Research Radar — 2026-06-25
Methods & AI
Computational
Zero-shot design of drug-binding proteins via neural iterative selection−expansion
Nature Published 2026-06-24 research article DOI: 10.1038/s41586-026-10670-w
deep learning protein design drug discovery
Summary: Presents a neural iterative selection-expansion framework for zero-shot design of small-molecule-binding proteins from scratch, without requiring pre-existing binding proteins as templates. Designing proteins that bind specific small molecules is a central challenge in computational protein design with applications in drug delivery, biosensing, and therapeutic sequestration. Current approaches typically require known binders as starting templates or depend on high-throughput screening. The authors develop a method that pairs two neural networks — a sequence design network and a binding prediction network — in an iterative optimization loop. Starting from random amino acid sequences, the design network proposes candidates, the binding network evaluates them, and feedback refines the next iteration. Using this approach, they successfully designed proteins that bind the drug digoxin with high affinity and specificity, with experimental validation confirming binding. The zero-shot capability — designing binders without any prior binder data for the target — represents a significant advance in protein design methodology.
Why it matters: The ability to design small-molecule-binding proteins from scratch without templates opens transformative possibilities for drug delivery, diagnostics, and synthetic biology. Instead of screening large libraries or engineering existing proteins, researchers could computationally design binders for any therapeutic molecule of interest — from drugs to toxins to metabolites. The iterative design-evaluate loop architecture also provides a framework that could be extended to other molecular design problems beyond protein-small molecule interactions, including protein-protein interface design and enzyme design.
Why for Yiru: Designed binding proteins for drug delivery could be engineered to deliver immunomodulatory payloads specifically to the TME, acting as molecular depots that release therapeutics in response to tumour-specific cues. The zero-shot design methodology is also computationally relevant: the neural iterative framework represents a general paradigm for molecular design that could be applied to design TME-targeted binding proteins, such as checkpoint decoys or cytokine traps. Understanding the architecture of this framework — how design and evaluation networks interact — may inform the development of computational tools for designing TME-relevant proteins.
Non-coding variant prioritization based on cell type, developmental stage and evolutionary constraint
Nature Genetics Published 2026-06-22 research article DOI: 10.1038/s41588-026-02618-7
deep learning gene regulation evolutionary constraint machine learning variant effect prediction
Summary: Presents a deep learning framework that predicts non-coding variant effects across the allele frequency spectrum in over 100 fetal and adult cell types, linking these predictions with evolutionary constraint to prioritize variants associated with complex diseases. Non-coding variants constitute the majority of GWAS-identified risk loci, but prioritizing which variants are functionally important remains a major challenge because most non-coding variants have no obvious functional annotation. The authors use deep learning sequence models trained on chromatin accessibility, histone modification, and transcription factor binding data across diverse cell types and developmental stages to predict the regulatory impact of any single nucleotide change. By integrating these predictions with measures of evolutionary constraint (phyloP, phastCons), they develop a prioritisation framework that identifies non-coding variants most likely to be causal for diseases including Alzheimer's disease, autism spectrum disorder, congenital heart disease, and schizophrenia. The integration of cell-type-specific and developmental-stage-specific predictions is a key advance, as many disease-relevant regulatory elements are active only in specific cell types or during specific developmental windows.
Why it matters: Most disease-associated variants from GWAS fall in non-coding regions whose functional impact is difficult to assess, creating a major bottleneck in translating genetic associations into mechanistic understanding. This study provides a comprehensive resource and computational framework for prioritizing non-coding variants across diverse cell types and developmental stages, addressing both the cell-type specificity and the evolutionary conservation dimensions of non-coding variant interpretation. The integration of deep learning predictions with evolutionary constraint is methodologically important: it leverages two independent sources of functional information — sequence-based regulatory predictions and cross-species conservation — to increase confidence in variant prioritisation.
Why for Yiru: Non-coding variant prioritisation is directly relevant to cancer genomics, where most GWAS loci for cancer risk and immunotherapy response fall in non-coding regions. The cell-type-specific framework could be applied to TME-relevant cell types (T cells, macrophages, CAFs) to identify non-coding variants that affect immune cell function in the tumour microenvironment. The integration of developmental-stage-specific predictions is also relevant: cancer progression often involves reactivation of developmental gene programs, and non-coding variants that affect these programs could influence tumour behaviour and immune evasion.
Longitudinal changes in DNA methylation in IDH-mutant glioma fuel disease progression through altered cell state differentiation
Nature Genetics Published 2026-06-22 research article DOI: 10.1038/s41588-026-02642-7
cancer genomics single-cell epigenetics statistical genetics
Summary: Uses single-cell multi-omics to interrogate longitudinal samples of IDH-mutant gliomas, revealing that disease progression is coupled with genome-wide DNA hypomethylation and changes in cell state differentiation topologies, with increased emergence of stem-like states in aggressive disease. IDH-mutant gliomas are a molecularly defined subtype of diffuse glioma characterised by mutations in isocitrate dehydrogenase (IDH1/2) and a distinct DNA hypermethylation phenotype. While it is known that these tumours eventually progress to higher-grade disease, the molecular mechanisms driving progression remain incompletely understood. The authors performed single-cell multi-omic profiling (DNA methylation, chromatin accessibility, and transcriptomics) of paired initial and recurrent IDH-mutant glioma samples from the same patients. They find that progression is accompanied by genome-wide loss of DNA methylation (hypomethylation), particularly at heterochromatic regions, and a reconfiguration of chromatin accessibility that promotes adoption of stem-like gene expression programs. The stem-like cells are characterised by increased expression of neural stem cell markers and decreased expression of differentiation genes, suggesting that progressive hypomethylation unlocks developmental plasticity that drives tumour aggressiveness.
Why it matters: IDH-mutant gliomas are among the best-studied epigenetically defined cancers, yet the molecular mechanisms driving their inevitable progression have remained unclear. This study provides direct longitudinal evidence that epigenetic dysregulation — specifically, progressive DNA hypomethylation — is a driver of malignant progression, not merely a passenger phenomenon. The finding that hypomethylation promotes the emergence of stem-like cell states is conceptually important: it suggests that epigenetic therapies (such as hypomethylating agents) might need to be carefully timed, as they could potentially accelerate the very process that drives progression. The single-cell multi-omic approach also sets a methodological standard for studying tumour evolution at the epigenetic level.
Why for Yiru: Epigenetic heterogeneity in the TME is increasingly recognised as a driver of immune evasion and therapy resistance. The finding that progressive hypomethylation in glioma unlocks stem-like programs raises the question of whether similar epigenetic plasticity operates in other TME contexts — for instance, whether epigenetic reprogramming of tumour cells in response to immunotherapy drives acquired resistance. The single-cell multi-omic framework (integrating methylation, chromatin, and transcriptome from the same cells) is directly applicable to studying TME heterogeneity, where coordinated epigenetic and transcriptional changes define functionally distinct cell states. More broadly, understanding how epigenetic changes reshape developmental potential in cancer could inform strategies to lock tumour cells into differentiated, less aggressive states.
Setting benchmarks for practical quantum utility of combinatorial optimization
Nature Computational Science Published 2026-06-23 research article DOI: 10.1038/s43588-026-00979-x
computational biology
Summary: Introduces SAGELD (Statistical test for Gene-Environment with Longitudinal Data), a method that leverages repeated measurements in longitudinal biobank studies to detect gene-environment interactions with substantially greater statistical power than cross-sectional approaches. Gene-environment interaction (GxE) analysis aims to identify genetic variants whose effects on a trait are modified by environmental exposures, which is critical for understanding why some individuals are more susceptible to environmental risk factors than others. Traditional GxE tests based on cross-sectional data suffer from limited statistical power because they use only a single measurement per individual. SAGELD leverages the repeated measurements available in large longitudinal cohorts (such as UK Biobank, All of Us) to model how genetic effects on trait trajectories are modified by time-varying environmental exposures. The method uses a linear mixed model framework that accounts for within-individual correlations and can detect interactions that would be missed in cross-sectional analyses. The authors demonstrate through simulations and real-data applications that SAGELD substantially increases power to detect GxE interactions, particularly for variants with small-to-moderate interaction effects.
Why it matters: Gene-environment interactions are likely ubiquitous in complex traits but have been difficult to detect due to limited statistical power and the complexity of modelling both genetic and environmental variation. The increasing availability of longitudinal data from large biobanks creates an opportunity to detect GxE interactions that were previously invisible, but requires statistical methods designed for longitudinal designs. SAGELD fills this methodological gap by providing a principled, computationally efficient framework for longitudinal GxE testing. The power gains from using repeated measurements are substantial — effectively converting a cross-sectional study into a longitudinal one multiplies the effective sample size. This could enable the discovery of GxE interactions that explain individual differences in response to environmental exposures including diet, pollution, medications, and lifestyle factors.
Why for Yiru: Longitudinal biobank data are increasingly used to study how genetic and environmental factors interact over time to influence cancer risk and treatment response. SAGELD's framework could be applied to identify genetic variants that modify how TME-related environmental factors — such as chronic inflammation, metabolic stress, or immunotherapy — affect tumour progression and immune cell dynamics. From a computational perspective, the linear mixed model framework for longitudinal data is a versatile tool: similar approaches could model how genetic variants modify immune cell state transitions over time in response to treatment, or how tumour microenvironmental exposures interact with tumour genetics to shape clonal evolution.
Biomedical discoveries
Biomedicine
Costimulation drives CAR-T cell division fate
Nature Immunology Published 2026-06-22 research article DOI: 10.1038/s41590-026-02559-7
computational biology
Summary: Reveals a reciprocal circuit linking cuproptosis — a recently discovered form of copper-dependent programmed cell death — and antitumour immunity. The authors show that CD8+ T cell-derived IFN-γ enhances tumour cell susceptibility to cuproptosis by upregulating copper import machinery and downregulating copper efflux transporters in tumour cells. Conversely, cuproptosis in tumour cells releases immunostimulatory molecules that promote dendritic cell activation and enhance T cell priming, creating a positive feedback loop. Engagement of this cuproptosis-immunity axis with PD-L1 blockade synergistically amplifies tumour cell killing and overcomes immunotherapy resistance in preclinical models. The study identifies FDX1 (ferredoxin 1) and LIAS (lipoic acid synthase) as key mediators of IFN-γ-induced cuproptosis sensitivity and demonstrates that cuproptosis induction combined with checkpoint blockade produces durable antitumour responses in otherwise resistant tumour models. This cuproptosis-immunity circuit represents a previously unrecognised mechanism by which T cells can directly influence tumour cell death modality and reveals cuproptosis induction as a potential therapeutic strategy for immunotherapy-resistant cancers.
Why it matters: The discovery of a mechanistic link between cuproptosis and antitumour immunity opens a new axis in cancer immunotherapy. Until now, cuproptosis has been studied primarily as a cell-intrinsic cell death mechanism triggered by copper overload, with unclear immunological consequences. This study demonstrates that cuproptosis is not merely a tumour cell-autonomous process but is actively regulated by T cell-derived IFN-γ and, in turn, shapes antitumour immune responses. The finding that cuproptosis induction synergises with PD-L1 blockade to overcome immunotherapy resistance is particularly significant clinically: it identifies a potential therapeutic strategy for the large population of patients who do not respond to current immunotherapies. The identification of specific molecular mediators (FDX1, LIAS) provides actionable targets for pharmacological intervention.
Why for Yiru: The cuproptosis-immunity circuit is directly relevant to understanding how cell death modalities in the TME shape antitumour immune responses. Different forms of programmed cell death (apoptosis, necroptosis, pyroptosis, ferroptosis, cuproptosis) have distinct immunological consequences, and this study adds cuproptosis to the repertoire of immunogenic cell death modalities that can be leveraged for therapy. The finding that IFN-γ from T cells regulates tumour cell cuproptosis sensitivity reveals a new dimension of TME immune regulation: T cells are not only killing tumour cells directly but also conditioning them to undergo specific forms of cell death with distinct immunological consequences. From a computational perspective, modelling the cuproptosis-immunity circuit as a dynamical system could reveal optimal intervention strategies for combining cuproptosis inducers with immunotherapy.
Cuproptosis-immunity crosstalk informs strategy to overcome immunotherapy resistance
Cell Published 2026-06-22 research article DOI:
cancer genomics
Summary: Demonstrates that asymmetric cell division in CAR-T cells is an early fate determinant that explains why 4-1BBζ CAR-T cells preferentially generate durable memory progeny while CD28ζ CAR-T cells skew toward short-lived effector fates. Chimeric antigen receptor (CAR) T cell therapy has transformed the treatment of haematological malignancies, but the durability of responses varies dramatically depending on the costimulatory domain used in the CAR construct. The 4-1BB (CD137) costimulatory domain is associated with longer persistence and more durable responses, while the CD28 domain drives rapid expansion but shorter persistence. This study reveals that the basis for this difference lies in how each costimulatory domain influences asymmetric cell division — the process by which a single CAR-T cell divides into two daughter cells with distinct fates. Using time-lapse microscopy and single-cell tracking, the authors show that 4-1BBζ CAR-T cells undergo asymmetric division that consistently generates one daughter cell with memory-like features and another with effector-like features. In contrast, CD28ζ CAR-T cells undergo symmetric division that produces two effector-like daughter cells, rapidly depleting the memory precursor pool. This asymmetric division program is established within the first few divisions after activation and acts as an early 'program selector' that determines long-term fate outcomes.
Why it matters: Understanding why 4-1BBζ CAR-T cells persist longer than CD28ζ CAR-T cells has been a central question in CAR-T biology with direct clinical implications. This study provides a mechanistic answer at the single-cell level: asymmetric cell division establishes distinct fates from the very first divisions. This mechanistic understanding opens the possibility of engineering CAR constructs or culture conditions that promote asymmetric division and memory formation, potentially improving the durability of CAR-T cell responses. The finding also has broader implications for T cell biology, revealing that costimulatory signals not only provide activation strength but actively instruct cell division programmes that determine fate outcomes.
Why for Yiru: CAR-T cell persistence is a key determinant of therapeutic efficacy in the TME, where suppressive signals and metabolic stress can rapidly exhaust adoptively transferred T cells. Understanding how costimulation programmes asymmetric division and memory formation could inform the design of CAR-T cells engineered for enhanced persistence in solid tumours. From a computational perspective, the asymmetric division framework could be modelled as a cell fate decision network, where costimulatory signals are inputs that bias the probability of symmetric vs. asymmetric division outcomes. Such models could predict how different CAR designs or culture conditions influence the balance between memory and effector formation in the TME.
Expansion and CAR engineering of granulocyte-monocyte progenitors for cellular immunotherapy
Cell Published 2026-06-19 research article DOI:
cancer genomics
Summary: Establishes granulocyte-monocyte progenitors (GMPs) as a renewable upstream platform for engineered myeloid cell therapy and introduces a CAR-Fc strategy that couples direct tumour targeting to host antigen-presenting cell activation. While CAR-T cell therapy has achieved remarkable success in haematological malignancies, its efficacy in solid tumours has been limited partly by the immunosuppressive tumour microenvironment and the difficulty of T cell infiltration. Myeloid cells — including macrophages, dendritic cells, and neutrophils — are naturally recruited to tumours and can be engineered to perform antitumour functions. However, engineered myeloid cell therapy has been hampered by the difficulty of obtaining sufficient numbers of functional cells. This study overcomes this limitation by engineering GMPs — the bone marrow progenitors that give rise to all myeloid lineages — with CAR constructs. CAR-engineered GMPs can be expanded ex vivo and then differentiate into functional CAR-expressing macrophages, dendritic cells, and neutrophils upon adoptive transfer. The CAR-Fc strategy incorporates an Fc domain that enables the CAR to both recognise tumour antigens and engage host Fcγ receptors on antigen-presenting cells, thereby coupling direct tumour killing with indirect activation of adaptive immune responses.
Why it matters: Myeloid cell therapy has been a largely untapped frontier in cellular immunotherapy compared to T cell and NK cell therapies. This study provides a practical platform that overcomes the key manufacturing bottleneck by engineering renewable progenitors rather than terminally differentiated myeloid cells. The CAR-Fc design is particularly innovative: it simultaneously provides direct tumour cytotoxicity (through CAR-mediated recognition) and indirect immune activation (through Fc-mediated antigen presentation), effectively bridging innate and adaptive immunity. If this platform translates to the clinic, it could provide a complementary approach to CAR-T therapy, particularly for solid tumours where myeloid cell infiltration is naturally more efficient than T cell infiltration.
Why for Yiru: Engineered myeloid cells represent a promising but underexplored modality for TME reprogramming. Macrophages and dendritic cells are abundant in the TME and can exert both pro- and anti-tumour functions depending on their activation state. The CAR-GMP platform could be used to deliver engineered myeloid cells that selectively localise to tumours and perform tailored functions — such as phagocytosis of tumour cells, antigen cross-presentation to T cells, or remodelling of the TME extracellular matrix. The CAR-Fc design is particularly relevant: it creates a bridge between engineered myeloid cells and the endogenous immune system, potentially amplifying antitumour immunity through multiple mechanisms operating in parallel.
PROTEUS trial heralds perioperative therapy for prostate cancer
Nature Medicine Published 2026-06-22 research article DOI: 10.1038/d41591-026-00032-4
cancer genomics
Summary: Introduces CANVAS (Cellular Architecture and Neighborhood-informed Virtual spAtial tumor profiling from hiStopathology), a computational platform that decodes spatial tumour ecosystems from routine haematoxylin and eosin (H&E) histopathology slides to enable population-level prognostic modelling and immunotherapy response prediction without the need for specialised spatial molecular profiling. Spatial biology technologies (spatial transcriptomics, imaging mass cytometry) have revolutionised our understanding of tumour organisation, but these technologies remain too expensive and technically demanding for routine clinical use. CANVAS addresses this gap by training deep learning models on large collections of H&E slides with paired spatial molecular data, learning to infer spatial cell-type distributions, tissue neighbourhood architectures, and cell-cell interaction patterns directly from routine histological images. The platform uses a graph neural network architecture that represents each tissue section as a graph of cellular neighbourhoods, learning to recognise tissue organisation patterns associated with clinical outcomes. CANVAS identifies spatially organised immune niches, tumour-stroma interfaces, and tertiary lymphoid structures from H&E alone and uses these features to predict immunotherapy response and patient survival across multiple cancer types. By operating on routinely collected H&E slides, CANVAS makes spatial tumour ecology analysis accessible at population scale.
Why it matters: The vast majority of tumour biopsies worldwide are examined by H&E histopathology, but the rich spatial information contained in these images — cellular organisation, tissue architecture, immune infiltration patterns — is only qualitatively assessed by pathologists. CANVAS bridges the gap between routine histopathology and cutting-edge spatial biology by learning to decode molecular-level spatial organisation from standard H&E images. If validated broadly, this approach could democratise spatial tumour analysis: any institution with H&E staining capability could generate spatial tumour ecology insights without expensive spatial profiling equipment. The ability to predict immunotherapy response from H&E is particularly impactful, as it could help identify patients most likely to benefit from immunotherapy using only routinely collected tissue sections.
Why for Yiru: CANVAS is directly relevant to TME research: it provides a computational framework for extracting spatial TME features — immune infiltration patterns, tumour-stroma organisation, tertiary lymphoid structures — from widely available H&E slides. For Yiru's computational methods work, CANVAS's graph neural network architecture for representing cellular neighbourhoods is a methodological template: similar architectures could be applied to other spatial data types or combined with molecular profiling data for multimodal TME analysis. The population-level scalability of CANVAS is also relevant: it enables spatial TME analysis across large clinical cohorts where spatial molecular profiling is infeasible, potentially identifying TME features that predict immunotherapy outcomes across thousands of patients.
Fusobacterium periodonticum promotes colorectal tumorigenesis via decanoic acid-driven neutrophil chemotaxis
Nature Communications Published 2026-06-24 research article DOI: 10.1038/s41467-026-74591-y
cancer genomics
Summary: Reveals that Fusobacterium periodonticum — an oral bacterium not previously linked to colorectal cancer — is enriched in CRC tumours and promotes colorectal tumorigenesis through decanoic acid-driven neutrophil chemotaxis. The gut microbiome has emerged as a critical modulator of colorectal cancer (CRC) development, with Fusobacterium nucleatum being the most well-established bacterial driver. Using a multi-omics approach combining metagenomics, metabolomics, and functional assays, the authors identify F. periodonticum as a novel CRC-associated bacterium. F. periodonticum is enriched in CRC tissues compared to adjacent normal tissue and correlates with elevated levels of decanoic acid (a medium-chain fatty acid). Mechanistically, F. periodonticum-derived decanoic acid drives neutrophil chemotaxis through a G-protein-dependent mechanism, recruiting neutrophils to the tumour microenvironment where they promote tumour growth through the release of protumourigenic factors including MMP-9 and reactive oxygen species. The study demonstrates that antibiotic treatment targeting F. periodonticum or pharmacological inhibition of neutrophil recruitment reduces tumour burden in preclinical CRC models, establishing a causal role for this bacterium in CRC promotion.
Why it matters: The discovery of a new CRC-associated bacterium expands our understanding of how the gut microbiome influences cancer development. The identification of decanoic acid as the specific molecular mediator — a metabolite that activates GPCR-dependent neutrophil chemotaxis — provides a clear mechanistic link from bacterial colonisation to tumour promotion. This is important because it identifies specific, targetable nodes in the bacteria-metabolite-immune axis: antibiotics targeting F. periodonticum, decanoic acid neutralisation, or inhibition of neutrophil recruitment could all be potential preventive or therapeutic strategies. The finding also highlights that Fusobacterium species beyond F. nucleatum can promote CRC, suggesting that the Fusobacterium genus exerts broader oncogenic effects than previously appreciated.
Why for Yiru: The bacteria-metabolite-immune axis in CRC is a model for understanding how the TME is shaped by extrinsic microbial factors. Neutrophils are abundant in many solid tumours but their role in cancer progression is context-dependent, with both pro- and anti-tumour functions reported. This study reveals that bacterial metabolites can recruit and reprogramme neutrophils toward a protumourigenic phenotype in the TME, adding to our understanding of TME immune regulation. From a computational perspective, the multi-omics integrative approach — combining metagenomics with metabolomics to identify causal bacterial metabolites — is a methodological framework applicable to studying how other microbial species influence the TME through metabolite-mediated immune modulation.
Cross-disciplinary watchlist
Other Fields
Arthritis susceptibility begins before birth
Nature Immunology Published 2026-06-23 research article DOI: 10.1038/s41590-026-02570-y
computational biology
Summary: Uses longitudinal single-cell multi-omics in SOD1-ALS mice to identify a disease-associated motor neuron (DM) state that precedes cell death. Amyotrophic lateral sclerosis (ALS) is a progressive neurodegenerative disease characterised by the selective loss of motor neurons, but the molecular events that precede motor neuron death have remained poorly defined, hampering the development of early intervention strategies. The authors performed longitudinal single-nucleus RNA-seq and ATAC-seq on spinal cord motor neurons from SOD1-G93A ALS mice at multiple time points spanning the disease course, from pre-symptomatic through end-stage. They identify a distinct disease-associated motor neuron state that emerges before detectable cell loss and is characterised by a specific transcription factor network involving ATF3, FOS, JUN, and STAT3. This DM state is conserved in human ALS post-mortem tissue and is enriched for ALS-associated genetic risk variants, suggesting that the transition to this state is a shared feature of motor neuron degeneration across ALS subtypes. The DM state includes activation of the integrated stress response, mitochondrial dysfunction signatures, and altered RNA processing, preceding activation of apoptotic pathways.
Why it matters: Identifying the earliest molecular changes that precede motor neuron death is critical for developing ALS therapies that can intervene before irreversible damage occurs. This study provides a detailed molecular roadmap of the pre-death motor neuron state, revealing that neurons pass through a definable disease-associated state before committing to cell death. The finding that this state is conserved between mouse models and human ALS suggests it is a fundamental feature of ALS pathogenesis. The identification of specific transcription factors (ATF3, FOS, JUN) that orchestrate this state provides potential therapeutic targets for intercepting disease progression. The conservation of ALS risk variant enrichment in the DM state suggests that genetic predisposition to ALS may act by modulating the transition into or out of this pathogenic state.
Why for Yiru: The concept of a disease-associated cell state that precedes cell death is broadly applicable beyond neurodegeneration — similar pre-death states likely exist in other pathological contexts including the TME, where tumour cells may enter stress-associated states before undergoing cell death in response to therapy. The longitudinal single-cell multi-omic approach used here is methodologically relevant: similar designs could be applied to map how TME cell states evolve during tumour progression or in response to treatment. Understanding the transcription factor networks that orchestrate the transition between cell states in the TME could identify therapeutic vulnerabilities.
An emergent disease-associated motor neuron state precedes cell death in ALS
Cell Published 2026-06-23 research article DOI:
statistical genetics
Summary: Reveals that the roots of rheumatoid arthritis susceptibility lie in prenatal tissue architecture, suggesting that why certain joints develop inflammation in RA while others escape is determined during development. Rheumatoid arthritis (RA) is a chronic autoimmune disease characterised by inflammatory arthritis that typically affects specific joints — most commonly the small joints of the hands and feet — while sparing others. The reason for this joint-specific susceptibility has been a longstanding puzzle. This study demonstrates that the anatomical pattern of joint involvement in RA is predetermined by the developmental origin of synovial tissue. Using developmental biology approaches in mouse models and human tissue analysis, the authors show that synovial joints arise from distinct embryonic lineages with different molecular programmes, and these developmental differences persist into adulthood, creating joint-specific differences in immune accessibility, stromal cell composition, and susceptibility to inflammatory triggers. Specifically, joints derived from the lateral plate mesoderm (such as hand and foot joints) exhibit distinct patterns of vascularisation, innervation, and stromal cell gene expression that make them more permissive to immune cell infiltration and synovial inflammation compared to joints derived from other embryonic lineages.
Why it matters: The finding that RA susceptibility is hardwired during development fundamentally reframes our understanding of why autoimmune inflammation targets specific tissues. This shifts the focus from purely immunological explanations — such as joint-specific antigens or biomechanical stress — to include developmental determinants of tissue permissiveness to immune attack. Understanding the molecular basis of joint-specific susceptibility could reveal new therapeutic strategies for RA that target the tissue-level determinants of inflammation rather than the immune system alone. More broadly, this study demonstrates that developmental programming of tissue architecture has lifelong consequences for disease susceptibility, a principle likely applicable to many autoimmune and inflammatory diseases with characteristic anatomical patterns.
Why for Yiru: The principle that developmental origins determine tissue-specific susceptibility to immune attack is relevant to understanding why different TMEs exhibit different immune responses. Just as synovial joints differ in their developmental programming and consequent immune permissiveness, different organs and tissues may have distinct developmental programmes that influence how their microenvironments respond to tumour initiation and immune surveillance. This could explain why certain cancers are more or less responsive to immunotherapy depending on the tissue of origin. More broadly, this study exemplifies how integrating developmental biology with immunology can reveal fundamental principles of tissue-immune interactions.
Shared neural geometries for bilingual semantic representations in human hippocampal neurons
Cell Published 2026-06-24 research article DOI:
computational biology
Summary: Demonstrates, through hippocampal recordings in bilingual speakers, that English and Spanish access a shared neural meaning space while being read out through distinct, language-specific neural axes. How the human brain represents and processes multiple languages without confusion has been a central question in neuroscience. This study provides direct evidence from single-neuron recordings in the human hippocampus of bilingual participants listening to and speaking in English and Spanish. The authors find that individual hippocampal neurons respond selectively to specific semantic concepts (e.g., a 'dog neuron' fires when the concept of dog is accessed in either language), indicating that the neural representation of meaning is shared across languages. However, the neural population activity patterns that encode this meaning are rotated into different readout axes depending on which language is being used, allowing the brain to access the same concept through different linguistic channels without interference. This shared meaning space with rotated readout axes provides an efficient neural architecture for multilingualism: the brain maintains a single semantic representation system that can be accessed by multiple languages through language-specific neural transformations.
Why it matters: Understanding how the brain handles multiple languages has profound implications for our understanding of cognition, development, and the neural basis of language. The finding of a shared neural meaning space with language-specific readout axes elegantly explains how bilingual individuals can operate in multiple languages without confusion — the semantic representation is unified, but the readout mechanism is language-specific. The single-neuron recording data from the human hippocampus is exceptionally rare and provides a unique window into the neural code for meaning. This neural architecture of shared representation with modality-specific readout may be a general principle of brain organisation, not limited to language but potentially applicable to other domains where the same information is accessed through different channels.
Why for Yiru: The shared-representation-with-rotated-readout neural architecture is conceptually interesting for computational modelling of multimodal data integration. Just as the brain maintains a shared semantic space accessible through different languages, computational models for multimodal biomedical data (e.g., integrating transcriptomics, proteomics, and imaging data) could be designed with a shared latent representation space with modality-specific readout projections. This architecture is mathematically elegant and could inspire new approaches to multi-omics integration in TME research, where different molecular modalities capture complementary but overlapping aspects of tumour biology.
Airway immune signatures of protection and disease progression in recent human tuberculosis household contacts
Nature Immunology Published 2026-06-24 research article DOI: 10.1038/s41590-026-02544-0
computational biology
Summary: Characterises the local airway immune factors that distinguish Mycobacterium tuberculosis-infected individuals who remain healthy from those who progress to active tuberculosis (TB). A central unresolved question in TB immunology is why only 5-10% of individuals infected with M. tuberculosis develop active disease, while the majority control the infection without clinical symptoms. This study provides a comprehensive analysis of the airway immune response in recently exposed household contacts of TB patients, comparing those who remained healthy with those who progressed to active disease. Using bronchoalveolar lavage samples and high-dimensional immune profiling (flow cytometry, transcriptomics, proteomics), the authors identify distinct immune signatures in the airways that predict protection vs. progression. Protective immunity is associated with sustained lung-resident memory T cell responses, particularly CD4+ and CD8+ tissue-resident memory T (TRM) cells producing IFN-γ, TNF-α, and IL-17. In contrast, progression to active disease is characterised by a shift toward myeloid cell-dominated inflammation with elevated neutrophils, inflammatory monocytes, and type I interferon signatures, alongside a loss of the TRM response. These airway immune signatures provide potential biomarkers for identifying individuals at risk of progression and candidates for preventive therapy.
Why it matters: TB remains one of the world's deadliest infectious diseases, killing over 1.5 million people annually. The ability to identify which infected individuals will progress to active disease is critical for targeted preventive therapy, as treating all 2 billion latently infected individuals is neither feasible nor necessary. This study provides the most comprehensive characterisation to date of the local airway immune responses that distinguish protection from progression. The finding that tissue-resident memory T cells are the key mediators of protection in the lung mucosa has important implications for TB vaccine design: vaccines that generate lung TRM responses may be more effective than those that rely solely on systemic immunity. The identification of progression-associated myeloid signatures also provides potential therapeutic targets for host-directed therapy.
Why for Yiru: The concept of tissue-resident immune responses determining protection vs. disease progression is directly applicable to the TME, where tissue-resident memory T cells in tumours have been associated with improved immunotherapy outcomes. The TRM-focused protective signature in TB parallels findings in cancer where CD8+ TRM cells in tumours predict response to checkpoint blockade. The methodological approach — comprehensive airway immune profiling correlating with clinical outcomes — is a template for studying how local tissue immune states determine disease outcomes. Understanding how tissue-resident immune populations are maintained and function in different tissue contexts (lung mucosa, TME) could reveal shared principles of tissue immunity.
Diversifying the antibody response to influenza virus with an mRNA-based vaccine
Nature Immunology Published 2026-06-23 research article DOI: 10.1038/s41590-026-02565-9
computational biology
Summary: Demonstrates that an mRNA-based influenza vaccine generates not only robust influenza virus-specific antibody titres but also prolonged germinal centre B cell responses in humans, resulting in enhanced breadth of antibody binding to antigenically diverse viral strains. Influenza virus poses a persistent public health challenge due to its high antigenic variability — seasonal drift and occasional pandemic shifts require annual vaccine updates with variable efficacy. mRNA vaccines, which were successfully deployed against SARS-CoV-2, have the potential to improve influenza vaccine efficacy, but their ability to generate broadly protective responses has been unclear. This study reports results from a phase 1 clinical trial of an mRNA-based influenza vaccine encoding haemagglutinin (HA) from multiple influenza strains. Compared to traditional inactivated influenza vaccines, the mRNA vaccine induced significantly higher and more sustained germinal centre B cell responses in draining lymph nodes, as measured by fine-needle aspiration and single-cell analysis of lymph node B cells. This prolonged germinal centre activity led to enhanced somatic hypermutation and the generation of antibodies with greater breadth of binding — recognising not only the vaccine-matched strains but also antigenically drifted strains that had emerged in subsequent seasons. The mRNA platform also enabled rapid inclusion of updated HA sequences to match circulating strains.
Why it matters: Improving influenza vaccine efficacy is a major public health priority, with current vaccines providing only 40-60% protection in good years and far less when the circulating strain is mismatched. The demonstration that mRNA vaccines generate more robust germinal centre responses and broader antibody breadth than traditional vaccines suggests that mRNA technology could substantially improve influenza protection. The prolonged germinal centre activity is particularly important: it is the cellular process that enables affinity maturation and the generation of antibodies capable of recognising diverse viral variants. If mRNA influenza vaccines prove to be broadly protective across seasons, they could eliminate the need for annual vaccine updates and dramatically reduce the global burden of seasonal influenza.
Why for Yiru: The capacity of mRNA vaccines to sustain germinal centre responses and generate broadly neutralising antibodies is directly relevant to cancer vaccine development. Personalised mRNA cancer vaccines aim to generate T cell and B cell responses against tumour neoantigens, and prolonged germinal centre activity could enhance the breadth and durability of anti-tumour antibody responses. The finding that mRNA vaccines outperform traditional platforms in generating protective breadth is also relevant to understanding how different vaccine platforms shape immune responses in the TME for therapeutic applications.