Research Radar — 2026-06-20
Methods & AI
Computational
Perturbation Curve models continuous transcriptional response trajectories and improves prediction of genetic modulations
bioRxiv Published 2026-06-19 preprint DOI: 10.64898/2026.06.16.732192
Perturb-seq single-cell CRISPR screen perturbation modeling transcriptional trajectory functional genomics computational method gene regulation
Summary: Presents Perturbation Curve (PertCurve), a nonlinear, curve-based computational framework that models continuous transcriptional response trajectories from single-cell CRISPR screens by explicitly incorporating diverse perturbation magnitudes and strengths. Single-cell CRISPR screens such as Perturb-seq have transformed functional genomics by linking genetic perturbations to transcriptomic phenotypes at single-cell resolution. However, current analytical frameworks treat perturbation assignments as discrete labels — a gene is either knocked out or not — ignoring the biological reality that perturbation strength varies continuously across cells due to differences in guide efficiency, copy number, and cellular context. This conflation of variable perturbation strength with diverse downstream responses obscures dose-response relationships and reduces power to detect subtle transcriptional effects. PertCurve addresses this by ordering cells along a continuous perturbation-strength axis and fitting nonlinear response curves for each gene, enabling the deconvolution of perturbation magnitude from response diversity. Applied to published Perturb-seq datasets, PertCurve recapitulates known dose-response relationships, identifies genes with graded versus switch-like responses to perturbation strength, and reveals that many transcriptional responses are better modelled as continuous functions of perturbation dose rather than discrete on/off states. The framework also improves the prediction of transcriptional outcomes for combinatorial perturbations by modelling how response curves compose across multiple genetic modulations.
Why it matters: The shift from discrete to continuous perturbation modelling addresses a fundamental limitation in how single-cell CRISPR screen data are analysed. Perturbation efficiency varies substantially across cells in any pooled screen — some cells receive multiple guides, some receive none, and guide activity varies — yet prevailing analysis methods collapse this rich quantitative variation into binary labels. PertCurve demonstrates that modelling perturbation as a continuous variable reveals dose-response relationships that are invisible to discrete methods, analogous to how dose-response curves in pharmacology reveal drug mechanisms that binary treated/untreated comparisons miss. This is particularly important as Perturb-seq and related technologies scale to genome-wide screens: the ability to extract quantitative perturbation-response relationships from each experiment dramatically increases the information yield per cell and per dollar.
Why for Yiru: Perturbation-response modelling is central to TME functional genomics. Understanding how genetic perturbations — knockout of checkpoint molecules, chemokine receptors, or metabolic enzymes — reshape the TME requires knowing not just whether a gene is perturbed but how strongly. PertCurve could be applied to TME Perturb-seq data to model how graded changes in immune checkpoint expression or cytokine signalling affect T cell activation states, macrophage polarization, or tumour cell immunogenicity. The continuous framework is also valuable for modelling pharmacological perturbations in the TME, where drug concentration varies spatially across the tumour. More broadly, the curve-based approach could be extended to model continuous environmental gradients in the TME — oxygen tension, nutrient availability, cytokine concentration — that shape immune cell states along continuous trajectories rather than discrete categories.
SteerAF — Distogram-based Steering of AlphaFold2 toward Alternative Conformations
bioRxiv Published 2026-06-19 preprint DOI: 10.64898/2026.06.19.733296
AlphaFold2 protein conformation alternative conformations distogram deep learning structural biology computational method
Summary: Introduces SteerAF, an inference-time optimization framework that steers AlphaFold2 toward predicting alternative protein conformations by leveraging information encoded in the distogram derived from deep multiple sequence alignments, without requiring model retraining. End-to-end structure predictors such as AlphaFold2 have revolutionized structural biology by predicting protein structures with near-experimental accuracy, but they typically output only the dominant conformational state — the structure most represented in the training data. Many proteins function through conformational changes (open/closed states in enzymes, active/inactive states in receptors, folded/unfolded states in intrinsically disordered regions), and predicting these alternative states is essential for understanding mechanism and designing drugs. Existing approaches for recovering alternative conformations — such as subsampling MSAs, running molecular dynamics, or training specialized models — are computationally expensive and offer limited interpretability. SteerAF takes a different approach: it operates at inference time on a pre-trained AlphaFold2 model, using gradient-based optimization to modify sparse features derived from the MSA distogram — a matrix encoding pairwise distance predictions between residues. By perturbing these distogram features in directions that correspond to specific conformational changes, SteerAF guides AlphaFold2 to predict alternative states. Across four benchmark datasets covering enzymes, transporters, and signalling proteins, SteerAF matches or exceeds existing methods while being orders of magnitude faster and providing interpretable feature-perturbation maps that reveal which residue-residue contacts drive conformational transitions.
Why it matters: The ability to predict alternative protein conformations from sequence alone would transform drug discovery, enzyme design, and our understanding of protein allostery. Many drug targets — kinases, GPCRs, ion channels — exist in multiple conformational states, and drugs often bind preferentially to specific states. SteerAF's inference-time approach is particularly attractive because it requires no retraining of the underlying AlphaFold2 model, making it immediately applicable as structure prediction models improve. The interpretability of the distogram perturbations — showing which specific residue contacts drive conformational changes — provides mechanistic insight that purely predictive methods lack. This work also demonstrates that the distogram, typically treated as an intermediate feature, encodes rich conformational information that can be mined for biological discovery.
Why for Yiru: Protein conformational dynamics are central to immune recognition in the TME. Immune checkpoint receptors (PD-1, CTLA-4), their ligands (PD-L1), and cytokine receptors all undergo conformational changes that regulate binding affinity and signalling. SteerAF could be used to predict alternative conformations of these immune receptors, potentially identifying cryptic binding sites for therapeutic antibodies or revealing how tumour mutations alter conformational dynamics to evade immune recognition. More broadly, the distogram-steering approach could be extended to predict how post-translational modifications (phosphorylation, ubiquitination) — which are abundant in TME signalling networks — alter protein conformation and function. The computational efficiency of inference-time steering also makes it practical for screening large numbers of mutations or modifications.
StickForStats — automated statistical assumption validation for reproducible computational biology
bioRxiv Published 2026-06-19 preprint DOI: 10.64898/2026.06.15.732278
statistical validation reproducibility computational biology software tool assumption checking open science web platform
Summary: Presents StickForStats, an open-source web platform that automates statistical assumption validation as a default precondition for computational biology analyses, addressing the widespread problem of unreported assumption violations. Reproducible computational biology depends on statistical decisions that routine workflows often skip: verifying that a differential-expression test's assumptions hold across all genes, that an ANOVA comparing analysis strategies is robust to non-normality, or that a meta-analysis is not distorted by publication bias. Surveys consistently find that fewer than 20% of published biomedical studies report checking these assumptions, and existing statistical software leaves validation to the analyst as an optional, often-ignored step. StickForStats reframes assumption validation as a default precondition through its Guardian system — a middleware pipeline of eight validators covering normality, variance homogeneity, independence, outliers, sample size adequacy, modality, linearity, and multicollinearity. The platform operates as a web application where users upload their data and specify their intended analysis; the Guardian system automatically runs all relevant validators, flags violations with specific recommendations (transformations, non-parametric alternatives, robust methods), and generates a validation report that can be included in publications. The system is designed to integrate with common computational biology workflows and provides programmatic API access for incorporation into automated pipelines.
Why it matters: Statistical assumption violations are a silent contributor to the reproducibility crisis in computational biology. When a t-test is applied to non-normal data, a linear regression is fitted to nonlinear relationships, or an ANOVA is run on heteroscedastic groups, the resulting p-values and effect sizes can be misleading — yet these violations are rarely checked or reported. StickForStats addresses this by making assumption validation automatic and mandatory rather than optional and ignored. By generating a standardized validation report, it also improves transparency: reviewers and readers can see exactly which assumptions were checked and whether any were violated. If widely adopted, tools like StickForStats could substantially improve the statistical rigour of computational biology, much as automated code testing improved software reliability.
Why for Yiru: TME computational analyses routinely apply a battery of statistical methods — differential expression testing across cell types, survival analyses stratified by immune infiltration scores, correlation analyses between TME features and clinical outcomes — each of which carries assumptions that are rarely explicitly validated. StickForStats could be integrated into TME analysis pipelines to automatically verify that the statistical methods applied to single-cell and spatial transcriptomics data are appropriate for the data distributions at hand. For example, many single-cell differential expression tools assume specific distributions (negative binomial, zero-inflated) that may not hold across all genes or cell types in TME data; automated validation would flag genes or comparisons where assumptions are violated. The programmatic API also enables incorporation into reproducible workflow systems like Nextflow or Snakemake.
segSHAPE — RNA secondary structure prediction from nanopore direct RNA sequencing
bioRxiv Published 2026-06-18 preprint DOI: 10.64898/2026.06.15.732177
RNA structure nanopore sequencing direct RNA sequencing computational method chemical probing single-molecule bioinformatics
Summary: Introduces segSHAPE, a probe-agnostic framework for RNA secondary structure prediction from nanopore direct RNA sequencing (DRS) data that improves signal alignment and modification-rate estimation over existing tools. Chemical probing coupled with nanopore DRS offers an attractive route to single-molecule RNA structural inference because the same read provides both sequence identity and modification information. However, current tools are limited by inaccurate signal-to-sequence alignment — the process of mapping raw nanopore current signals to nucleotide positions — which degrades modification-rate estimation and downstream structure prediction. segSHAPE addresses this through three innovations: improved signal alignment using prior information from basecalling and per-read signal baseline shift correction; learning of position-specific k-mer raw signal parameters to account for sequence context effects on nanopore signals; and estimation of per-nucleotide modification rates using an unsupervised anomaly detection approach that does not require a modification-free control sample. The method is validated on both RNA002 and RNA004 nanopore chemistries and across multiple chemical probing strategies (SHAPE, DMS, CMCT), demonstrating improved accuracy in modification detection and secondary structure prediction compared to existing tools. The probe-agnostic design means segSHAPE can be applied to data from any chemical probing experiment without retraining.
Why it matters: RNA structure is increasingly recognized as a critical layer of gene regulation — affecting splicing, translation, localization, and degradation — yet experimental determination of RNA structures at scale remains challenging. Nanopore DRS with chemical probing offers a path to high-throughput, single-molecule structure determination, but its utility has been limited by computational tools that introduce errors at the signal-alignment stage. segSHAPE addresses this bottleneck with a principled, probe-agnostic approach that should improve structure prediction accuracy across diverse experimental protocols. The ability to work without a modification-free control sample is practically important because it reduces experimental complexity and cost. As nanopore DRS becomes more widely adopted for transcriptomics, methods that extract structural information from the same reads — effectively getting two modalities (expression and structure) for the price of one — will become increasingly valuable.
Why for Yiru: RNA structure and RNA-binding proteins are emerging as important regulators in the TME. RNA structural elements control the translation of key immune modulators (cytokines, checkpoint molecules), and RNA-binding proteins that recognize specific structural motifs can alter immune cell function. Nanopore DRS could be applied to T cells or tumour cells from the TME to simultaneously profile gene expression and RNA structure, revealing how structural changes contribute to immune activation or exhaustion. segSHAPE's probe-agnostic framework is valuable for TME studies where multiple probing strategies might be needed to capture different structural features across diverse RNA species. The computational improvements in signal alignment also benefit any TME study using nanopore DRS for transcript-level analyses.
Leveraging longitudinal data to boost statistical power for gene–environment interaction analysis (SAGELD)
Nature Computational Science Published 2026-06-16 research article DOI: 10.1038/s43588-026-01002-z
gene-environment interaction longitudinal data statistical power GWAS biobank computational method epidemiology
Summary: Introduces SAGELD, a statistical method that leverages longitudinal data to detect gene-environment interactions with substantially greater statistical power than conventional cross-sectional approaches, enabling new discoveries in large biobank studies. Gene-environment (G×E) interactions — where a genetic variant's effect on a trait depends on an environmental exposure — are thought to be widespread but have been notoriously difficult to detect in genome-wide studies due to limited statistical power. Conventional G×E methods use cross-sectional data (one measurement per person) and require extremely large sample sizes to achieve adequate power, particularly when testing millions of genetic variants. SAGELD exploits the fact that many biobanks now contain longitudinal measurements — repeated assessments of the same trait in the same individuals over time. By modelling how genetic effects change as a function of time-varying environmental exposures within individuals, SAGELD extracts more information from the same number of participants. The method is evaluated through extensive simulations and applied to UK Biobank data, where it identifies novel G×E interactions — including genetic variants whose effects on blood pressure and metabolic traits are modified by age, medication use, and lifestyle factors — that were undetectable with cross-sectional methods. The framework is computationally efficient and scales to biobank-sized datasets with millions of variants and hundreds of thousands of participants.
Why it matters: The detection of gene-environment interactions has been a persistent challenge in human genetics, with many researchers suspecting that G×E effects are important but acknowledging that current methods lack the power to find them. SAGELD addresses this by exploiting an underutilized feature of modern biobanks: longitudinal data. As biobanks continue to accumulate repeated measurements over decades of follow-up, methods like SAGELD that can harness this temporal dimension will become increasingly powerful. The computational efficiency is also critical — a method that requires prohibitive computation for biobank-scale data is not practically useful regardless of its statistical properties. By enabling the discovery of G×E interactions at scale, SAGELD could identify genetic variants that modulate environmental risk factors for common diseases, with implications for personalized risk prediction and targeted prevention.
Why for Yiru: While SAGELD is developed in the context of human genetics and biobanks, the statistical principle — leveraging repeated measurements to detect interactions — is broadly applicable. In TME research, longitudinal data are becoming available through sequential biopsies, liquid biopsies during treatment, and time-course experiments in model systems. A method analogous to SAGELD could be used to detect gene-treatment interactions in clinical trials: identifying genetic variants (germline or somatic) that modify how the TME responds to immunotherapy over time. More generally, the framework demonstrates how to extract interaction signals from longitudinal data, a principle that could be adapted to single-cell time-course experiments tracking how genetic perturbations interact with temporal environmental changes in the TME.
Biomedical discoveries
Biomedicine
Killer-cell dominance dichotomy governs tumor immune networks and stratifies inflamed cancers
bioRxiv Published 2026-06-19 preprint DOI: 10.64898/2026.06.15.732326
tumour microenvironment killer cell CD8 T cell NK cell immune network pan-cancer scRNA-seq spatial transcriptomics
Summary: Uncovers a conserved framework in which terminal cytotoxic immunity in tumours diverges into mutually exclusive states dominated by either exhausted CD8+ T cells (Tex) or CD56dimCD16hi NK (NK1) cells, redefining the prevailing hot-cold tumour paradigm. Cancer immunotherapy has been most successful in "hot" tumours with high lymphocyte infiltration, yet even among hot tumours, response rates vary dramatically. The prevailing model assumes that hot tumours are characterized by coordinated infiltration of multiple cytotoxic lineages working together. Analysing nearly 5,000 pan-cancer single-cell RNA-seq samples with orthogonal validation by spectral cytometry and spatial transcriptomics, the authors discovered that this assumption is incorrect: at the population level, terminal cytotoxic immunity in individuals and malignancies consistently diverges into states dominated by either exhausted CD8+ T cells or cytotoxic NK cells, with few tumours showing high levels of both simultaneously. This Tex-NK1 dominance axis governs the primary dimension of tumour-intrinsic and tumour-extrinsic variance and is shaped by distinct tumour molecular features, microbiome signatures, and clinical metadata. The dichotomy has therapeutic implications: Tex-dominant tumours are enriched for immune checkpoint expression and MHC-I presentation, suggesting sensitivity to checkpoint blockade, while NK1-dominant tumours are enriched for stress-ligand expression and MHC-I loss, suggesting sensitivity to NK-based therapies. The authors propose a revised tumour classification framework that stratifies inflamed cancers based on their killer-cell dominance state rather than simply the quantity of infiltrating lymphocytes.
Why it matters: This study fundamentally revises our understanding of anti-tumour immunity by showing that the simple hot-cold paradigm — which has guided immunotherapy research and clinical trial design for over a decade — misses a critical axis of variation. The finding that terminal cytotoxicity is dominated by either T cells or NK cells, but rarely both, has immediate therapeutic implications: it suggests that some "hot" tumours fail checkpoint blockade because they are NK-dominant and lack the T cell targets that checkpoint inhibitors act upon, while others may respond poorly to NK-based therapies because they are Tex-dominant and have downregulated NK-activating ligands. The pan-cancer scale and multi-omic validation make this a robust finding likely to influence how clinical trials stratify patients and how combination immunotherapies are designed.
Why for Yiru: This work is directly relevant to TME computational research because it provides a new, data-driven framework for classifying the immune contexture of tumours. The Tex-NK1 dominance axis could be used as a computational phenotype to stratify TME samples in any single-cell or spatial transcriptomics study, replacing the simple hot-cold classification. The finding that this axis is shaped by tumour molecular features opens opportunities for computational modelling: predicting a tumour's killer-cell dominance state from its mutational profile, gene expression, or histopathology could guide therapy selection. The spatial transcriptomics validation also suggests that the Tex-NK1 axis may manifest spatially — with NK-dominant and Tex-dominant regions potentially coexisting within the same tumour — an important consideration for spatial TME analysis. The methodology — identifying a dominant axis of variation from large-scale multi-omic data — provides a template for discovering other organizing principles of the TME.
CD4+ T cells impair tumor growth through IL-3 and TNF-dependent vascular damage
Science Published 2026-06-18 research article DOI: 10.1126/science.ads7910
CD4 T cell tumour vasculature IL-3 TNF myeloid cell vascular damage tumour stroma immunotherapy
Summary: Identifies a tumour stroma-targeting mechanism in which tumour antigen-specific CD4+ T cells inhibit tumour growth not through direct cytotoxicity but through myeloid cell- and TNF-dependent vascular damage triggered by T cell-derived interleukin-3 (IL-3). Most cancer immunotherapy strategies focus on CD8+ cytotoxic T lymphocytes that directly kill tumour cells, with CD4+ T cells viewed primarily as helpers that support CD8+ responses. Using multiplex immunofluorescence and single-cell and tissue transcriptomics in mouse tumour models, the authors discovered that CD4+ T cells can exert potent anti-tumour effects through a fundamentally different mechanism. Upon recognizing tumour antigens, CD4+ T cells produce IL-3, which acts on tumour-infiltrating myeloid cells to drive their differentiation into classically activated macrophages that cluster around tumour blood vessels. These perivascular macrophage clusters produce high local concentrations of TNF, which directly damages tumour endothelial cells, disrupts blood supply, and causes ischemic tumour cell death. This vascular damage mechanism is spatially restricted to the tumour, sparing normal tissue vasculature. The effect is independent of CD8+ T cells and direct cytotoxicity, revealing a parallel anti-tumour axis mediated by CD4+ T cells that targets the tumour stroma rather than tumour cells themselves. The IL-3-TNF-vascular damage pathway represents a therapeutically tractable mechanism that could be harnessed independently of or in combination with existing immunotherapies.
Why it matters: This study expands the repertoire of anti-tumour immune mechanisms beyond direct cytotoxicity, showing that CD4+ T cells can starve tumours by destroying their blood supply. This is conceptually important because it suggests that tumours can be attacked through their stromal infrastructure rather than exclusively through direct tumour cell killing — analogous to besieging a city by cutting its supply lines rather than storming its walls. The identification of a specific molecular pathway (IL-3 → myeloid cells → TNF → endothelial damage) provides clear therapeutic targets: IL-3 or TNF delivery to tumours, engineering of CD4+ T cells to enhance IL-3 production, or combination strategies that simultaneously engage CD8+ direct killing and CD4+ vascular damage. The tumour specificity of the vascular damage is also practically important, as systemic TNF delivery is prohibitively toxic.
Why for Yiru: This work reveals that the TME is not just a battleground for direct tumour-immune cell combat but also a target for stromal disruption. For computational TME research, this suggests that TME characterization should include features of vascular integrity and perivascular myeloid organization — not just immune cell counts and activation states. Spatial transcriptomics data could be analysed specifically for signatures of IL-3 signalling, TNF production, and endothelial damage, identifying tumours where this mechanism is naturally active or could be therapeutically induced. The myeloid-T cell-endothelial axis also represents a multicellular interaction network that could be computationally modelled to predict conditions under which vascular damage is maximized while sparing normal tissue. More broadly, this study exemplifies how single-cell and spatial data can reveal unexpected multicellular mechanisms of immune action.
The atypical IκB factor IκBδ enhances CD8 T cell accumulation and effector functions in solid tumors
bioRxiv Published 2026-06-19 preprint DOI: 10.64898/2026.06.17.732005
IκBδ Nfkbid CD8 T cell tumour-infiltrating lymphocyte NF-κB T cell exhaustion immunotherapy solid tumour
Summary: Identifies IκBδ (encoded by Nfkbid), a poorly characterized member of the IκB family of NF-κB regulators, as a molecular lever that simultaneously overcomes two major barriers to anti-tumour CD8+ T cell immunity: constrained T cell accumulation in tumours and restrained cytotoxic effector function. Tumours defeat anti-tumour immunity through two prominent mechanisms — limiting the ability of T cells and CAR T cells to survive and expand within the tumour microenvironment, and suppressing their capacity to sustain full cytotoxic activity. The authors found that Nfkbid is an NFAT target gene expressed in CD8+ effector T cells and at modest levels in tumour-infiltrating lymphocytes (TILs). Nfkbid depletion impaired TIL accumulation and exacerbated solid tumour growth, while ectopic IκBδ overexpression enhanced TIL expansion, reduced expression of exhaustion-associated transcription factors (TOX, NR4A1) and inhibitory receptors (PD-1, TIM-3, LAG-3), and increased production of effector cytokines (IFN-γ, TNF, granzyme B). Mechanistically, IκBδ functions as an atypical IκB that, unlike classical IκBs, does not simply inhibit NF-κB but instead modulates the dynamics of NF-κB signalling — promoting sustained, oscillatory NF-κB activity that supports T cell effector programs while suppressing the chronic NF-κB signalling associated with exhaustion. The discovery that a single factor can simultaneously enhance T cell accumulation and effector function while suppressing exhaustion makes IκBδ an attractive target for engineering more effective T cell therapies.
Why it matters: Current strategies for improving T cell therapy — checkpoint blockade, cytokine support, metabolic reprogramming — typically address only one aspect of T cell dysfunction at a time. IκBδ is remarkable because it appears to act as a coordinated molecular switch that simultaneously promotes accumulation, effector function, and resistance to exhaustion. If IκBδ overexpression can be translated to CAR T cells or TCR-engineered T cells, it could produce T cell products that are both more abundant in tumours and more functionally potent — a combination that has been difficult to achieve. The mechanistic link to NF-κB dynamics is also significant because it suggests that the qualitative pattern of NF-κB signalling (oscillatory vs. sustained) — not just its magnitude — determines T cell fate, a concept that could inform the design of synthetic NF-κB signalling circuits in engineered T cells.
Why for Yiru: T cell exhaustion in the TME is a central barrier to effective immunotherapy, and the identification of factors that regulate the balance between effector function and exhaustion is directly relevant to TME research. IκBδ expression or activity could be used as a biomarker to identify TILs that are poised for effective anti-tumour responses versus those destined for exhaustion. Computational analysis of single-cell TME data could examine the relationship between NF-κB signalling dynamics (inferred from target gene expression patterns) and T cell functional states, testing whether the oscillatory-vs- sustained NF-κB model applies across different TME contexts. The IκBδ pathway also connects to NFAT signalling, which is central to T cell activation and is regulated by calcium signalling — a pathway that can be modulated by the ionic microenvironment of tumours.
Azacytidine restores T cell function in AML by modulating DNA methylation
bioRxiv Published 2026-06-17 preprint DOI: 10.64898/2026.06.14.732148
AML azacytidine DNA methylation T cell exhaustion immunotherapy epigenetics leukemia hypomethylating agent
Summary: Demonstrates that azacytidine, an FDA-approved hypomethylating agent, restores T cell function in acute myeloid leukemia (AML) by reversing DNA methylation-driven T cell exhaustion, providing a mechanistic rationale for combining epigenetic therapy with immunotherapy in AML. AML is an aggressive blood cancer with poor outcomes, and while chemotherapy remains standard, relapse is common and immunotherapy has had limited success. T cell dysfunction and exhaustion are prominent features of AML but remain poorly characterized compared to solid tumours. DNA methylation is a major driver of T cell exhaustion — de novo DNA methyltransferases (DNMT3A/B) establish methylation patterns that silence effector gene loci and enforce exhaustion programs — and inhibition of DNA methylation can block exhaustion and restore T cell function in chronic viral infections and some solid tumour models. Using a spontaneous AML mouse model and primary samples from AML patients, the authors show that azacytidine treatment reduces DNA methylation at key effector loci in T cells, reverses exhaustion-associated chromatin states, and restores T cell proliferation and cytokine production. The effects are most pronounced in bone marrow-resident T cells, which are in direct contact with leukemic cells and experience the most severe exhaustion. Azacytidine also synergizes with checkpoint blockade in the AML model, suggesting a combination strategy in which epigenetic reprogramming of T cells sensitizes them to checkpoint inhibitor-mediated reactivation.
Why it matters: The limited success of checkpoint blockade in AML, despite evidence of T cell infiltration, has been puzzling. This study provides a mechanistic explanation: AML-associated T cell exhaustion is driven by epigenetic silencing that checkpoint blockade alone cannot reverse — the T cells need to be epigenetically "unlocked" before they can respond to checkpoint inhibitors. Azacytidine is already FDA-approved for myelodysplastic syndromes and AML, making this a readily translatable combination. The concept of epigenetic pre-conditioning of T cells before immunotherapy — using hypomethylating agents to reverse exhaustion-associated methylation patterns — may also apply beyond AML to other cancers where T cell exhaustion has a strong epigenetic component, including solid tumours with chronic antigen exposure.
Why for Yiru: Epigenetic regulation of T cell exhaustion is highly relevant to the TME, where chronic antigen stimulation and immunosuppressive signals drive progressive epigenetic silencing of effector gene loci. The finding that hypomethylating agents can reverse this silencing suggests that TME immunosuppression may be therapeutically tractable through epigenetic interventions, not just through checkpoint blockade or cytokine support. Computational methods could be developed to infer T cell exhaustion-associated methylation states from single-cell transcriptomic or ATAC-seq data from the TME, identifying patients whose T cells are epigenetically "locked" in exhaustion and might benefit from hypomethylating agent pre-treatment. The bone marrow TME in AML also provides an interesting contrast to solid tumour TMEs, and comparative analysis could reveal which features of T cell exhaustion are shared across haematological and solid malignancies.
Cell Cycle Sensing Shapes Human T Cell Fate and Exhaustion Programs
bioRxiv Published 2026-06-12 preprint DOI: 10.64898/2026.06.11.731737
T cell cell cycle exhaustion differentiation mass cytometry single-cell human immunology
Summary: Uses high-throughput single-cell mass cytometry with parallel measurement of cell cycle state, receptor signalling, and differentiation markers to disentangle how cell cycle dynamics shape human T cell fate decisions and exhaustion programs. The relationship between cell division and T cell differentiation has been recognized since the earliest studies of lymphocyte activation — T cells must divide to acquire effector functions, but excessive division drives terminal differentiation and exhaustion — yet the precise connections between cell cycle progression, signalling, and fate remain poorly understood. The authors leverage mass cytometry to simultaneously measure cell cycle markers (Ki-67, cyclins, phospho-Rb), TCR signalling (phospho-CD3ζ, phospho-ZAP70, phospho-ERK), and differentiation markers (transcription factors, surface receptors) in primary human T cells. By systematically modulating cell cycle progression with inhibitors and varying TCR signal strength, they identify specific cell cycle phases that are permissive or restrictive for particular fate decisions. A key finding is that the G1/S transition acts as a critical checkpoint: T cells that progress rapidly through G1 into S phase are biased toward effector differentiation, while those with prolonged G1 are biased toward memory or exhaustion fates, depending on the signalling context. In exhausted T cells from chronic stimulation models, cell cycle progression is slowed but not arrested — creating a state of "frustrated cycling" that reinforces exhaustion-associated transcriptional programs.
Why it matters: Despite decades of research on T cell differentiation, the role of cell cycle dynamics — as distinct from the simple fact of division — has been underappreciated. This study provides a systematic, quantitative framework linking specific cell cycle phases to fate outcomes, with practical implications for T cell manufacturing. If the G1/S transition is a fate-determination checkpoint, then engineering T cells to modulate G1 duration — through cytokine cocktails, genetic modifications, or small molecules — could bias differentiation toward desired fates (effector, memory, or stem-like) and away from exhaustion. The concept of "frustrated cycling" in exhausted T cells is also provocative: it suggests that exhaustion is not simply a failure to divide but an actively maintained state of dysregulated cell cycle progression, which might be targetable by cell cycle-modulating drugs.
Why for Yiru: T cell fate in the TME — whether infiltrating T cells become functional effectors, exhausted, or memory-like — is a central determinant of immunotherapy outcomes. The finding that cell cycle dynamics at the G1/S transition influence these fate decisions suggests that TME factors that affect T cell proliferation (IL-2 availability, nutrient competition, checkpoint signalling) may shape anti-tumour immunity partly through their effects on cell cycle progression. Computational analysis of TME single-cell data could incorporate cell cycle phase as a covariate when modelling T cell differentiation trajectories, potentially revealing fate biases that are obscured when cell cycle is treated as a confounder. The mass cytometry approach also provides a template for high-dimensional profiling of T cell states in TME samples where the number of available cells is limited.
Cross-disciplinary watchlist
Other Fields
Runx–CBFβ regulates the development of tolerogenic Thetis cells
Nature Immunology Published 2026-06-19 research article DOI: 10.1038/s41590-026-02566-8
Runx CBFβ Thetis cells RORγt antigen-presenting cell immune tolerance gut immunology intestinal immunity
Summary: Reports that the Runx–CBFβ transcription factor complex plays a crucial role in the development of intestinal RORγt+ antigen-presenting cells known as Thetis cells, which are required to promote immune tolerance to food antigens and gut commensal microorganisms. The intestinal immune system faces a unique challenge: it must mount protective responses against pathogens while maintaining tolerance to the enormous load of dietary antigens and commensal bacteria. Thetis cells are a recently identified subset of RORγt+ antigen-presenting cells in the intestinal lamina propria that are distinct from conventional dendritic cells and are implicated in inducing regulatory T cells (Tregs) specific for oral antigens. The authors show that genetic deletion of Runx transcription factors or their obligate cofactor CBFβ specifically in RORγt+ cells leads to a profound loss of Thetis cells in the small intestine, accompanied by impaired generation of intestinal Tregs and breakdown of oral tolerance. Mechanistically, Runx–CBFβ directly activates Thetis cell signature genes and maintains the chromatin landscape required for their identity. The study establishes Runx–CBFβ as a lineage-defining transcription factor complex for these tolerogenic cells, connecting transcriptional regulation of antigen-presenting cell development to the maintenance of intestinal immune homeostasis.
Why it matters: The discovery of Thetis cells and their transcriptional regulators fills an important gap in our understanding of how oral tolerance is maintained. While Foxp3+ Tregs are recognized as the primary mediators of tolerance, the antigen-presenting cells that induce them in the gut have been less well defined. The identification of Runx–CBFβ as a master regulator of Thetis cell development provides a genetic entry point for studying these cells and their role in diseases of intestinal immune dysregulation — food allergy, inflammatory bowel disease, and celiac disease — where oral tolerance breaks down. Understanding how tolerogenic antigen-presenting cells are generated could inform strategies to therapeutically induce or restore oral tolerance.
Why for Yiru: The gut immune system and the TME share interesting parallels: both are sites where the immune system must navigate the balance between tolerance and immunity in the presence of a high antigenic load. The transcriptional circuits that drive tolerogenic antigen-presenting cell development in the gut — including Runx–CBFβ — may have counterparts in the TME, where tumour-associated dendritic cells and macrophages often adopt tolerogenic phenotypes that suppress anti-tumour immunity. Understanding how Thetis cells induce Tregs could reveal mechanisms that are co-opted by tumours to generate immunosuppressive Tregs in the TME. Computational comparison of the transcriptional programs of gut tolerogenic APCs and tumour-associated APCs could identify shared tolerance-promoting circuits.
mRNA-based influenza vaccine expands the B cell response breadth in humans
Nature Immunology Published 2026-06-15 research article DOI: 10.1038/s41590-026-02569-5
mRNA vaccine influenza B cell germinal center antibody memory B cell vaccinology human immunology
Summary: Compares the humoral immune response of individuals receiving conventional inactivated influenza vaccines to those receiving an mRNA-based quadrivalent influenza vaccine, demonstrating that the mRNA platform induces more robust and durable B cell responses. While mRNA vaccines have transformed the response to COVID-19, their application to influenza — a virus for which seasonal vaccines already exist — provides a unique opportunity to compare vaccine platforms head-to-head for the same pathogen. The authors conducted a detailed analysis of B cell responses in individuals receiving either the standard inactivated influenza vaccine or an mRNA-based quadrivalent vaccine encoding the same four influenza strains. Using lymph node fine-needle aspiration to sample germinal centers — the sites where B cells undergo affinity maturation — they found that the mRNA vaccine induced prolonged germinal center responses lasting months rather than weeks, generated higher frequencies of influenza-specific memory B cells, and produced antibodies with greater breadth of viral strain recognition. The enhanced germinal center response was associated with sustained antigen expression from the mRNA, which provided continued B cell receptor stimulation, in contrast to the bolus of pre-formed antigen delivered by inactivated vaccines. The mRNA platform also induced stronger T follicular helper cell responses, which are essential for germinal center maintenance and affinity maturation.
Why it matters: Seasonal influenza vaccines have modest and variable effectiveness, in part because they elicit narrow antibody responses that are easily evaded by viral mutation. This head-to-head comparison provides strong evidence that the mRNA platform can overcome this limitation by driving more prolonged and higher-quality B cell responses. The mechanistic insight — that sustained antigen expression from mRNA prolongs germinal center reactions — explains why mRNA vaccines generate superior B cell immunity and has implications for vaccine design against other rapidly mutating viruses (HIV, hepatitis C, coronaviruses). As mRNA vaccine technology matures and costs decrease, these findings support the case for replacing inactivated influenza vaccines with mRNA-based formulations.
Why for Yiru: While vaccine biology may seem distant from TME research, the B cell response mechanisms elucidated here — prolonged germinal center reactions driven by sustained antigen exposure, T follicular helper cell support, and affinity maturation — are relevant to understanding tertiary lymphoid structures (TLS) in tumours. TLS are organized immune structures that form in some tumours and are associated with better responses to immunotherapy; they contain germinal center-like regions where B cells undergo affinity maturation against tumour antigens. The factors that sustain productive germinal center reactions in vaccine responses — antigen persistence, Tfh cell support — may similarly determine whether TLS in tumours produce high-affinity anti-tumour antibodies. Computational analysis of TLS in spatial transcriptomics data could examine whether the molecular features of productive vaccine-induced germinal centers are recapitulated in tumour-associated TLS.
Single cell in vivo analysis of type I IFN and NK cell-mediated control of B cell infection densities during acute gammaherpesvirus infection
bioRxiv Published 2026-06-16 preprint DOI: 10.64898/2026.06.15.732291
gammaherpesvirus type I interferon NK cell B cell single-cell in vivo imaging viral infection innate immunity
Summary: Uses a gammaherpesvirus reporter system to visualize infected B cells in lymph nodes and dissect how type I interferons and NK cells control infection at the single-cell level in vivo. Human gammaherpesviruses — Epstein-Barr virus (EBV) and Kaposi's sarcoma-associated herpesvirus (KSHV) — establish lifelong latent infections in B cells and can cause severe disease in immunocompromised individuals. While type I IFNs and NK cells are known to be important for controlling these infections, how they operate at the level of individual infected cells within lymphoid tissues has been unclear. Using murine gammaherpesvirus 68 (MHV68) expressing a fluorescent reporter, the authors tracked infected B cells in the lymph nodes of wild-type and type I IFN receptor-deficient mice. They found that type I IFN signalling acts locally to suppress the viral load within individual infected B cells, reducing the number of viral genomes per cell rather than simply eliminating infected cells. NK cells, by contrast, function through a different mechanism: they preferentially recognize and kill B cells with the highest viral loads, acting as a "ceiling" that prevents any single cell from harbouring excessive virus. The two innate defence layers operate with distinct spatial dynamics — IFN-mediated control is diffuse and affects all infected cells in the lymph node, while NK-mediated killing is focused on high-burden cells that upregulate NK-activating ligands. This layered defence explains why loss of either IFN signalling or NK cells alone leads to increased infection but loss of both is catastrophic.
Why it matters: This study provides a rare single-cell resolution view of innate immune control of viral infection in intact lymphoid tissues — a level of detail that has been difficult to achieve for B cell-tropic viruses. The finding that type I IFN and NK cells operate through distinct, complementary mechanisms (viral load suppression vs. high-burden cell elimination) explains the functional redundancy that has complicated genetic studies of these pathways. The layered defence model also has clinical implications: immunodeficiencies affecting both IFN and NK pathways (as can occur after hematopoietic stem cell transplantation) would be expected to produce particularly severe gammaherpesvirus disease, while deficiencies in only one pathway might be partially compensated.
Why for Yiru: The concept of layered immune defence — with one mechanism controlling the "average" infection level and another eliminating extreme outliers — may apply to immune surveillance of other cell types harbouring intracellular pathogens or oncogenic viruses. In the TME, tumour cells expressing viral antigens (e.g., HPV+ or EBV+ tumours) may be subject to analogous layered control by IFNs and NK cells, with IFN suppressing viral gene expression and NK cells eliminating cells with the highest antigen loads. The single-cell in vivo tracking approach used here — combining fluorescent viral reporters with quantitative imaging — could inspire similar approaches for tracking immune-tumour cell interactions at single-cell resolution in TME models.
Trypanosoma brucei infection remodels the uterine immune environment and drives neuroendocrine dysfunction
bioRxiv Published 2026-06-12 preprint DOI: 10.64898/2026.06.11.731423
Trypanosoma brucei female reproductive tract uterus immunology neuroendocrine infection tissue immunology
Summary: Reveals that Trypanosoma brucei, the parasite causing African sleeping sickness, infiltrates the uterine lining and substantially remodels the immune environment of the female reproductive tract, with downstream neuroendocrine consequences. Human African Trypanosomiasis (HAT) is a systemic parasitic infection associated with immunological, metabolic, and neurological pathology. Reproductive dysfunction has been clinically recognized in HAT patients, but whether parasites directly infiltrate the female reproductive tract and how infection reshapes its immune landscape has been unknown. Using a murine model, the authors demonstrate that T. brucei parasites localize within the uterine endometrium during both acute and chronic infection. Chronic infection drives progressive remodelling of the uterine immune compartment, with shifts in macrophage, dendritic cell, and lymphocyte populations, accompanied by disrupted estrous cycling and altered hypothalamic-pituitary-gonadal axis function. The findings establish the female reproductive tract as a site of active parasite infection and immune restructuring in trypanosomiasis, linking tissue-level immune changes to systemic neuroendocrine dysfunction.
Why it matters: The female reproductive tract has been an understudied compartment in systemic infectious diseases, with most research focusing on sexually transmitted infections that directly target reproductive organs. This study demonstrates that a blood-borne parasite can establish residency in the uterine lining and induce local immune remodelling with systemic hormonal consequences. This has implications beyond trypanosomiasis: other systemic infections associated with reproductive dysfunction (tuberculosis, chronic viral infections) may similarly involve direct pathogen infiltration of reproductive tissues. The neuroendocrine consequences also highlight the interconnectedness of immune and endocrine systems in infection — tissue-level immune changes can have organism-level physiological effects through disruption of hormonal axes.
Why for Yiru: Tissue-specific immune remodelling by pathogens provides a natural experiment for understanding how local immune environments are shaped and maintained. The uterine immune compartment undergoes dramatic physiological remodelling during the reproductive cycle, and understanding how infection disrupts this process could reveal general principles of tissue immune homeostasis. For TME research, the concept that a pathogen can establish a tissue niche and progressively remodel the local immune environment is analogous to how tumours establish and reshape their microenvironment — recruiting suppressive immune cells, altering cytokine milieus, and disrupting normal tissue architecture. The neuroendocrine axis disruption also resonates with emerging evidence that tumours can influence systemic physiology through neural and hormonal pathways.
Decoding common and rare noncoding variant effects across cellular and developmental contexts
Nature Genetics Published 2026-06-15 research article DOI: 10.1038/s41588-026-02619-6
noncoding variant chromatin accessibility gene regulation deep learning functional genomics resource human genetics
Summary: Contributes a resource of predicted regulatory effects for noncoding genetic variants across diverse cellular and developmental contexts, and introduces a method to identify variants with extreme regulatory effects for disease variant discovery. The vast majority of disease-associated genetic variants from GWAS lie in noncoding regions of the genome, where they are presumed to affect gene regulation by altering transcription factor binding sites, chromatin accessibility, or enhancer activity. However, predicting which noncoding variants are functional — and in which cell types and developmental stages they act — remains a major challenge. The authors generated a large-scale resource of predicted chromatin accessibility effects for millions of common and rare noncoding variants across hundreds of human cell types and developmental contexts, using deep learning models trained on large chromatin accessibility datasets (ATAC-seq, DNase-seq). They also developed a statistical method to identify variants with outlier regulatory effects — those that cause changes in predicted chromatin accessibility far outside the normal range of variation at a given genomic position. Applying this framework to variants associated with immune-mediated diseases, neuropsychiatric disorders, and developmental conditions, they identify specific noncoding variants with strong predicted regulatory effects in disease-relevant cell types, providing candidates for functional validation.
Why it matters: Noncoding variant interpretation is one of the major bottlenecks in human genetics. GWAS have identified thousands of disease-associated loci, but pinpointing the causal variant(s) and their target gene(s) at each locus remains laborious. A comprehensive resource of predicted regulatory variant effects across cell types and developmental stages accelerates this process by providing testable hypotheses: which variants are likely to be functional, in which cell types they act, and which genes they regulate. The focus on outlier effects is statistically principled — most noncoding variants have negligible regulatory effects, and prioritizing those with extreme predicted effects enriches for true functional variants. The resource will be valuable for any researcher trying to connect noncoding GWAS hits to molecular mechanism.
Why for Yiru: Noncoding regulatory variation is relevant to the TME because tumour cells, immune cells, and stromal cells all acquire noncoding mutations (somatic and germline) that can affect TME gene expression programs. The variant effect resource could be used to annotate noncoding mutations found in tumour genomes or in GWAS of immunotherapy response: are there noncoding variants predicted to alter chromatin accessibility of immune checkpoint genes, cytokine loci, or MHC genes specifically in T cell or myeloid cell types? The method for identifying variants with extreme regulatory effects could also be adapted to identify noncoding mutations in tumours that have outsized effects on TME gene regulation. More broadly, the deep learning framework for predicting chromatin accessibility from sequence is relevant to any computational study of gene regulation in the TME.