Research Radar — 2026-06-01
Methods & AI
Computational
FLOWR: flow matching for structure-aware de novo, interaction- and fragment-based ligand generation
Nature Computational Science Published 2026-05-28 research article DOI: 10.1038/s43588-026-00998-8
drug design flow matching deep learning ligand generation structure-based equivariant computational chemistry generative model
Summary: Introduces FLOWR, a structure-based deep learning framework for generating and optimizing three-dimensional ligands that integrates continuous and categorical flow matching with equivariant optimal transport and efficient protein pocket conditioning. Structure-based drug design requires generating ligand molecules that fit a target protein pocket with high affinity and favorable drug-like properties, but existing diffusion- and flow-based methods often produce geometrically invalid molecules or struggle with fragment-based and interaction-constrained generation. FLOWR addresses these challenges through three innovations: (1) a unified flow matching formulation that handles both continuous atom coordinates and categorical atom/bond types, (2) equivariant optimal transport that aligns generated and reference structures for more stable training, and (3) pocket conditioning that efficiently encodes the protein binding site. Alongside FLOWR, the authors introduce SPINDR, a curated dataset of ligand–pocket cocrystal complexes designed to address data quality issues in existing benchmarks. FLOWR surpasses current state-of-the-art methods on PoseBusters validity, pose accuracy, and drug-likeness metrics. The framework supports de novo generation, fragment-based elaboration, and interaction-constrained design — generating ligands that satisfy specified protein–ligand interaction patterns.
Why it matters: Generative models for drug design have progressed rapidly, but generating physically realistic, synthesizable molecules remains the central challenge. FLOWR's flow matching approach offers a principled alternative to diffusion models, with the equivariant optimal transport component explicitly addressing a known weakness of flow-based methods in high-dimensional molecular spaces. The interaction-constrained generation mode is particularly valuable for medicinal chemistry — it allows chemists to specify required hydrogen bonds or hydrophobic contacts and generate molecules that satisfy those constraints. The SPINDR dataset also addresses the well-known problem of low-quality training data in molecular AI, which has led to inflated benchmark performance.
Why for Yiru: Computational drug design tools that can generate molecules satisfying specific protein–ligand interaction constraints could be applied to TME targets — for example, generating ligands that selectively bind immunosuppressive checkpoints or metabolic enzymes in the TME while avoiding systemic homologs. The fragment-based elaboration mode is especially relevant for optimizing hits from TME-focused phenotypic screens. More broadly, FLOWR's flow matching framework could potentially be extended to other structured prediction problems in computational biology, such as protein–protein interface design or peptide–MHC binding prediction.
Decoding Condition-Specific Cellular Crosstalk in Spatial Omics via Bilinear Edge Classification
preprint Published 2026-05-06 preprint DOI: 10.64898/2026.05.03.722470
spatial omics cell-cell communication bilinear classification tissue organization graph neural network niche computational biology spatial transcriptomics
Summary: Presents a computational framework for decoding condition-specific cell–cell communication from spatial omics data using bilinear edge classification. Tissues are multicellular communities whose function emerges from both individual cell characteristics and their spatial organization — yet most cell–cell communication inference methods operate on dissociated single-cell data, losing the spatial context that determines which cells actually interact. This work models the problem as edge classification on a spatial graph, where nodes are cells (or spots) and edges represent potential communication axes. The bilinear formulation jointly models the sender cell state, receiver cell state, and their interaction, enabling the classifier to identify communication edges that are condition-specific — active in disease but not in health, or in one tissue region but not another. The method is demonstrated on spatial transcriptomics data, identifying condition-specific ligand–receptor interactions that are spatially restricted to specific tissue niches. By focusing on differential communication rather than absolute communication, the framework naturally filters out ubiquitous housekeeping interactions and highlights context-dependent signaling.
Why it matters: Spatial omics technologies (Visium, MERFISH, Xenium, CosMx) are generating unprecedented data on tissue architecture, but methods for extracting cell–cell communication from these data have lagged behind the technology. Most existing tools either ignore spatial context (using dissociated scRNA-seq communication inference) or use simple distance-based co-localization. The bilinear edge classification framework provides a principled statistical approach for identifying which communication axes are active and condition-specific, directly addressing the core question spatial biologists want to answer: 'Which cells are talking to which other cells, and how does this change in disease?' This is essential for understanding tissue reorganization during disease progression, ageing, and therapeutic response.
Why for Yiru: The TME is defined by spatially organized cellular communication: tumour cells signal to immune cells, stromal cells recruit and exclude lymphocytes, and immune niches form around specific chemokine gradients. A bilinear edge classification approach could be applied to spatial transcriptomics data from tumour sections to systematically map condition-specific TME communication — comparing treated vs. untreated, responder vs. non-responder, or primary vs. metastatic tumours. Identifying which communication edges are altered by immunotherapy could reveal mechanisms of response and resistance that are invisible to non-spatial analyses.
Multi-ancestry transcriptome-wide association studies uncover insights into breast cancer genetics and biology
Nature Communications Published 2026-05-30 research article DOI: 10.1038/s41467-026-73801-x
TWAS transcriptome-wide association study breast cancer multi-ancestry GWAS gene regulation alternative splicing computational genomics
Summary: Conducts multi-ancestry transcriptome-wide association studies (TWAS) to discover breast cancer susceptibility genes by integrating GWAS data from 178,534 cases and 248,300 controls with ancestry-specific genetic models of gene expression, alternative splicing, and 3′ UTR alternative polyadenylation. GWAS have identified over 200 risk loci for breast cancer, but pinpointing the causal genes at these loci has been challenging because most risk variants are non-coding and their target genes are unknown. The authors develop ancestry-specific prediction models using genomic and transcriptomic data from 652 normal female breast tissue samples, then apply these models to GWAS summary statistics from multiple ancestry groups. The multi-ancestry design is critical because genetic architecture differs across populations, and single-ancestry TWAS can miss associations that are present only in populations not included in the reference panel. The study identifies known and novel breast cancer susceptibility genes, validates findings across independent datasets, and demonstrates that integrating multiple molecular phenotypes — not just total gene expression, but also isoform-level splicing and 3′ UTR usage — significantly increases discovery power.
Why it matters: Breast cancer GWAS have been enormously successful at identifying risk loci, but translating these loci into biological understanding and clinical utility requires identifying the effector genes. TWAS bridges this gap by using genetically predicted expression to implicate specific genes at GWAS loci. The multi-ancestry design is particularly important for health equity — without it, genetic risk prediction and biological insight are biased toward European populations. The inclusion of splicing and polyadenylation phenotypes is also noteworthy; many disease-associated variants affect isoform usage rather than total expression, and traditional expression-only TWAS would miss these effects.
Why for Yiru: TWAS methodology is directly transferable to other cancer types and immune phenotypes. The same framework could be applied to identify genes whose genetically predicted expression in immune cells affects immunotherapy response or TME composition. The multi-ancestry aspect is also relevant — if TME features have population-specific genetic architecture, single-ancestry analyses would miss important biology. More broadly, the splicing-aware TWAS approach could be adapted to study how genetically regulated alternative splicing in tumour or immune cells affects cancer outcomes.
Protein language models for structural biology
Nature Computational Science Published 2026-05-28 comment / review DOI: 10.1038/s43588-026-00993-z
protein language model structural biology deep learning protein design AlphaFold evolution computational biology review
Summary: A perspective piece discussing how protein language models — large-scale deep learning models trained on hundreds of millions of protein sequences — can effectively decode the grammar of protein evolution, making structure prediction and design scalable. Protein language models learn the statistical patterns of amino acid co-evolution across evolutionary time, implicitly capturing the physical constraints that govern protein folding. Unlike structure-based approaches like AlphaFold that require multiple sequence alignments, language models can make predictions from single sequences, dramatically expanding the scope of predictable proteins to include orphan sequences, designed proteins, and viral proteins with few homologs. The piece surveys recent advances including ESM-3, which integrates sequence, structure, and function modalities, and discusses how these models are accelerating biological discovery and protein engineering across scales — from understanding variant effects in human disease to designing novel enzymes for industrial applications.
Why it matters: Protein language models represent a paradigm shift in structural biology: rather than explicitly modeling physics or co-evolutionary constraints, they learn these principles implicitly from massive sequence data. This single-sequence capability is transformative for studying proteins that lack deep multiple sequence alignments — a category that includes many disease-relevant human proteins, viral proteins, and engineered sequences. The convergence of protein language models with structure prediction (ESMFold) and design (ProteinMPNN) is creating a unified computational framework for protein science that will accelerate both basic discovery and therapeutic development.
Why for Yiru: Protein language models can be applied to TME-relevant proteins including immune checkpoints, chemokine receptors, and tumour-specific neoantigens. Single-sequence prediction is particularly valuable for studying somatic mutations in cancer — tumour-specific mutations often occur in domains where evolutionary information is sparse, making traditional methods unreliable. Embeddings from protein language models could also serve as features for predicting peptide–MHC binding, TCR–pMHC recognition, and antibody–antigen interactions — all critical for computational immunology in the TME.
Biomedical discoveries
Biomedicine
Pan-cancer spatial atlas of tertiary lymphoid structures
Science Published 2026-05-28 research article DOI: 10.1126/science.adz2742
tertiary lymphoid structure TLS spatial transcriptomics pan-cancer tumour microenvironment B cell T cell immune niche cancer immunotherapy
Summary: Constructs a comprehensive pan-cancer spatial transcriptomics atlas of tertiary lymphoid structures (TLSs) spanning 12 cancer types, characterizing TLS spatial architecture, maturation states, and clinical relevance. TLSs are organized aggregates of B cells, T cells, and dendritic cells that form in non-lymphoid tissues during chronic inflammation and cancer, functioning as local sites of adaptive immune priming. Their presence is generally associated with favorable prognosis and better immunotherapy responses, but their spatial organization across different cancer types has not been systematically characterized. This study analyzes spatial transcriptomics data to map the cellular composition, spatial architecture, and maturation trajectories of TLSs across a diverse set of tumour types. The authors find that TLS maturation is accompanied by coordinated remodeling of distinct niche cell populations and distance-dependent gradients in tumour programs — more mature TLSs are associated with spatially restricted immunosuppressive programs in immediately adjacent tumour cells, suggesting a dynamic interplay between immune activation and tumour adaptation. The atlas reveals both conserved features of TLS biology across cancers and cancer-type-specific differences in TLS composition and spatial organization.
Why it matters: TLSs have emerged as one of the most promising biomarkers for immunotherapy response, but their biology has been studied largely in individual cancer types. This pan-cancer atlas provides a unified framework for understanding TLS formation, maturation, and function, revealing both universal principles and cancer-specific variations. The finding that TLS maturation correlates with distance-dependent immunosuppressive programs in adjacent tumour cells is particularly important — it suggests that TLSs are not simply 'good' immune structures but exist in a dynamic equilibrium with tumour counter-adaptation. This has therapeutic implications: strategies to promote TLS maturation may need to be combined with interventions that block tumour adaptation to maintain therapeutic benefit.
Why for Yiru: TLSs are a quintessential TME structure — they represent the immune system's attempt to organize an adaptive response within the tumour itself. Understanding TLS spatial organization computationally is directly aligned with Yiru's interest in spatial multi-omics and computational immunology. The distance-dependent gradient analysis could be extended using cell–cell communication inference to identify the ligand–receptor pairs that mediate the TLS–tumour interaction. This atlas also provides a reference for benchmarking computational methods that detect and characterize TLSs from spatial transcriptomics data.
MAGE-A4/MAGE-A8-targeted TCR-based bispecific T cell engager in recurrent and/or refractory solid tumors: a phase 1 trial
Nature Medicine Published 2026-05-31 clinical trial (phase 1) DOI: 10.1038/s41591-026-04455-x
T cell engager bispecific TCR-based MAGE-A4 MAGE-A8 solid tumour immunotherapy phase 1 trial IMA401
Summary: Reports a prespecified interim analysis of a phase 1 first-in-human trial of IMA401, a T cell receptor (TCR)-based next-generation bispecific T cell engaging receptor (TCER) targeting an HLA-A*02:01-presented peptide derived from MAGE-A4/MAGE-A8. IMA401 incorporates a high-affinity TCR-based targeting domain specific for the MAGE-A4/A8 peptide–HLA complex, a low-affinity T-cell-recruiting domain (anti-CD3) to minimize cytokine release syndrome, and an optimized Fc domain to prolong half-life. In this trial, 61 patients with advanced solid tumours received intravenous IMA401 (0.0066 mg to 2.5 mg) with or without pembrolizumab. The primary endpoint was determination of the maximum tolerated dose. The TCER platform represents an evolution beyond traditional bispecific T cell engagers (BiTEs): using a TCR-like targeting domain rather than an antibody fragment allows targeting of intracellular proteins presented as peptide–HLA complexes, vastly expanding the repertoire of addressable tumour antigens. MAGE-A4 and MAGE-A8 are cancer-testis antigens expressed in multiple solid tumour types but not in normal adult tissues (except testis), making them attractive immunotherapy targets.
Why it matters: T cell engagers have transformed the treatment of hematologic malignancies but have had limited success in solid tumours due to target antigen selection, on-target off-tumour toxicity, and the immunosuppressive TME. IMA401 addresses two of these challenges: the TCR-based targeting domain accesses intracellular antigens (expanding beyond surface proteins), and the low-affinity CD3 engager is designed to reduce cytokine release syndrome — a major dose-limiting toxicity of T cell engagers. If this phase 1 trial demonstrates a manageable safety profile and early signals of efficacy, it would validate the TCER platform and open the door to targeting other cancer-testis antigens and neoantigens in solid tumours.
Why for Yiru: MAGE-A4 is one of the most well-characterized cancer-testis antigens and is frequently expressed in tumours with poor T cell infiltration. A computational analysis of MAGE-A4 expression patterns across the TME — in different tumour regions, across different cancer types, and in relation to immune infiltration — could identify which patients and tumour subtypes are most likely to benefit from MAGE-A4-targeted therapies. More broadly, the TCR-based targeting paradigm is directly relevant to computational neoantigen prediction: identifying which peptide–HLA complexes are tumour-specific and amenable to TCR-based targeting is a core problem in computational immunology.
Pan-cancer single-cell atlases of mouse and human tumor-associated dendritic cells
Nature Communications Published 2026-05-30 research article DOI: 10.1038/s41467-026-73721-w
dendritic cell DC single-cell RNA-seq pan-cancer atlas tumour microenvironment antigen presentation cancer immunotherapy
Summary: Generates pan-cancer single-cell RNA-seq atlases of mouse and human tumour-associated dendritic cells (TADCs), encompassing 14 mouse tumour models and 10 human cancer types, to achieve a comprehensive mapping of DC subsets and states in the cancer context. Dendritic cells are the professional antigen-presenting cells of the immune system and are critical for initiating and sustaining anti-tumour T cell responses, yet their heterogeneity across tumour types has not been systematically characterized. The study identifies several lineage-defined DC subsets (cDC1, cDC2, pDC, moDC) along with maturation and functional states within each subset. Key findings include: TADCs acquire an inflammatory transcriptional profile with tumour progression; tumour-mediated reprogramming occurs not only in DCs within the tumour but also in DCs from tumour-draining lymph nodes, indicating systemic effects; and specific DC states correlate with T cell infiltration and checkpoint immunotherapy response across cancer types. The atlas provides a reference for understanding DC biology in cancer and identifies conserved and cancer-type-specific DC programs.
Why it matters: DCs are the gatekeepers of anti-tumour immunity — without functional DCs, T cells cannot be primed against tumour antigens, and checkpoint blockade cannot work. Yet DC biology in cancer has been relatively understudied compared to T cells and macrophages. This pan-cancer atlas fills a major gap by systematically characterizing DC heterogeneity across tumour types, providing both a reference map and testable hypotheses about which DC subsets and states are most important for anti-tumour immunity. The finding that tumour-draining lymph node DCs are also reprogrammed has clinical implications: DC-targeted therapies may need to address both intratumoural and lymphoid DCs.
Why for Yiru: DCs are the bridge between innate and adaptive immunity in the TME. Computationally, this atlas can be used as a reference for annotating DC subsets in new single-cell datasets, for deconvolving DC states from bulk transcriptomic data, and for identifying DC gene signatures associated with immunotherapy response. The cross-species comparison (mouse vs. human) is particularly valuable for evaluating the translational relevance of preclinical mouse models. More broadly, integrating this DC atlas with T cell and macrophage atlases could enable computational inference of the complete antigen presentation axis in the TME.
Systemic multi-omics analysis reveals interferon response heterogeneity and links lipid metabolism to immune alterations in severe COVID-19
Genome Medicine Published 2026-05-29 research article DOI: 10.1186/s13073-026-01677-z
interferon multi-omics lipid metabolism COVID-19 immune dysfunction transcriptomics metabolomics systems immunology
Summary: Performs a systemic multi-omics analysis of hospitalized COVID-19 patients to characterize interferon response heterogeneity and its immunometabolic consequences. While interferons play a central role in antiviral defense, their dysregulation contributes to inflammation and immune dysfunction in severe respiratory viral infections. The study finds that interferon-stimulated gene (ISG) expression is highly heterogeneous across patients and is not simply proportional to disease severity — some severely ill patients have paradoxically low ISG signatures. Through integrated transcriptomic, proteomic, and metabolomic profiling, the authors identify that this ISG heterogeneity is linked to distinct metabolic states, particularly alterations in lipid metabolism. Specific lipid species correlate with immune cell dysfunction, including impaired T cell and NK cell activity. The findings suggest that interferon response quality — not just quantity — determines clinical outcomes, and that the metabolic context in which interferon signaling occurs shapes its immunological effects.
Why it matters: This study addresses a clinically important paradox: why do some patients with severe viral infections have poor interferon responses while others have excessive inflammation? The multi-omics integration reveals that the answer lies at the intersection of interferon signaling and cellular metabolism — the same ISG can have different functional consequences depending on the metabolic state of the cell. This has therapeutic implications: interferon-based therapies may be more effective in patients with specific metabolic profiles, and metabolic interventions could modulate interferon response quality. While the study is in COVID-19, the principles likely apply to other settings of chronic interferon exposure including cancer and autoimmune disease.
Why for Yiru: Chronic interferon signaling is a hallmark of the TME — type I and type II interferons shape T cell exhaustion, macrophage polarization, and tumour cell immunogenicity. The finding that lipid metabolism modulates interferon response quality is directly relevant to the TME, where lipid availability and metabolism are highly dysregulated. Computational integration of TME transcriptomic and metabolomic data could reveal whether similar interferon–metabolism interactions shape anti-tumour immunity. The multi-omics integration framework used here could also serve as a template for analyzing TME multi-omics datasets.
Cross-disciplinary watchlist
Other Fields
Unstructured transcription factor interactions enable emergent specificity
Science Published 2026-05-28 research article DOI: 10.1126/science.aeb6487
transcription factor intrinsically disordered region IDR chromatin single-molecule gene regulation phase separation systems biology
Summary: Investigates how intrinsically disordered regions (IDRs) of transcription factors influence their chromatin binding and nuclear organization using proximity-assisted photoactivation (PAPA), a single-molecule protein–protein interaction sensor. A long-standing puzzle in gene regulation is how transcription factors achieve specific genomic targeting given that their DNA binding domains often recognize short, degenerate sequence motifs that occur millions of times in the genome. This study reveals that IDRs — long considered passive linkers between functional domains — actively reshape the TF binding landscape through weak, multivalent interactions. Focusing on Sp1, the authors find that the isolated DNA binding domain interacts poorly with chromatin and does not colocalize with full-length Sp1. However, the IDR enables Sp1 to form dynamic interaction networks with other TFs and chromatin-associated proteins, effectively creating an emergent binding specificity that is not encoded in the DNA recognition motif alone. The PAPA single-molecule sensor allows direct observation of these weak, transient interactions that are invisible to conventional methods like co-immunoprecipitation. The findings establish IDR-mediated interaction networks as a general mechanism for achieving regulatory specificity in mammalian genomes.
Why it matters: This work fundamentally reframes how we think about transcription factor specificity. The prevailing model — that TFs find their targets through DNA sequence recognition alone — is incomplete. IDRs, which constitute a large fraction of most TFs but have been mechanistically enigmatic, emerge as specificity determinants through their ability to form weak interaction networks. This explains why TFs with similar DNA binding domains can regulate completely different gene sets, and why mutations in IDRs are increasingly found in developmental disorders and cancer. The PAPA technique also represents a methodological advance for studying weak, transient protein interactions that are central to biological regulation but difficult to capture experimentally.
Why for Yiru: Transcription factors are key regulators of immune cell differentiation and TME remodeling — T cell exhaustion, macrophage polarization, and CAF activation are all driven by specific TF programs. The finding that IDRs determine TF specificity through interaction networks suggests that the same TF may have different target gene sets in different TME cell types depending on which interaction partners are available. This could be computationally modeled using single-cell multi-omics data that simultaneously measures TF expression, chromatin accessibility, and gene expression across TME cell types.
Selective autophagy fine-tunes plant immunity to promote cell survival during viral infection
Science Published 2026-05-28 research article DOI: 10.1126/science.adu9554
autophagy plant immunity viral infection EDS1 cell death Arabidopsis selective degradation cell biology
Summary: Reveals that Arabidopsis thaliana activates selective autophagy during viral infection not to degrade viral components — as might be expected — but to selectively remove the immune regulator Enhanced Disease Susceptibility 1 (EDS1) to prevent cell death. RNA viruses co-opt host endomembranes to form replication complexes, triggering cellular stress and immune responses. The canonical view is that autophagy degrades viral proteins or replication complexes to restrict infection. This study shows a surprising alternative: autophagy selectively targets EDS1, a central immune regulator, for degradation to prevent excessive immune activation that would otherwise kill the infected cell. The selective degradation is mediated by oligomeric metabolic enzymes that function as autophagy receptors, and is activated by viruses targeting mitochondria, chloroplasts, and the endoplasmic reticulum. By removing EDS1, autophagy dampens immunity just enough to allow infected cells to survive — a fine-tuning mechanism that balances pathogen defense against self-destruction.
Why it matters: This study inverts the standard narrative of autophagy in infection — it's not always about degrading the pathogen. The discovery that autophagy can selectively remove immune signaling components to prevent immunopathology reveals a previously unappreciated role for autophagy in calibrating immune response intensity. This has implications beyond plants: if similar mechanisms operate in mammalian cells, selective autophagy of immune regulators could be a general mechanism for preventing excessive inflammation during infection and possibly in sterile inflammatory conditions including cancer. The identification of metabolic enzymes as selective autophagy receptors is also notable — it connects metabolism directly to immune regulation through the autophagy machinery.
Why for Yiru: While plant immunity differs from mammalian immunity in mechanistic details, the principle — that selective degradation of immune regulators calibrates response intensity — is likely conserved. In the TME, chronic inflammatory signaling can drive both tumour-promoting inflammation and immune cell exhaustion. Selective autophagy of specific immune signaling components could represent a mechanism by which TME cells (both tumour and immune) modulate their inflammatory set point. Computational analysis of autophagy-related gene expression patterns across TME cell types could reveal whether similar calibration mechanisms operate in cancer.
A high-throughput selection system for fast-acting covalent protein drugs
Science Published 2026-05-28 research article DOI: 10.1126/science.adv3081
covalent inhibitor protein drug yeast display directed evolution high-throughput screening drug discovery protein engineering biotechnology
Summary: Develops a yeast display platform coupled with chemoselective modification that enables high-throughput selection of fast-acting covalent protein drugs. Covalent protein drugs — proteins that form irreversible bonds with their targets — offer therapeutic advantages including prolonged target engagement and the ability to inhibit challenging targets, but their development has been limited by slow reaction kinetics and the absence of high-throughput selection platforms. The challenge is multidimensional: rapid covalent binding requires coordinated optimization of binding affinity, protein stability, and warhead geometry, which cannot be addressed by screening these properties separately. The authors engineer a yeast display system in which covalent bond formation between the displayed protein and a target is coupled to a selectable phenotype, enabling FACS-based screening of millions of variants. Critically, the system selects for fast binding kinetics without increasing intrinsic warhead reactivity — which would cause off-target labeling. Using this platform, they evolve covalent proteins with dramatically improved binding rates against clinically relevant targets.
Why it matters: Covalent drugs — from aspirin to ibrutinib to the COVID-19 antiviral Paxlovid — have been among the most successful therapeutics, but the covalent drug discovery process has been largely empirical. This platform brings directed evolution and high-throughput screening to covalent protein drug development, potentially accelerating the discovery of covalent biologics for targets that have been undruggable by conventional approaches. The ability to select for fast kinetics without increasing promiscuous reactivity addresses the central safety concern of covalent drugs. As the line between small molecules and biologics continues to blur, platforms that can engineer covalent proteins with drug-like properties will become increasingly valuable.
Why for Yiru: Covalent protein drugs could target extracellular TME components with exceptional durability — for example, a covalent protein that irreversibly blocks an immunosuppressive cytokine or checkpoint ligand in the TME could provide sustained relief from immune suppression without requiring continuous dosing. The yeast display platform could theoretically be adapted to evolve covalent proteins that selectively react with TME-specific post-translational modifications or proteolytic fragments, adding a layer of tumour selectivity.
Optically detected and radio wave-controlled spin chemistry in flavoproteins
Nature Biotechnology Published 2026-05-29 research article DOI: 10.1038/s41587-026-03158-5
spin chemistry flavoprotein quantum sensing radio wave cryptochrome magnetoreception biophysics optically addressable
Summary: Demonstrates that photogenerated spin-correlated radical pairs in certain flavoproteins — specifically cryptochrome and improved light-oxygen-voltage (iLOV) protein — can be manipulated by radio waves, enabling magnetic field sensing and spatial modulation of photoluminescence. Optically addressable spin systems, such as nitrogen-vacancy centers in diamond, have been widely studied for quantum sensing applications, but biological systems have not been considered viable platforms. This study shows that flavoproteins, which are ubiquitous in nature and already function as magnetoreceptors in migratory birds (cryptochromes), can serve as optically addressable spin systems. Radiofrequency pulses and magnetic field gradients can manipulate the spin state of photogenerated radical pairs, producing detectable changes in fluorescence. This establishes proteins as a new platform for optically addressable quantum sensing, with potential advantages over solid-state systems including genetic encodability, room-temperature operation, and the ability to function in aqueous cellular environments.
Why it matters: This is a conceptual breakthrough at the interface of quantum physics and biology. The demonstration that proteins can function as quantum sensors opens an entirely new direction for biological sensing and imaging. Genetically encoded quantum sensors could be expressed in specific cell types or subcellular compartments, enabling quantum-enhanced measurements of magnetic fields, temperature, and other physical parameters in living systems. The connection to magnetoreception in migratory birds is also fascinating — it suggests that nature may already use quantum sensing principles for navigation, and that these principles can be harnessed for engineered applications.
Why for Yiru: While this is fundamental biophysics rather than TME biology, optically addressable quantum sensors in proteins could eventually enable new modalities for studying the TME. Genetically encoded magnetic field sensors could report on iron accumulation in tumour-associated macrophages, oxygen radical production during immune activation, or metabolic activity through spin-labeled metabolites. More broadly, the study exemplifies how fundamental biophysical discoveries can generate unexpected tools for biological investigation.