Research Radar — 2026-06-14
Methods & AI
Computational
Large-scale, spatially resolved panoramic CRISPR screening in native tissue environments using Perturb-DBiT
Nature Biotechnology Published 2026-06-11 research article DOI: 10.1038/s41587-026-03127-y
spatial CRISPR screening total RNA sequencing Perturb-DBiT tumour microenvironment metastasis non-coding RNA functional genomics single-guide RNA
Summary: Presents Perturb-DBiT, a method for co-sequencing of spatial total RNA whole transcriptomes and single guide RNAs (sgRNAs) on the same tissue section in situ. Spatially resolved CRISPR screening in vivo has been limited to small perturbation panels and subsets of protein-coding RNAs. Perturb-DBiT breaks through these constraints by enabling large (80,000+) sgRNA panels across tumour colonies in multiple consecutive tissue sections with their corresponding total RNA transcriptomes. In a human cancer metastatic colonization model, the authors linked perturbations affecting long noncoding RNA covariation, microRNA–mRNA interactions, and distinct amino acid-specific tRNA alterations to tumour migration and growth. By integrating transcriptional pseudotime trajectories, they further observed the impact of perturbations on clonal dynamics and cooperation between tumour cells. In an immune-competent syngeneic mouse model, investigation of the tumour immune microenvironment indicated distinct, synergistic effects on immune infiltration and suppression. Perturb-DBiT thus provides a spatially resolved comprehensive view of perturbation responses in complex tissues, including small and large RNA regulation, tumour proliferation, migration, metastasis, and immune interactions. The method represents a convergence of spatial transcriptomics with functional genomics at unprecedented scale and molecular breadth.
Why it matters: CRISPR screens have transformed functional genomics by enabling systematic discovery of gene function, but traditional screens operate on dissociated cells and lose all spatial context — they cannot reveal how a gene perturbation affects tissue organization, cell localization, or intercellular communication. Perturb-DBiT solves this by reading out both the perturbation identity and the spatial total RNA transcriptome (including lncRNAs, miRNAs, and tRNAs, not just mRNAs) of each cell in intact tissue. The scale — 80,000+ sgRNAs — is orders of magnitude beyond previous spatially resolved screens. The ability to observe how perturbations affect non-coding RNA regulation, clonal dynamics, and immune infiltration simultaneously in the same tissue section is unprecedented. This technology could fundamentally change how we study gene function in multicellular contexts like tumours, developing organs, and immune tissues.
Why for Yiru: Perturb-DBiT is directly applicable to studying the tumour microenvironment at a resolution that has never been possible. One could systematically perturb immune checkpoint genes, chemokine receptors, or metabolic enzymes in tumour-immune co-cultures or in vivo models and map how each perturbation reshapes the spatial organization of the TME — including effects on non-coding RNA networks that have been largely invisible to previous perturbation screens. The miRNA–mRNA and tRNA perturbation linkages are particularly interesting because these regulatory layers are increasingly implicated in immune cell function and tumour immune evasion but have been difficult to study systematically. The computational challenges are substantial — analyzing spatial total RNA perturbation data requires methods for integrating multiple RNA species, spatial autocorrelation, and pseudotime trajectories — representing a rich area for methodological development. The clonal dynamics analysis also provides a template for studying how perturbations affect tumour evolution and cooperation between subclones.
Scaling up training dataset size for transcriptomic AI models is much pain with little gain
Nature Methods Published 2026-06-09 research article DOI: 10.1038/s41592-026-03119-5
single-cell foundation model transcriptomics dataset scaling model evaluation RNA benchmarking deep learning
Summary: Systematically evaluates the role of training dataset size and diversity in the performance of single-cell foundation models. The prevailing assumption in the field — inherited from the success of large language models — has been that more training data invariably leads to better models, driving efforts to aggregate ever-larger single-cell atlases comprising tens of millions of cells. This study directly tests that assumption by training single-cell foundation models on datasets of varying sizes and compositions and measuring performance across a comprehensive suite of downstream tasks including cell-type annotation, batch integration, gene program discovery, and perturbation response prediction. The authors find that increasing dataset size beyond a relatively modest set point yields surprisingly little performance gain across most tasks, with diminishing returns setting in far earlier than expected. Dataset diversity — the range of tissues, species, and experimental protocols represented — matters more than raw cell count. The study provides concrete guidance on the cost-benefit tradeoffs of scaling training data versus investing in data quality and diversity, and challenges the "scale is all you need" narrative that has dominated the field.
Why it matters: The single-cell foundation model field has been in an arms race to train on increasingly massive datasets, with computational costs scaling superlinearly. This study provides the first rigorous evidence that this race may be misguided — beyond a certain point, adding more cells to the training set produces marginal returns. If validated across model architectures and tasks, this finding could redirect substantial computational resources from data aggregation toward improving model architectures, training objectives, and data quality. It also has implications for equity: if dataset size requirements are lower than assumed, researchers with more modest computational resources can still train competitive models, democratizing access to foundation model development. The emphasis on diversity over size aligns with biological intuition — a model that has seen 50 well-chosen tissues likely understands transcriptomic variation better than one trained on 500 redundant samples from the same tissue type.
Why for Yiru: This finding is directly relevant to building TME-specific foundation models. If dataset size yields diminishing returns, then curating a high-quality, diverse TME dataset (spanning multiple cancer types, immune contexts, and treatment conditions) may be more valuable than aggregating every available single-cell dataset. This could accelerate the development of TME-focused models by lowering the data barrier. The study also provides a benchmarking framework for evaluating what foundation models actually learn from their training data — an approach that could be applied to assess whether existing models capture TME-relevant biology or merely memorize cell-type labels. For computational biologists building or fine-tuning foundation models, the concrete guidance on data scaling versus diversity tradeoffs is immediately actionable.
Bridging three-dimensional molecular structures and artificial intelligence with a conformation description language
Nature Machine Intelligence Published 2026-06-11 research article DOI: 10.1038/s42256-026-01250-8
molecular conformation language model 3D molecular generation conformer prediction representation learning cheminformatics structure-based AI
Summary: Introduces ConfSeq, a molecular conformation description language that converts three-dimensional molecular structures into text-like sequences, enabling language models to perform 3D molecular modelling tasks with strong performance. A fundamental challenge in molecular AI is the mismatch between the sequential nature of language models and the inherently three-dimensional character of molecular structures — conformations, stereochemistry, and spatial arrangements of atoms. ConfSeq bridges this gap by developing a formal language that encodes molecular 3D conformations as structured text sequences, preserving spatial information such as bond lengths, angles, dihedral angles, and non-covalent interactions. Once encoded, standard transformer language models can be trained on these ConfSeq representations to perform tasks including conformer prediction (predicting the 3D shape of a molecule from its 2D graph), 3D molecular generation (generating new molecules with desired 3D properties), and molecular representation learning (learning embeddings that capture 3D structural features). The authors demonstrate that language models trained on ConfSeq representations achieve strong performance across these tasks, often matching or exceeding specialized 3D-aware architectures while benefiting from the scalability and pre-training advantages of the language model ecosystem.
Why it matters: The representation of 3D molecular structure has been a persistent bottleneck in AI-driven drug discovery. Graph neural networks and equivariant architectures have been the dominant approaches, but they require specialized model designs and are not easily integrated with the broader language model ecosystem that is rapidly advancing. ConfSeq offers an elegant alternative: by converting 3D structure into a language that standard transformer models can process, it brings the full power of large-scale language model pre-training, transfer learning, and multimodal integration to 3D molecular tasks. The ability to use off-the-shelf language model architectures and training pipelines could dramatically accelerate progress in structure-based drug design, virtual screening, and molecular property prediction. The concept of a "conformation description language" also opens the door to multimodal models that jointly reason about molecular structure and biological text.
Why for Yiru: Structure-based approaches are increasingly relevant to computational immunology and cancer biology. Predicting how tumour neoantigens fold and present to T cell receptors, understanding how small-molecule immunotherapies bind their targets, and designing molecules that disrupt checkpoint protein-protein interactions all require reasoning about 3D molecular structure. ConfSeq could enable language model-based approaches to these problems, potentially integrating structural information with the vast immunological knowledge captured in biomedical text corpora. The framework also provides a blueprint for representing other complex biological structures (such as protein complexes or chromatin architectures) as language sequences that transformer models can process. For those working at the intersection of AI and biology, ConfSeq demonstrates how careful representation design can unlock the power of general-purpose models for domain-specific problems.
From virtual experiments to biomedical insight with synthetic data
Nature Machine Intelligence Published 2026-06-11 perspective DOI: 10.1038/s42256-026-01244-6
synthetic data machine learning biomedical AI simulation sim-to-real gap benchmark immunology
Summary: Presents a comprehensive perspective on the role of synthetic datasets in biomedical machine learning, focusing on the persistent simulation-to-reality gap that limits how well synthetic performance predicts real-world performance. Synthetic data — computationally generated datasets that mimic real biological data — has become increasingly important for training and benchmarking biomedical ML models, particularly in domains where real data is scarce, expensive, or privacy-restricted. The authors systematically survey the landscape of synthetic data generation approaches in biomedicine, from mechanistic simulations of cellular processes to deep generative models that learn to produce realistic synthetic omics data. They identify the sim-to-real gap as the central unresolved challenge: models that perform excellently on synthetic benchmarks often fail on real biological data because synthetic data cannot capture the full complexity, noise, and confounding factors present in real biological systems. The authors propose a framework for evaluating and reducing the sim-to-real gap, including adversarial validation, domain adaptation strategies, and hybrid approaches that combine synthetic data with small amounts of real data. The perspective also addresses the ethical dimensions of synthetic biomedical data, including the risks of perpetuating biases present in the real data used to train generative models.
Why it matters: Synthetic data is increasingly proposed as a solution to data scarcity, privacy, and reproducibility challenges in biomedical ML. However, the community lacks a systematic understanding of when synthetic data is sufficient and when it leads to misleading conclusions. This perspective provides the most comprehensive treatment of the sim-to-real gap in biomedical ML to date, offering both a diagnostic framework and practical strategies for mitigation. The implications are broad: from regulatory decisions about whether ML models can be validated on synthetic data, to methodological choices in tool development, to the design of benchmark competitions. The emphasis on domain adaptation and hybrid approaches also provides a practical path forward — rather than choosing between real and synthetic data, use both strategically.
Why for Yiru: Synthetic single-cell and spatial transcriptomics data is increasingly used for benchmarking computational methods, including TME analysis tools that infer cell-cell communication, identify spatial niches, and predict perturbation responses. The sim-to-real gap analysis in this perspective is directly relevant: methods that perform well on synthetic TME data generated by simulators may fail on real tumour biopsies where cell-type proportions, spatial arrangements, and transcriptomic noise are far more complex. The proposed adversarial validation framework could be applied to evaluate whether synthetic TME data generators produce realistic enough data for benchmarking. For those developing computational methods for TME analysis, this perspective provides essential guidance on when to trust synthetic benchmarks and how to design evaluation strategies that include real-data validation.
CAREPath — Semantic Context-Aware Reasoning Paths with Mechanism-Augmented Embeddings for Drug Repurposing
bioRxiv Published 2026-06-12 preprint DOI:
drug repurposing knowledge graph large language model biomedical AI mechanism reasoning graph neural network
Summary: Proposes CAREPath, a KG-LLM framework that combines knowledge graph traversal with large language model embeddings for mechanism-aware drug repurposing. Biomedical knowledge graphs (BKGs) that include drugs, genes, and diseases can support drug repurposing by connecting drugs to diseases through gene-mediated multi-hop paths, but deeper traversal does not necessarily improve mechanistic reasoning — long paths grow combinatorially and frequently pass through hub genes, producing irrelevant gene regulatory signals. CAREPath addresses this with two complementary modules inspired by graph search strategies: a DFS-like module that constrains traversal to short disease-gene-drug paths, converts each path into structured prompts, and encodes them with a biomedical language model to generate semantic path embeddings; and a BFS-like module that constructs entity-level mechanism-context embeddings from one-hop gene neighborhoods and enriches them through similarity-guided augmentation using pharmacologically related drugs and gene-signature-similar diseases. Across five biomedical KGs, CAREPath achieves the best overall AUPRC among 18 baselines, improving performance by up to 3.8%. Semantic short-path encoding contributes most to performance, while mechanism-context augmentation improves robustness under sparse evidence and strengthens Gene Ontology functional agreement. Case studies and recently FDA-approved indications demonstrate practical relevance.
Why it matters: Drug repurposing — finding new uses for existing drugs — is one of the most promising applications of biomedical AI because it can dramatically reduce the time and cost of bringing treatments to patients. However, most computational repurposing methods either lack mechanistic interpretability (pure ML approaches) or lack scalability (pure network approaches). CAREPath bridges this gap by using language models to encode mechanistic paths as semantic embeddings, combining the scalability of embedding-based methods with the interpretability of path-based reasoning. The DFS/BFS hybrid approach is conceptually elegant and addresses a fundamental tension in knowledge graph reasoning: depth vs. breadth. The finding that short paths contribute most to performance is practically important — it suggests that drug repurposing signals are captured by relatively local graph structure, simplifying computational requirements.
Why for Yiru: Drug repurposing is directly relevant to cancer immunotherapy, where identifying existing drugs that could enhance immune checkpoint inhibitor efficacy or overcome resistance is an active area of research. CAREPath could be applied to identify drugs that modulate TME-relevant pathways — for example, drugs that reprogramme immunosuppressive macrophages, reduce T cell exhaustion, or disrupt tumour-stromal interactions. The path-based reasoning framework could also incorporate TME-specific knowledge (cell-type-specific gene expression, spatial organization, signalling networks) to prioritize drugs expected to be effective in specific TME contexts. The mechanism-context augmentation strategy provides a template for enriching computational predictions with biological domain knowledge — an approach broadly applicable to TME-focused ML models.
Biomedical discoveries
Biomedicine
The PLK4 inhibitor RP-1664 demonstrates potent efficacy in neuroblastoma preclinical models through a dual mechanism of sensitivity
Nature Communications Published 2026-06-13 research article DOI: 10.1038/s41467-026-74061-5
neuroblastoma PLK4 inhibitor targeted therapy TRIM37 synthetic lethality centrosome paediatric cancer preclinical model
Summary: Demonstrates that the novel PLK4 inhibitor RP-1664 has potent efficacy in neuroblastoma through a dual mechanism of sensitivity involving both TRIM37-dependent and TRIM37-independent pathways. It was recently shown that inhibition of polo-like kinase 4 (PLK4) induces synthetic lethality in cancers with chromosome 17q-encoded TRIM37 copy number gain due to cooperative regulation of centriole duplication and mitotic spindle nucleation. This study shows that chromosome 17q/TRIM37 gain is a defining feature of high-risk neuroblastoma and renders patient-derived cell lines hypersensitive to RP-1664. At high doses, RP-1664 causes centrosome depletion in a TRIM37-dependent manner. Critically, at low doses, RP-1664 causes cell death in a TRIM37-independent fashion via centrosome amplification — neuroblastoma cells succumb to multipolar mitoses due to an inability to cluster or inactivate supernumerary centrosomes. CRISPR screens and live cell imaging confirmed this dual mechanism. RP-1664 monotherapy showed robust anti-tumour activity in 14 of 15 human neuroblastoma-derived xenograft models and significantly extended survival in a transgenic MYCN-driven murine model. Remarkably, RP-1664 combined with GD2-directed chemoimmunotherapy resulted in maintained complete responses in 6 of 9 mice with established MYCN-driven tumours. These data support clinical development of PLK4 inhibitors for high-risk neuroblastoma.
Why it matters: High-risk neuroblastoma remains one of the most difficult-to-cure paediatric cancers, with survival rates below 50% despite intensive multimodal therapy. The discovery of a dual mechanism of sensitivity to PLK4 inhibition — operating through both TRIM37-dependent centrosome depletion and TRIM37-independent centrosome amplification — is mechanistically fascinating and therapeutically important. The TRIM37-independent mechanism means that even tumours without TRIM37 gain may be sensitive, broadening the potential patient population. The near-universal efficacy across PDX models (14/15) is exceptional and suggests a fundamental dependency of neuroblastoma on centrosome homeostasis. The synergy with GD2-directed chemoimmunotherapy is particularly exciting because GD2 antibodies are already standard of care for high-risk neuroblastoma, providing a clear clinical development path. The complete responses in the combination arm are the kind of result that motivates rapid translation.
Why for Yiru: Centrosome biology intersects with cancer immunology in several ways that are relevant to the TME. Centrosome amplification can trigger innate immune sensing pathways (cGAS-STING) through micronuclei formation, and mitotic errors can generate neoantigens through chromosomal instability. The finding that PLK4 inhibition causes centrosome amplification at low doses suggests an opportunity to combine PLK4 inhibitors with immunotherapy — the mitotic chaos induced by centrosome amplification could increase tumour immunogenicity while the direct cytotoxicity reduces tumour burden. The PDX model results also demonstrate the value of preclinical models that preserve tumour heterogeneity for evaluating targeted therapies, a principle that applies broadly to TME research where patient-derived models are essential for capturing tumour-immune interactions. The dual mechanism finding — where the same drug has different effects at different doses — is a reminder that dose optimization is critical and that mechanism of action can be dose-dependent.
YBX1 confers immunosuppressive bone metastatic traits in non-small cell lung cancer
Nature Communications Published 2026-06-13 research article DOI: 10.1038/s41467-026-73931-2
bone metastasis NSCLC YBX1 immunosuppression glycosylation immunotherapy resistance Icaritin
Summary: Identifies the transcription factor YBX1 as a central regulator that simultaneously drives bone metastasis and establishes an immunosuppressive microenvironment in non-small cell lung cancer (NSCLC). Bone metastases from lung cancer are a devastating complication often linked to resistance against immunotherapy, but the mechanistic connection between metastasis and immune evasion has been poorly understood. This study shows that YBX1 achieves its dual function by activating distinct signalling pathways: IL-6 to drive bone metastasis and CCL5 to create an immunosuppressive environment. Mechanistically, YBX1 protein levels are controlled by glycosylation that marks it for autophagic degradation — reduced YBX1 glycosylation was observed in highly bone-metastatic NSCLC cells, suggesting that loss of this post-translational modification stabilizes YBX1 and unleashes its pro-metastatic and immunosuppressive programmes. In a key translational advance, the authors identify the small molecule Icaritin as an agent that boosts YBX1 glycosylation, leading to its degradation. Icaritin treatment simultaneously inhibited bone metastasis and re-sensitized tumours to immune attack in preclinical models. This work reveals YBX1 as a single therapeutic target for combating both metastatic spread and immunotherapy resistance.
Why it matters: Bone metastasis is a common and devastating complication of NSCLC and many other cancers, causing severe pain, fractures, and neurological complications. Immunotherapy has limited efficacy in bone metastases, and the mechanisms underlying this resistance have been obscure. This study provides a unified molecular explanation — YBX1 coordinates both the physical spread to bone and the immune evasion that makes these metastases resistant to immunotherapy. The identification of a single transcription factor that controls both processes is conceptually important and therapeutically attractive because it means one intervention could address two clinical problems simultaneously. The discovery that YBX1 stability is regulated by glycosylation-directed autophagy adds a novel post-translational regulatory mechanism. Most importantly, the identification of Icaritin as a drug that induces YBX1 degradation provides an immediately actionable therapeutic strategy. Icaritin is a prenylated flavonoid that has already shown clinical activity in hepatocellular carcinoma, suggesting a potential path to rapid clinical testing.
Why for Yiru: The connection between metastasis and immune evasion is central to understanding why some tumours respond to immunotherapy while others do not. The YBX1-IL-6-CCL5 axis identified here provides a concrete molecular mechanism that could be interrogated in other cancer types and metastatic sites. The concept that a single transcription factor coordinates both tissue tropism and immune microenvironment remodelling has parallels in other settings — for example, transcription factors that drive liver metastasis may also suppress anti-tumour immunity in the liver microenvironment. The glycosylation-dependent degradation mechanism highlights the importance of post-translational regulation in cancer, an area that is often understudied relative to transcriptional and genetic alterations. For computational analysis, YBX1 target gene signatures could be used to identify tumours at high risk of bone metastasis and immunotherapy failure from transcriptomic data, potentially guiding treatment decisions.
Spatial cellular order underlies locally-confined mechanisms of immune resistance in oropharyngeal cancer
Nature Communications Published 2026-06-13 research article DOI: 10.1038/s41467-026-74318-z
oropharyngeal cancer HPV immune escape spatial transcriptomics tumour microenvironment T cell evasion hypoxia
Summary: Uses single-cell spatial gene expression profiling to characterize cellular organization and mechanisms of immune resistance in HPV-associated oropharyngeal squamous cell carcinomas (HPV-OPSCCs). Despite expressing abundant viral antigens that should make these tumours highly immunogenic, HPV-OPSCCs frequently evade immune destruction. Through spatial analysis, the authors describe distinct tumour-parenchymal immune foci that differ in cytokine expression, spatial location, immune cell infiltration, and cancer cell states. These immune foci display profound differences in co-inhibitory receptor signalling and immunosuppressive myeloid cells, suggesting that different regions of the same tumour are dominated by distinct, locally-confined mechanisms of immunosuppression rather than a single global immune escape strategy. A critical finding is the identification of senescent-like HPV-OPSCC cells that lack HPV transcripts (HPVoff cells) — these cells are evident across the tumour parenchyma and are able to evade HPV-specific T cell-mediated immunity in vitro. HPVoff cells are enriched within hypoxic regions and near IFN-γ-producing T cells, suggesting that both metabolic stress (hypoxia) and immune pressure (IFN-γ) can promote this immune-evasive cellular state. This reveals a complex spatial interplay where the very immune response intended to eliminate the tumour may inadvertently select for antigen-loss variants.
Why it matters: HPV-positive oropharyngeal cancers are a growing epidemic, particularly in younger patients, and while they generally have a better prognosis than HPV-negative head and neck cancers, a significant fraction recur and become fatal. Understanding how these tumours evade immunity despite expressing foreign viral antigens is critical for improving immunotherapy outcomes. This study makes two major contributions: first, it demonstrates that immune escape mechanisms are spatially compartmentalized — different regions of the same tumour use different immunosuppressive strategies, which has profound implications for combination immunotherapy design (one drug may not be sufficient if different regions require different interventions). Second, the discovery of HPVoff cells that arise under immune pressure provides a mechanism for acquired resistance to T cell-based immunotherapies and suggests that therapeutic vaccines targeting multiple HPV antigens may be needed to prevent antigen escape. The hypoxia connection also suggests that targeting tumour oxygenation could reduce the emergence of immune-evasive variants.
Why for Yiru: The finding that immune escape mechanisms are spatially compartmentalized within individual tumours is a paradigm-shifting observation for TME research. Most TME studies either average across entire tumours (bulk analysis) or treat the TME as a single ecosystem, but this work shows that a single tumour contains multiple distinct immune microenvironments operating under different rules. This has direct implications for computational TME analysis: methods that assume spatial homogeneity or that average immune features across a tumour section will miss this critical heterogeneity. The HPVoff cell state — where immune pressure selects for antigen-loss variants — parallels mechanisms of immune escape under checkpoint therapy and suggests that spatial tracking of antigen presentation heterogeneity could predict immunotherapy failure. The IFN-γ-driven selection of immune-evasive states is a phenomenon that likely occurs across many cancer types and merits systematic computational investigation using spatial transcriptomics data.
Clonal evolution and mutational trajectories of metastatic colorectal cancer shaped by anticancer therapies
Nature Communications Published 2026-06-13 research article DOI: 10.1038/s41467-026-74384-3
colorectal cancer clonal evolution metastasis whole-genome sequencing therapy resistance single-cell tumoroid mutational signature
Summary: Investigates the evolutionary changes in cancer genomes under treatment by performing whole-genome sequencing of 58 single-cell-derived tumoroids and 18 matched bulk tumours from 6 metastatic colorectal cancer patients. While the genomic landscape of primary colorectal cancer has been extensively characterized, the evolutionary trajectories that shape metastatic disease — particularly under the selective pressure of chemotherapy and targeted therapy — remain incompletely understood. By comparing matched primary and metastatic samples at single-cell resolution, the authors reconstruct the clonal architecture and mutational history of each patient's cancer. The study reveals that metastatic dissemination can occur early or late in tumour evolution, with different patients showing distinct patterns. Critically, anticancer therapies impose strong selective pressures that reshape the clonal composition of metastases, with therapy-resistant subclones expanding under treatment. The authors identify therapy-associated mutational signatures, including signatures of platinum-based chemotherapy and specific targeted agents, that are enriched in post-treatment metastases. The tumoroid approach allows functional validation of mutations identified by sequencing, connecting genomic alterations to cellular phenotypes.
Why it matters: Understanding how metastases evolve under therapy is essential for designing treatment strategies that prevent or overcome resistance. This study provides one of the most detailed views of metastatic colorectal cancer evolution to date, combining single-cell-derived tumoroids with whole-genome sequencing to achieve both depth and functional validation. The finding that therapy-associated mutational signatures are enriched in post-treatment metastases is important: it suggests that the treatments themselves contribute to the mutational burden of surviving cancer cells, potentially creating new vulnerabilities or resistance mechanisms. The variation in metastatic timing (early vs. late dissemination) has clinical implications — patients with early dissemination may benefit from more aggressive systemic therapy at diagnosis, while those with late dissemination may be candidates for metastasis-directed local therapy.
Why for Yiru: The clonal evolution framework used here is directly applicable to studying how the TME shapes and is shaped by tumour evolution. Metastatic sites have distinct microenvironments (liver, lung, peritoneum) that impose different selective pressures on arriving cancer cells — understanding which clones succeed in which environments could reveal TME-specific vulnerabilities. The therapy-associated mutational signatures identified in this study could be used as biomarkers of treatment-induced evolution in other cancer types. For computational biologists, the combination of single-cell-derived tumoroids with whole-genome sequencing provides a powerful template for connecting genomic evolution to functional phenotypes — an approach that could be extended to study how immune pressure shapes tumour evolution by incorporating T cell recognition assays into the tumoroid platform.
Spatial architecture of autism pathogenesis reveals mosaic structural disarray during early development
Nature Communications Published 2026-06-13 research article DOI: 10.1038/s41467-026-74320-5
autism brain organoid spatial transcriptomics neurodevelopment progenitor cell neuronal organization disease model
Summary: Uses spatial and single-cell transcriptomics of patient-derived brain organoids to reveal disrupted progenitor–neuron organization and local neuronal disarray in autism spectrum disorder, implicating spatially mosaic pathogenesis in disease heterogeneity. Autism is a highly heterogeneous neurodevelopmental disorder, and understanding how genetic risk factors translate into cellular phenotypes during early brain development has been challenging. The authors generated cerebral organoids from induced pluripotent stem cells derived from individuals with autism and controls, then applied spatial transcriptomics and single-cell RNA sequencing to map cellular organization at high resolution. They found that autism organoids exhibit spatially mosaic structural disarray — regions of disrupted progenitor zone organization and neuronal layering interspersed with apparently normal regions — rather than uniform developmental abnormalities. This mosaicism may explain both the heterogeneity of autism phenotypes and the observation that many autism risk genes are expressed in specific progenitor populations during narrow developmental windows. The spatial disarray affects the radial organization of neuronal migration and the formation of distinct cortical layer-like structures, suggesting that early defects in progenitor behaviour propagate to produce widespread but patchy disruption of cortical architecture.
Why it matters: Autism research has been dominated by genetics, but translating genetic findings into mechanistic understanding of brain development has been extremely difficult. This study bridges that gap by showing how autism-associated genetic backgrounds produce specific, measurable disruptions in tissue architecture during early neurodevelopment. The finding of spatially mosaic defects — patches of disrupted organization next to normal regions — is mechanistically novel and helps explain why autism phenotypes are so variable: the specific location and extent of disrupted regions may determine which neural circuits are affected. The organoid platform also demonstrates how patient-derived models can be used to study neurodevelopmental disorders that are impossible to investigate in living human brains. This approach could be extended to other neuropsychiatric conditions and used to screen for interventions that restore normal tissue architecture.
Why for Yiru: The spatial transcriptomics approach used here is methodologically instructive for TME research. The discovery of spatially mosaic defects — where pathology is patchy rather than uniform — parallels observations in tumours, where different regions of the same tumour can have distinct genetic alterations, immune infiltrates, and drug sensitivities. The analytical framework for identifying spatially restricted transcriptional programmes could be applied to map immune exclusion zones, metabolic niches, and treatment-resistant regions within tumours. The organoid platform also demonstrates the power of combining patient-derived models with spatial omics, an approach that is increasingly used in cancer research with patient-derived tumour organoids co-cultured with immune cells to study TME organization.
Cross-disciplinary watchlist
Other Fields
The embryonic origins of site-specific arthritis
Nature Immunology Published 2026-06-08 research article DOI: 10.1038/s41590-026-02542-2
arthritis embryonic development fibroblast joint site-specific inflammation single-cell RNA-seq synovium
Summary: Maps the cellular and structural composition of developing human finger joints to uncover why specific joints are preferentially affected by inflammatory arthritis while adjacent joints are spared. In human fingers, proximal interphalangeal (PIP) joints are commonly affected by inflammatory arthritis, whereas distal interphalangeal joints are relatively spared — this provides a natural model to investigate the anatomical basis of inflammation susceptibility. Using single-cell RNA sequencing, imaging, and X-ray tomography, the authors examined cellular composition, spatial organization, and structure of finger joints during fetal development. They discovered that PIP joints have a larger synovial volume and are specifically enriched for PI16+ "universal" fibroblasts — a mesenchymal population located in perivascular regions and at developing tendon–ligament interfaces. Critically, these PI16+ fibroblasts exhibited both a shared inflammatory and cell-type-specific response to cytokine stimulation, suggesting that their combination of spatial location and transcriptional responsiveness promotes inflammation. The authors propose that differences in the stoichiometry of mesenchymal cells established during embryonic development — particularly the abundance and positioning of PI16+ fibroblasts — is a general principle that drives inflammation susceptibility across different tissues. This "positional memory" established in utero may explain why certain anatomical sites are recurrently affected by inflammatory diseases throughout life.
Why it matters: Why inflammatory diseases affect specific anatomical sites while sparing adjacent tissues has been a long-standing mystery in rheumatology and immunology. The dominant paradigm has focused on local triggers — mechanical stress, infection, or injury — but this study proposes a fundamentally different explanation: site-specific disease susceptibility is programmed during embryonic development through the stoichiometry and positioning of tissue-resident mesenchymal cells. The identification of PI16+ fibroblasts as a developmentally established population that confers inflammation susceptibility is a major conceptual advance. If validated in other tissues and diseases, the "positional memory" hypothesis could explain site-specific patterns in many inflammatory conditions — why certain skin regions develop psoriasis, why specific gut segments are affected in Crohn's disease, or why particular joints are targeted in different forms of arthritis. Therapeutically, targeting PI16+ fibroblasts or their inflammatory programmes could provide site-specific anti-inflammatory therapy without systemic immunosuppression.
Why for Yiru: The concept of developmentally programmed tissue susceptibility to inflammation has intriguing parallels in tumour immunology. The TME is shaped not only by the tumour itself but by the pre-existing cellular composition of the tissue in which it arises — tissue-resident fibroblasts, macrophages, and extracellular matrix established during development may create permissive or restrictive environments for tumour growth and immune infiltration. The finding that PI16+ fibroblasts have both shared and cell-type-specific inflammatory responses suggests that understanding tissue-specific fibroblast biology is essential for predicting how different tissues respond to immunotherapies. The single-cell and spatial analysis framework used here — combining developmental biology with immunology — provides a template for studying how tissue architecture established early in life influences disease susceptibility decades later, an approach that could be applied to understand why certain organs are more susceptible to cancer or metastasis.
Single-cell spatial pharmacobiology for imaging antibody-based therapies in solid tumors
Nature Biotechnology Published 2026-06-08 research article DOI: 10.1038/s41587-026-03171-8
spatial proteomics antibody therapy pharmacobiology tumour microenvironment drug delivery imaging phase 1 trial
Summary: Introduces single-cell spatial pharmacobiology (SSP), a platform that combines in situ imaging of a systemically infused fluorescent therapeutic antibody with high-plex spatial proteomics to map antibody delivery, target engagement, and microenvironmental barriers at single-cell resolution in human tumours. A fundamental challenge in antibody-based cancer therapy is that even when a therapeutic antibody is present in the bloodstream, its ability to reach and engage its target on tumour cells is highly heterogeneous and depends on local tissue architecture. SSP addresses this by fluorescently labelling the therapeutic antibody, infusing it into patients, and then performing high-plex spatial proteomics on tumour biopsies to simultaneously visualize where the antibody goes, which cells it engages, and what stromal barriers limit its distribution. Applied to head and neck and pancreatic tumours from patients treated in phase 1 clinical trials, SSP revealed marked spatial heterogeneity in antibody delivery and target engagement. Conserved stromal barriers — including dense extracellular matrix, perivascular fibroblast accumulation, and abnormal vasculature — systematically limited antibody penetration, creating tumour regions that were effectively untreated despite adequate systemic drug levels. This platform transforms antibody therapy from a "black box" where only systemic pharmacokinetics are measured to a spatially resolved understanding of drug action at the cellular level.
Why it matters: Antibody-based therapies — including checkpoint inhibitors, antibody-drug conjugates, and bispecific antibodies — are among the most important classes of cancer drugs, yet we have remarkably little understanding of where these drugs actually go inside human tumours. Current pharmacokinetic measurements are limited to blood levels, which may have little relationship to intratumoural drug concentrations. SSP directly visualizes the "last mile" problem of antibody therapy: does the drug reach its target cells, and if not, what barriers are responsible? The finding that conserved stromal barriers limit antibody delivery across different tumour types suggests that stromal targeting strategies (e.g., hyaluronidase, anti-fibrotic agents) may be broadly applicable for improving antibody therapy efficacy. The integration of SSP into phase 1 trials is a model for how pharmacodynamic biomarkers can be incorporated early in drug development to understand mechanism of action and identify resistance mechanisms before large phase 3 trials.
Why for Yiru: SSP is directly relevant to understanding why immunotherapies succeed or fail in the TME. Checkpoint inhibitors must reach specific immune cells within tumours, bispecific T cell engagers must bridge T cells and tumour cells in close proximity, and antibody-drug conjugates must penetrate tumour tissue to deliver their payload. SSP could reveal which TME features predict antibody penetration and which stromal barriers should be targeted to improve delivery. For computational analysis, SSP data provides a unique opportunity to build spatial models of antibody transport through tumour tissue — combining imaging data with mathematical models of diffusion, binding, and cellular uptake could predict optimal dosing strategies and identify patients most likely to benefit from specific antibodies. The finding that pancreatic tumours (which are notoriously immunotherapy-resistant) have particularly severe antibody delivery barriers provides a mechanistic explanation for clinical failures and a rationale for stroma-targeting combination strategies.
Hybrid solid−liquid optics enable scalable, high-resolution light-sheet microscopy across diverse immersion media
Nature Biotechnology Published 2026-06-09 research article DOI: 10.1038/s41587-026-03172-7
light-sheet microscopy optics imaging cleared tissue scalable imaging neuroimaging
Summary: Presents a hybrid solid–liquid optics design for light-sheet fluorescence microscopy that enables scalable, high-resolution imaging of intact biological samples across diverse immersion media, including cleared tissues, expanded samples, and living specimens. Light-sheet microscopy is a powerful technique for imaging large intact biological samples with minimal photodamage, but current implementations face tradeoffs between resolution, field of view, and compatibility with different sample preparations. The hybrid optics approach combines solid immersion lenses (which provide high numerical aperture for high resolution) with liquid immersion objectives (which provide long working distances and compatibility with diverse refractive index media). This design enables a single instrument to image samples ranging from cleared mouse brains to live zebrafish embryos to expanded human tissue sections, without the need to change optical components. The authors demonstrate applications including whole-brain imaging of neuronal circuits in cleared mouse brains, developmental imaging of living zebrafish, and high-resolution imaging of human clinical biopsy specimens. The scalable design means that larger samples can be imaged at the same resolution by tiling, and the open-source hardware specifications enable other laboratories to build comparable instruments.
Why it matters: The ability to image intact biological specimens at high resolution across scales — from subcellular structures to whole organs — is fundamental to modern biology. However, the field has been fragmented by incompatible instrumentation: an instrument optimized for cleared tissue cannot image live samples, and vice versa. The hybrid solid–liquid optics design solves this by creating a universal light-sheet platform that works across sample types and preparation methods. This could accelerate discovery by enabling correlated imaging — for example, imaging the same sample live and then after clearing to relate dynamic processes to static anatomy. The open-source approach also democratizes access to high-end microscopy, which has historically been concentrated in well-funded labs and core facilities. For clinical applications, the ability to image intact human biopsy specimens could enable 3D pathology, where tissue architecture is assessed in three dimensions rather than on thin sections.
Why for Yiru: Advanced microscopy is increasingly important for TME research, where understanding the spatial organization of tumours and immune cells requires imaging across scales — from subcellular signalling events to millimetre-scale tissue architecture. The hybrid optics approach could enable correlated live and fixed imaging of the same tumour samples, revealing how dynamic processes like T cell migration relate to static features like vascular architecture and extracellular matrix organization. The compatibility with diverse sample preparations is particularly valuable for TME studies, which use a wide range of models including organoids, tissue slices, and intact cleared tumours. The scalability of the design also means that multi-sample imaging experiments — comparing treated vs. untreated tumours, or primary vs. metastatic sites — become more practical.
A k-mer-based genome-wide association study approach empowering gene mining in polyploids
Nature Genetics Published 2026-06-12 research article DOI: 10.1038/s41588-026-02641-8
GWAS polyploid k-mer plant genomics sugarcane statistical genetics crop breeding
Summary: Presents KMERIA, a k-mer-based GWAS framework specifically designed for polyploid organisms where traditional SNP-based approaches fail due to multiple homeologous genomes. Standard GWAS relies on aligning sequencing reads to a reference genome and calling variants, but in high-ploidy species such as sugarcane (which can have over 100 chromosomes and ploidy levels exceeding 10x), read alignment is ambiguous, variant calling is unreliable, and dosage estimation is nearly impossible. KMERIA bypasses these challenges by operating directly on k-mer counts — short DNA sequence words — without requiring reference alignment or variant calling. The method associates k-mer presence/absence or abundance patterns with phenotypes, then maps significant k-mers back to genomic features for biological interpretation. The authors demonstrated KMERIA's power by applying it to wild sugarcane (Saccharum spontaneum), identifying genetic variants associated with key agronomic traits including biomass yield, sugar content, and stress tolerance that were invisible to conventional GWAS. The method showed substantially higher statistical power and computational efficiency compared to alignment-based approaches in polyploid contexts.
Why it matters: Many of the world's most important crops are polyploid — wheat, potato, cotton, coffee, strawberry, and sugarcane — yet they have been largely excluded from the genomics revolution that has transformed breeding in diploid crops. KMERIA solves this by fundamentally rethinking the GWAS paradigm: instead of forcing polyploid data into diploid analysis frameworks, it works directly with raw sequencing data in a way that naturally accommodates multiple genomes and unknown ploidy. For global food security, enabling genetic improvement of polyploid crops could have massive impact — sugarcane alone provides about 80% of the world's sugar and is a major biofuel feedstock. Beyond agriculture, the k-mer approach is applicable to any setting with complex genomic architectures, including cancer genomes with extensive aneuploidy.
Why for Yiru: The k-mer GWAS approach has direct applications in cancer genomics where tumour genomes are often highly aneuploid with complex copy number alterations, making traditional alignment-based variant calling unreliable — exactly the same challenge that polyploid plant genomics faces. K-mer-based association methods could identify somatic mutations or copy number alterations associated with drug response or metastatic potential directly from raw sequencing data without the biases of alignment and variant calling pipelines. The computational efficiency of KMERIA is also noteworthy: as single-cell and spatial sequencing datasets grow to include thousands of samples, alignment-free approaches that operate on k-mer spectra could enable analyses that are currently computationally prohibitive. This framework exemplifies how methods developed for challenging genomes in one field can cross-pollinate to address analogous problems in cancer biology.
Prion-based protein self-assembly tunes mutagenesis to enable rapid adaptation
Cell Published 2026-06-09 research article DOI:
prion protein self-assembly mutagenesis adaptation drug resistance genome maintenance evolvability
Summary: Reveals that prion-based protein self-assembly enables reversible switching of genome-maintenance pathways during rapid adaptation and the emergence of drug resistance. Prions — proteins that can adopt alternative, self-propagating conformations — are best known for their role in neurodegenerative diseases, but a growing body of work has shown that prion-like protein switches can serve adaptive functions in yeast and other organisms. This study demonstrates that specific proteins can form prion-like assemblies that globally alter the balance of DNA repair and mutagenesis pathways. When environmental stress is encountered, the prion switch activates, increasing mutation rates and generating genetic diversity that fuels adaptation. Critically, once adaptation has occurred, the prion state can be reversed, restoring normal genome maintenance and preventing the accumulation of deleterious mutations that would result from sustained hypermutation. The authors show that this mechanism contributes to the rapid emergence of antifungal drug resistance in pathogenic yeast, with the prion switch enabling a transient "mutator phenotype" that generates resistance mutations without permanently compromising genome stability. This represents a regulated, reversible mechanism for increasing evolvability — a capacity that has been theorized but rarely demonstrated with such mechanistic clarity.
Why it matters: The idea that organisms can regulate their own mutation rates in response to stress — effectively controlling their own evolvability — has been controversial in evolutionary biology. This study provides one of the clearest demonstrations of regulated, reversible hypermutation through a specific molecular mechanism. The prion-based switch is elegant because it couples environmental sensing (stress triggers prion formation) to a phenotypic output (increased mutagenesis) with built-in reversibility (prions can be cleared when stress subsides). The finding that this mechanism contributes to drug resistance in pathogenic fungi has immediate clinical relevance: if fungi use prion switches to generate antifungal resistance, targeting the switch itself could prevent resistance emergence. More broadly, the concept of regulated mutagenesis challenges the assumption that mutation rates are passive and suggests that therapeutic strategies that block stress-induced hypermutation could complement drugs that directly kill pathogens or cancer cells.
Why for Yiru: The concept of regulated mutagenesis is directly relevant to cancer evolution and therapy resistance. Cancer cells exposed to chemotherapy or targeted therapy experience profound stress, and there is increasing evidence that stressed cancer cells upregulate error-prone DNA repair pathways and suppress high-fidelity repair — essentially inducing a mutator phenotype that accelerates the emergence of resistance mutations. The prion-based switch mechanism described here provides a molecular precedent for how such stress-induced hypermutation could be regulated. If analogous protein-conformation-based switches operate in cancer cells, they could represent therapeutic targets for preventing or delaying drug resistance. The reversibility of the prion switch is particularly interesting — if cancer cells use similar reversible mechanisms, then "drug holidays" might allow restoration of normal genome maintenance and resensitization to therapy. Computational methods for detecting signatures of stress-induced mutagenesis in tumour genomes could identify patients whose tumours are actively using such mechanisms.