Research Radar — 2026-05-29
Methods & AI
Computational
UcTCRp: a TCRβ-based framework for quantitative MAIT- and iNKT-associated repertoire-state profiling
bioRxiv Published 2026-05-28 preprint DOI: 10.64898/2026.05.25.727598
TCR repertoire MAIT iNKT transformer immune repertoire deep learning unconventional T cell computational immunology
Summary: Presents UcTCRp, a computational framework that profiles mucosal-associated invariant T (MAIT) and invariant natural killer T (iNKT) cell states using only TCRβ chain sequences — without requiring paired α chains, tetramer staining, or transcriptomic data. MAIT and iNKT cells are unconventional T cells with semi-invariant TCRs that play critical roles in tissue immunity, cancer surveillance, and infectious disease, but their identification in public immune-repertoire datasets has been impossible because most archived datasets contain only TCRβ sequences. UcTCRp solves this by training a transformer-based model on over one million TCRβ sequences with curated cross-species MAIT, iNKT, and conventional T cell labels. The framework learns conserved TCRβ features — V-gene usage patterns and CDR3β sequence motifs — that distinguish unconventional from conventional T cells, employing V-matched negative sampling to prevent the model from relying on germline-segment shortcuts. In paired scRNA-seq/scTCR-seq datasets, UcTCRp accurately recovered transcriptome-defined MAIT and iNKT populations and identified additional MAIT-like candidates with receptor-level evidence that were missed by expression-based annotation alone. Calibration against paired single-cell references and synthetic spike-in experiments established robust operating characteristics for repertoire-level abundance estimation, enabling archived TCRβ-only datasets to be repurposed for systems-level studies of unconventional T cell biology across tissues, diseases, and therapeutic contexts.
Why it matters: The vast majority of existing immune-repertoire datasets are TCRβ-only, representing thousands of samples from cancer, autoimmune disease, and infectious disease cohorts that have been effectively inaccessible for unconventional T cell analysis. UcTCRp unlocks this massive archived resource, enabling retrospective discovery of MAIT and iNKT associations with clinical outcomes, treatment responses, and disease states at a scale that has never been possible. This is especially timely given growing recognition that unconventional T cells are important mediators of checkpoint immunotherapy response and tissue-specific immune surveillance.
Why for Yiru: Unconventional T cells — particularly MAIT cells — are abundant in mucosal tissues and the liver and are increasingly recognized as TME-resident populations with both pro- and anti-tumour functions. The ability to quantify MAIT and iNKT cell abundance from bulk TCRβ data means existing cancer cohort datasets can be mined for unconventional T cell correlates of immunotherapy response, tumour stage, and survival without any new experimental data generation. This is precisely the kind of computational immunology tool that enables biological discovery from archived data.
Individual-Specific Gaussian Graphical Models for Heterogeneous Populations with Application to Epigenetic Gene Regulation in Lung Adenocarcinoma
bioRxiv Published 2026-05-28 preprint DOI: 10.64898/2026.05.25.727641
graphical model precision matrix empirical Bayes lung adenocarcinoma multi-omics patient-specific heterogeneity gene regulation
Summary: Introduces SIREN (Sample-specific Inference via Regularized Empirical-Bayes Networks), a method that estimates patient-specific partial correlation networks from multi-omics data by combining a population-level empirical Bayes prior with a rank-1 individual-specific update. The core insight is that population-average networks obscure the inter-patient molecular heterogeneity that drives differential clinical outcomes — two patients with the same cancer subtype may have fundamentally different regulatory wiring that population-level methods cannot capture. Since a sample-specific precision matrix cannot be estimated from a single observation, SIREN uses a conjugate Inverse Wishart prior whose mean is the Oracle Approximating Shrinkage (OAS) estimator, yielding closed-form individual-specific posteriors without MCMC — making it computationally scalable to large cohorts. On simulated heterogeneous populations, SIREN achieves superior edge recovery over OAS, Ledoit-Wolf, and graphical Lasso, while remaining competitive in homogeneous settings. Applied to paired transcriptomic and methylomic profiles from lung adenocarcinoma (LUAD), SIREN identifies individual-specific gene-methylation regulatory edges that stratify patients by survival in ways invisible to population-level analysis, implicating chromatin remodeling and WNT signaling in epigenetic heterogeneity. The method is released as a Python package.
Why it matters: Precision medicine requires knowing not just which genes are aberrant in a patient's tumour but how their regulatory relationships differ from population norms — a mutated gene may have different downstream consequences depending on the patient's specific regulatory network context. SIREN provides a principled statistical framework for inferring these patient-specific networks, bridging the gap between population-level multi-omics analyses and truly personalized regulatory models. The closed-form solution without MCMC makes it practical for clinical-scale cohorts.
Why for Yiru: TME heterogeneity is a central challenge — two patients' tumours may have similar cellular compositions but different regulatory wiring that determines immunotherapy response. SIREN could be applied to paired single-cell multi-omic data from the TME to identify patient-specific regulatory networks that explain differential outcomes, particularly for identifying which gene-methylation edges distinguish responders from non-responders in immuno-oncology cohorts.
ATLAS: a scverse-compatible package for multi-omic single-cell trajectory inference integration
bioRxiv Published 2026-05-27 preprint DOI: 10.64898/2026.05.23.727175
trajectory inference single-cell multi-omics scRNA-seq scATAC-seq pseudotime scverse chromatin accessibility
Summary: Presents ATLAS (Advanced Trajectory Learning from multi-omics At Single-cell resolution), a scverse-compatible framework for trajectory inference in paired single-cell RNA-seq and ATAC-seq data. Most existing trajectory inference methods use only transcriptomic information, missing the chromatin-level regulatory dynamics that drive cell-state transitions. ATLAS integrates both modalities through Weighted Nearest Neighbor graphs, enabling transcriptomic and chromatin accessibility data to jointly inform pseudotime estimation, terminal-state identification, and fate probability inference within a unified multi-omic representation. The framework also enables joint exploration of transcription factor expression and target gene activity along pseudotime, providing direct access to the regulatory programs and chromatin-associated transitions that are invisible to transcriptome-only methods. Across synthetic and real datasets, ATLAS reconstructs coherent developmental trajectories, captures progressive fate commitment, and resolves biologically meaningful lineage structures that single-modality methods miss.
Why it matters: Cell-state transitions — differentiation, activation, exhaustion — are driven by coordinated changes in chromatin accessibility and gene expression. Methods that only look at one modality are seeing half the picture. ATLAS bridges this gap within the scverse ecosystem, making multi-omic trajectory inference accessible to the broader single-cell community. This is particularly important as paired scRNA+scATAC data becomes increasingly common through platforms like 10x Multiome.
Why for Yiru: T cell exhaustion and macrophage polarization in the TME are trajectory problems — cells transition through intermediate states that determine functional outcomes. Multi-omic trajectory inference that simultaneously models transcriptomic and epigenomic changes could identify the chromatin-level regulatory events that lock in dysfunctional T cell states or immunosuppressive macrophage phenotypes, revealing intervention points that transcriptome-only analysis would miss.
BIFO: A Biological Information Flow Ontology for Directed Propagation in Heterogeneous Biomedical Knowledge Graphs
bioRxiv Published 2026-05-28 preprint DOI: 10.64898/2026.05.25.727605
knowledge graph ontology information flow biomedical graph propagation random walk computational method
Summary: Introduces the Biological Information Flow Ontology (BIFO), a graph-agnostic specification defining which directed relationships in heterogeneous biomedical knowledge graphs are biologically admissible for computational signal propagation. Biomedical KGs integrate diverse entity types (genes, proteins, drugs, diseases, pathways) connected by many relationship types, but algorithms that propagate signal across these graphs — random walks, diffusion, message passing — implicitly assume that every edge can carry biological signal. In reality, hierarchical, lexical, and purely statistical edges do not represent admissible directed state transformations, and traversing them propagates signal along paths that are not biologically meaningful. BIFO defines fourteen entity classes, a taxonomy of flow classes organized around the central dogma backbone (Gene→RNA→Protein→Pathway→Cell→Phenotype→Disease), admissibility constraints, and a two-level CURIE mapping that can be applied to any graph. A four-step conditioning protocol converts a raw property graph into a conditioned propagation graph retaining only admissible, direction-aware edges. Applied to the Data Distillery Knowledge Graph, BIFO conditioning retained 70.7% of 33.6M edges, cleanly separating mechanistic from observational associations.
Why it matters: Knowledge graph-based methods are increasingly used for drug repurposing, target identification, and mechanism of action prediction, but their reliability suffers when propagation algorithms traverse non-biological edges. BIFO provides a principled, reusable framework for ensuring that computational analyses of biomedical KGs operate only on biologically meaningful paths, improving both interpretability and false-positive rates. This is infrastructure-level work that benefits all downstream graph-based biomedical AI.
Why for Yiru: Many computational TME analyses involve integrating heterogeneous data types — gene expression, protein interactions, pathway annotations, drug-target relationships — into graph-based frameworks. BIFO's approach to distinguishing mechanistic from observational edges could be applied to TME-specific knowledge graphs to ensure that computational predictions of drug effects, cell-cell interactions, or signalling pathways are grounded in admissible biological information flow rather than statistical associations.
FLOWR: Flow Matching for Structure-Aware De Novo, Interaction- and Fragment-Based Ligand Generation
Nature Computational Science Published 2026-05-28 research article DOI:
flow matching drug design ligand generation structure-based generative model deep learning computational chemistry
Summary: Describes FLOWR, a flow matching-based generative framework for structure-aware de novo ligand design that jointly models molecular interactions and fragment-based assembly. Flow matching — an emerging generative modeling paradigm that learns to transform simple prior distributions into complex data distributions through learned velocity fields — has recently shown promise for molecular generation, offering more stable training and better mode coverage than diffusion models. FLOWR adapts this framework to the structure-based drug design setting, conditioning ligand generation on protein binding pocket geometries and incorporating interaction-aware scoring during the generative process. The model supports both de novo generation from scratch and fragment-based elaboration, where a starting fragment is grown into a complete ligand while maintaining favorable binding interactions. Benchmarking against existing structure-based generative models demonstrates competitive or superior performance on standard drug design metrics including binding affinity, synthetic accessibility, and molecular drug-likeness, with particular advantages in generating ligands that maintain key protein-ligand interactions identified in co-crystal structures.
Why it matters: Structure-based de novo drug design — computationally generating novel molecules that bind a target protein — has the potential to dramatically accelerate early-stage drug discovery, but current methods suffer from generating molecules that are either not synthesizable, not drug-like, or that fail to maintain key binding interactions. Flow matching offers a mathematically elegant alternative to diffusion and reinforcement learning approaches, and FLOWR demonstrates that this new paradigm can produce high-quality, interaction-aware ligand designs suitable for medicinal chemistry follow-up.
Why for Yiru: While not directly TME-related, computational drug design tools are increasingly relevant to immuno-oncology — identifying small molecules that modulate immune checkpoints, metabolic enzymes, or signalling pathways in the TME. Being aware of state-of-the-art generative methods helps contextualize the growing pipeline of computationally designed immunomodulatory compounds that may eventually impact TME-targeted therapies.
Biomedical discoveries
Biomedicine
ATR Kinase Inhibitors Induce Mitochondrial Fission in CD8+ T Cells and Impair Immune Memory In Vivo
bioRxiv Published 2026-05-26 preprint DOI: 10.64898/2026.05.25.727628
ATR kinase CD8+ T cell mitochondrial fission immune memory DNA damage response immunotherapy clinical trial toxicity
Summary: Reveals a critical and previously unrecognized toxicity of ATR kinase inhibitors: they potently trigger unscheduled mitochondrial fission in actively dividing CD8+ T cells, causing loss of mitochondrial mass that persists in memory CD8+ T cells and impairs the formation of durable immune memory. ATR (ataxia telangiectasia and Rad3-related) kinase restrains CDK1 activity during S and G2 phases, confining CDK1-driven processes — including mitochondrial fission — to mitosis. ATR inhibitors were developed to potentiate chemotherapy by preventing DNA damage-induced cell cycle arrest, and are currently being evaluated in multiple clinical trials in combination with chemotherapy, radiation, and immune checkpoint inhibitors. This study shows that ATR inhibition disrupts cell cycle organization even in undamaged cells, causing premature CDK1 activation that triggers mitochondrial fragmentation outside of mitosis. In a mouse model of LCMV Armstrong infection, ATR inhibitor treatment during the peak of CD8+ T cell expansion significantly impaired the formation of functional memory CD8+ T cells. These findings carry immediate clinical implications: trials combining ATR inhibitors with immunotherapy — where durable anti-tumour T cell memory is the mechanism of action — may be undermining the very immune memory responses they aim to generate.
Why it matters: This is the kind of finding that can change clinical trial design. ATR inhibitors are in active clinical development with substantial industry investment, and their combination with immune checkpoint blockade is based on the rationale that increased tumour DNA damage will enhance immunogenicity. But if ATR inhibitors simultaneously cripple the T cells needed for durable response, the combination may be self-defeating. The study identifies mitochondrial fission as the mechanism, suggesting that mitigation strategies (e.g., mitochondrial fusion promoters) or alternative dosing schedules could preserve immune memory while retaining anti-tumour efficacy.
Why for Yiru: The T cell-intrinsic effects of cancer therapies are a major and underappreciated dimension of treatment response — most studies focus on how drugs kill tumour cells while neglecting effects on immune cells. This study exemplifies why cell-type-specific pharmacodynamics matter, particularly for drugs that target fundamental cell cycle machinery present in both tumour and T cells. Computational integration of drug-target networks with T cell state atlases could systematically predict such immune-toxicities before clinical trials.
Lipogenesis-Driven EGFR Palmitoylation Enables Metastatic Immune Evasion in Triple-Negative Breast Cancer
bioRxiv Published 2026-05-26 preprint DOI: 10.64898/2026.05.21.726063
lipogenesis EGFR palmitoylation FASN MHC-I immune evasion triple-negative breast cancer metastasis CD8+ T cell
Summary: Demonstrates that de novo lipogenesis drives metastatic immune evasion in triple-negative breast cancer (TNBC) through a previously unknown mechanism: fatty acid synthase (FASN)-mediated palmitoylation of EGFR creates a lipid-dependent membrane signaling scaffold that sustains PI3K-AKT-mTOR signaling, which in turn suppresses MHC-I antigen presentation through post-translational regulation. Metastasis is the leading cause of TNBC death, yet how disseminating tumour cells evade CD8+ T cell surveillance has been unclear. This study shows that FASN — the enzyme that synthesizes palmitate, the 16-carbon saturated fatty acid — attaches palmitate to EGFR (a post-translational lipid modification), anchoring EGFR in lipid raft membrane domains. This palmitoylated EGFR sustains PI3K-AKT-mTOR signaling independently of MAPK compensation, and this signaling actively suppresses MHC-I surface expression, allowing metastatic cells to escape CD8+ T cell killing. Genetic or pharmacological FASN inhibition, or expression of palmitoylation-deficient EGFR mutants, restores MHC-I surface expression, unleashes CD8+ T cell activation, and markedly impairs lung metastasis without affecting primary tumour growth. Importantly, this lipid-dependent axis is not rescued by exogenous lipids — the tumour cell must synthesize its own palmitate — making FASN a non-redundant metabolic vulnerability specifically for the metastatic immune evasion phenotype. Clinically, FASN inhibitors are advancing through clinical trials, providing a direct translational path.
Why it matters: Most cancer metabolism research focuses on how metabolic rewiring supports proliferation, but this study reveals that lipid metabolism is also a direct regulator of immune visibility — tumour cells use FASN not just to build membranes but to actively hide from T cells. The FASN→EGFR palmitoylation→MHC-I suppression axis is a concrete, targetable immunometabolic checkpoint, and since FASN inhibitors are already in clinical development, the translational path from this discovery to trials could be rapid. The finding that this mechanism is specific to metastasis — not primary tumour growth — also suggests that FASN inhibition may be particularly effective in the adjuvant setting to prevent metastatic recurrence.
Why for Yiru: Immunometabolism in the TME — how metabolic pathways in tumour cells directly regulate immune recognition — is a rapidly growing field. This study provides a mechanistic link from a specific metabolic enzyme (FASN) to a specific post-translational modification (palmitoylation) to a specific immune phenotype (MHC-I suppression and CD8+ T cell evasion). This kind of detailed mechanistic chain is exactly what computational TME models need to move beyond correlation to causation, and the FASN-EGFR palmitoylation axis could be interrogated in spatial multi-omic data to assess its relevance across cancer types and metastatic sites.
Fibroblast TGF-β3 Promotes Tissue-Residency and Survival of CD8 T Cells in Barrier Tissues and Tumors
bioRxiv Published 2026-05-22 preprint DOI: 10.64898/2026.05.20.726599
CD8+ T cell tissue-resident memory TRM fibroblast TGF-β3 tumour microenvironment immunotherapy stroma
Summary: Identifies fibroblast-derived TGF-β3 as a conserved stromal niche factor that specifically sustains CD8+ tissue-resident memory T cells (TRM) in barrier tissues and tumours, challenging the prevailing view that TGF-β signaling is primarily immunosuppressive. CD8+ TRM cells provide frontline protection against pathogens and contribute to tumour control, but the microenvironmental signals that maintain their persistence have been poorly defined. Through human single-cell cross-tissue atlases, the authors found that CD8+ TRM abundance preferentially correlates with fibroblast abundance across healthy barrier tissues and multiple tumour types, with TGFB3 emerging as a top fibroblast-enriched candidate mediator. In genetic mouse models, fibroblast-specific deletion of Tgfb3 reduced CD8+ TRM across barrier tissues at steady state and impaired antigen-specific TRM formation following viral infection. Critically, in tumour models, loss or antibody-mediated neutralization of TGF-β3 impaired CD8+ T cell residency and cytotoxicity, induced proteotoxic stress and apoptotic programs, and accelerated tumour growth — opposite to the effect expected from blocking an immunosuppressive cytokine. This isoform-specific finding explains the limited efficacy of pan-TGF-β blockade in cancer: global TGF-β inhibition may simultaneously block the TGF-β3 signal that sustains beneficial TRM populations, creating a therapeutic trade-off.
Why it matters: Pan-TGF-β blockade has been a major focus in immuno-oncology, but clinical results have been disappointing. This study provides a mechanistic explanation: not all TGF-β isoforms are immunosuppressive, and TGF-β3 from fibroblasts is actually required to keep tumour-fighting TRM cells alive and functional. This suggests that next-generation TGF-β-targeting therapies need isoform selectivity — blocking TGF-β1 (immunosuppressive) while sparing TGF-β3 (TRM-sustaining). More broadly, it reframes fibroblasts as active supporters rather than just barriers to anti-tumour immunity.
Why for Yiru: The fibroblast-T cell axis is an emerging frontier in TME biology that has been overshadowed by the focus on tumour-immune and myeloid-immune interactions. This study provides a specific molecular mechanism (fibroblast TGF-β3 → CD8+ TRM survival) that can be interrogated in spatial transcriptomic data by examining co-localization of TGFB3-expressing fibroblasts with TRM gene signatures. The isoform-specific finding also highlights the importance of transcript-level resolution in computational TME analyses — bulk TGFB signatures obscure functionally opposite isoform effects.
CA19-9 Induces Microenvironment Remodeling in Pancreatic Ductal Adenocarcinoma
bioRxiv Published 2026-05-26 preprint DOI: 10.64898/2026.05.22.727290
CA19-9 glycan pancreatic cancer PDAC tumour microenvironment cancer-associated fibroblast Treg IL-1α TGF-β
Summary: Elucidates how the aberrant glycan CA19-9 — the most widely used clinical biomarker for pancreatic ductal adenocarcinoma (PDAC) — is not merely a passive biomarker but an active driver of immunosuppressive TME remodeling through paracrine signaling. CA19-9 (sialyl Lewis A) is a carbohydrate antigen elevated in most PDAC patients and used clinically for monitoring disease progression, but its biological function in the TME has been unclear. Using a genetically engineered mouse model with inducible CA19-9 expression, the authors mapped TME changes upon CA19-9 elevation and found expansion of antigen-presenting cancer-associated fibroblasts (apCAFs) and regulatory T cells (Tregs) — both strongly immunosuppressive populations. Antibody blockade of CA19-9 restored normal histology and reduced apCAF and Treg abundance. Mechanistically, CA19-9 induces IL-1α and TGF-β expression in tumour cells, which reprogram pancreatic mesothelial cells into apCAFs; these apCAFs then directly ligate naive CD4+ T cells to drive Treg differentiation. Antibody blockade of IL-1α and TGF-β in mice reduced apCAF and Treg differentiation, phenocopying CA19-9 blockade. The study establishes CA19-9 as a glycan master regulator of the PDAC TME, bridging tumour cell glycobiology to stromal and immune reprogramming.
Why it matters: CA19-9 has been used clinically for decades as a disease monitoring biomarker, but this study reveals it is a biologically active molecule that actively shapes the immunosuppressive PDAC microenvironment. This transforms CA19-9 from a passive correlate of tumour burden into a potential therapeutic target — anti-CA19-9 antibodies could both block TME remodeling and serve as tumour-targeting agents. The identification of the IL-1α/TGF-β axis downstream of CA19-9 provides additional druggable nodes for intercepting this signaling cascade.
Why for Yiru: PDAC is one of the most immunotherapy-resistant cancers, and understanding the molecular drivers of its profoundly immunosuppressive TME is critical. The CA19-9→IL-1α/TGF-β→apCAF→Treg cascade is a multi-step signaling chain that could be interrogated computationally through cell-cell communication analysis of PDAC single-cell and spatial data. This kind of ligand-receptor cascade modeling is directly aligned with TME computational methods that infer intercellular signaling from multi-omic data.
Immunomodulatory Impact of PTPN11/SHP2-Based Vertical RAS-MAPK Pathway Inhibition in Pancreatic Cancer
bioRxiv Published 2026-05-27 preprint DOI: 10.64898/2026.05.23.727401
SHP2 PTPN11 RAS-MAPK MEK inhibitor pancreatic cancer PDAC TME immunomodulation adaptive resistance
Summary: Comprehensively maps the dynamic immunomodulatory effects of SHP2-based vertical RAS pathway inhibition in pancreatic cancer, from early treatment response through adaptive resistance. Allosteric SHP2 inhibitors are being evaluated in clinical trials as part of vertical RAS pathway combinations for KRAS-mutant malignancies including PDAC, but their effects on the tumour immune microenvironment have not been systematically characterized. Using human and murine PTPN11 knockout PDAC cell lines, the KPC autochthonous mouse model, and patient-derived organoids, this study finds that short-term dual MEK/SHP2 inhibition induces a mixed immunological response: increased T cell infiltration and reduced immunosuppressive M2-like macrophages, but simultaneously decreased mature dendritic cells and expanded monocytic myeloid-derived suppressor cells (M-MDSCs). These changes are associated with tumour-intrinsic upregulation of CXCR3 ligands and TGF-β, and increased expression of TIGIT and TIM-3 checkpoint ligands. With prolonged treatment and transition to adaptive resistance, the initial immune-activating effects are lost while immunosuppressive features prevail — M2 macrophages re-accumulate, DC maturation remains impaired, and checkpoint ligand expression is further enhanced. Dual SHP2/RAS inhibition recapitulates these effects, identifying a conserved mechanism. The study provides a rationale for combining SHP2-based regimens with TGF-β blockade, TIGIT/TIM-3 checkpoint inhibitors, or CD40 agonism to counteract emerging immune suppression.
Why it matters: RAS pathway-targeted therapies are finally reaching the clinic for PDAC, but durable responses are rare. Understanding how these therapies reshape the immune contexture — both the initial beneficial effects and the subsequent immunosuppressive adaptation — is essential for designing rational combinations. The finding that SHP2 inhibition creates a window of T cell infiltration that closes as resistance emerges suggests that immunotherapy combinations should be timed to coincide with early treatment, not added after progression. The specific checkpoint axes identified (TIGIT, TIM-3) provide actionable targets for clinical trial design.
Why for Yiru: The temporal dynamics of therapy-induced TME remodeling — how the immune landscape evolves during treatment — is a critical dimension that is often missed in static single-timepoint analyses. This study's longitudinal characterization of TME changes during targeted therapy provides a template for computational time-course analyses that could predict optimal immunotherapy combination windows from serial biopsy or liquid biopsy data.
Cross-disciplinary watchlist
Other Fields
PD-1 Remodels SHP2 Dynamics and Drives the Non-Catalytic Inhibition of T Cell Activation
bioRxiv Published 2026-05-22 preprint DOI: 10.64898/2026.05.20.725130
PD-1 SHP2 T cell activation condensate phase separation TCR signaling CD28 biophysics immunotherapy
Summary: Reveals that PD-1 suppresses T cell activation not primarily through SHP2's phosphatase activity — as the canonical model holds — but through a non-catalytic biophysical mechanism: PD-1 engagement remodels SHP2's conformation, exposing its tandem SH2 domains to directly dismantle TCR and CD28 signaling condensates. Using an integrated approach combining X-ray crystallography, single-molecule magnetic tweezers, supported lipid bilayers, TIRF imaging, and liquid-liquid phase separation assays, the study determines the crystal structure of SHP2 bound to dually phosphorylated PD-1 cytoplasmic motifs, revealing a ligand-induced open conformation distinct from the oncogenic E76K state. Single-molecule measurements directly resolve the closed-to-open conformational transition of individual SHP2 molecules — an 8-nm amplitude change — and show that phosphorylated PD-1 accelerates this transition by at least 94-fold. Critically, the open conformation exposes the tandem SH2 domains as a physical barrier that directly dismantles phase-separated pTCR and pCD28 signaling condensates, independent of phosphatase activity. This dual-layered inhibitory mechanism — enzymatic (phosphatase) plus biophysical (condensate disruption) — explains why PD-1 inhibition of proximal TCR/CD28 signaling is faster and more robust than a purely enzymatic model would predict, and suggests new strategies for immunotherapy design that target the condensate-disrupting function.
Why it matters: This study fundamentally revises our understanding of how the most clinically important immune checkpoint works. The finding that PD-1 operates via condensate disruption rather than purely through dephosphorylation opens a new dimension for therapeutic intervention — drugs that block SHP2's conformational opening or its condensate-disrupting activity could reverse T cell suppression without the side effects of complete SHP2 inhibition. For immunotherapy, this suggests that anti-PD-1 antibodies work not just by blocking ligand binding but by preventing SHP2 recruitment and subsequent condensate dismantling. The single-molecule biophysics approach also demonstrates how mechanistic structural biology can directly inform cancer immunotherapy.
Why for Yiru: Biomolecular condensates are increasingly recognized as organizing principles in T cell signaling — the TCR and CD28 form phase-separated signaling clusters upon activation, and PD-1's ability to physically disrupt these condensates represents a new mechanism of immune regulation. Understanding how the TME's biophysical properties (pH, crowding, metabolite concentrations) affect condensate stability could reveal why some tumours are more immunosuppressive than others, connecting TME metabolism to immune synapse biophysics through condensate biology.
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. Using genetic models and integrated multi-omics approaches (ATAC-seq, RNA-seq, metabolomics), the study shows that 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. ACLY-deficient B cells exhibit profound defects in TLR and BCR signaling ex vivo despite normal development and homeostasis, with a more pronounced impact on chromatin accessibility than on transcription — suggesting that 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 and glycolysis, 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 that boost ACLY activity), autoimmune disease (ACLY inhibition as a B cell-targeted therapy), and understanding why some B cell malignancies are sensitive to metabolic interventions.
Why for Yiru: B cells and tertiary lymphoid structures (TLS) are increasingly recognized as positive prognostic factors in the TME, associated with better immunotherapy responses. The metabolic requirements for B cell activation and antibody production in the TME may be distinct from those in secondary lymphoid organs due to the metabolically hostile tumour environment. Understanding ACLY's role in B cell chromatin regulation could inform whether TLS-resident B cells in tumours face metabolic constraints that limit their anti-tumour function, and whether metabolic interventions could boost intratumoural humoral immunity.
cGAS Bends Unpaired DNA to Form an Unconventional Structure That Hyperactivates the Innate Immune Response
bioRxiv Published 2026-05-25 preprint DOI: 10.64898/2026.05.22.726336
cGAS innate immunity DNA sensing cryo-EM single-molecule FRET phase separation STING structural biology
Summary: Discovers that cGAS — the cytosolic DNA sensor that triggers type I interferon responses through STING — recognizes DNA with unpaired regions ('bubble DNA') in a fundamentally different structural mode than fully complementary dsDNA, leading to cGAS hyperactivation. Canonically, cGAS forms 2:2 dimers on dsDNA and then oligomerizes into phase-separated condensates that potently activate STING. This study shows that DNA containing unpaired regions — such as those generated during transcription, recombination, or replication — causes cGAS to bind more tightly to the catalytic domain but suppresses condensation. Cryo-EM and single-molecule FRET reveal that cGAS forms 2:1 complexes by bending bubble DNA into a V-shape, using the unpaired region as a hinge, which limits oligomerization. Paradoxically, this non-canonical binding mode results in cGAS hyperactivation — the 2:1 complexes are more catalytically active per unit than the oligomeric 2:2 complexes, despite lacking phase separation. This uncovers a novel mode of pattern recognition where the same receptor produces different activation levels depending on the structural features of its ligand, adding a new dimension to innate immune sensing beyond the simple presence or absence of DNA.
Why it matters: The cGAS-STING pathway is a central innate immune sensor with roles in antiviral immunity, antitumour immunity, autoinflammatory disease, and cellular senescence. The finding that cGAS activation levels depend on DNA topology — not just DNA presence — has implications for all these contexts. During transcription, replication, and DNA repair, bubble DNA is transiently generated in the nucleus; if this DNA leaks to the cytosol (e.g., due to genomic instability in cancer cells or mitochondrial stress), it could produce a qualitatively different — and stronger — innate immune response than sheared genomic DNA. This may explain why certain types of DNA damage are more immunogenic than others, with direct relevance to understanding why some tumours are spontaneously immunogenic while others are not.
Why for Yiru: cGAS-STING signaling in the TME is a double-edged sword — chronic STING activation can promote T cell priming but sustained interferon signaling can drive immune suppression. The finding that different DNA structures produce quantitatively different cGAS activation levels suggests that the type of DNA released in the TME (from mitotic errors, micronuclei, mitochondrial damage, or dying cells) may determine whether the resulting innate immune response is beneficial or detrimental. This adds a structural biology dimension to TME innate immunity that could be interrogated through analysis of DNA damage patterns and cGAS activation signatures across tumour types.
NF1 Splicing Reprograms ERα Signaling to Promote Luminal Breast Cancer Progression
bioRxiv Published 2026-05-26 preprint DOI: 10.64898/2026.05.21.726647
alternative splicing NF1 neurofibromin ERα breast cancer endocrine resistance RNA-binding protein post-transcriptional regulation
Summary: Identifies an alternatively spliced NF1 isoform lacking the nuclear localization sequence (NLS) as a driver of luminal breast cancer progression and endocrine therapy resistance through reprogramming of ERα signaling. NF1 (neurofibromin) is a tumour suppressor and negative regulator of RAS, but its splicing regulation in cancer has been largely unexplored. Analysis of TCGA and AURORA cohorts revealed that the NF1 NLS-skipped isoform is enriched in metastatic tumours, preferentially in luminal subtypes, and associated with decreased overall survival independent of NF1 genomic mutations. CRISPR-engineered MCF7 models expressing the NLS-skipped isoform show abolished nuclear neurofibromin localization, enhanced ERK signaling, and resistance to endocrine and MAPK-targeted therapies. Surprisingly, despite increased ERα protein levels, canonical estrogen response programs were suppressed while KRAS, EMT, and inflammatory pathways were activated. Mechanistically, NLS-skipped NF1 expression increases RNA-bound ERα and reprograms RNA-binding protein networks (CELF, ESRP1, SRSF families), leading to widespread alternative splicing — defining a post-transcriptional mechanism by which a splicing event in one gene (NF1) cascades into transcriptome-wide splicing changes through ERα relocalization.
Why it matters: Alternative splicing is an emerging driver of cancer but has been studied primarily in the context of oncogenic fusion proteins or splice site mutations. This study reveals that splicing of a tumour suppressor can have far-reaching effects through relocalization of a transcription factor (ERα) to RNA regulatory functions — a mechanism that would be invisible to genomic sequencing and conventional RNA-seq analysis. The finding that NF1 NLS skipping reprograms ERα from a DNA-binding transcription factor to an RNA-binding splicing regulator is a novel mode of oncogenic pathway activation that may explain some cases of endocrine resistance without ER mutations.
Why for Yiru: The concept that a transcription factor (ERα) can be reprogrammed to function as an RNA-binding protein through a splicing event in an upstream tumour suppressor (NF1) is a striking example of cross-omic regulatory complexity. This kind of splicing-driven rewiring would be missed by standard gene-level differential expression analysis and requires isoform-level or splicing-aware computational approaches. For TME computational work, this underscores the importance of analyzing splicing and isoform switching — not just gene expression — when studying regulatory reprogramming in tumour and immune cells.
WTAP-Mediated Epitranscriptomic Program in Alveolar Macrophages Confers Prolonged Protection Against Postinfluenza Bacterial Pneumonia
bioRxiv Published 2026-05-27 preprint DOI: 10.64898/2026.05.23.727154
epitranscriptomics m6A WTAP trained immunity alveolar macrophage influenza RNA modification phagocytosis
Summary: Demonstrates that N6-methyladenosine (m6A) RNA modification orchestrates trained immunity in alveolar macrophages following influenza infection, conferring prolonged protection against secondary bacterial pneumonia. Trained immunity — the phenomenon whereby innate immune cells develop a form of memory after initial exposure — is typically studied through the lens of epigenetic reprogramming (histone modifications). This study reveals an epitranscriptomic mechanism: influenza A virus (IAV)-trained alveolar macrophages maintain low WTAP (Wilms tumor 1-associated protein) expression and reduced global m6A deposition for over two months post-infection. This m6A reduction stabilizes mRNAs encoding phagocytic and metabolic genes, enhancing the macrophages' antibacterial capacity. Pharmacological or genetic reduction of m6A faithfully recapitulates the trained immunity phenotype, improving phagocytosis and protecting mice from secondary bacterial pneumonia. Clinically, elevated WTAP in alveolar macrophages correlates with impaired phagocytosis and disease severity in COVID-19 and COPD patients, suggesting that the WTAP-m6A axis is a therapeutically tractable node for modulating innate immune memory in respiratory disease.
Why it matters: Trained immunity has emerged as a paradigm-shifting concept in innate immunology, but its molecular basis — beyond histone modifications — has been unclear. This study identifies m6A RNA modification as a new layer of trained immunity regulation, where reduced RNA methylation enhances the stability of functionally important transcripts. The two-month persistence of this epitranscriptomic memory suggests that RNA modifications can encode longer-lasting immune states than previously appreciated. The clinical correlation with COVID-19 and COPD severity positions WTAP as a potential biomarker and therapeutic target for respiratory infections where trained immunity is protective.
Why for Yiru: The innate immune compartment of the TME — particularly tumour-associated macrophages — shares functional programs with trained macrophages, including enhanced phagocytic capacity and altered metabolic states. The WTAP-m6A axis could be explored in TAMs to understand whether tumour-derived signals induce a particular epitranscriptomic state that either promotes or inhibits anti-tumour macrophage functions. More broadly, epitranscriptomic regulation is an underexplored dimension of TME biology that could reveal new regulatory mechanisms controlling immune cell function in tumours.
Friday delivery
BioTech News Delivery
CRISPR 2.0: First In Vivo Base Editing Trial Shows Encouraging Early Results in Cardiovascular Disease
FierceBiotech Published 2026-05-28 industry news DOI:
CRISPR base editing in vivo cardiovascular gene therapy clinical trial biotech
Summary: Verve Therapeutics and Beam Therapeutics reported encouraging early data from their ongoing Phase 1b studies of in vivo base editing therapies targeting PCSK9 and other cardiovascular genes. Base editing — a precision CRISPR technology that chemically converts one DNA base to another without creating double-strand breaks — is being administered systemically via lipid nanoparticles to permanently lower cholesterol and treat atherosclerotic cardiovascular disease. Early data show dose-dependent, durable reductions in target protein levels with a manageable safety profile. This marks a significant milestone for in vivo base editing, moving beyond ex vivo applications (like sickle cell disease) to systemic delivery, and if successful, could establish a new paradigm of 'one-and-done' genetic medicines for common chronic diseases.
Why it matters: In vivo base editing represents the next frontier of genetic medicine — moving from ex vivo cell therapies and rare disease applications to systemic treatments for common diseases. Success in cardiovascular disease would validate the platform for a massive patient population and accelerate development for other indications including metabolic diseases and cancer predisposition syndromes. The safety bar is exceptionally high for treating otherwise healthy individuals with genetic medicines, making these early safety signals critically important.
Why for Yiru: While genetics-focused, the progress of in vivo delivery technologies — particularly LNPs that can target specific tissues beyond the liver — has implications for TME-targeted therapies. If LNPs can be engineered to target tumour-resident immune cells or stromal cells, the same delivery technology could enable in vivo genetic reprogramming of the TME, such as transiently engineering TAMs or CAFs for anti-tumour phenotypes.
ADC Deal-Making Accelerates: Major Pharma Doubles Down on Antibody-Drug Conjugates with Multi-Billion Dollar Partnerships
STAT Published 2026-05-28 industry news DOI:
antibody-drug conjugate ADC biotech pharma oncology deal-making M&A
Summary: The antibody-drug conjugate (ADC) space continues to see intense deal activity, with multiple billion-dollar-plus partnerships announced this week. Major pharmaceutical companies are racing to build ADC portfolios, with particular interest in next-generation payloads (topoisomerase inhibitors, immunomodulators), improved linkers with better stability, and novel targets beyond HER2 and TROP2. The deals span both clinical-stage assets and discovery platforms, with some agreements structured around AI-designed ADCs that use machine learning to optimize target selection, linker chemistry, and payload combinations. This investment wave reflects the clinical validation of ADCs (Enhertu's success in HER2-low breast cancer has been transformative) and the expectation that ADCs will become a dominant oncology modality alongside checkpoint inhibitors and targeted therapies.
Why it matters: ADCs are moving from a niche oncology modality to a mainstream therapeutic class, with implications for how cancer is treated across tumour types. The shift toward immunomodulatory payloads — ADCs that deliver immune-stimulating rather than cytotoxic agents — is particularly noteworthy, as it blurs the line between targeted therapy and immunotherapy. AI-designed ADCs represent a convergence of computational drug design and biologics that could accelerate the identification of optimal target-payload-linker combinations.
Why for Yiru: ADCs are fundamentally TME-targeting agents — their efficacy depends on tumour-selective antigen expression and efficient payload release in the TME. Understanding how TME features (pH, protease activity, hypoxia, immune infiltration) affect ADC distribution, linker cleavage, and bystander killing is a computational problem well-suited to spatial TME analysis. The growing interest in immunomodulatory ADC payloads that activate rather than kill immune cells also directly intersects with TME immunology.
AI-Native Biotech Fundraising Surges: Foundation Model Companies Raise Record Rounds for Drug Discovery
Endpoints News Published 2026-05-27 industry news DOI:
AI drug discovery foundation model biotech fundraising venture capital protein design
Summary: AI-native biotechnology companies — those building proprietary foundation models for drug discovery rather than applying existing AI tools — raised record-breaking venture rounds this quarter. Companies developing large language models for protein design, generative models for antibody optimization, and multimodal foundation models integrating genomics, proteomics, and clinical data are attracting valuations previously reserved for clinical-stage therapeutics companies. The shift reflects growing confidence that AI-designed molecules can reach the clinic faster and with higher success rates than traditionally discovered drugs, with several AI-designed candidates now in Phase 2 trials. However, the space is becoming crowded, and differentiation is shifting from model architecture to data scale — companies with access to proprietary, high-quality biological data are commanding premium valuations over those relying on public datasets.
Why it matters: The AI drug discovery space is transitioning from proof-of-concept to clinical validation, with several AI-designed molecules now in mid-stage trials. If these trials succeed, it would fundamentally change pharmaceutical R&D economics and the competitive landscape. The emphasis on proprietary data as the key differentiator is also reshaping how academic-industry partnerships are structured, with implications for how academic labs share and monetize their data.
Why for Yiru: The computational tools described in this week's Research Radar — from TCR repertoire analysis to structure-based ligand generation — are the academic counterparts of the industrial AI platforms now attracting billions in investment. Understanding the commercial landscape helps contextualize which computational methods are likely to be rapidly industrialized (protein design, molecular generation) versus which remain academic tools (specialized single-cell analysis, niche ontology frameworks). This is useful for strategic career decisions about which computational skills to develop.
FDA Approves First Subcutaneous PD-1 Inhibitor, Expanding Access and Reducing Treatment Burden
FierceBiotech Published 2026-05-26 industry news DOI:
FDA PD-1 subcutaneous immunotherapy regulatory biotech access
Summary: The FDA approved the first subcutaneous formulation of a PD-1 inhibitor, allowing patients to receive immunotherapy via a brief injection rather than hours-long intravenous infusion. The approval was based on pharmacokinetic bridging studies demonstrating non-inferior exposure compared to the intravenous formulation, plus patient preference data showing strong preference for subcutaneous administration. This regulatory milestone is expected to significantly reduce healthcare system burden (infusion chair time, nursing resources) and improve patient quality of life. Multiple other checkpoint inhibitors with subcutaneous formulations are in late-stage development, suggesting that the era of infusion-center-dependent immunotherapy may be coming to an end. This also opens possibilities for at-home administration and expanded access in low-resource settings where infusion infrastructure is limited.
Why it matters: While scientifically incremental, the practical impact of subcutaneous immunotherapy is enormous: shorter treatment times, reduced healthcare costs, and potentially expanded global access to checkpoint inhibitors. This is a reminder that delivery innovation — not just drug discovery — can transform patient care. The move toward self-administration also blurs the line between biologics and oral small molecules, potentially changing how immunotherapies are positioned relative to targeted therapies in treatment algorithms.
Why for Yiru: The shift to subcutaneous immunotherapy highlights an underexplored dimension of cancer pharmacology: drug pharmacokinetics and biodistribution in the TME. Subcutaneous administration results in different drug concentration-time profiles than IV, which could affect intratumoural drug penetration, receptor occupancy dynamics, and the balance between tumour and lymphoid tissue exposure. Computational pharmacokinetic-pharmacodynamic modeling of TME drug distribution is directly relevant to understanding whether subcutaneous checkpoint blockade achieves equivalent intratumoural target engagement.