarkers of biological ageing rather than passive correlates. The signature is enriched for genes that extend lifespan when modulated in model organisms, suggesting that the identified transcriptomic changes are causal drivers of ageing rather than consequences. Remarkably, long-lived species (naked mole-rat, bat) show attenuated versions of the same signature, suggesting that longevity evolves by slowing or resisting these universal ageing-associated transcriptomic changes.
Why it matters: Ageing is the dominant risk factor for most chronic diseases including cancer, cardiovascular disease, and neurodegeneration — yet ageing itself has not been a direct therapeutic target because we lack robust, actionable biomarkers of biological age. A universal transcriptomic ageing clock that predicts mortality risk provides both a tool for evaluating anti-ageing interventions (do they reverse the transcriptomic signature?) and a roadmap for identifying the core pathways that drive ageing across species. The cross-species conservation of the signal is powerful evidence that these transcriptomic changes are fundamental to the ageing process, not artefacts of any particular model system.
Why for Yiru: Age is a major confounder in cancer immunology — older patients respond differently to immunotherapy, and the aged TME is distinct from the young TME in ways that are poorly understood. A universal transcriptomic ageing signature could be used to 'age-normalize' TME gene expression data, distinguishing age-related changes from cancer-specific changes. This is particularly relevant for understanding why immunotherapy efficacy varies with age and whether age-associated immune dysfunction in the TME can be therapeutically reversed.
Spatiotemporal Transcriptome Atlas of Human Embryos After Gastrulation
Nature Published 2026-05-27 research article DOI: 10.1038/s41586-026-10680-8
human embryo gastrulation spatial transcriptomics developmental biology organogenesis cell atlas single-cell
Summary: Presents a comprehensive spatiotemporal transcriptome atlas of human embryonic development from gastrulation (Carnegie stage 7, ~post-conception day 16) through early organogenesis (Carnegie stage 13, ~post-conception day 30), capturing the emergence of all major organ systems at single-cell resolution with spatial context. Human embryonic development during the first month after gastrulation — when the three germ layers (ectoderm, mesoderm, endoderm) are specified and begin forming organ primordia — has been largely inaccessible to molecular analysis due to the small size of embryos and ethical constraints on research. This study overcomes these limitations by applying spatial transcriptomics and single-cell RNA-seq to a rare collection of intact human embryos spanning this critical developmental window. The resulting atlas maps the molecular programs that drive specification of the neural tube, heart, gut tube, somites, limb buds, and primordial germ cells, revealing both conserved features shared with mouse development and human-specific innovations. Key findings include: the identification of transient progenitor populations that exist only during this window and have no adult counterpart; the discovery that organ-specific transcriptional programs are initiated earlier than previously appreciated, often before morphological structures are visible; and the mapping of signaling centers (organizers) that pattern adjacent tissues through secreted morphogens — including novel human-specific signaling interactions between the developing neural tube and adjacent mesoderm. The atlas is made available as an interactive resource for the developmental biology community.
Why it matters: The period immediately after gastrulation is when the human body plan is established and when most major congenital abnormalities originate — yet it has been a 'black box' of human biology. This atlas illuminates the molecular events of early human organogenesis at unprecedented resolution, providing a reference for understanding birth defects, for guiding stem cell-derived organoid differentiation toward authentic human cell types, and for identifying developmental pathways that are aberrantly reactivated in cancer. The discovery of transient human-specific progenitor populations could explain why some human tissues regenerate poorly compared to their mouse counterparts.
Why for Yiru: Developmental pathways are frequently reactivated in cancer — tumours hijack embryonic programs for proliferation, migration, and immune evasion. A comprehensive map of which developmental programs are active in which cell lineages during human organogenesis could serve as a reference for identifying which embryonic programs are aberrantly active in specific tumour types, potentially revealing developmental vulnerabilities that could be therapeutically targeted. This is particularly relevant for cancers with stem-like features and for understanding tumour cell plasticity.
Cellular Water-Potential Sensing Through Biomolecular Condensation
Nature Published 2026-05-27 research article DOI: 10.1038/s41586-026-10676-4
water potential biomolecular condensate osmotic stress phase separation osmosensing cell volume regulation intrinsically disordered protein
Summary: Discovers that cells sense and respond to changes in water potential — the thermodynamic availability of water — through the regulation of biomolecular condensates, establishing a direct physical mechanism for cellular osmosensing. All cells must maintain appropriate volume and hydration in the face of osmotic challenges, and the molecular mechanisms by which cells sense water status have been surprisingly elusive — unlike sensing of specific solutes or metabolites, water itself has been considered difficult to detect directly. This study shows that biomolecular condensates formed by intrinsically disordered proteins (IDPs) are exquisitely sensitive to cellular water potential because the phase separation that drives their formation depends on the effective concentration of macromolecules, which changes as water enters or leaves the cell. Under hyperosmotic stress (water leaving the cell, increasing macromolecular crowding), condensates form or grow; under hypoosmotic stress (water entering, decreasing crowding), condensates dissolve. The authors identify specific IDPs whose condensation state serves as a cellular water-potential sensor — their condensation triggers downstream signaling cascades that activate volume recovery mechanisms including ion transporter regulation and organic osmolyte synthesis. Disruption of condensate-based osmosensing impairs cellular volume regulation and survival under osmotic stress. The system functions as a direct physical sensor — no dedicated receptor or signaling cascade is required for the initial detection, which is an emergent property of the cytoplasmic phase separation landscape responding to changes in molecular crowding.
Why it matters: Osmotic stress is a universal cellular challenge — cells in the kidney medulla, skin, and intestine experience extreme osmotic fluctuations, and tumour cells in the TME face osmotic stress from necrosis, inflammation, and aberrant metabolism. The discovery that cells sense water status through the physical behavior of condensates rather than through dedicated receptor proteins represents a fundamentally different paradigm for cellular sensing — one based on the collective material properties of the cytoplasm rather than on lock-and-key molecular recognition. This 'physical sensing' modality may represent a general principle by which cells detect and respond to mechanical, osmotic, and crowding-related perturbations.
Why for Yiru: The TME is characterized by abnormal physical properties — increased interstitial pressure, altered osmolarity due to necrosis and metabolic activity, and changes in extracellular matrix stiffness that affect cellular crowding. If condensate-based water-potential sensing operates in TME-resident cells, then osmotic and physical perturbations in tumours could be directly altering gene expression programs in both cancer and immune cells through condensate dynamics — adding a biophysical dimension to TME signaling that is invisible to standard transcriptomic analyses.
Mechanism of Age-Related Accumulation of mtDNA Mutations in Human Blood
Nature Published 2026-05-27 research article DOI: 10.1038/s41586-026-10672-8
mitochondrial DNA mtDNA ageing mutation clonal expansion hematopoiesis replication
Summary: Elucidates the molecular mechanism by which mitochondrial DNA (mtDNA) mutations clonally expand in human blood cells with age — resolving a long-standing puzzle in mitochondrial genetics. Somatic mtDNA mutations accumulate with age in many tissues, and certain mutations can expand to high levels within individual cells (reaching homoplasmy) and across cell populations through clonal expansion of the host cell. In the hematopoietic system, age-related clonal expansion of mtDNA mutations is associated with altered immune function and increased risk of hematologic malignancy, but the mechanism driving clonal expansion has been debated — proposals include replicative advantage of cells with certain mtDNA mutations, random genetic drift, or selective advantage through altered metabolism. This study uses single-cell multi-omics of human blood cells across ages to show that mtDNA mutations accumulate primarily through replication errors during hematopoietic stem cell (HSC) division, and that the mutations that clonally expand are those that confer a subtle proliferative advantage to HSCs — not through metabolic changes but through a mitochondria-to-nucleus retrograde signaling pathway that modulates chromatin state and self-renewal gene expression. The specific mtDNA mutations that expand are those affecting complex I of the electron transport chain, which trigger a modest increase in NADH/NAD+ ratio that activates the SIRT1 deacetylase, leading to epigenetic changes that bias HSC division toward self-renewal over differentiation. This mechanism explains why specific mtDNA mutations (rather than random ones) dominate in aged blood and provides a direct link between mitochondrial genetics and stem cell behavior.
Why it matters: Clonal hematopoiesis — the age-related expansion of blood cell clones with somatic mutations — is a major risk factor for hematologic malignancy, cardiovascular disease, and all-cause mortality. While most attention has focused on nuclear DNA mutations (DNMT3A, TET2, ASXL1), this study shows that mtDNA mutations drive a parallel form of clonal hematopoiesis through a distinct mechanism — mitochondrial retrograde signaling to the epigenome — that converges on similar self-renewal pathways. This expands the clonal hematopoiesis landscape and identifies the mtDNA-NADH-SIRT1 axis as a potential therapeutic target for modulating HSC aging.
Why for Yiru: Mitochondrial function and mtDNA mutations are increasingly studied in the TME — mitochondrial state influences T cell differentiation, macrophage polarization, and tumour cell metabolism. The finding that mtDNA mutations can bias stem cell behavior through retrograde epigenetic signaling suggests that mtDNA mutations in TME stem-like cells (cancer stem cells, T memory stem cells) could similarly influence their self-renewal and differentiation, adding a mitochondrial dimension to cellular hierarchy dynamics in tumours.