Research Radar — 2026-06-11
Methods & AI
Computational
CytoSignal detects locations and dynamics of ligand–receptor signaling at cellular resolution from spatial transcriptomic data
Nature Genetics Published 2026-06-10 research article DOI: 10.1038/s41588-026-02624-9
ligand-receptor spatial transcriptomics cell-cell communication RNA velocity CytoSignal tissue biology computational method signaling inference
Summary: Introduces CytoSignal, a computational method that infers ligand-receptor (LR) interactions at cellular resolution directly from spatial transcriptomic data, and extends the analysis to predict temporal signaling dynamics by incorporating RNA velocity. A central challenge in spatial transcriptomics is that while technologies like Visium, MERFISH, and Xenium can measure gene expression with spatial coordinates, inferring which cells are actually communicating — and through which LR pairs — requires computational methods that account for spatial proximity, co-expression patterns, and downstream transcriptional responses. CytoSignal addresses all three: it models the spatial distance between potential sender and receiver cells, scores LR interactions based on co-expression of ligand and receptor genes in neighboring cells, and uses the expression of known LR target genes as an orthogonal validation signal. The key innovation is the integration of RNA velocity — which estimates the future transcriptional state of each cell from splicing kinetics — to predict not just which cells are currently signaling, but how LR interactions will shift over developmental or disease time. The authors demonstrate CytoSignal on multiple spatial transcriptomic platforms and biological contexts, including mouse brain development, tumour-immune interfaces, and lung fibrosis, revealing spatially organized signaling niches that are invisible to traditional non-spatial LR methods.
Why it matters: Cell-cell communication is the fundamental language of multicellular biology, and LR interactions are its vocabulary. Existing LR inference methods (CellPhoneDB, NicheNet, CellChat) operate on dissociated single-cell data and lose all spatial context — they can tell you that two cell types express complementary ligand and receptor, but not whether those cells are actually adjacent in the tissue. CytoSignal solves this by operating directly on spatial data, enabling the discovery of spatially organized communication hubs. The RNA velocity extension is particularly innovative — it transforms LR inference from a static snapshot into a dynamic prediction, potentially revealing how communication networks evolve during development, disease progression, or therapy. As spatial transcriptomics becomes a standard tool, methods like CytoSignal will be essential for extracting functional insight from these rich datasets.
Why for Yiru: CytoSignal is directly applicable to TME spatial analysis. The TME is defined by spatially organized cell-cell communication — checkpoint interactions at the immune synapse, chemokine gradients guiding immune infiltration, growth factor signaling at the tumour-stroma interface. Applying CytoSignal to TME spatial transcriptomics data could identify which LR pairs define immune-infiltrated vs. immune-excluded niches, reveal how communication networks reorganize after immunotherapy, and predict which signaling pathways drive T cell exhaustion or macrophage polarization. The RNA velocity extension could be used to predict how TME communication will evolve after treatment — for example, whether checkpoint blockade will shift the TME toward a more inflammatory communication state. Computationally, CytoSignal also provides a framework for integrating spatial LR inference with other spatial modalities such as spatial proteomics.
Explicit dynamic cross-strand interactions for DNA sequence language modelling
Nature Machine Intelligence Published 2026-06-04 research article DOI: 10.1038/s42256-026-01249-1
DNA language model genomics deep learning double-strand DNA CrossDNA non-coding variant regulatory genomics sequence modeling
Summary: Presents CrossDNA, a parameter-efficient language model for DNA sequence modeling that explicitly incorporates the double-stranded, cross-strand interaction dynamics of DNA — a fundamental biological property ignored by most genomic language models. Current DNA language models (DNABERT, Enformer, HyenaDNA) treat genomic DNA as a single-stranded sequence of A, C, G, T tokens, much like text in natural language processing. This ignores the fact that DNA exists as a double helix where the two strands interact through Watson-Crick base pairing, creating structural and functional constraints that influence transcription factor binding, chromatin organization, and regulatory element function. CrossDNA addresses this by designing a cross-attention mechanism that models the forward strand alongside its reverse complement, allowing the model to learn strand-asymmetric features (such as transcription on one strand) while also capturing strand-symmetric constraints (such as the structural properties of the double helix). The model achieves strong performance on benchmarks including promoter prediction, enhancer identification, and chromatin accessibility prediction, while using significantly fewer parameters than single-stranded models of comparable performance. The authors demonstrate that CrossDNA's cross-strand attention weights highlight regulatory regions and can be used to prioritize non-coding variants associated with disease.
Why it matters: Genomic language models are increasingly used for variant effect prediction, regulatory element annotation, and sequence design, but their single-stranded architecture is a fundamental limitation. DNA is physically and functionally double-stranded — transcription occurs on one strand, replication forks involve both, and structural properties like bendability and supercoiling depend on the double helix. CrossDNA demonstrates that explicitly modeling this double-stranded nature improves performance and enables new analyses (such as strand-specific regulatory prediction) while being more parameter-efficient. This is a conceptually important step toward genomic models that respect the physical reality of DNA, and the parameter efficiency makes the approach practical for labs with limited computational resources.
Why for Yiru: Non-coding variant interpretation is critical for cancer genomics — many GWAS hits for cancer risk and immunotherapy response fall in non-coding regions, and understanding which variants affect gene regulation in the TME requires accurate regulatory sequence models. CrossDNA could be applied to prioritize non-coding variants in TME-relevant genes (checkpoint molecules, cytokines, chemokines) that may influence tumour-immune interactions. The strand-specific modeling is also relevant for understanding the regulatory architecture of immune genes, which often have complex, strand-asymmetric regulatory landscapes with enhancers and promoters operating on different strands. More broadly, the cross-attention design principle could inspire analogous architectural innovations for modeling other biological sequence features with intrinsic symmetry constraints.
Navigating molecular OOD-ness
Nature Machine Intelligence Published 2026-06-09 news and views DOI: 10.1038/s42256-026-01251-7
out-of-distribution molecular machine learning drug discovery distribution shift generalization chemical space benchmarking
Summary: Highlights a new metric for quantifying chemical distribution shift and evaluating the generalization capability of molecular machine learning models — a critical but often overlooked issue in AI-driven drug discovery. Machine learning models for molecular property prediction are typically trained on known chemical space and evaluated on held-out molecules from the same distribution. However, in real drug discovery, the goal is to find molecules that are structurally novel — molecules that are, by definition, out-of-distribution (OOD) relative to the training data. The inability of current models to reliably predict the properties of OOD molecules is a major barrier to their practical utility. The highlighted work introduces a quantitative framework for measuring how OOD a given molecule is relative to a training set, and for evaluating model performance as a function of OOD-ness. This enables researchers to identify when model predictions can be trusted and when they degrade into essentially random guesses. The news and views piece contextualizes this advance within the broader challenge of bridging the gap between in silico predictions and experimental validation in drug discovery pipelines.
Why it matters: The OOD problem is perhaps the single greatest barrier to the practical impact of AI in drug discovery. Pharmaceutical companies invest heavily in AI-driven molecular design, but if models cannot predict whether a novel molecule will actually be active, selective, and safe, the entire pipeline breaks down. A rigorous framework for quantifying and communicating prediction uncertainty for OOD molecules could transform how computational chemists use ML models — moving from blind trust in model predictions to calibrated decision-making that accounts for the novelty of each candidate. This is also relevant to regulatory science, where demonstrating the reliability of AI predictions for novel chemical entities will be essential for acceptance by agencies like the FDA.
Why for Yiru: While Yiru's primary focus is TME biology rather than drug discovery, the OOD generalization problem is universal in biomedical machine learning. Any model trained on one set of biological conditions (specific cancer types, mouse models, in vitro systems) and applied to another (different cancer types, human patients, in vivo) faces an OOD challenge. The framework for quantifying distribution shift could be adapted to assess whether a TME prediction model — for example, one that predicts immunotherapy response from transcriptional data — is operating within or outside its training distribution. This is essential for clinical translation, where predictions on patients from different populations or with different tumour types must be accompanied by uncertainty estimates.
MIXPRS enables multi-population and multi-method polygenic risk scores using summary statistics
Nature Genetics Published 2026-06-09 research article DOI: 10.1038/s41588-026-02637-4
polygenic risk score PRS multi-population GWAS summary statistics statistical genetics MIXPRS health equity
Summary: Introduces MIXPRS, a computational framework for constructing polygenic risk scores (PRS) that combine information from multiple populations and multiple PRS methods using only GWAS summary statistics — addressing a critical equity gap in genomic risk prediction. PRS aggregate the effects of thousands of genetic variants to estimate an individual's genetic risk for a disease, and they have shown promise for risk stratification in research settings. However, a major limitation is that PRS developed in European-ancestry populations perform poorly when applied to individuals of non-European ancestries, due to differences in linkage disequilibrium patterns, allele frequencies, and effect sizes. MIXPRS addresses this by combining GWAS summary statistics from multiple ancestry groups, leveraging the complementary information in each population to produce risk scores that are more accurate across diverse ancestries. Unlike methods that require individual-level data — which are rarely available due to privacy concerns — MIXPRS operates entirely on publicly available summary statistics. The authors demonstrate that MIXPRS-derived scores outperform single-population and single-method PRS across multiple ancestries for a range of complex diseases.
Why it matters: The poor portability of PRS across populations is one of the most pressing equity issues in genomic medicine. As PRS move toward clinical implementation for diseases including coronary artery disease, breast cancer, and type 2 diabetes, ensuring that these tools work for all patients — not just those of European ancestry — is both a scientific and ethical imperative. MIXPRS provides a practical solution that works with existing data resources, requiring no individual-level data sharing. The multi-method aspect is also important: different PRS methods (clumping-and-thresholding, LDpred, PRS-CS) have different strengths, and MIXPRS optimally combines them. This framework could accelerate the development of equitable PRS for clinical use worldwide.
Why for Yiru: While PRS are not Yiru's primary research focus, the multi-population framework is conceptually relevant to TME research. As single-cell and spatial atlases of the TME are built across diverse patient populations, methods that can integrate data from multiple cohorts with different genetic backgrounds will be essential for identifying generalizable TME features vs. population-specific ones. The MIXPRS framework for combining summary statistics across populations could inspire analogous methods for meta-analysis of TME single-cell data across ancestries, or for building risk models that account for both germline genetic risk and TME composition.
Chromatix: a differentiable, GPU-accelerated wave-optics library
Nature Methods Published 2026-06-08 research article DOI: 10.1038/s41592-026-03121-x
wave optics computational microscopy GPU differentiable programming simulation imaging open source Chromatix
Summary: Introduces Chromatix, an open-source library of differentiable wave-optics models that enables scalable, GPU-accelerated simulation of light propagation through optical systems — a critical computational tool for designing and optimizing modern microscopes. Many advanced microscopy methods, including light-sheet microscopy, structured illumination, and adaptive optics, rely on computational models of wave propagation to reconstruct images or to design optical components. However, existing wave-optics simulation tools are typically slow (CPU-based), not differentiable (precluding gradient-based optimization), and implemented as one-off research code that is difficult to reproduce or extend. Chromatix addresses all three limitations: it is built on JAX, a differentiable programming framework, enabling gradient-based optimization of optical designs; it runs on GPUs, achieving up to 22× speedup compared to typical research code; and it is modular and well-documented, allowing researchers to compose standard optical elements (lenses, apertures, gratings, propagation models) into complex simulations. The differentiability is particularly powerful — it means that optical systems can be optimized end-to-end using gradient descent, jointly designing the optical hardware and the computational reconstruction algorithm.
Why it matters: Computational microscopy — where image formation is treated as a joint optimization of optics and algorithms — is one of the most exciting frontiers in biological imaging. However, progress has been slowed by the lack of standardized, efficient simulation tools. Chromatix provides a common foundation that could accelerate the field in the same way that deep learning frameworks (TensorFlow, PyTorch) accelerated AI research. The differentiability is the key innovation: researchers can now optimize microscope designs using the same gradient-based methods that revolutionized deep learning, potentially discovering optical configurations that human designers would never consider. The open-source nature and modular design also promote reproducibility and community contributions.
Why for Yiru: Advanced microscopy — including spatial transcriptomics, multiplexed imaging, and intravital microscopy — is central to TME research. Better computational optics could improve the resolution, throughput, and depth of TME imaging, enabling more detailed mapping of tumour-immune interactions in 3D. The differentiable optimization approach could be applied to design TME-specific imaging systems — for example, optimizing light-sheet microscopy for cleared tumour samples, or designing adaptive optics for deep-tissue imaging of tumours in living animals. Chromatix could also facilitate the integration of computational optics with downstream image analysis, enabling end-to-end optimization of the entire imaging-to-insight pipeline.
Biomedical discoveries
Biomedicine
A prognostic human brain network for diffuse midline glioma
Nature Published 2026-06-10 research article DOI: 10.1038/s41586-026-10631-3
diffuse midline glioma brain tumour tumour network neuron-glioma synapse prognosis brain connectivity paediatric cancer functional MRI
Summary: Develops tumour network mapping to define a conserved brain-wide connectivity profile for diffuse midline glioma (DMG) — a near-universally lethal childhood brain tumour — and demonstrates that this network is independently prognostic of patient survival. Building on the groundbreaking discovery that gliomas form functional synapses with neurons and integrate into neural circuits to promote their own growth, the authors ask whether these tumour-integrated networks can be mapped in the living human brain and whether they have clinical significance. Using resting-state functional MRI from patients with pontine and thalamic DMG, they compute the brain-wide functional connectivity of each tumour and identify a conserved DMG-associated network that is remarkably consistent across patients and tumour locations. Tumour functional connectivity with this network was independently predictive of overall survival in two external validation cohorts. The DMG network overlaps with brain regions that show peak neurometabolic changes during development at ages that match the peak incidence of DMG — suggesting that developmental maturation of specific circuits creates a permissive environment for tumour initiation or growth. Strikingly, incidental surgical resection of tumour tissue in high-connectivity thalamic regions conferred a survival advantage, while single-nucleus RNA sequencing confirmed enrichment of synaptic gene expression programs in high-connectivity tumour regions.
Why it matters: This study represents a major advance in understanding how brain tumours hijack normal neural circuitry. The concept that cancers integrate into host tissue networks is not limited to gliomas — recent work has shown that tumours in other organs can co-opt neuronal signaling, and that cancer cells form gap-junction-coupled networks. This study provides the first systematic, clinically validated map of such a tumour network in humans, demonstrating that the degree of network integration directly impacts patient survival. The finding that the DMG network aligns with developmentally regulated brain regions also provides a potential explanation for why DMG occurs at characteristic ages and in characteristic locations. Clinically, the DMG network could serve as an imaging biomarker for risk stratification, and the network nodes themselves could be therapeutic targets — disrupting tumour-network integration might slow tumour growth.
Why for Yiru: The concept of tumour-host network integration extends naturally to the TME. Just as gliomas integrate into neural circuits, solid tumours integrate into stromal, vascular, and immune networks within their tissue of origin. Computational methods analogous to tumour network mapping could be developed for the TME — using spatial transcriptomics or multiplexed imaging to map how individual tumours are integrated into the surrounding cellular ecosystem, and whether the strength of this integration predicts outcomes or treatment responses. The finding that network connectivity predicts survival independently of traditional clinical factors suggests that tumour "social behavior" — how a tumour interacts with its environment — may be as important as its intrinsic molecular features.
Mutation-dependent responses to sleep and exercise in clonal haematopoiesis
Nature Published 2026-06-10 research article DOI: 10.1038/s41586-026-10634-0
clonal haematopoiesis sleep exercise lifestyle mutation-dependent aging blood CHIP
Summary: Reveals that two fundamental lifestyle factors — sleep and exercise — have mutation-dependent effects on the expansion and suppression of blood cell clones carrying different driver mutations in clonal haematopoiesis, the age-related phenomenon where haematopoietic stem cells acquire mutations and expand to dominate blood production. Clonal haematopoiesis of indeterminate potential (CHIP) is common in older adults and is associated with increased risk of haematologic cancers, cardiovascular disease, and all-cause mortality. However, it has been unclear whether lifestyle factors can influence the dynamics of mutant clones. This study uses longitudinal blood sequencing data from large population cohorts to track how sleep patterns and physical activity levels associate with the growth or contraction of clones carrying specific driver mutations (DNMT3A, TET2, ASXL1, JAK2, etc.). The findings are striking and mutation-specific: clones carrying certain mutations (such as TET2) expand more rapidly under conditions of sleep deprivation, while clones carrying other mutations (such as DNMT3A) are relatively insensitive to sleep. Exercise shows similarly mutation-dependent effects, with some clones suppressed and others unaffected. Mechanistically, the authors link these differential responses to the distinct effects of each mutation on inflammatory signaling, metabolic stress responses, and interactions with the bone marrow niche.
Why it matters: This study adds an important lifestyle dimension to clonal haematopoiesis, which has largely been viewed as an inevitable consequence of aging. The finding that sleep and exercise can differentially affect the growth of specific mutant clones suggests that lifestyle interventions might be used to suppress high-risk clones while sparing low-risk ones — a form of "clonal management" rather than blanket clone elimination. The mutation specificity is particularly important: it means that blanket lifestyle recommendations may not be appropriate, and that knowing which mutation a person carries could guide personalized lifestyle advice. More broadly, this study establishes that environmental and behavioural factors can modulate somatic evolution in vivo, connecting lifestyle epidemiology to molecular clonal dynamics.
Why for Yiru: The concept of mutation-dependent environmental responses is directly relevant to cancer biology and the TME. Just as specific CHIP mutations respond differently to sleep and exercise, specific tumour mutations may respond differently to TME conditions — hypoxia, nutrient availability, immune pressure, cytokine milieu. This could explain why tumours with different driver mutations thrive in different TME contexts and respond differently to therapies that modify the TME (such as checkpoint blockade or anti-angiogenic therapy). Computational methods could be developed to predict, from mutational and transcriptional data, which environmental factors will promote or suppress specific tumour clones — enabling environmentally informed precision oncology.
Hypoxia shapes both therapeutic response and resistance in metastatic clear cell renal cell carcinoma
Cancer Cell Published 2026-06-04 research article DOI:
renal cell carcinoma hypoxia VEGFR-TKI immune checkpoint inhibitor tumour microenvironment SPP1 TAM metastasis drug resistance
Summary: Comprehensively profiles the tumour microenvironment in mouse models and human tumour samples of clear cell renal cell carcinoma (ccRCC) after treatment with VEGFR tyrosine kinase inhibitors (TKIs) and immune checkpoint inhibitors (ICIs), revealing that hypoxia is a central determinant of both therapeutic response and the emergence of resistance. ccRCC is characterized by inactivation of the VHL tumour suppressor, leading to constitutive activation of the hypoxia-inducible factor (HIF) pathway and a highly angiogenic and hypoxic TME. The standard of care combines VEGFR-TKIs (which target angiogenesis) with ICIs (which activate anti-tumour immunity), but responses are heterogeneous and resistance is nearly universal. The authors find that true tissue hypoxia — as opposed to the pseudo-hypoxia driven by VHL loss — recruits SPP1+ tumour-associated macrophages (TAMs), and that the presence of these SPP1+ TAMs serves as a biomarker of therapeutic response. However, chronic treatment with VEGFR-TKIs exacerbates hypoxia in surviving tumour regions, which in turn promotes more aggressive tumour behaviour and increased metastasis in mouse models. This highlights a therapeutic paradox: the very treatment intended to starve the tumour of blood supply may, by intensifying hypoxia, select for more malignant tumour cells and promote metastatic dissemination.
Why it matters: This study illuminates the double-edged nature of anti-angiogenic therapy and provides a mechanistic framework for understanding treatment failure in ccRCC. The identification of SPP1+ TAMs as a hypoxia-driven biomarker of response is clinically actionable — it suggests that patients with high SPP1+ TAM infiltration may benefit most from combination therapy, while those with low infiltration may need alternative approaches. The finding that chronic anti-angiogenic therapy can promote metastasis is sobering and aligns with clinical observations that some patients develop more aggressive disease after prolonged TKI treatment. More broadly, the study demonstrates that hypoxia is not just a passive consequence of tumour growth but an active driver of TME remodeling, immune evasion, and therapeutic resistance — a principle that applies across many solid tumour types.
Why for Yiru: Hypoxia is a hallmark of the TME that affects virtually all aspects of tumour biology — immune cell function, metabolism, angiogenesis, invasion, and drug delivery. This study provides a detailed map of how hypoxia shapes the TME in the context of therapy, identifying specific cell types (SPP1+ TAMs) and signaling pathways that mediate hypoxia-driven resistance. For computational TME research, this suggests that integrating hypoxia signatures with spatial and single-cell data could improve predictions of therapy response and identify patients at risk for hypoxia-driven metastasis. The SPP1+ TAM biomarker could also be explored in other tumour types where hypoxia and TAM infiltration are prominent — including pancreatic cancer, glioblastoma, and hepatocellular carcinoma. Methodologically, the multi-modal approach combining mouse models with human tumour profiling provides a template for studying TME dynamics under therapeutic pressure.
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 evolution phenotypic switching
Summary: Discovers that prion-based protein self-assembly can reversibly tune the activity of genome-maintenance pathways, enabling cells to rapidly switch between high-fidelity and error-prone DNA repair states to accelerate adaptation under stress — including the emergence of drug resistance. Prions are proteins that can exist in multiple conformational states, including self-templating amyloid aggregates that can be transmitted from mother to daughter cells. While prions are best known for their role in neurodegenerative diseases, a growing body of work has shown that prion-like protein switches can serve adaptive functions by creating heritable phenotypic diversity without genetic change. This study identifies a prion-forming protein that, upon switching to its aggregated state, suppresses high-fidelity DNA repair pathways while upregulating error-prone translesion synthesis polymerases. The result is a transient increase in the mutation rate — a mutagenesis burst that generates genetic diversity upon which natural selection can act. Critically, this switch is reversible: when the stress is removed, the prion disaggregates and the high-fidelity repair machinery is restored. The authors demonstrate that this mechanism enables rapid adaptation to antifungal drugs in pathogenic fungi, with implications for how cancer cells might similarly tune mutagenesis to evade therapy.
Why it matters: The idea that cells can actively regulate their mutation rate in response to stress challenges the traditional view that mutations are purely stochastic. This study provides a molecular mechanism — prion-based switching of DNA repair fidelity — that enables cells to "turn up" mutagenesis when they need genetic diversity most, then "turn it down" when the crisis passes. This has profound implications for understanding drug resistance in cancer: if tumour cells can similarly boost their mutation rate in response to chemotherapy or targeted therapy, then the emergence of resistance may not be a passive process of selection from pre-existing variants, but an active, regulated response. The prion-based mechanism also suggests that targeting the switch itself — preventing the prion from forming, or forcing it to disaggregate — could suppress adaptive mutagenesis and delay resistance.
Why for Yiru: Drug resistance is a central challenge in cancer therapy, including immunotherapy. While this study focuses on fungal drug resistance, the principles are directly applicable to cancer: tumour cells are under constant selective pressure from the immune system, chemotherapy, and targeted agents, and they evolve resistance through both genetic and non-genetic mechanisms. The concept of regulated mutagenesis — that cells can actively control their mutation rate — suggests that the TME may influence the evolvability of tumour cells. For example, TME stresses such as hypoxia, nutrient deprivation, or immune attack could trigger mutagenic programs that accelerate the emergence of resistant clones. Computationally, this suggests new models of tumour evolution where mutation rates are not constant but dynamically regulated by TME conditions — models that could better predict when and how resistance emerges.
Replaying germinal center evolution on a quantified affinity landscape
Cell Published 2026-06-05 research article DOI:
germinal center affinity maturation B cell antibody evolution parallel replay somatic hypermutation immunology
Summary: Performs a "parallel replay" experiment on germinal center B cells to quantitatively dissect the evolutionary forces that produce predictable increases in antibody affinity during the immune response. Antibody affinity maturation is one of the most striking examples of somatic evolution in vertebrates: B cells in germinal centers undergo repeated rounds of somatic hypermutation and selection for improved antigen binding, producing antibodies with orders-of-magnitude higher affinity over the course of weeks. However, the relative contributions of mutation, selection, competition, and stochastic drift to this process have been difficult to disentangle because each germinal center reaction is unique. The authors solve this by performing "parallel replays": they take B cells from a single germinal center at an early time point, split them into multiple replicate cultures, and allow each to evolve independently while quantifying the affinity landscape (the relationship between antibody sequence and binding affinity) at high resolution. By comparing the evolutionary trajectories across replicates, they measure the predictability of affinity maturation and identify which steps are deterministic (driven by strong selection for specific mutations) vs. stochastic (driven by random mutation order or clonal competition). The results reveal that affinity maturation is highly predictable at the phenotypic level — all replicates converge on high affinity — but the genotypic routes they take are diverse, with different sets of mutations achieving similar affinity gains.
Why it matters: Understanding the rules of antibody affinity maturation has direct implications for vaccine design. The goal of most vaccines is to elicit germinal center responses that produce high-affinity, broadly neutralizing antibodies, but we currently lack the ability to predict or control which antibody lineages will emerge from a given immunization. The parallel replay approach provides a quantitative framework for measuring the predictability and constraints on affinity maturation, which could inform the design of immunogens that reliably guide B cell evolution toward desired antibody specificities. More broadly, this study demonstrates that somatic evolution in vertebrates can be as predictable as organismal evolution when the selective landscape is well-characterized.
Why for Yiru: Germinal center biology is relevant to tumour immunology because tertiary lymphoid structures (TLS) — ectopic lymphoid aggregates that form in the TME — can support germinal center-like reactions that produce anti-tumour antibodies and support T cell responses. Understanding the evolutionary dynamics of affinity maturation could inform strategies to enhance TLS function in tumours. The parallel replay experimental design is also conceptually relevant to studying the evolution of tumour cell populations under therapy — one could imagine analogous experiments where tumour cells from a single lesion are split into replicate cultures and evolved under drug pressure to measure the predictability of resistance mechanisms. At a computational level, the framework for quantifying evolutionary predictability from parallel trajectories could be applied to analyse longitudinal tumour sequencing data from patients on therapy.
Cross-disciplinary watchlist
Other Fields
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 3D imaging neuroscience solid-liquid optics scalable
Summary: Presents hybrid solid-liquid optics, a new approach for light-sheet microscopy that overcomes a fundamental limitation of current systems: the need to match the refractive index of the immersion medium to the sample, which restricts the types of samples and clearing methods that can be used. Light-sheet microscopy has transformed 3D imaging of intact biological specimens — whole mouse brains, tumour biopsies, organoids — by illuminating a thin plane of tissue and capturing the emitted fluorescence orthogonally, enabling fast, low-phototoxicity volumetric imaging. However, current light-sheet systems require the sample to be immersed in a medium whose refractive index precisely matches the optical design, limiting the technique to specific tissue-clearing protocols and preventing imaging across diverse sample types in the same system. The authors solve this by designing hybrid optics that combine solid optical elements (lenses) with liquid elements whose refractive properties can be dynamically tuned. This allows a single microscope to image samples prepared with different clearing methods — from organic-solvent-based clearing (which requires high-refractive-index media) to aqueous-based expansion microscopy. The system achieves near-diffraction-limited resolution across centimetre-scale specimens and is demonstrated on cleared mouse brains, axolotl brains, and human tissue samples.
Why it matters: This advance removes a major practical barrier to the widespread adoption of light-sheet microscopy for diverse biological applications. Currently, labs that want to image both solvent-cleared and hydrogel-expanded samples need multiple microscope systems or compromise on image quality. Hybrid solid-liquid optics enable a single, scalable platform that works across sample types, reducing cost and complexity. As tissue clearing and expansion methods continue to proliferate — and as large-scale projects like BRAIN Initiative and Human Cell Atlas generate thousands of intact specimens — the need for flexible, high-throughput volumetric imaging will only grow.
Why for Yiru: Volumetric imaging of intact tumours and tumour-bearing organs is essential for understanding the 3D architecture of the TME — how immune infiltrates are distributed, where hypoxic and necrotic regions form, and how therapeutic agents penetrate. Hybrid solid-liquid optics could enable imaging of cleared tumour samples prepared with different protocols in a single system, facilitating comparative studies across tumour types and treatment conditions. For spatial TME analysis, the ability to image large specimens at high resolution without refractive index mismatch artifacts would improve the accuracy of 3D cell segmentation, vascular tracing, and immune cell mapping — all critical for computational TME modeling.
Linking GWAS risk genes to transcriptional features of major depressive disorder via in vivo Perturb-seq
Nature Genetics Published 2026-06-10 research article DOI: 10.1038/s41588-026-02638-3
Perturb-seq GWAS major depressive disorder in vivo AAV single-cell functional genomics psychiatric genetics
Summary: Optimizes and applies in vivo AAV-Perturb-seq to systematically profile GWAS-linked major depressive disorder (MDD) risk genes and their corresponding transcriptomic features in the mouse brain — bridging the gap between genetic association and functional mechanism for psychiatric disease. GWAS have identified hundreds of genomic loci associated with MDD risk, but the vast majority of these variants fall in non-coding regions and their target genes, cell types, and molecular mechanisms remain unknown. Perturb-seq — which combines CRISPR-based genetic perturbations with single-cell RNA-seq readout — has been transformative for functional genomics in vitro, but applying it in vivo, especially in the brain, has been challenging. The authors develop an optimized AAV-based delivery system for in vivo Perturb-seq in the mouse prefrontal cortex, enabling simultaneous knockout of multiple MDD risk genes and transcriptomic profiling of the resulting cellular changes at single-cell resolution. They identify cell-type-specific transcriptional programs associated with each risk gene, revealing that different MDD risk genes affect distinct cell types — some primarily affect neurons, others affect glia, and some have broad effects across multiple cell types. The study also identifies shared transcriptional signatures across risk genes that converge on pathways implicated in synaptic function, neuroinflammation, and stress response.
Why it matters: This study represents a major methodological advance for functional validation of GWAS hits in complex brain diseases. The gap between genetic association and biological mechanism is perhaps widest in psychiatry, where the relevant cell types, circuits, and molecular processes are extraordinarily complex and difficult to study experimentally. In vivo Perturb-seq in the brain provides a systematic, scalable approach to functionally characterize risk genes in their native tissue context, revealing not just what each gene does, but which cell types it affects and which pathways it regulates. This framework is immediately applicable to other brain disorders (schizophrenia, autism, Alzheimer's) and could transform how we translate GWAS findings into biological insight and therapeutic targets.
Why for Yiru: While MDD is outside Yiru's primary research area, the in vivo Perturb-seq methodology is broadly relevant. Applying similar approaches in the TME — for example, using in vivo Perturb-seq to functionally characterize genes associated with immunotherapy response or resistance in tumour models — could systematically map which genes in which cell types (tumour, immune, stromal) mediate therapeutic outcomes. The optimized AAV delivery system for in vivo perturbations could be adapted for intratumoural delivery, enabling functional genomics directly within the TME. Computationally, the framework for linking perturbation effects to cell-type-specific transcriptional programs provides a template for analyzing Perturb-seq data from complex tissues.
The embryonic origins of site-specific arthritis
Nature Immunology Published 2026-06-08 research article DOI: 10.1038/s41590-026-02542-2
arthritis fibroblast developmental biology PI16 synovial joint inflammation single-cell
Summary: Combines single-cell RNA sequencing, advanced imaging, and X-ray tomography of developing human finger joints to reveal that the predilection of specific joints for inflammatory arthritis is established during fetal development through differences in fibroblast stoichiometry. A long-standing puzzle in rheumatology is why certain joints — such as the proximal interphalangeal (PIP) joints of the fingers — are preferentially affected by rheumatoid and psoriatic arthritis, while adjacent distal interphalangeal (DIP) joints are often spared. This study maps the cellular and structural composition of developing finger joints and discovers that PIP joints have a larger synovial volume and are enriched for PI16+ "universal" fibroblasts compared to DIP joints. These PI16+ fibroblasts are located in perivascular regions and at tendon-ligament interfaces, positioning them at sites of mechanical stress. Critically, PI16+ fibroblasts exhibit a heightened inflammatory response to cytokine stimulation compared to other fibroblast populations, suggesting that their enrichment in PIP joints creates an inherently more pro-inflammatory environment that predisposes these joints to arthritis. The authors propose that differences in mesenchymal cell stoichiometry established in utero represent a general principle driving site-specific inflammation susceptibility across tissues.
Why it matters: This study provides an elegant explanation for one of the most puzzling clinical observations in rheumatology — why arthritis affects some joints and not others — and traces the answer all the way back to embryonic development. The concept that tissue-specific disease susceptibility is "programmed" during development through differences in stromal cell composition has profound implications beyond arthritis. It suggests that many diseases with characteristic anatomical distributions — inflammatory bowel disease affecting specific gut segments, psoriasis affecting specific skin sites, myocarditis affecting specific heart regions — may similarly be explained by developmentally determined differences in tissue-resident mesenchymal cell populations.
Why for Yiru: The concept of developmentally programmed tissue susceptibility is directly relevant to cancer biology. Different tissues have vastly different cancer susceptibilities that cannot be fully explained by mutation rates or environmental exposures — for example, the small intestine is far less cancer-prone than the colon despite having a higher cell turnover rate. Differences in tissue-resident fibroblast and immune cell composition established during development could create permissive or restrictive environments for tumour initiation. In the TME context, the PI16+ fibroblast population identified here is analogous to tissue-resident fibroblasts in other organs that may influence tumour susceptibility and the composition of the tumour stroma. Computationally, single-cell atlases of developing and adult tissues could be mined to identify developmentally determined stromal cell populations that influence disease susceptibility across organs.
Multiplexed, precise genome engineering in monocots with twin prime editing systems
Nature Biotechnology Published 2026-06-05 research article DOI: 10.1038/s41587-026-03174-5
prime editing genome engineering multiplexed monocots plant biotechnology CRISPR twin prime editing agriculture
Summary: Develops a twin prime editing system that enables multiplexed, precise genome engineering in monocot plants — including the major crops rice, wheat, and maize — overcoming the low efficiency that has limited prime editing applications in these economically critical species. Prime editing is a "search-and-replace" genome editing technology that can install precise base substitutions, small insertions, and small deletions without requiring double-strand DNA breaks or donor templates, making it more versatile and potentially safer than traditional CRISPR-Cas9 editing. However, prime editing efficiency in plants, particularly monocots (grasses including the world's most important food crops), has been too low for practical applications. The authors solve this by engineering a "twin" prime editing system that uses paired prime editing guide RNAs (pegRNAs) targeting both DNA strands at the same locus, which dramatically improves editing efficiency. They further optimize the system for multiplexing — simultaneously editing multiple genomic loci — by designing compatible pegRNA expression cassettes. The system achieves efficient multiplexed editing in rice protoplasts and stable transgenic plants, demonstrating simultaneous edits at up to four loci with efficiencies suitable for practical crop improvement.
Why it matters: Precise genome editing in crops is essential for addressing global challenges in food security, climate resilience, and sustainable agriculture. While CRISPR-Cas9 has been widely adopted for gene knockout in plants, many agriculturally important traits require precise nucleotide changes (to improve enzyme activity, alter regulatory sequences, or introduce known beneficial alleles from wild relatives) rather than gene disruption. Prime editing enables these precise edits, and the twin prime editing system brings it to practical efficiency levels in the world's most important food crops. The multiplexing capability is particularly valuable because most agronomic traits are polygenic — improving yield, drought tolerance, or disease resistance typically requires editing multiple genes simultaneously.
Why for Yiru: While plant genome editing is outside Yiru's immediate research area, the technological principles — particularly the twin prime editing strategy for improving editing efficiency through dual-strand targeting — could inform genome editing approaches in mammalian cells. For TME research, efficient multiplexed genome editing could enable the simultaneous perturbation of multiple genes in tumour cells, immune cells, or stromal cells to study combinatorial effects on TME composition and function. The pegRNA design principles and multiplexing strategies developed for plants could be adapted for mammalian systems, particularly for in vivo Perturb-seq applications where editing efficiency and multiplexing capacity are critical.
Pleiotropic shared heritability quantifies the shared genetic variance of common diseases
Nature Genetics Published 2026-06-09 research article DOI: 10.1038/s41588-026-02607-w
pleiotropy heritability genetic correlation GWAS complex disease UK Biobank statistical genetics PHBC
Summary: Introduces pleiotropic shared heritability with bias correction (PHBC), a statistical method that estimates the genetic variance of a target disease that is shared with a set of auxiliary diseases, revealing that roughly half of common disease heritability is pleiotropic with a broad range of conditions. While genetic correlation studies have identified pairs of diseases that share genetic risk factors, estimating the total proportion of a disease's genetic architecture that is shared across many conditions simultaneously has been technically challenging due to sampling noise in genetic correlation matrices. PHBC solves this using a Monte Carlo bias correction procedure that accounts for estimation error, producing unbiased estimates of pleiotropic heritability. Applied to 15 diseases across seven disease categories in the UK Biobank, PHBC reveals that on average 27% of SNP heritability is shared with four other diseases in the same category, and this rises to 48% when expanding to 62 auxiliary diseases and traits — meaning that nearly half of the genetic risk for any given common disease is shared with other conditions. This pervasive pleiotropy spans disease categories: cardiovascular, autoimmune, metabolic, and neuropsychiatric diseases all share substantial genetic components. The authors also find that genetic effects are more pleiotropic than non-genetic risk factors, suggesting that shared biology at the molecular level is a fundamental feature of complex disease.
Why it matters: This study quantifies what many clinicians and geneticists have suspected: common diseases are not genetically distinct entities but share substantial genetic architecture. The finding that nearly half of disease heritability is pleiotropic has profound implications for drug development, risk prediction, and our understanding of disease biology. For drug development, it suggests that targeting shared pathways could yield therapies effective across multiple diseases — as already seen with JAK inhibitors (effective in multiple autoimmune conditions) and PCSK9 inhibitors (affecting both lipid levels and cardiovascular risk). For risk prediction, it suggests that multi-disease PRS that leverage shared genetic architecture could outperform single-disease scores. For basic biology, it underscores that the genetic boundaries between diseases are blurred — a fact that disease-specific research programs often ignore.
Why for Yiru: Pleiotropy between cancer and non-cancer diseases is increasingly recognized — for example, autoimmune disease risk genes also influence cancer susceptibility, and cardiovascular risk factors share genetic architecture with certain cancers. PHBC could be applied to quantify the shared genetic architecture between cancer types and between cancer and immune-mediated diseases, potentially identifying TME-relevant pathways that influence both cancer risk and immune function. More broadly, the finding that genetic effects are highly pleiotropic while environmental effects are more disease-specific has implications for TME research: germline genetic variation may create a "baseline" TME state that influences susceptibility to multiple cancer types, while environmental and somatic factors determine which specific cancer develops.