Research Radar — 2026-05-31
Methods & AI
Computational
HELIX: a scalable model for predicting context-dependent regulation of RNA splicing and isoform usage
Nature Computational Science Published 2026-05-19 research article DOI: 10.1038/s43588-026-00988-w
alternative splicing deep learning RNA-binding protein isoform transcriptomics long-read sequencing computational biology gene regulation
Summary: Presents HELIX, a hierarchical deep learning framework that predicts tissue- and condition-specific alternative splicing patterns and transcript isoform usage by jointly modeling pre-mRNA sequence features and RNA-binding protein (RBP) expression profiles. Alternative splicing is a fundamental mechanism of gene regulation that generates transcriptomic diversity — most human genes produce multiple isoforms — but predicting how splicing changes across tissues and disease states has remained a major challenge. HELIX addresses this by integrating two streams of information: the cis-regulatory sequence code embedded in pre-mRNA (splice sites, branch points, regulatory motifs) and the trans-regulatory environment encoded by RBP expression levels. The model leverages both short-read RNA-seq (for splicing quantification) and long-read sequencing data (for full-length isoform resolution) during training, enabling it to simultaneously predict splicing event outcomes and full transcript isoform usage. HELIX outperforms existing splicing prediction models on held-out tissue and condition benchmarks, and its hierarchical architecture allows it to generalize to unseen biological contexts. The model also provides interpretability — attention weights over sequence motifs and RBP contributions reveal which regulatory features drive predictions in specific contexts.
Why it matters: The splicing code — the rules by which sequence and regulatory proteins determine splice site selection — has been one of the great unsolved problems in molecular biology. HELIX represents a significant advance by making these predictions at scale across tissues and conditions, moving beyond the single-gene or single-event models that have dominated the field. This has immediate applications in rare disease diagnostics (predicting whether a variant of unknown significance disrupts splicing), cancer biology (where aberrant splicing drives oncogenesis), and therapeutic development (antisense oligonucleotides and small molecules that modulate splicing). The integration of long-read data for isoform-level prediction is particularly timely as long-read sequencing becomes more widely adopted.
Why for Yiru: Alternative splicing is increasingly recognized as a major source of neoantigens in cancer — tumour-specific splice variants can generate peptides presented by MHC-I that are recognized by T cells. A tool like HELIX could predict which splicing events are likely to generate TME-specific neoantigens by comparing tumour and normal tissue predictions. More broadly, the TME is characterized by altered RBP expression patterns in both tumour and immune cells; HELIX could be used to systematically predict how these RBP changes reshape the splicing landscape across all cell types in the TME.
HESpotEx: a dual-stream deep learning framework for spot-level gene expression prediction from histological images
Nature Computational Science Published 2026-05-15 research article DOI: 10.1038/s43588-026-00992-0
spatial transcriptomics digital pathology deep learning histology graph neural network gene expression prediction computational pathology
Summary: Introduces HESpotEx, a dual-stream multimodal deep learning framework that predicts spatial gene expression patterns directly from whole-slide histopathological images (WSIs), without requiring physical spatial transcriptomics assays. Spatial transcriptomics (ST) methods like Visium and MERFISH can map gene expression onto tissue architecture, but they remain expensive, low-throughput, and destructive — you cannot run ST on every clinical sample. HESpotEx solves this by learning the mapping from tissue morphology to gene expression: given an H&E-stained WSI, it predicts spot-level expression for thousands of genes. The dual-stream architecture uses a graph attention autoencoder to capture tissue morphology features and a graph convolution network decoder to predict spatial gene expression patterns, with a self-supervised training objective that aligns predicted and measured expression. The model can predict up to 5,457 genes at single-spot resolution, recovering known spatial patterns of immune infiltration, tumour-stroma boundaries, and functional tissue zones. Critically, HESpotEx generalizes across cancer types and tissue sources, demonstrating that histological morphology contains sufficient information to infer the underlying molecular landscape.
Why it matters: This is a practical tool that bridges the gap between the vast archive of existing H&E-stained slides in pathology departments worldwide and the new world of spatial transcriptomics. Every cancer patient has H&E slides taken; if those slides can be computationally transformed into spatial gene expression maps, it would unlock retrospective spatial analysis at unprecedented scale. This has immediate translational implications: predicting immune infiltration patterns, tumour subtype, and therapy response biomarkers from routine clinical images without additional assays. The approach also reduces the cost barrier to spatial biology, particularly for resource-limited settings.
Why for Yiru: The TME is defined by spatial organization — immune cells, stromal cells, and tumour cells are not randomly distributed but organized into niches that determine therapeutic outcomes. HESpotEx could be applied to large retrospective cohorts of H&E-stained TMA or biopsy slides to infer spatial immune landscapes, correlating predicted immune infiltration patterns with clinical outcomes across thousands of patients. This is exactly the kind of computational pathology approach that could reveal TME spatial features predictive of immunotherapy response without requiring prospective spatial assays.
Scoring gene importance by interpreting single-cell foundation models
Nature Biotechnology Published 2026-05-27 research article DOI: 10.1038/s41587-026-03112-5
single-cell foundation model gene importance attribution interpretability scRNA-seq deep learning computational biology
Summary: Introduces SIGnature, a framework for scoring gene functional importance by computing attribution scores from single-cell RNA-seq foundation models — large pretrained transformer models that have learned generalizable representations of gene expression across millions of cells. A fundamental problem in single-cell analysis is that expression level alone is a poor indicator of a gene's regulatory or functional importance: some highly expressed genes (e.g., housekeeping genes) are uninformative about cell state, while lowly expressed transcription factors can be master regulators. Foundation models, trained on massive scRNA-seq corpora, learn contextual representations where a gene's importance is reflected not in its absolute expression but in how much it contributes to the model's predictions. SIGnature extracts these attribution scores, demonstrating that they reduce technical noise, emphasize known regulatory genes, and enable robust cross-dataset comparisons — a persistent challenge in scRNA-seq analysis where batch effects confound gene-level analyses. The framework is released as a Python package supporting rapid gene set searches and cross-study queries, and is validated on rare disease cohorts where SIGnature attributions successfully identify disease-relevant genes that expression-based methods miss.
Why it matters: Single-cell foundation models are becoming the default starting point for scRNA-seq analysis, but most users treat them as black boxes that produce embeddings for clustering and visualization. SIGnature opens the black box, extracting biologically meaningful gene importance scores that can be used for biomarker discovery, drug target identification, and disease gene prioritization. This transforms foundation models from dimensionality reduction tools into hypothesis-generating engines. The approach is model-agnostic and can be applied to any scRNA-seq foundation model, making it broadly useful as the field converges on a few dominant pretrained models.
Why for Yiru: TME single-cell atlases contain hundreds of cell types and states across dozens of patients, and identifying which genes are functionally important — as opposed to just differentially expressed — is critical for distinguishing drivers from passengers. SIGnature could be applied to TME scRNA-seq atlases to identify the genes that foundation models consider most important for defining T cell exhaustion, macrophage polarization, or CAF activation states, providing a principled complement to conventional differential expression analysis.
TxPert: using multiple knowledge graphs for prediction of transcriptomic perturbation effects
Nature Biotechnology Published 2026-05-01 research article DOI: 10.1038/s41587-026-03113-4
perturbation prediction knowledge graph transcriptomics deep learning drug discovery gene regulation transfer learning
Summary: Presents TxPert, a latent-transfer-based deep learning method that uses multiple biomedical knowledge graphs to predict how genetic perturbations (knockouts, knockdowns, overexpression) will affect the transcriptome — a core problem in functional genomics and drug discovery. The challenge is that the space of possible perturbations across genes, cell types, tissues, and species is enormous; experimental screening cannot cover all combinations, so computational methods that generalize to unseen perturbations are essential. TxPert addresses this by encoding gene-gene relationships from multiple knowledge graphs (protein-protein interaction, pathway, gene ontology, co-expression networks) into a shared latent space, then training a predictor that transfers perturbation effects from well-characterized genes and cell types to novel contexts. The multi-KG approach allows the model to leverage complementary information — a gene's pathway membership may predict its perturbation effect even when direct experimental data is absent. TxPert outperforms existing methods on held-out perturbation prediction benchmarks and demonstrates practical utility in identifying genes whose perturbation would reverse disease-associated transcriptional signatures.
Why it matters: Predicting perturbation effects is the computational equivalent of genetic screening at scale — if we can accurately predict what happens when any gene is perturbed in any cell type, we can prioritize drug targets, anticipate side effects, and design combination therapies computationally. TxPert's multi-KG approach is notable because it leverages the vast curated knowledge in public biomedical databases, integrating structured biological knowledge with deep learning in a way that improves generalization. This is particularly important for rare cell types and understudied genes where direct perturbation data is sparse.
Why for Yiru: TME perturbation prediction — what happens to the tumour-immune ecosystem when we inhibit a specific kinase, block a checkpoint, or delete a metabolic enzyme — is directly relevant to understanding therapy mechanisms and identifying synergy. TxPert could be configured to predict how perturbations of TME-relevant genes (e.g., FASN, SHP2, ATR) affect the transcriptional state of tumour cells, T cells, and macrophages, enabling computational screening of potential TME-targeting interventions.
Partially shared multi-modal embedding learns holistic representation of cell state
Nature Computational Science Published 2026-02-25 research article DOI: 10.1038/s43588-025-00948-w
multi-omics single-cell representation learning autoencoder data integration modality latent space computational biology
Summary: Introduces APOLLO (Autoencoder with Partially Overlapping Latent space learned through Latent Optimization), a computational framework for multi-modal single-cell data integration that automatically determines how much information should be shared versus kept modality-specific. Current multi-omics integration methods typically force all modalities into a fully shared latent space (losing modality-specific information) or keep them entirely separate (missing cross-modal relationships). APOLLO learns a partially overlapping latent space where some dimensions are shared across modalities (capturing conserved cell state features) and others are modality-specific (preserving modality-unique information). The framework uses an autoencoder architecture with a novel optimization scheme that balances reconstruction fidelity against cross-modal alignment, tested on paired single-cell RNA-seq and ATAC-seq data, as well as imaging-based multi-modal measurements. The partially shared representation outperforms both fully shared and fully separate approaches on biological benchmarks including cell type classification and trajectory inference.
Why it matters: Single-cell multi-omics is rapidly becoming standard — 10x Multiome, CITE-seq, and spatial multi-omics platforms routinely measure multiple modalities from the same cell. But integration methods have lagged behind the data, forcing users to choose between losing modality-specific biology or missing cross-modal relationships. APOLLO's partially shared approach offers a principled middle ground, acknowledging that some biological information is shared across modalities (e.g., cell identity) while other information is modality-specific (e.g., chromatin state vs. protein expression). This is conceptually aligned with how biologists think about multi-modal data.
Why for Yiru: The TME is increasingly studied with multi-modal single-cell technologies that simultaneously measure transcriptomes, chromatin accessibility, and surface proteins from the same cells. APOLLO could be applied to TME multi-omics data to identify which aspects of immune cell states are shared across modalities (e.g., core exhaustion programs) versus modality-specific (e.g., chromatin priming for future activation), providing a more nuanced view of TME cell states than transcriptome-only analysis.
Biomedical discoveries
Biomedicine
Kupffer cell calibration of T cell responses via VSIG4–CD5 interaction promotes tumor evasion
Nature Immunology Published 2026-04-29 research article DOI: 10.1038/s41590-026-02510-w
Kupffer cell VSIG4 CD5 liver metastasis T cell immune evasion checkpoint tumour microenvironment
Summary: Reveals that VSIG4, an immune checkpoint molecule predominantly expressed by liver-resident Kupffer cells, engages CD5 on T cells to calibrate the sensitivity of T cell receptor (TCR) signaling — and that this physiological calibration mechanism is hijacked by liver metastases to evade T cell-mediated killing. The liver is the most common site of metastasis for gastrointestinal cancers and is notoriously resistant to immunotherapy, but the mechanisms underlying this resistance have been unclear. This study shows that the VSIG4–CD5 interaction has an antigen-affinity-dependent effect: it impedes activation of low-affinity CD8+ T cells (which would otherwise attack healthy liver tissue inappropriately) while permitting high-affinity T cell responses against strong antigens. This represents a physiological mechanism for preventing autoimmunity in the liver — a tissue constantly exposed to gut-derived antigens. However, liver metastases exploit this by presenting weakly immunogenic antigens that fall below the affinity threshold, effectively rendering tumour-specific T cells invisible. Genetic deletion or antibody blockade of VSIG4 restores T cell-mediated control of liver metastases in mouse models, identifying VSIG4 as a liver-specific immune checkpoint and a therapeutic target for hepatic metastases.
Why it matters: Liver metastases represent a major clinical challenge — they are common, respond poorly to immunotherapy, and are associated with dismal prognosis. The discovery that Kupffer cells actively calibrate T cell responses through VSIG4–CD5 explains why the liver is an immunologically privileged site for metastasis and provides a actionable target. Anti-VSIG4 antibodies could specifically reverse immune suppression in the liver without systemic immune activation, potentially transforming the treatment of hepatic metastases from colorectal, pancreatic, and other cancers. This is also a beautiful example of how physiological immune regulation (preventing liver autoimmunity) can be co-opted by cancer cells.
Why for Yiru: The liver TME is distinct from other metastatic sites and remains poorly understood at the computational level — most TME atlases focus on primary tumours or lung metastases. This study identifies a specific, targetable molecular axis (VSIG4–CD5) that could be interrogated in liver metastasis single-cell and spatial data to assess its relevance across cancer types and patients. The antigen-affinity-dependent mechanism also suggests opportunities for computational prediction of which tumour neoantigens are most likely to be affected by VSIG4-mediated calibration.
Tumor-derived branched-chain α-keto acids activate Notch signaling in tumor-associated macrophages to limit immunity
Nature Immunology Published 2026-04-14 research article DOI: 10.1038/s41590-026-02484-9
BCKA branched-chain amino acid Notch tumor-associated macrophage TAM metabolism tumour microenvironment immune suppression
Summary: Demonstrates that tumour cells actively secrete branched-chain α-keto acids (BCKAs) — the deaminated catabolites of branched-chain amino acids (BCAAs) — into the tumour microenvironment, where they are taken up by tumour-associated macrophages (TAMs) and activate Notch signaling to drive immunosuppressive macrophage polarization. BCAA metabolism is well-known to support tumour cell growth through mTORC1 activation, but the fate and function of the downstream BCKA metabolites in the TME has been largely ignored. Using clinical samples and genetically engineered mouse models, this study shows that tumour cells overexpress the branched-chain aminotransferase BCAT1, which generates BCKAs from BCAAs, and then secrete these α-keto acids through monocarboxylate transporters. TAMs take up BCKAs, which directly activate Notch intracellular domain (NICD) signaling — not through the canonical ligand-receptor mechanism but through a metabolite-driven pathway — reprogramming TAMs toward an immunosuppressive phenotype that promotes tumour progression. Genetic or pharmacological interruption of BCKA secretion or Notch signaling in TAMs restores anti-tumour immunity and suppresses tumour growth. This establishes BCKAs as immunometabolites that function as intercellular signals in the TME, linking tumour BCAA metabolism to macrophage-mediated immune suppression.
Why it matters: Immunometabolism has focused primarily on how metabolic pathways within immune cells affect their function (e.g., glycolysis in effector T cells, fatty acid oxidation in Tregs). This study expands the paradigm to intercellular metabolic communication: tumour cells don't just compete with immune cells for nutrients, they actively export specific metabolites that reprogram immune cells. The BCKA–Notch axis is a concrete, targetable mechanism — inhibiting BCAT1 or blocking BCKA transporters could disrupt this immunosuppressive communication channel without directly affecting immune cell metabolism. This also suggests that dietary BCAA restriction might have immunomodulatory effects beyond tumour cell growth inhibition.
Why for Yiru: Tumour-immune metabolic crosstalk is an emerging dimension of TME biology that is poorly captured by standard transcriptomic analyses. The BCKA–Notch axis is a specific, testable metabolic communication mechanism that could be computationally modeled using metabolic flux analysis integrated with cell-cell communication inference from single-cell data. Identifying which metabolites mediate which immunosuppressive signals in the TME is a frontier for computational immunometabolism.
Integration of donor microbiota following FMT correlates with anti-PD-1 response in melanoma
Nature Communications Published 2026-05-30 research article DOI: 10.1038/s41467-026-73465-7
fecal microbiota transplantation FMT microbiome anti-PD-1 melanoma immunotherapy metagenomics strain-resolved
Summary: Performs a strain-resolved metagenomic meta-analysis across three independent clinical trials of fecal microbiota transplantation (FMT) combined with anti-PD-1 therapy in melanoma (n=41 total), revealing that therapeutic benefit is associated with successful integration of donor microbiota rather than engraftment of any specific bacterial species. FMT has shown promise in overcoming resistance to checkpoint immunotherapy, but identifying the microbial features that mediate this effect has been challenging — different studies have identified different 'responder-associated' species, leading to confusion about which bacteria matter. This meta-analysis resolves the discrepancy by shifting the analytical framework from species-level presence/absence to strain-level engraftment dynamics: responders consistently showed greater acquisition of donor-derived strains, higher post-FMT similarity to their donor's microbiome, and more stable microbiota composition over time. No single species or community type predicted response across all cohorts; instead, the degree of donor microbiota integration — how well the recipient's gut ecosystem accepted and maintained the transplanted community — was the consistent correlate of clinical benefit. This suggests that FMT efficacy depends less on delivering specific 'good' bacteria and more on the ecological process of community integration, which may reflect host factors like immune status, diet, and baseline microbiome configuration.
Why it matters: The microbiome-immunotherapy field has been plagued by inconsistent results across studies, with different 'responder-associated' taxa identified in different cohorts. This meta-analysis reframes the question: it's not about which species you give, but whether the recipient's ecosystem successfully integrates the donor community. This has profound implications for FMT trial design — donor-recipient compatibility may matter more than donor identity, and pre-FMT conditioning to facilitate engraftment may be more important than selecting the 'right' donor. The strain-resolved analytical approach used here also sets a methodological standard for microbiome biomarker studies.
Why for Yiru: The gut microbiome–TME axis is an area where computational methods can bridge distinct data types — integrating metagenomic data with tumour transcriptomic and immune profiling data to understand how gut microbial metabolites, immune-modulatory molecules, and systemic immune tone affect the TME. Strain-resolved metagenomics is technically demanding but provides resolution that species-level analysis misses; adopting these methods for TME-microbiome correlation studies could reveal mechanistic links that are invisible at coarser taxonomic resolution.
αKG-mediated carnitine synthesis drives DNA repair via histone acetylation
Nature Published 2026-05-27 research article DOI: 10.1038/s41586-026-10584-7
α-ketoglutarate αKG carnitine DNA repair histone acetylation homologous recombination metabolism cancer
Summary: Reveals a metabolic pathway linking the TCA cycle intermediate α-ketoglutarate (αKG) to DNA repair through carnitine synthesis and histone acetylation, with direct implications for cancer therapy. Homologous recombination (HR) deficiency sensitizes cancers to DNA-damaging agents like PARP inhibitors and platinum chemotherapy, but in HR-proficient cancers, the metabolic determinants of DNA repair competency have been unclear. This study identifies that αKG — a metabolite at the intersection of the TCA cycle, amino acid metabolism, and dioxygenase activity — is required for carnitine synthesis. Carnitine-derived acetyl groups fuel histone acetylation at DNA repair loci, opening chromatin to allow repair machinery access. When αKG is depleted, carnitine synthesis drops, histone acetylation at repair sites decreases, chromatin becomes inaccessible, and HR-proficient cells become functionally HR-deficient and sensitized to DNA-damaging agents. Critically, this effect is independent of αKG's canonical role as a dioxygenase cofactor, representing a distinct metabolic-epigenetic mechanism. The pathway can be therapeutically targeted by inhibiting enzymes in the carnitine synthesis pathway, providing a strategy to induce chemical HR deficiency in otherwise resistant tumours.
Why it matters: Inducing HR deficiency in HR-proficient tumours has been a major goal in oncology because it would expand the population of patients who benefit from PARP inhibitors and platinum agents. This study identifies a metabolic route to achieve this — targeting carnitine synthesis — that is mechanistically distinct from genetic HR deficiency. The αKG-carnitine-histone acetylation axis also connects systemic metabolism (TCA cycle activity, nutrient availability) to genomic integrity, suggesting that dietary or pharmacological interventions affecting αKG levels could modulate DNA repair capacity. This is a compelling example of how metabolism, epigenetics, and DNA repair are integrated.
Why for Yiru: The TME is metabolically heterogeneous — different regions of a tumour have different nutrient availability, oxygenation, and metabolic activity. If αKG availability varies spatially within a tumour, then DNA repair capacity may also vary spatially, creating regions of relative HR deficiency that could be preferentially targeted by DNA-damaging therapies. Computational integration of spatial metabolomics and spatial transcriptomics data could reveal whether metabolic heterogeneity creates drug sensitivity heterogeneity in the TME.
Purine salvage pathway protects CD8+ T cells from metabolic stress
Nature Immunology Published 2026-04-13 research article DOI: 10.1038/s41590-026-02491-w
CD8+ T cell purine salvage metabolic stress high-fat diet oxidative stress antitumor immunity metabolism immunometabolism
Summary: Shows that even temporary exposure to a high-fat diet (HFD) causes lasting metabolic damage to CD8+ T cells that impairs anti-tumour immunity for weeks after returning to a normal diet, and identifies the purine salvage pathway as a critical protective mechanism. The study found that HFD-induced metabolic stress causes persistent changes in the CD8+ T cell metabolome, including accumulation of peroxidation-sensitive phospholipids and depletion of antioxidants, which compromise CD8+ T cell survival and effector function. Under oxidative stress, CD8+ T cells upregulate the xanthine salvage pathway to produce guanosine and other purine nucleotides that protect against oxidative damage. The lasting nature of the metabolic impairment — persisting 10 weeks after diet normalization — suggests that HFD induces a form of metabolic memory in CD8+ T cells that is not easily reversed. Importantly, the purine salvage pathway can be therapeutically enhanced to protect CD8+ T cells during oxidative stress, suggesting strategies to preserve anti-tumour immunity in metabolically compromised hosts.
Why it matters: Obesity and metabolic syndrome are major risk factors for cancer and are associated with impaired immunotherapy responses, but the mechanisms linking systemic metabolism to T cell function have been unclear. This study provides a mechanistic explanation — HFD induces lasting oxidative damage in CD8+ T cells — and identifies a specific protective pathway (xanthine salvage) that could be therapeutically targeted. The finding that metabolic damage persists long after diet normalization is clinically significant, suggesting that patients with a history of obesity may have impaired T cell function even after weight loss, and may benefit from interventions that enhance purine salvage.
Why for Yiru: T cell metabolism in the TME is shaped by both local tumour-derived metabolic stress (hypoxia, nutrient depletion, lactate) and systemic host metabolism (obesity, diabetes, diet). Understanding how these systemic and local metabolic stressors interact to impair T cell function is critical for predicting immunotherapy outcomes. The purine salvage pathway represents a metabolic vulnerability/protection axis that could be incorporated into computational models of T cell metabolic fitness in the TME.
Cross-disciplinary watchlist
Other Fields
Human haematopoietic stem cells remember inflammatory stress
Nature Published 2026-05-27 research article DOI: 10.1038/s41586-026-10522-7
haematopoietic stem cell HSC inflammation epigenetic memory single-cell multiomics ageing haematopoiesis trained immunity
Summary: Demonstrates that human haematopoietic stem cells (HSCs) encode lasting epigenetic memory of inflammatory stress through stable chromatin remodeling, fundamentally changing how they respond to future challenges. While it has long been known that inflammation affects haematopoiesis — driving emergency myelopoiesis at the expense of lymphopoiesis — whether human HSCs retain a durable record of past inflammatory exposures has been unknown. Using xenograft inflammation-recovery models combined with single-cell multiomics (scRNA-seq + scATAC-seq), the authors identified two transcriptionally and epigenetically distinct HSC subpopulations that emerge after inflammatory challenge and persist after the inflammation resolves. These 'memory HSCs' maintain open chromatin at inflammatory response gene loci and closed chromatin at quiescence and self-renewal loci, producing a sustained myeloid-biased differentiation output. The epigenetic memory is stable through cell division and can be transmitted to daughter cells, representing a form of cellular learning at the level of the most primitive blood-forming cell. The study provides a mechanistic foundation for understanding how chronic inflammation — from infections, autoimmune disease, or ageing ('inflammageing') — progressively erodes HSC function and biases the hematopoietic system toward myeloid output.
Why it matters: The concept that stem cells can 'remember' past experiences has profound implications for understanding ageing, disease susceptibility, and therapeutic strategies. If HSCs accumulate epigenetic scars from every infection and inflammatory episode, this could explain why the hematopoietic system becomes progressively myeloid-biased with age, why older individuals have impaired immune responses to new infections and vaccines, and why chronic inflammatory diseases are associated with increased risk of hematologic malignancies. The finding that this memory is epigenetically encoded raises the possibility of epigenetic 'rejuvenation' strategies that could erase harmful inflammatory memories from aged HSCs.
Why for Yiru: The TME is a chronically inflamed environment, and tumour-infiltrating immune cells — including HSCs mobilized from the bone marrow — are exposed to sustained inflammatory signaling. If HSCs acquire and transmit inflammatory memory, this could affect the quality of immune cells recruited to the TME, potentially biasing toward myeloid-derived suppressor cells rather than effective anti-tumour lymphocytes. Understanding how systemic inflammation shapes the HSC compartment could inform strategies to improve hematopoietic function in cancer patients and enhance immunotherapy responses.
β-Arrestin condensates regulate G-protein-coupled receptor function
Nature Published 2026-05-27 research article DOI: 10.1038/s41586-026-10539-y
β-arrestin GPCR phase separation condensate signal transduction biophysics cell signaling drug target
Summary: Discovers that β-arrestins — the multifunctional adaptor proteins that regulate signaling of G-protein-coupled receptors (GPCRs), the largest class of drug targets — undergo liquid-liquid phase separation to form biomolecular condensates that spatially organize and diversify GPCR signaling. β-arrestins are known to interact with a remarkably wide array of signaling effectors at many different GPCRs, mediating functions ranging from receptor desensitization to MAP kinase activation, but how a single adaptor protein family orchestrates such diverse outcomes has been a long-standing puzzle. This study shows that β-arrestins form condensates upon GPCR activation, and that these condensates function as signaling hubs: they concentrate specific signaling effectors while excluding others, effectively creating distinct signaling environments around different GPCRs. The condensate composition is determined by the specific GPCR and by post-translational modifications on β-arrestin, providing a mechanistic basis for how the same adaptor can produce different signaling outputs at different receptors. Disruption of β-arrestin condensation — through mutation of key domain interfaces — impairs specific GPCR signaling pathways while leaving others intact, demonstrating that the condensate is functionally required for signal diversification.
Why it matters: GPCRs are targeted by approximately 34% of all FDA-approved drugs, making them the single most important class of drug targets. Understanding how β-arrestins generate signaling diversity has direct implications for biased agonism — the development of drugs that activate beneficial GPCR pathways while avoiding harmful ones. The condensate model provides a new framework for understanding biased signaling: different ligands may stabilize different β-arrestin condensate compositions, explaining why chemically distinct agonists produce different biological effects at the same receptor. This has implications for designing better GPCR drugs with fewer side effects.
Why for Yiru: Biomolecular condensates are emerging as organizing principles across cell biology, and several recent papers have implicated condensates in T cell receptor signaling, innate immune sensing, and now GPCR signaling. Chemokine receptors — which are GPCRs that direct immune cell migration in the TME — signal through β-arrestins, and differential condensate formation at different chemokine receptors could explain how T cells integrate multiple chemotactic signals in the TME. Understanding condensate-level regulation of immune cell migration could reveal new intervention points for controlling T cell trafficking to tumours.
ACLY integrates metabolism and chromatin accessibility to enable B cell activation and humoral immunity
bioRxiv Published 2026-05-27 preprint DOI: 10.64898/2026.05.24.727510
ACLY acetyl-CoA metabolism chromatin accessibility B cell humoral immunity epigenetics germinal center plasmablast
Summary: Demonstrates that ATP-citrate lyase (ACLY) — the enzyme that converts citrate to acetyl-CoA, linking glucose metabolism to lipid synthesis and histone acetylation — is a critical gatekeeper of B cell activation and antibody production through its control of chromatin accessibility. B cell activation triggers a coordinated metabolic-epigenetic reprogramming program in which ACLY-derived acetyl-CoA fuels histone acetylation at specific genomic loci to open chromatin and enable the transcriptional programs required for activation, survival, and differentiation into antibody-secreting cells. Using genetic models and integrated multi-omics (ATAC-seq, RNA-seq, metabolomics), the study shows that ACLY-deficient B cells exhibit profound defects in TLR and BCR signaling despite normal development and homeostasis. The impact on chromatin accessibility is more pronounced than on transcription, suggesting ACLY establishes a permissive epigenetic landscape rather than directly driving gene expression. In vivo, B cell-intrinsic ACLY loss impairs antigen-specific antibody production, reduces germinal center and plasmablast formation, and deletion after activation still reduces plasmablast generation, indicating an ongoing requirement beyond the initial activation phase.
Why it matters: The metabolic requirements of lymphocyte activation have focused heavily on T cells, but B cell metabolism — particularly the metabolic-epigenetic interface — has been relatively neglected. ACLY sits at the nexus of glucose metabolism, lipid synthesis, and histone acetylation, and this study establishes it as a central integrator for humoral immunity. This has implications for vaccine design (metabolic adjuvants boosting ACLY activity), autoimmune disease (ACLY inhibition as B cell-targeted therapy), and understanding B cell malignancies where metabolic reprogramming is a hallmark.
Why for Yiru: B cells and tertiary lymphoid structures (TLS) in the TME are increasingly recognized as positive prognostic factors associated with better immunotherapy responses. The metabolic requirements for B cell activation and antibody production in the TME may be distinct from secondary lymphoid organs due to the metabolically hostile tumour environment. Understanding ACLY's role could inform whether TLS-resident B cells face metabolic constraints limiting their anti-tumour function, and whether metabolic interventions could boost intratumoural humoral immunity.
Chemokine-defined macrophage niches establish spatial organization of tumor immunity
Nature Immunology Published 2026-03-23 research article DOI: 10.1038/s41590-026-02445-2
macrophage chemokine spatial transcriptomics tumour microenvironment lung cancer immune niche tissue-resident single-cell
Summary: Uncovers a division of labor between tissue-resident and recruited macrophage subsets in lung cancer, revealing that chemokine-expressing interstitial macrophages (IMs) establish spatially organized immune niches that shape the anti-tumour immune response. Using single-cell and spatial transcriptomics, the study identifies two functionally opposing tissue-resident IM subsets: Cxcl13+CD206hi IMs that organize lymphoid-like niches supporting T and B cell responses, and Cxcl9+CD206lo IMs that are associated with immune exclusion. Recruited macrophages (Ly6c2+Fn1+Vcan+ recMacs) occupy distinct spatial niches and contribute to immunosuppression. The chemokine expression profile of each macrophage subset determines which immune cells are recruited to their niche, effectively establishing a chemokine-based spatial code that organizes tumour immunity. Depletion or reprogramming of specific macrophage subsets alters the spatial organization of T cells and affects tumour growth, demonstrating that macrophage identity — not just abundance — determines immune outcomes.
Why it matters: Macrophages are among the most abundant immune cells in solid tumours, but the relationship between macrophage heterogeneity and anti-tumour immunity has been unclear. This study reveals that macrophage function is intimately tied to spatial organization — it's not just what type of macrophage is present, but where it is located and what chemokine signals it produces that determines immune outcomes. This reframes macrophage-targeted therapy from simply depleting 'bad' macrophages to reprogramming their spatial organization, with chemokine axes representing specific, druggable targets for reshaping the TME spatial architecture.
Why for Yiru: Spatial organization is the defining feature of the TME, and chemokine-mediated cell recruitment is the mechanism that establishes this organization. This study provides a detailed map of the chemokine code used by different macrophage subsets, which is directly relevant for computational models of TME spatial organization. Cell-cell communication inference from spatial transcriptomics data could leverage these chemokine-receptor pairs to predict how perturbing specific macrophage subsets would reorganize the immune landscape.
Cellular water-potential sensing through biomolecular condensation
Nature Published 2026-05-27 research article DOI: 10.1038/s41586-026-10591-8
water potential biomolecular condensate phase separation osmotic stress SAM domain biophysics cell biology
Summary: Identifies a mechanism by which cells sense changes in water potential — the thermodynamic availability of water — through a protein that undergoes water-potential-dependent biomolecular condensation. Water is the universal solvent of life, and cells must constantly adapt to changes in water availability during osmotic stress, desiccation, and normal physiological processes, but how cells actually sense water potential at the molecular level has been unknown. The study identifies SAM8, a sterile alpha motif (SAM)-containing protein, that forms condensates in response to decreased water potential both in vivo (in plants during drought) and in vitro (in reconstituted systems). The condensation threshold is tuned to physiologically relevant water potential ranges, and SAM8 condensates are required for hyperosmotic stress tolerance and normal seed germination. Biophysical characterization reveals that SAM8 condensation is driven by the same physical principles that govern all phase-separating systems, but the condensation propensity is directly modulated by water activity — as water becomes less available, the effective concentration of SAM8 increases, crossing the phase separation threshold. This represents a direct physical mechanism for sensing water availability, distinct from canonical osmosensing pathways that rely on membrane tension or kinase cascades.
Why it matters: This is a conceptually novel finding — the idea that a cell can sense its hydration state through the physical chemistry of protein phase separation rather than through dedicated signaling pathways. Water potential affects every biochemical reaction in the cell, and a direct physical sensor would allow rapid, graded responses to osmotic stress. While the study is in plants, the physical principles are universal, and similar mechanisms likely operate in animal cells, particularly in tissues that experience osmotic fluctuations (kidney, skin, mucosa). The finding also connects two active fields — biomolecular condensates and cellular stress responses — in an unexpected way.
Why for Yiru: The TME is characterized by abnormal physical and chemical conditions including hypoxia, acidosis, and altered osmolarity due to necrosis, inflammation, and aberrant vascularization. If water potential sensing through condensates operates in mammalian cells, it could affect immune cell function, tumour cell survival, and drug distribution in the TME. Understanding how TME physical properties regulate condensate formation across cell types could reveal new dimensions of TME biology that are invisible to standard transcriptomic and proteomic analyses.