Germinal centers are dynamic immune structures in which B cells repeatedly alternate between a proliferative dark zone and a selection-dominated light zone. During this cycle, centroblasts divide and diversify their immunoglobulin receptors, while selected centrocytes can receive survival signals, return to the dark zone and undergo further rounds of mutation and selection. Some cells eventually leave the reaction as prememory B cells or plasmablasts, which can give rise to memory B cells or plasma cells. Because single-cell sequencing destroys the cells it profiles, the sequence and regulation of these transitions must be inferred from a static snapshot, making it difficult to distinguish cell-fate drivers from correlations and to predict how a mutation will alter trajectories.

DynaVelo addresses that problem by treating RNA expression and transcription-factor motif accessibility as two connected dynamical systems. Variational encoders first map the two data types into lower-dimensional latent spaces, where a shared neural ordinary differential equation learns a vector field describing cellular change over an inferred latent time. Separate decoders translate that field back into gene-expression and motif-accessibility space, allowing the model to estimate RNA velocities for all modeled genes as well as velocities for transcription-factor motifs. Training combines reconstruction of the observed data, regularization of initial states and latent times, alignment with available scVelo RNA-velocity estimates for a subset of genes, and a consistency term requiring RNA and motif velocities to imply similar trajectories. The resulting model is generative, so it can create cells along inferred trajectories and simulate loss-of-function changes without requiring new cells for every perturbation.

Applied to 7,166 wild-type mouse germinal-center B cells, DynaVelo reconstructed the major branches of the reaction. Its model-derived vector field captured the centroblast cell cycle, the transition from centroblasts to light-zone centrocytes, the return of selected centrocytes to the dark zone, and exits toward prememory cells and plasmablasts. Its latent-time ordering more closely followed these biological transitions than the ordering produced by scVelo, which did not identify the bifurcation between dark-zone cycling and light-zone differentiation. A comparison with the multiome velocity method MultiVelo likewise found differences: MultiVelo did not recover the dominant centroblast-to-centrocyte direction and placed plasmablasts near the beginning rather than at a terminal exit. These results describe dynamics inferred from sampled cells and model outputs, not direct continuous observation of cells moving between zones.

Generated trajectories also exposed cell-state-specific regulatory patterns. Motif accessibility for Spi1 and Nfkb1 increased during the centroblast-to-centrocyte transition, whereas Spi1 remained stable and Nfkb1 decreased during dark-zone cycling. Along the route toward prememory cells, the model identified temporal changes in factors including Runx1 and Irf8, while the plasmablast branch showed accessibility and expression patterns involving Prdm1, Irf4 and Xbp1. The distinction between expression and motif accessibility was important: Prdm1 motifs were accessible in both prememory and plasmablast states, but Prdm1 RNA was expressed only in plasmablasts. The analysis therefore separated RNA-expression patterns from the accessibility of regulatory motifs, while the factors identified as possible drivers remain interpretations of temporal patterns rather than proof of causation in every inferred transition.

The same framework revealed how heterozygous deletions of Arid1a or Ctcf altered the inferred germinal-center landscape in samples containing 6,716 and 4,688 cells, respectively. Both mutants retained recognizable cell-cycle and differentiation branches but lacked the light-zone-to-dark-zone recycling path. Prememory cells accounted for 5.42% of wild-type cells, compared with 20.09% in Arid1aHet and 25.45% in CtcfHet germinal centers; the study also examined an Arid1a homozygous sample, in which the proportion was 44.96%. These are differences among genotypes, not measurements presented as a decline over time. The model associated the loss of recycling with altered activity of factors including Foxp1, Bach2, Spi1 and Ctcf, and with loss of a Foxo1-to-Cxcr4 regulatory relationship associated with migration back to the dark zone.

DynaVelo inferred dynamic gene-regulatory networks through two related calculations. Jacobian matrices measured the local derivative of one gene’s RNA velocity with respect to another gene’s expression, revealing how regulatory relationships varied among individual cells and along trajectories. Finite in silico knockouts instead set a transcription factor’s RNA and motif accessibility to their minimum observed values and calculated the resulting changes in velocity. The two approaches were not identical: Jacobians represented infinitesimal local changes, whereas simulated knockouts imposed finite changes across cells. Together, they recovered multiple previously reported relationships in wild-type cells while identifying edges that were missing or changed in the mutants. One example was the Spi1-to-Bcl-6 relationship, which disappeared along the centroblast-to-centrocyte-to-prememory trajectory in both heterozygotes. The sign of the modeled Bcl-6 influence on Bcl-2 also differed between wild-type and mutant cells.

To evaluate the model as a predictor of genetic loss, the researchers simulated Arid1a loss in wild-type cells and compared the predicted velocity changes with differential velocities inferred from Arid1aHet cells. Agreement was generally stronger for motif velocities than for RNA velocities across cell types. The simulation predicted reduced connectivity between centroblasts and centrocytes, loss of recycling, and a stronger route from centrocytes to prememory cells, matching the altered mutant structure. The study then perturbed candidate genes in the Arid1aHet model and searched for changes in RNA and motif velocities that pointed toward wild-type values. Ctcf, Stat3 and Bcl6 were among the strongest candidates, alongside Tox and Bcor. A velocity was counted as rescued when its direction of change matched the direction from mutant to wild-type in that cell state, making these computational predictions distinct from the subsequent in vivo outcomes.

Experimental profiling of Arid1aHet;CtcfHet double-heterozygous germinal centers showed a partial reversal of the rapid centrocyte-to-prememory exit characteristic of the Arid1a mutant. In centrocytes, the measured rescue rates were 38% for RNA expression, 64% for RNA velocity, 51% for motif accessibility and 46% for motif velocity. The defining Arid1aHet phenotype was an elevated dark-zone/light-zone ratio. Two newly generated compound-heterozygous models, Arid1aHet;Bcl6Het and Arid1aHet;Stat3Het, each restored that ratio to wild-type levels, providing direct functional support for two predictions. Together with the Ctcf double heterozygote, the experiments validated all three nominated targets. The results show that DynaVelo can nominate experimentally testable genetic interactions in this mouse system, but the study does not establish clinical effectiveness, and its predictions remain dependent on the quality of the training data, chosen genes, velocity guidance and model settings.

The experimental work used age- and sex-matched mice, with both males and females represented in each group, and the researchers reported no sex-dependent effects. Germinal-center B cells were profiled by paired single-cell RNA sequencing and ATAC sequencing, while flow cytometry measured cell-surface markers and the dark-zone/light-zone composition of germinal centers. The computational inputs included 2,000 highly variable genes, approximately 400–1,000 genes with scVelo velocity estimates, about 200 expressed transcription factors and roughly 20 genes of interest, with the union used for modeling. The study used separate wild-type, Arid1aHet, CtcfHet and compound-heterozygous samples, and it trained each model on a 90:10 training and validation split with early stopping. These design features support the model’s use as a trajectory and perturbation-analysis framework for this particular germinal-center system, while the experimental rescue results concern mouse phenotypes rather than human disease outcomes or therapeutic interventions.

Subject of Research: Computational modeling of mouse germinal-center B-cell dynamics and validation of predicted genetic rescue targets

Article Title: Deep dynamical models of single-cell multiomic velocities predict loss-of-function and rescue perturbations in B cells

Article References: Karbalayghareh, A., Pelzer, B., Chin, C. R., Melnick, A., Barisic, D., & Leslie, C. S. (2026). Deep dynamical models of single-cell multiomic velocities predict loss-of-function and rescue perturbations in B cells. Nature Biotechnology. https://doi.org/10.1038/s41587-026-03348-1

Image Credits: AI Generated

DOI: 10.1038/s41587-026-03348-1

Keywords: single-cell multiome, germinal-center B cells, neural differential equations, gene-regulatory networks, in silico perturbation, Arid1a, Ctcf, Bcl6, Stat3, epigenetic regulation, Deep, dynamical

The work focused on chimeric RNAs, or chRNAs, formed when sequences from separate genomic loci are combined within one transcript. Researchers had previously linked some chimeric transcripts to cancer-associated chromosomal rearrangements, but earlier work had also shown that RNA-level trans-splicing can generate chRNAs and functional fusion proteins without corresponding changes in genomic structure. Isolated examples had been reported in normal contexts, including JAZF1-JJAZ1 in endometrial fibroblasts and PAX3-FOXO1 during muscle differentiation. The new study, described in a research highlight published in Cell Research, asks whether additional trans-spliced RNAs occur widely in mammalian tissues and whether they can make functional proteins in normal immune cells.

To map chRNAs in mouse macrophages, the researchers combined long-read Oxford Nanopore direct RNA sequencing with short-read sequencing, computational analysis, PCR and Sanger sequencing. They also used NanoString junction-specific profiling, an orthogonal hybridization-based method for testing predefined RNA junctions. More than 30,000 candidate chimeric RNAs were detected. Many joined genes on different chromosomes or separated by large genomic distances, and numerous transcripts changed after inflammatory stimulation. That regulation argues that chRNA formation is not merely random noise generated during transcription. Instead, the joining of RNA sequences appears to respond to the physiological state of the cell, although the researchers did not directly test the function of every candidate transcript.

Validating such molecules presents a central technical problem. Short-read sequencing usually involves reverse transcription, and template switching during reverse transcription or library preparation can create artificial sequence junctions that resemble genuine RNA fusions. Direct RNA sequencing avoids that particular source of error, while the independent NanoString assay and PCR-based tests provide additional ways to confirm selected junctions. Agreement among these methods strengthened the case that the identified chRNAs are endogenous. The technologies did not recover identical transcript sets, however. Abundance, isoform structure, sequencing chemistry and assay sensitivity can all affect which low-level chRNAs are detected, leaving standardized validation criteria as an important requirement for future studies.

The analysis also connected chRNA production with the three-dimensional organization of the genome. Inflammatory stimulation increased contacts between several parental genomic loci involved in forming chRNAs. Depleting CTCF, a protein that helps organize chromatin loops, reduced those contacts and decreased chRNA production without substantially altering expression of the parent genes. This supports a model in which the physical arrangement of chromosomes helps determine which independently transcribed RNAs can be connected. Proximity alone is unlikely to be sufficient, though. The study did not resolve how two transcripts are selected, aligned and joined, or which splice-site features, RNA-binding proteins, transcriptional timing or regulatory sequences guide the process. Genomic architecture may therefore facilitate RNA-level gene joining without being its sole determinant.

The strongest functional evidence came from the Gsdmd-Tmem106a chimera. Its RNA joined exons 1–2 of Gsdmd with exons 6–9 of Tmem106a without an accompanying chromosomal rearrangement. Translation produced an unannotated GSDMD-TMEM106A fusion protein containing the membrane-associated N-terminal region of GSDMD and a shifted C-terminal region derived from TMEM106A. GSDMD normally acts as an executor of pyroptosis: inflammatory caspase-mediated cleavage releases its N-terminal domain, which oligomerizes in membranes to form pores. Those pores promote membrane permeabilization and IL-1β release. The fusion protein localized to the plasma membrane and associated with activated GSDMD-NT domains, adding a regulatory layer to an established inflammatory pathway.

Experiments showed that the effect was not simply a consequence of making more GSDMD. The chimeric RNA was expressed at less than 10% of the level of parental Gsdmd. Adding an amount of GSDMD-TMEM106A-encoding messenger RNA equal to that lower level nearly doubled IL-1β release when supplied alongside GSDMD-encoding messenger RNA, whereas a comparable increase in GSDMD had little effect. Removing the TMEM106A-derived C-terminal segment abolished the enhancement. These results indicate that the fusion protein performs a specific function rather than acting merely as an additional source of GSDMD. The work also provides a concrete example of why conventional expression measurements alone can miss biologically important activity: a relatively scarce transcript can encode a protein with a large effect on cellular behavior.

Complementary mouse genetic models established relevance during infection and systemic inflammation. A GSDMD-TMEM106A-MYC knock-in allowed the fusion protein to be detected directly, while a Gsdmd-Tmem106aStop model selectively prevented translation of the fusion while preserving both parental genes. Eliminating GSDMD-TMEM106A increased bacterial burden after infection with Salmonella Typhimurium but protected mice from lethal endotoxemia. Increasing fusion activity produced the opposite pattern, improving bacterial clearance while worsening inflammatory pathology. The chimera therefore appears to help tune a trade-off: preserving an effective response against intracellular bacteria can also intensify inflammatory injury. The experiments were performed in mice, so they do not establish that the same quantitative balance operates in people.

The findings may not transfer directly from mice to humans. The researchers identified more than 900 chRNAs in human monocyte-derived macrophages and detected chRNAs involving the same parent genes in both mouse and human cells, but they did not detect GSDMD-TMEM106A itself in the human macrophages examined. Conservation of the genes from which a chimera is assembled therefore does not guarantee conservation of the assembled RNA. Species differences in exon structure, splice-site compatibility, regulatory sequences, RNA-binding factors and genome organization could all affect whether a particular chRNA is generated. More broadly, the study suggests that cellular responses may depend not only on how strongly individual genes are expressed, but also on which transcripts become connected, when they are translated and whether the resulting proteins are functional.

That implication reaches beyond immune biology. Stimulus-responsive chRNAs could contribute to regulatory networks in development, nervous-system function, tissue repair, aging and disease, but the present study does not demonstrate such roles. Its strongest conclusion is that mammalian immune cells contain a large, regulated landscape of chimeric RNAs and that at least one member can encode a functional fusion protein in vivo. By linking chromatin contacts and CTCF-dependent genome organization to RNA joining, and then to the control of pyroptosis, the work reframes the transcriptome as more than a collection of transcripts encoded one gene at a time. It also exposes a deeper question for future research: how far the three-dimensional organization of the genome determines which genes become functionally connected at the RNA level, and how often that connection creates a protein with a biological role that would be missed by measuring parent-gene expression alone.

Subject of Research: Inflammation-induced chimeric RNAs in mammalian immune cells and fusion-protein control of pyroptosis

Article Title: Functional chimeric RNAs expand the mammalian proteome

Article References: Xie, Z., & Li, H. (2026). Functional chimeric RNAs expand the mammalian proteome. Cell Research. https://doi.org/10.1038/s41422-026-01309-6

Image Credits: AI Generated

DOI: 10.1038/s41422-026-01309-6

Keywords: chimeric RNA, pyroptosis, inflammation, immune cells, fusion proteins, CTCF, genome organization, macrophages, IL-1β, RNA sequencing, Functional, chimeric

The stakes of this research are enormous. Genome organization is intimately connected to essential cellular functions such as gene expression, DNA replication, and DNA repair. Enhancers, the distal regulatory elements that boost gene activity, may lie millions of DNA letters away from the genes they control along the linear chromosome, yet they must physically find and contact their targets within the crowded nuclear space. Understanding how the chromatin fiber brings the right pieces together requires more than biology alone; it requires quantitative models in which chromosomes are treated as polymer chains and proteins as interacting particles whose collective behavior can be simulated, perturbed, and compared against experimental data from microscopy and chromosome-conformation assays.

The first pillar of the framework discussed by Fontana and Chiariello is phase separation, the same process that causes oil to bead up in water, here operating on proteins inside the nucleus. Many regulatory proteins carry intrinsically disordered regions that allow them to interact with copies of themselves, forming stable condensates. Well-studied examples include Mediator and RNA polymerase II, which cluster together at active genes, as well as coactivators that assemble at super-enhancers and transcription factors whose activation domains can condense to drive gene expression. At larger scales, phase separation helps build Cajal bodies and the nucleolus, a multi-phase structure assembled from distinct molecular species of proteins and RNA. Crucially, when the phase-separation properties of these proteins go wrong, the result can be disease, ranging from rare genetic disorders involving nucleolar dysfunction to cancer driven by aberrant condensates.

To capture this phenomenon computationally, theorists have built models in which chromatin is represented as a polymer and proteins are represented as particles that bind to it. Some of the simplest versions include only polymer-protein interactions, leaving out direct protein-protein attraction. Even so, these bridging models give rise to what is known as polymer-polymer phase separation, in which the polymer itself nucleates condensates of bound proteins that then glue distant chromatin segments together. Despite their minimalism, such models have successfully reconstructed the three-dimensional structures of real genomic regions, described chromatin remodeling across different cell types, and even tracked how genome architecture is rearranged following viral infections, including SARS-CoV-2.

More realistic systems require explicit protein-protein interactions alongside chromatin-protein affinity, and, in the most complete versions, loop extrusion as well. In loop extrusion, molecular motors of the cohesin or condensin families slide along the chromatin fiber, enlarging DNA loops until they are halted at anchor points, often corresponding to binding sites of the architectural protein CTCF. Single-molecule imaging has now captured this process in real time, and computational studies show that extrusion contributes to global chromosome condensation during mitosis as well as to the local loops and domains that organize the interphase genome. When all these ingredients coexist, the behavior of the system is governed by a set of control parameters, including protein concentration, self-interaction affinity, chromatin affinity, and motor density and processivity, and the resulting equilibrium states fall into complex phase diagrams reminiscent of those studied in condensed matter physics.

One of the most striking predictions to emerge from large-scale molecular dynamics simulations concerns a scenario now called polymer-assisted condensation, or PAC. Simulations revealed that phase-separated condensates can arise in two distinct ways: either spontaneously through strong protein-protein affinity, independently of the polymer, or in a polymer-assisted route in which protein-protein attraction alone is too weak to drive condensation, but binding to chromatin concentrates the proteins enough to tip the system over the threshold. This second mechanism, first predicted computationally and later explored theoretically, has been proposed as a potentially general principle of chromatin organization, with implications for heterochromatin formation and epigenetic memory.

The power of the integrated approach was demonstrated in recent work on PHF13, a protein that binds chromatin enriched with the histone methylation marks H3K4me3 and H3K27me3 and can also self-associate through specific N-terminal sub-domains. Molecular dynamics simulations showed that when both chromatin-protein and protein-protein interactions were enabled at sufficient PHF13 concentration, the modeled polymer collapsed into a compact, roughly spherical globule decorated with PHF13 clusters, mirroring microscopy images of interphase cells overexpressing PHF13, in which chromosomes condense into rounded structures. When the self-interaction domains were disabled in silico, compaction was reduced and condensation disappeared, exactly matching the behavior of engineered PHF13 variants lacking those domains in living cells. The close agreement between simulation and experiment provides a compelling validation of the physics-based framework.

The same modeling philosophy is now being applied to disease. The chimeric protein NUP98-HOXA9, an oncofusion associated with leukemia, can undergo phase separation and bind chromatin, promoting the formation of aberrant loops that bring oncogenes such as PBX3 into contact with distant regulatory elements. Models of the relevant genomic regions accurately recapitulated the aberrant looping patterns observed in experimental Hi-C contact maps from human cells and in single-molecule mobility measurements, while also capturing the interplay between phase separation and loop extrusion that underlies the pathological rewiring. Because altered phase separation has been implicated in cancers and rare genetic diseases, such models offer a route to understanding how molecular-scale misfolding translates into gene misregulation.

The third pillar of the framework concerns the nuclear envelope. The nuclear lamina, a protein network embedded in the inner membrane, adsorbs peripheral chromatin, spatially segregating specific genomic regions from the nuclear interior and repressing their activity. Computational models represent the lamina as an adsorbing membrane that attracts or sterically excludes chromatin segments, with interactions parameterized from DamID profiles that quantify locus-specific lamina association. When loop extrusion and lamina adsorption are combined in a single simulation, the framework accurately reproduces the organization of peripheral loci such as the mouse Sox9 and Kcnj2 region, and tuning chromatin-lamina affinity allows different degrees of peripheral association and distinct cell populations to be described. Simulations of genomic editing, including deletion of CTCF boundary elements, quantitatively matched experimental outcomes, establishing a link between loop extrusion and lamina association even though the two processes act independently.

Fontana and Chiariello conclude that the great advantage of these physics-based models is their versatility: phase separation, loop extrusion, and membrane adsorption can be combined within a single computational framework to describe increasingly detailed biological contexts, at the cost of larger parameter spaces and more complex phase diagrams that molecular dynamics simulations are well suited to explore. Looking forward, the authors point to promising extensions, including modeling pathogenic protein variants, the chromatin changes of ageing, and the architectural disruptions induced by external pathogens such as viruses. As genome-wide datasets grow richer and simulations grow faster, the dream of predicting, and perhaps one day correcting, the physical misfolding events behind human disease moves steadily closer to reality.

Subject of Research: Physics-based computational modeling of chromatin 3D organization through phase separation, loop extrusion, and nuclear lamina adsorption

Article Title: On the physics of chromatin molecular complex: from phase-separation to membrane adsorption

Article References: Fontana, A., & Chiariello, A. M. (2026). On the physics of chromatin molecular complex: from phase-separation to membrane adsorption. Frontiers in Molecular Biosciences, 13, Article 1951961. https://doi.org/10.3389/fmolb.2026.1951961

Image Credits: AI Generated

DOI: 10.3389/fmolb.2026.1951961

Keywords: chromatin, 3D genome organization, phase separation, loop extrusion, polymer physics, nuclear lamina, molecular dynamics, gene regulation, biomolecular condensates, CTCF, cohesin, leukemia

The team attacked the problem with an unusually rich experimental resource: bulk Hi-C contact maps generated from the ventral hippocampus, a brain region central to memory and emotion, of eight inbred mouse strains. Five of these were classical laboratory strains and three were wild-derived, capturing a meaningful slice of the genetic diversity within a single species. The resulting dataset comprises 14.55 billion valid contact pairs, an enormous catalog of who-touches-whom information across the mouse genome. By comparing chromatin architecture across strains that differ at millions of sequence positions, the researchers could ask which features of genome folding are stable enough to survive both genetic drift and deliberate breeding, and which are labile.

Before any evolutionary claims could be made, however, the authors confronted a stubborn methodological problem. The most popular analytical tools treat TADs as discrete objects: an algorithm scans a Hi-C contact map and draws sharp lines where it detects a boundary. When the researchers applied such discrete calling across replicate samples, they found poor concordance between replicates, meaning the same biological sample analyzed twice could yield noticeably different sets of called domains. That instability makes discrete TAD calls a shaky foundation for cross-strain or cross-species comparison, because apparent differences might reflect algorithmic noise rather than genuine biology.

The solution was to abandon the binary view. Instead of asking whether a boundary exists at a given position, the team measured separation scores, a continuous metric that quantifies how strongly a genomic location insulates the regions on either side. Rather than a yes-or-no verdict, every point along the genome receives a graded insulation value, and weakly insulated boundaries can be distinguished from strongly insulated ones. This shift from categorical to continuous measurement substantially improved reproducibility between replicates, giving the researchers a reliable yardstick for comparing chromatin architecture across strains and, ultimately, across species. The lesson is one the field is likely to absorb broadly: when a biological property varies in degree, measuring it as a degree rather than a category can dissolve apparent irreproducibility.

Armed with this framework, the researchers identified 2,857 highly insulated genomic regions whose insulation is conserved across all eight mouse strains. These are places where the folding of the chromosome creates a particularly impermeable wall, and where that wall appears in every strain examined, from the most domesticated laboratory line to the wildest-derived strain. The consistency across such genetically diverse backgrounds suggests that these boundaries are not accidents of recent breeding history but features actively maintained by the cell.

What kind of genomic real estate hosts these conserved, strongly insulated boundaries? The team found they are markedly enriched for multiple binding sites of CTCF, the architectural protein best known for stitching DNA loops together and anchoring the boundaries of chromatin domains. They also sit in accessible, open chromatin, lie near super-enhancers, the dense clusters of regulatory elements that drive expression of genes defining cell identity, and are found close to housekeeping genes, the essential maintenance machinery every cell must keep running. In other words, the most insulated boundaries are not scattered randomly through the genome; they cluster around the regulatory and structural elements that biologists already consider among the most functionally consequential.

The evolutionary signal extends below the level of chromatin folding into the DNA sequence itself. The conserved highly insulated regions show greater DNA sequence conservation across placental mammals than would be expected by chance, indicating that the nucleotide letters underlying these boundaries have been constrained by purifying selection over deep evolutionary time. They also display more stable DNA methylation patterns, suggesting that their epigenetic state is buffered against the genetic perturbations that distinguish one mouse strain from another. Together, these converging lines of evidence, from protein binding, chromatin accessibility, sequence conservation and epigenetic stability, paint the strongly insulated boundary as a multiply protected feature of the mammalian genome.

The most striking result emerges when the comparison is pushed beyond mice. In a cross-species analysis, the researchers measured how often insulated regions occupy the same genomic positions in human and mouse, two lineages separated by roughly 90 million years of evolution. They found that the positional overlap between the two species increases steadily with insulation strength, climbing to 80.7 percent for the most highly insulated regions. That figure is remarkable: it means that for the strongest boundaries, the vast majority of locations are shared between human and mouse, implying that the folding architecture at these sites has been maintained essentially intact since the common ancestor of the two species. Weakly insulated boundaries, by contrast, overlap far less, consistent with the idea that only the most robust architectural walls are under strong selective constraint.

Not every difference in chromatin conformation between strains carries deep meaning, however. The study also examined structural genomic rearrangements, deletions, duplications and inversions of DNA segments that differ among mouse strains, and found that these are the primary driver of chromatin conformation differences between strains. Yet when the team searched for functional consequences of those conformational differences, they found only limited evidence that they matter for gene regulation or phenotype. This is a subtle but important nuance: much of the strain-to-strain variation in genome folding may be passive, a mechanical consequence of DNA sequence rearrangement rather than a cause of biological divergence. The functional story, the data suggest, lives in the conserved, strongly insulated boundaries rather than in the variable ones.

The practical payoff of the study is a prioritization tool. Because insulation strength is continuous, reproducible and predictive of conservation, researchers screening a disease-associated genomic variant can now consult the insulation landscape to judge whether a nearby TAD boundary is likely to be an ancient, functionally critical architectural element or a labile, probably inconsequential one. Given that disruptions of TAD boundaries have been implicated in developmental disorders and cancers, for example when structural variants allow an enhancer to reach a gene it was never meant to touch, a quantitative framework for ranking boundary importance could sharpen the interpretation of human genomic data. The work also reframes the long-running TAD debate: rather than asking whether TADs as a class are functionally meaningful, the field can now ask which specific boundaries carry the strongest evidence of constraint, with insulation strength serving as the guide. Funded in part by the National Institutes of Health and the Howard Hughes Medical Institute’s Hanna H. Gray Fellows Program, the study demonstrates that a careful rethinking of measurement, from discrete calls to continuous scores, can turn a contested question in genome biology into a tractable, quantitative one.

Subject of Research: Evolutionary conservation of topologically associating domain boundaries measured by Hi-C derived insulation strength in mouse and human genomes

Article Title: Hi-C derived insulation strength predicts evolutionary conservation of topologically associating domains

Article References: Soto, D. C., Chen, P. B., Huang, H., Galasso, J., Chen, Z. J., Luo, C., Ernst, J., & Flint, J. (2026). Hi-C derived insulation strength predicts evolutionary conservation of topologically associating domains. Genome Biology. https://doi.org/10.1186/s13059-026-04298-2

Image Credits: AI Generated

DOI: 10.1186/s13059-026-04298-2

Keywords: TADs, Hi-C, chromatin architecture, insulation strength, CTCF, genome evolution, mouse strains, separation scores, super-enhancers, DNA methylation, comparative genomics, Genome Biology

Writing in PLOS Computational Biology, a research team led by Lu Chai, Jie Gao and colleagues introduces DeCTCF, an integrative framework that decodes CTCF binding sequences by combining experimental data with predictions from a powerful artificial intelligence model. The work analyzed 236,552 CTCF binding sequences drawn from chromatin immunoprecipitation sequencing, or ChIP-seq, experiments across 118 human cell lines. Rather than treating all of these binding sites as a uniform population, the researchers asked whether they could be sorted into meaningful functional groups, and whether those groups could explain the protein’s remarkably diverse roles across tissues and cell types.

The key innovation lies in how DeCTCF characterizes each binding site. Instead of relying solely on the raw DNA sequence underneath a CTCF peak, the framework taps into Sei, a pretrained deep learning model trained on vast collections of regulatory elements. Sei can take a stretch of DNA and predict a rich panel of epigenomic features for it, effectively forecasting what kinds of regulatory activities that sequence is likely to support in different cellular contexts. By feeding predicted epigenomic profiles into the analysis, DeCTCF gains a layer of functional information that sequence alone cannot provide, allowing binding sites with similar sequence motifs to be distinguished by their predicted regulatory surroundings.

Using this approach, the team clustered the hundreds of thousands of CTCF binding sites into 20 distinct groups. Each cluster represents a population of binding sites that share not only sequence characteristics but also predicted epigenomic behavior. The clusters were not arbitrary statistical artifacts. When the researchers examined their properties, they found that the groups could be annotated into coherent functional modules, revealing an underlying modular organization in how CTCF operates across the genome. One module stood out as particularly prominent: a major cluster associated with three-dimensional chromatin architecture, consistent with CTCF’s well-established role as an anchor point for chromatin loops.

Beyond the architectural module, the analysis uncovered three lineage-associated modules, clusters of CTCF binding sites whose predicted features tie them to particular cellular lineages rather than to generic genome folding. These lineage modules are especially intriguing because they suggest that subsets of CTCF binding sites are deployed in a cell-type-specific manner, potentially recruiting different molecular partners depending on the developmental or tissue context. In other words, the same architectural protein may carry out different regulatory conversations in different cell types, and the binding sites it occupies carry telltale signatures of those conversations.

The study provides a concrete example of this principle in stem cells. Several clusters enriched in the three stem cell lines included in the dataset also showed enrichment of ZIC-family proteins, a group of transcription factors known to be active in early development. These same clusters were associated with gene sets related to pluripotency and neurodevelopment, the biological programs that define what stem cells are and what they can become. This convergence of evidence, linking a CTCF cluster, a candidate co-factor family and a coherent gene expression program, illustrates how the framework can generate testable hypotheses about which co-factors cooperate with CTCF in specific cellular contexts.

The researchers also examined the shape of CTCF ChIP-seq signals at the cluster level and found striking associations with chromatin loop annotations. Binding sites displaying a single-peak signal profile were reproducibly associated with higher loop interaction scores, suggesting that these sites act as strong anchors for chromatin loops. In contrast, sites with double- and triple-peak profiles showed distinct loop-pairing preferences, hinting that the local arrangement of CTCF occupancy, not merely its presence or absence, influences how distant regions of the genome are brought into contact. This observation adds a geometric dimension to the functional classification, connecting the epigenomic identity of a binding site to its role in physical genome architecture.

Technically, the study demonstrates the growing power of transfer learning in genomics. Training a model from scratch to interpret hundreds of thousands of regulatory sequences across more than a hundred cell types would require enormous labeled datasets that few laboratories possess. By leveraging Sei’s pretrained representations, DeCTCF shows that knowledge embedded in large-scale epigenomic prediction models can be repurposed for downstream biological questions, in this case the functional dissection of a single architectural protein’s binding landscape. The strategy turns a general-purpose sequence model into a specialized analytical instrument, a pattern that is likely to become increasingly common as pretrained genomic foundation models mature.

The broader significance of the work lies in reframing how scientists think about CTCF. Rather than a monolithic factor performing a single genome-folding task, CTCF emerges as a modular organizer whose binding sites fall into functionally distinguishable classes, each with its own predicted epigenomic context, candidate co-factors and architectural behavior. This modular view helps explain how one protein can participate simultaneously in chromosome compaction, enhancer-promoter communication and lineage-specific gene regulation without those roles collapsing into one another. It also provides a practical resource: the 20-cluster classification offers other researchers a framework for interpreting CTCF binding data in their own cell types of interest.

There remain important caveats and open questions. The lineage associations are computational predictions and statistical enrichments, not demonstrations of mechanism, and the candidate co-factor relationships, such as the link between ZIC-family proteins and stem cell clusters, will need experimental validation in the laboratory. The reliance on predicted epigenomic features also means that the classification inherits any biases of the underlying Sei model. Nevertheless, by systematically mapping CTCF’s modular organization across 118 human cell lines and connecting binding-site classes to chromatin loops, co-factors and developmental gene programs, DeCTCF delivers a comprehensive and testable picture of how the genome’s master organizer achieves its remarkable contextual versatility. The study points toward a future in which the regulatory grammar of the genome can be read, cluster by cluster, directly from its sequence.

Subject of Research: Computational classification of CTCF binding sites using predicted epigenomic features to reveal modular genome organization

Article Title: DeCTCF: Decoding CTCF binding sequences by leveraging predicted epigenomic features

Article References: Chai, L., Gao, J., Guo, T., Ba, T., Li, Z., Liu, J., Wang, Y., & Zhang, L. (2026). DeCTCF: Decoding CTCF binding sequences by leveraging predicted epigenomic features. PLOS Computational Biology, 22(10), e1014848. https://doi.org/10.1371/journal.pcbi.1014848

Image Credits: AI Generated

DOI: 10.1371/journal.pcbi.1014848

Keywords: CTCF, DeCTCF, epigenomics, chromatin architecture, Sei model, deep learning, ChIP-seq, genome organization, transcription factors, stem cells, ZIC proteins, computational biology

The core difficulty is architectural. The noncoding genome is not read linearly: cis-regulatory elements such as enhancers can lie more than a million base pairs away from the genes they control, often buried inside the introns of unrelated genes. Enhancers, bound by clusters of transcription factors, physically loop through three-dimensional space to contact promoters near transcription start sites, and the specificity of these contacts is shaped by CTCF-bound boundary elements operating through cohesin-mediated loop extrusion. Chromatin structure is therefore a key determinant of gene expression, but conventional chromosome conformation capture (3C) methods, which rely on restriction enzymes that cut only at specific sequence motifs, cannot resolve contacts below roughly 500 base pairs and cannot generate single-allele data for most variants. Only about 30 percent of variants lie close enough to a restriction cut site to be studied at all.

The Oxford group, led by James Davies together with Jim Hughes and Thomas Milne, developed a variant-to-function platform called Micro Capture-C variant-to-function, or MCCv, built on their earlier Micro Capture-C method. Instead of restriction enzymes, the technique uses micrococcal nuclease to fragment chromatin, allowing data to be generated from any site of interest. Oligonucleotide probes are designed directly over the variant being tested, and ultra-short-range ligation junctions of less than 800 base pairs define the nanoscale topology at the variant itself, while long-range junctions spanning one kilobase to one megabase define the structure of the whole locus. The result is the highest-resolution 3C data produced to date, capable of resolving chromatin architecture at the level of individual transcription factor binding sites while simultaneously identifying the target genes contacted by each regulatory element.

The single-allele capability is what makes the platform powerful for genetics. In cells that are heterozygous for a variant, the reference and alternate chromosomes can be separated computationally, so each allele effectively serves as an internal control for the other, minimizing experimental noise. In benchmarking experiments on 42 heterozygous gain-of-function enhancer variants in Jurkat cells, inactive reference alleles displayed a regular nucleosomal patterning, whereas active alternate alleles showed distinctive fine-scale contact patterns consistent with transcription factor interactions within nucleosome-depleted regions. The method captured 86.7 percent of reads for allelic analysis, and analysis of the rheumatoid arthritis-associated variant rs2793109 revealed that the disease allele creates a RUNX1 motif with a complex pattern of short-range contacts, while simultaneously showing allele-specific changes in long-range interactions with the promoters of ZEB1 and ZNF438. By phasing the captured variant with other heterozygous SNPs across the locus, the team linked the risk allele to increased ZEB1 expression using intron-rich RNA sequencing.

Applied systematically to immune-mediated inflammatory disease genetics, the platform delivered striking results. Single-cell ATAC-seq analysis showed that CD4-positive T cells, particularly central memory T cells, carry the highest burden of IMID-associated variation. The researchers imputed 3,104 independent haplotypes from 400 studies and tested each variant within them for molecular mechanisms, assigning a potential causal variant and mechanism to 47 percent of haplotypes, including 405 unique enhancer variants. Deep MCCv capture in activated primary CD4-positive T cells then mapped enhancer-promoter contacts with a sensitivity far beyond genome-wide methods: the strongest contact for each variant was supported by a mean of roughly 839 ligation junctions, compared with an average of just 2.5 junctions from Micro-C, and a fivefold better signal-to-noise ratio. At the 1q24.3-25.1 locus, four independent disease signals mapped to four distinct regulatory elements, with the eczema enhancer contacting FASL and the vitiligo, celiac disease and Crohn’s disease enhancers contacting TNFSF14 while skipping the intervening TNFSF18 promoter.

Ranking promoters by contact frequency allowed the team to nominate 251 unique causal genes, and these clustered in key immune pathways and were enriched for tractable drug targets and rare primary immunodeficiency mutations. Notably, only 60.5 percent of variants contacted their nearest gene most strongly, underscoring how misleading nearest-gene assumptions can be. Single-allele analysis of the 405 captured regulatory elements showed that 54.6 percent of heterozygous disease-associated variants had at least one significantly altered contact, and disease variants showed far greater architectural changes than matched controls for cell type, haplotype and disease association. When the researchers paired these contact data with phased RNA sequencing, variants with detectable changes in allelic contact frequency were roughly threefold enriched for allelic imbalance in expression of their nominated genes, with directional concordance between the chromatin and RNA effects at up to 88.2 percent of well-phased pairs.

The centerpiece discovery came from the rs4409785 variant at the 11q21 locus, associated with multiple sclerosis, rheumatoid arthritis and several other autoimmune conditions. Motif analysis showed that the disease-associated allele creates an entirely new CTCF binding site, a neo-CTCF motif, and the reference allele is conserved throughout hominin and primate evolution, confirming that the risk allele is a gain of function. MCCv footprinting revealed a precise 20-base-pair footprint over the CTCF motif only on the alternate allele, and CTCF ChIP-seq and ATAC-seq from heterozygous donors confirmed increased CTCF binding and open chromatin there. To prove causality, the team coupled the platform with genome editing: using the cytosine base editor BE4max in primary CD4-positive T cells from a heterozygous donor, they mutated three critical bases of the CTCF motif while leaving the variant itself intact. A single experiment yielded six separable profiles, and editing the motif on the risk allele reverted both the long-range contacts to convergent downstream CTCF sites and the nanoscale chromatin structure to the inert nucleosomal pattern of the reference chromosome.

The mechanistic consequence was unexpected. Neither allele of rs4409785 contacts any gene promoter directly; instead, the neo-CTCF site forms a subdomain that walls off a T cell-specific super-enhancer lying between the FAM76B and SESN3 genes. Phased MCCv from the SESN3 promoter showed that the risk allele significantly reduces contacts between that super-enhancer and the SESN3 promoter, and phased RNA sequencing linked the variant to a 36.6 percent reduction in SESN3 transcript on the risk chromosome. SESN3 encodes Sestrin 3, the least-characterized member of a family known to inhibit mTOR signaling, a central pathway controlling immune cell activation. In primary human T cells, SESN3 knockout left mTOR signaling unchanged under normal conditions but increased it under amino acid starvation. Using an endogenously tagged cell line and co-immunoprecipitation, the team showed that SESN3 interacts with the GATOR2 complex during starvation and that this interaction is disrupted by tryptophan but not by other amino acids, identifying SESN3 as a tryptophan sensor for the mTOR pathway in T cells.

In vivo work confirmed the physiological relevance. Transplanting genome-edited hematopoietic stem and progenitor cells into mice and inducing experimental autoimmune encephalomyelitis, a model of multiple sclerosis, the researchers observed accelerated onset of disease symptoms and weight loss in animals lacking Sesn3, along with increased infiltrating activated T cells in the central nervous system. Together, the findings demonstrate that a common noncoding variant can promote autoimmunity through a gain-of-function CTCF mechanism that silences a metabolic sensor of immune homeostasis, a mechanism the authors note could only have been fully resolved with high-resolution, allele-specific chromatin mapping. Because genetic evidence is known to increase the probability of clinical trial success roughly 2.6-fold, the team argues that MCCv, which can already capture 405 sites in parallel and could plausibly exceed 1,000 targets per experiment, offers a scalable route from GWAS catalogs to validated causal variants, genes and therapeutic targets.

Subject of Research: Allele-specific nanoscale chromatin architecture mapping to identify causal noncoding regulatory variants in inflammatory disease

Article Title: Single-allele nanoscale mapping of regulatory variants

Article References: Hamley, J. C., Zhang, W., Willmott, D., Chen, L. Y., Sharlandjieva, V., Li, H., Dalgleish, J. L. T., Denny, N., Agarwal, G., Hentges, L., Ozcan, B., Doll, R. M., Wei, Y., Riva, S. G., Venkatesh, S. S., Arachi, M., Agarwal, D., Akkuratov, E. E., Baxter, M., … Davies, J. O. J. (2026). Single-allele nanoscale mapping of regulatory variants. Nature Genetics, 58(10), 2660-2672. https://doi.org/10.1038/s41588-026-02776-8

Image Credits: AI Generated

DOI: 10.1038/s41588-026-02776-8

Keywords: MCCv, Micro Capture-C, noncoding variants, chromatin architecture, CTCF, super-enhancer, SESN3, mTOR signaling, autoimmunity, GWAS, CD4 T cells, genome editing

The genes in question, HOXD1 and HOXD8, belong to the homeobox family, an ancient set of 39 transcription factors arranged in four clusters—HOXA, HOXB, HOXC, and HOXD—that orchestrate body patterning along the head-to-tail axis during embryonic development. In vertebrates, these genes are expressed in a strictly ordered sequence known as collinearity, and their spatiotemporal control depends heavily on the physical architecture of the chromosome. In recent decades, scientists have realized that the same developmental programs can be hijacked or corrupted in cancer, with individual HOX genes acting as oncogenes in some contexts and tumor suppressors in others. What remained unclear was whether subtype-specific regulatory patterns within the HOXD cluster—particularly those governed by chromatin structure—might explain why some breast cancers behave so much more viciously than others.

To find out, the team combined large-scale computational mining of patient datasets from The Cancer Genome Atlas with laboratory experiments in breast cancer cell lines and validation in mouse models. Scanning all HOXD family members across the four major molecular subtypes of breast cancer, they found that HOXD1 and HOXD8 stood out with the most consistent, subtype-specific downregulation in triple-negative tumors, while neighboring cluster genes such as HOXD3, HOXD4, HOXD9, HOXD10, HOXD11, HOXD12, and HOXD13 showed no statistically significant subtype-specific patterns. Survival analysis reinforced the clinical weight of the finding: patients with low expression of either gene had significantly worse overall survival, and the combination of low expression of both genes sharpened the poor-prognosis curve even further.

The correlation extended into the laboratory. Quantitative PCR and western blotting confirmed that triple-negative cell lines such as MDA-MB-231 expressed far less HOXD1 and HOXD8 protein than non-triple-negative lines like MCF7 and BT474. When the researchers silenced the two genes with RNA interference in the non-aggressive cells, the cells became markedly more invasive, more migratory, and more proliferative. The reverse experiment proved equally telling: forcing HOXD1 or HOXD8 expression in triple-negative cells suppressed invasion, migration, and proliferation. Together, these gain- and loss-of-function results painted HOXD1 and HOXD8 as tumor suppressors whose loss licenses the aggressive phenotype that defines triple-negative disease.

But why would two genes that are not physically adjacent within the cluster—and that do not regulate each other’s expression directly—switch off in concert? The answer, the researchers discovered, lies in the three-dimensional organization of the genome. The HOXD cluster sits near the boundary of a subtopologically associating domain, a discrete chromatin neighborhood, and publicly available CTCF ChIA–PET data revealed partitioned interaction patterns extending in two directions across the locus. Circularized chromosome conformation capture showed that HOXD1 and HOXD8 physically contact each other in the folded chromosome, forming a loop that skips over the intervening genes. Chromosome conformation capture assays then demonstrated that these HOXD1–HOXD8 interactions were far more prominent in MCF7 cells, where both genes are active, than in MDA-MB-231 cells, where both are silenced.

The architect of that loop is CTCF, the CCCTC-binding factor, a highly conserved protein often described as the master weaver of the genome. CTCF can act as an activator, repressor, or insulator depending on context, and its binding to DNA is famously sensitive to methylation of CpG dinucleotides within its recognition sequence. ChIP-seq data showed two CTCF-binding sites near HOXD1 and four near HOXD8 in MCF7 cells, with visibly reduced occupancy at the same sites in triple-negative MDA-MB-231 cells. Critically, total CTCF levels were identical between subtypes—the difference was not how much CTCF a cell makes, but whether it can gain access to its landing pads at the HOXD locus.

The methylation connection soon came into focus. Chromatin immunoprecipitation revealed that DNMT3b, one of the enzymes that writes DNA methylation, accumulated more heavily at the HOXD1 and HOXD8 regions in triple-negative cells than in non-triple-negative cells. Bisulfite-based analyses confirmed the consequence: MCF7 cells carried a hypomethylated, open chromatin state at the HOXD1 and HOXD8 promoters, whereas MDA-MB-231 cells were distinctly hypermethylated. When the researchers treated triple-negative cells with 5-Aza-2′-deoxycytidine, a demethylating agent, CTCF binding was restored at the HOXD1 and HOXD8 sites, chromatin interaction patterns within the cluster partially reorganized, and expression of both genes climbed back up. Methylation, in other words, appears to be the upstream switch that evicts CTCF, collapses the regulatory loop, and silences the tumor-suppressive pair.

The team then tested causality directly. Using CRISPR–Cas9, they deleted specific CTCF-binding sites near HOXD1 and HOXD8 in MCF7 cells. The edited cells lost CTCF enrichment at the targeted regions, showed reduced HOXD1 and HOXD8 expression, and became more invasive. In mouse xenografts, tumors derived from the deletion clones progressed faster, while re-expression of HOXD1 and HOXD8 attenuated the aggressive in vivo phenotype. Molecular profiling of the tumors showed that deletion clones carried increased expression of proliferation- and invasion-associated genes, including epithelial-to-mesenchymal transition markers, and that forced HOXD1/HOXD8 expression largely reversed these signatures. The loop, the data suggest, is not merely correlated with tumor behavior—it helps drive it.

Finally, the researchers traced the downstream pathway. Correlation analysis across patient and cell line datasets identified a set of co-dysregulated genes, narrowed to six that distinguished triple-negative tumors, including ERGIC1, TFF3, and FOXA1. FOXA1 emerged as the most responsive target: it was strongly upregulated when HOXD1 or HOXD8 were overexpressed in triple-negative cells, downregulated when the two genes were depleted in non-triple-negative cells, and restored in xenografts where HOXD1 and HOXD8 were re-introduced. Bioinformatic scanning found nine HOX consensus motifs in the FOXA1 promoter, and luciferase reporter assays showed that HOXD1 or HOXD8 overexpression activated a FOXA1 promoter construct spanning positions −2000 to −1, but not a shorter −1020 to −1 construct, pinpointing the regulatory elements to a discrete upstream region. Low FOXA1 expression, like low HOXD1/HOXD8, correlated with poor overall survival in patients.

The study is, according to its authors, the first to identify coordinated, subtype-specific downregulation of HOXD1 and HOXD8 in triple-negative breast cancer linked to altered CTCF occupancy and DNA methylation at the HOXD locus, defining a suppressed regulatory axis—CTCF, HOXD1, HOXD8, FOXA1—that parallels the well-characterized chromatin-based dysregulation of HOXA9 in acute myeloid leukemia. The implications are twofold. Scientifically, the work demonstrates how spatial genome organization can selectively silence specific genes within a cluster while sparing their neighbors, likely through CTCF-mediated loops that connect distant regulatory elements. Clinically, it suggests that the HOXD1/HOXD8 regulatory axis, and the epigenetic machinery that controls it, could serve as both biomarkers of aggressive disease and targets for intervention—raising the tantalizing possibility that demethylating strategies capable of restoring CTCF binding and re-folding the HOXD locus might one day help rein in one of the deadliest forms of breast cancer.

Subject of Research: CTCF-mediated chromatin architecture and epigenetic silencing of HOXD1 and HOXD8 in triple-negative breast cancer progression

Article Title: CTCF-associated chromatin changes between HOXD1 and HOXD8 contribute to TNBC progression

Article References: Oh, J. H., Jeong, D. S., & Kim, C. Y. (2026). CTCF-associated chromatin changes between HOXD1 and HOXD8 contribute to TNBC progression. Experimental & Molecular Medicine. https://doi.org/10.1038/s12276-026-01861-6

Image Credits: AI Generated

DOI: 10.1038/s12276-026-01861-6

Keywords: triple-negative breast cancer, HOXD1, HOXD8, CTCF, chromatin architecture, DNA methylation, FOXA1, tumor suppressor genes, HOX genes, 3D genome, epigenetics, CRISPR

The research team, led by Ziyao Liu of the Second Affiliated Hospital of Xi’an Jiaotong University, began with a simple premise. Oxidative stress, the imbalance that lets reactive oxygen species overwhelm cellular antioxidant defenses, is a well-established contributor to optic nerve degeneration. But which of the hundreds of oxidation-related genes actually matter in POAG? To find out, the investigators assembled a comprehensive set of 885 oxidative stress-related genes from the GeneCards and MSigDB databases, then mined five public transcriptomic datasets from the Gene Expression Omnibus, using optic nerve head astrocyte samples from the GSE9963 cohort as their discovery set.

The screening pipeline was deliberately redundant. Weighted gene co-expression network analysis grouped genes into modules correlated with disease status, while two independent machine learning algorithms, LASSO regression and support vector machine recursive feature elimination, each flagged candidate genes under ten-fold cross-validation. When the three approaches were intersected, only two genes survived: FXN and CYP1B1. Of these, FXN proved the sharper diagnostic blade. Receiver operating characteristic analysis in the discovery cohort yielded an area under the curve of 0.945 for FXN, far exceeding CYP1B1’s 0.700, and the result held up across four independent validation datasets with AUC values ranging from 0.708 to 0.946.

A diagnostic meta-analysis pooling the five datasets strengthened the case further. Elevated FXN expression was significantly associated with POAG status, with an overall odds ratio of 6.35. Notably, FXN levels tracked with disease sample type but not with race, gender, or age, suggesting the signal reflects disease biology rather than demographic confounders. Bioinformatic characterization placed the protein in the cytosol and mitochondria, consistent with its known role in mitochondrial iron transport and respiration, and the team retrieved its three-dimensional structure from the AlphaFold database to support future drug design efforts.

The immune dimension added another layer of intrigue. CIBERSORT deconvolution showed that POAG samples with high FXN expression differed in estimated immune cell abundance, with M1 and M2 macrophage subtypes significantly reduced in the discovery cohort. Single-sample gene set enrichment analysis linked high FXN expression to signatures including antigen-presenting cell function, chemokine receptor interaction, and inflammation promotion, although these immune associations were less consistent across the validation cohorts. Metabolic profiling told a clearer story: gene set enrichment analysis repeatedly tied high FXN expression to glycolysis and pyruvate metabolism gene sets, hinting that FXN may participate in the metabolic reprogramming that accompanies glaucomatous degeneration.

Bioinformatics alone, however, could not establish causation. The team turned to an in vitro model in which retinal ganglion cell-derived R28 cells are injured by a 12-hour exposure to 200 micromolar hydrogen peroxide, a widely used simulation of oxidative stress injury in glaucoma research. Interestingly, while FXN was upregulated in patient tissue samples, hydrogen peroxide sharply decreased FXN expression in the cultured cells. That inverse response suggested a defensive role: the stressed cells were losing a protective factor. The researchers tested this by manipulating FXN with overexpression vectors and short hairpin RNAs, verifying the changes by quantitative PCR and western blot.

The functional assays were strikingly consistent. CCK8 and EdU assays showed that hydrogen peroxide suppressed cell survival, that FXN overexpression blunted this loss of viability, and that FXN knockdown made it worse. Flow cytometry using the BODIPY-C11 probe revealed that FXN overexpression significantly reduced the reactive oxygen species surge induced by peroxide treatment, while silencing FXN accelerated ROS accumulation. Annexin V staining and TUNEL assays confirmed the anti-apoptotic effect, with FXN overexpression defending cells against both early and late apoptosis. Western blots added a mechanistic clue: FXN overexpression increased levels of the antioxidant proteins NRF2, HO-1, and SOD1, while knockdown decreased them, indicating that frataxin bolsters the cell’s endogenous antioxidant defenses.

The deeper question was who controls FXN in the first place. Mining the hTFtarget and GeneCards databases produced four candidate transcription factors: MAX, YY1, RAD21, and CTCF. Cross-referencing public knockdown data from the KnockTF database and expression correlations in the discovery dataset narrowed the field, and PCR experiments in R28 cells delivered the verdict: only silencing CTCF downregulated FXN. CTCF, a zinc-finger protein famous for organizing chromatin into loops and insulating genes from inappropriate regulatory influences, emerged as the upstream switch. Chromatin immunoprecipitation sealed the argument, showing that CTCF antibody pulled down the FXN promoter sequence spanning bases 1797 to 1815 upstream of the transcription start site, one of four predicted binding sites.

The rescue experiments tied the axis together. CTCF expression itself dropped in the hydrogen peroxide model, mirroring FXN. Overexpressing CTCF preserved cell viability, reduced ROS accumulation, and curbed apoptosis just as FXN overexpression did, but when the researchers simultaneously knocked down FXN, much of that protection evaporated. In other words, CTCF shields retinal ganglion cells largely by switching on frataxin. Because CTCF is a global regulator, the team also intersected their 125 oxidative stress-related differentially expressed genes with predicted CTCF targets and found 79 overlaps, suggesting the CTCF-FXN axis may sit within a broader oxidative stress transcriptional network in POAG.

The authors are candid about the limitations. There is no animal model yet, the acute hydrogen peroxide model cannot fully reproduce the chronic oxidative environment of human glaucoma, and the regulatory mechanism still needs confirmation through promoter mutation assays and in vivo work. The immune findings require experimental validation, and whether FXN directly modulates mitochondrial ROS production remains open. Even so, the study delivers something glaucoma research has lacked: a biomarker with validated diagnostic performance across multiple cohorts, a plausible mechanistic story connecting oxidative stress, mitochondrial biology, and chromatin regulation, and a therapeutic hypothesis that could be tested. If frataxin’s protective power holds up in living retinae, the CTCF-FXN axis may become more than a molecular curiosity. It could become a target, and for a disease that currently can only be slowed, that would be a genuinely new frontier.

Subject of Research: The role of the CTCF-FXN regulatory axis in oxidative stress, diagnosis, and progression of primary open-angle glaucoma

Article Title: CTCF-FXN regulatory axis plays critical roles in clinical assessments and disease progression of primary open-angle glaucoma

Article References: Cai, D., Yu, Y., Ma, J., Li, T., & Liu, Z. (2026). CTCF-FXN regulatory axis plays critical roles in clinical assessments and disease progression of primary open-angle glaucoma. Journal of Molecular Medicine, 104(1), Article 114. https://doi.org/10.1007/s00109-026-02717-2

Image Credits: AI Generated

DOI: 10.1007/s00109-026-02717-2

Keywords: primary open-angle glaucoma, FXN, frataxin, CTCF, oxidative stress, retinal ganglion cells, reactive oxygen species, biomarker, transcriptional regulation, machine learning, mitochondria, neurodegeneration

The brake in question sits in the 5ʹ untranslated region, the stretch of RNA that precedes the protein-coding sequence of a transcript. Although it does not encode protein, this region exerts powerful control over how efficiently ribosomes translate the main message. One of its most potent repressive elements is the upstream open reading frame, or uORF, a short alternative reading frame that begins at an upstream start codon and diverts ribosomes away from the true protein. When a ribosome initiates at a uORF and terminates before reaching the main coding sequence, that translation event is wasted. Genes carrying strong uORFs therefore tend to make less protein than their mRNA abundance would predict, and in a haploinsufficient disease that shortfall can tip an individual over the threshold of clinical disease.

Earlier proposals for exploiting this biology focused on steric-block antisense oligonucleotides, synthetic DNA-like molecules that bind to the uORF start codon and physically obstruct ribosome initiation. The approach has shown promise for a handful of specific genes, but its broad applicability has been debated, because the geometry of each uORF differs and blocking initiation is not always feasible or effective. The Oxford-led team, working with colleagues at the International Centre for Genetic Engineering and Biotechnology in Trieste, the University of Exeter and industry partners, pursued a different tactic. Instead of masking the uORF, they asked whether the exon that contains it could be spliced out of the mature mRNA altogether, converting an inhibitory 5ʹUTR into a shorter, permissive one through the cell’s own RNA processing machinery.

To test how widely such an approach might apply, the researchers systematically screened human transcript annotations. Using MANE transcript definitions, the matched annotations agreed upon by NCBI and EMBL-EBI as the definitive reference set for each human gene, they catalogued exons located entirely within 5ʹ untranslated regions whose removal would preserve the reading frame and regulatory logic of the transcript. This analysis yielded 2,210 potentially skippable 5ʹUTR exons. The critical next step was to determine which of those exons actually contain functional uORF start codons, since a start codon annotated in the genome is only repressive if ribosomes genuinely engage with it in living cells.

To make that determination, the team mined ribosome profiling data generated from 13 human tissues and cell lines, including brain, heart and skeletal muscle. Ribosome profiling captures snapshots of ribosome positions across the transcriptome, allowing researchers to distinguish start codons that are actively used from those that are silent. Applying this filter, the researchers identified 1,056 skippable 5ʹUTR exons harbouring translated uORF start codons. Crucially, 79 of these exons sit in genes already classified as haploinsufficient monogenic disease genes in expert-curated resources such as ClinGen, the Gene Curation Coalition and Gene2Phenotype. In other words, a defined, immediately clinically relevant target list already exists, and the authors suggest it could serve as a roadmap for precision medicines across dozens of rare disorders.

From that list the team prioritised six candidate exons in genes linked to neurodevelopmental disorders, a therapeutic area where haploinsufficiency is common and where the blood–brain barrier complicates conventional protein replacement. They constructed dual luciferase reporter assays in which the native 5ʹUTR of each target gene drives a measurable enzyme, then compared translation when the candidate exon was removed. For four of the six genes—CTCF, GRIN2B, KRIT1 and TSC1—skipping the target exon significantly boosted downstream protein production, with increases ranging from 1.4-fold to 5.5-fold. That magnitude matters, because many haploinsufficient disorders are thought to respond to even modest restoration of gene dosage, and a twofold increase in protein output would be transformative for conditions in which patients carry essentially half of the normal complement.

A key mechanistic question was whether the benefit came from removing the uORF itself or from other regulatory elements embedded in the skipped exons, such as RNA structures, microRNA sites or RNA-binding protein motifs that might independently suppress translation. To disentangle these effects, the researchers created reporters in which only the uORF start codons were mutated, leaving the rest of the exon sequence intact. Removing the start codons alone increased translation to comparable or even greater levels, between 1.4-fold and 7.9-fold, indicating that the uORF is the dominant repressive element in these exons. For TSC1, the gene mutated in a subset of tuberous sclerosis complex cases, the team went further and showed that multiple uORFs act additively, with each additional upstream reading frame further depressing protein output from the main coding sequence.

The most clinically significant experiment involved splice-switching antisense oligonucleotides, shortened to SSOs. These chemically modified oligonucleotides bind pre-mRNA sequences such as splice donor or acceptor sites and trick the cellular splicing machinery into excluding a chosen exon. The researchers designed SSOs against the prioritised TSC1 5ʹUTR exon and demonstrated that treatment induced the intended exon skipping in cells and up-regulated levels of the endogenous TSC1 protein, produced from the cell’s own genomic copy rather than from a reporter construct. This is the decisive proof of concept: it shows that the strategy works not just on engineered sequences but on the authentic genomic context of a real disease gene, using a drug modality—antisense oligonucleotides—that already has an established clinical track record in spinal muscular atrophy, Duchenne muscular dystrophy and other conditions.

Splice-switching antisense technology brings genuine advantages for rare disease. Because ASOs can be designed rationally from the genomic sequence, a bespoke candidate can be developed quickly for an individual gene or even an individual patient, an approach that has already produced landmark personalised treatments for ultra-rare genetic conditions. The Oxford team’s systematic catalogue of 1,056 uORF-bearing skippable exons effectively converts that bespoke potential into a scalable pipeline: identify the haploinsufficient gene, check whether it carries a skippable 5ʹUTR exon with an active uORF, design an SSO, and titrate protein output back toward normal. The authors argue that their findings support the broad application of 5ʹUTR exon skipping as a general therapeutic mechanism for upregulating protein production from clinically relevant haploinsufficient genes.

Important caveats remain before patients benefit. The four of six prioritised exons that did not respond uniformly remind the field that uORF architecture varies, and that some 5ʹUTR exons may contain activating elements whose loss could be harmful. Delivery to the brain and other tissues, dose control, and the safety of chronically boosting dosage of tumour-suppressor-class genes such as TSC1 and CTCF will all require careful preclinical and clinical evaluation. The translational potential has nonetheless attracted commercial interest: several of the authors hold patents on uORF-targeting technologies and exon-skipping approaches licensed to a University of Oxford spin-out, and the work draws on genomic data from hundreds of thousands of participants in the UK’s National Genomic Research Library. If the approach clears those hurdles, the humble untranslated region—a stretch of RNA long treated as a footnote to the genetic code—may become one of the most promising drug targets in rare disease medicine.

Subject of Research: Using splice-switching antisense oligonucleotides to skip uORF-containing 5ʹ untranslated region exons and upregulate protein expression from haploinsufficient disease genes.

Article Title: Modulating splicing in 5ʹ untranslated regions to treat rare haploinsufficient disease

Article References: Beer Wells, E. S., De Conti, L., Kim, H. C., Rohani, N., Chundru, K., Svrzikapa, N., McClorey, G., Watts, L. M., Dawes, R., Chen, Y., Martin-Geary, A. C., Griffiths, M. J., Scott, S., Bamford, R. A., Wood, M. J., Roberts, T. C., Mill, J., Wright, C. F., Baralle, M., … Whiffin, N. (2026). Modulating splicing in 5ʹ untranslated regions to treat rare haploinsufficient disease. Genome Medicine. https://doi.org/10.1186/s13073-026-01773-0

Image Credits: AI Generated

DOI: 10.1186/s13073-026-01773-0

Keywords: rare disease, haploinsufficiency, antisense oligonucleotides, 5ʹ untranslated region, upstream open reading frames, splicing, therapeutic upregulation, TSC1, CTCF, GRIN2B, neurodevelopmental disorders, Genome Medicine

The research, led by Zihao Deng, Lifang He, and colleagues under corresponding author Yukun Cui at the Cancer Hospital of Shantou University Medical College, tackles a phenomenon known as lineage plasticity. This is the capacity of cancer cells to shed their original identity and adopt alternative cell states when pressured by therapy, a behavior increasingly recognized as a major driver of treatment resistance in prostate cancer and other tumors. In the androgen receptor-low or androgen receptor-null state, prostate cancer cells build extensive epigenetic barriers: their chromatin, the packaged form of DNA inside the nucleus, becomes inaccessible, and the three-dimensional architecture of the genome locks into a repressive topology that keeps the androgen receptor gene and its regulatory elements silenced. How that locking-in happens, and whether it can be reversed, has remained poorly understood.

KIF11, also known as kinesin family member 11, is best known as a mitotic kinesin, a motor protein that drives spindle assembly during cell division and has long been pursued as a chemotherapy target. Existing KIF11 inhibitors tend to work at cytotoxic doses that halt division in any rapidly proliferating cell. The Shantou team took a different angle. Combining public transcriptomic and epigenomic datasets with their own RNA sequencing, ATAC-seq chromatin accessibility profiling, RNA interference, low-dose pharmacological inhibition with the KIF11 inhibitor ispinesib, chromatin immunoprecipitation, chromosome conformation capture, DNA methylation assays, and reporter and therapeutic-response experiments, they asked what happens to prostate cancer cells when KIF11 activity is reduced gently rather than destroyed completely.

The answer surprised them. In androgen receptor-low and androgen receptor-null prostate cancer models, including PC3 and DU145 cells, either RNAi-mediated depletion or low-dose pharmacological inhibition of KIF11 activated androgen receptor-associated transcriptional programs. The cells began, in effect, to remember who they were supposed to be. Chromatin accessibility increased at androgen receptor regulatory regions, and the team measured higher occupancy of FOXA1 and GRHL2, two pioneering transcription factors that act as gatekeepers of the androgen receptor enhancer, at that key regulatory element. At the same time, the co-activator EP300 and the transcription elongation factor SUPT5H showed increased occupancy at androgen receptor regulatory elements, signaling that the locus was not merely open but actively engaged in productive transcription.

The mechanism went deeper than transcription factor recruitment. Using chromosome conformation capture techniques, the researchers showed that KIF11 inhibition reduced DNA methylation at CTCF motifs, the DNA sequences recognized by the CCCTC-binding factor, an architectural protein that anchors loops in the genome. With methylation lifted, CTCF occupancy rose, and the physical communication between enhancers and the androgen receptor promoter strengthened. In other words, KIF11 inhibition did not simply sprinkle activating marks on the genome; it rebuilt the three-dimensional wiring of the androgen receptor locus, constructing enhancer-promoter loops that bring distant regulatory elements into direct contact with the gene they control.

That topological resetting proved remarkably stable. Once the enhancer-promoter architecture was established, it locked in a sustained androgen receptor transcriptional program and created a feed-forward reinforcement circuit in which the reactivated pathway maintained its own expression. Functionally, this drove what the authors describe as a lineage return: androgen-insensitive, androgen receptor-low or null prostate cancer cells transitioned back into an androgen receptor-responsive transcriptional state. The change was not cosmetic. With the androgen receptor machinery restored, the cells regained the very dependency that antiandrogen drugs exploit.

The therapeutic consequences were tested across multiple model systems. In cell-line experiments, patient-derived castration-resistant prostate cancer organoids, and xenograft models in mice, low-dose KIF11 inhibition combined with enzalutamide or apalutamide produced robust growth suppression in tumors that would normally ignore these drugs. Crucially, the KIF11 inhibition used in these experiments was deliberately kept at low, non-cytotoxic doses, avoiding the broad toxicity that would come with hitting every dividing cell. The strategy is therefore conceptually distinct from conventional chemotherapy: rather than killing cancer cells directly, the treatment reprograms their regulatory architecture so that a second, well-tolerated class of drugs can work again.

The findings carry significant implications for how lineage plasticity-driven resistance might be overcome in the clinic. If a subset of patients with androgen receptor-low or null disease could be identified, perhaps through biopsy-based profiling of KIF11 levels, CTCF methylation status, or androgen receptor locus topology, they might become candidates for a re-sensitization strategy in which short-term KIF11 inhibition converts their tumors into androgen receptor-dependent ones before initiating or resuming antiandrogen therapy. The authors note that Y. Cui and Z. Deng have filed a patent application related to the findings, suggesting momentum toward translational development. At the same time, the work remains at the preclinical stage, and questions about dosing windows, patient selection, and safety in humans will need to be addressed in future studies.

Scientifically, the study adds an important dimension to a growing body of work showing that cancer is as much a disease of genome architecture as of genome sequence. The idea that inhibiting a mitotic motor protein could demethylate CTCF sites, reopen chromatin, reorganize enhancer-promoter loops, and ultimately restore a whole lineage identity illustrates how deeply intertwined cell division machinery, epigenetics, and three-dimensional genome organization really are. For patients with castration-resistant prostate cancer whose tumors have gone androgen-receptor silent, the research offers a conceptually elegant possibility: the door that cancer slammed shut can, with the right molecular key, be opened again, and the treatments already on the shelf may once more find their target.

Subject of Research: KIF11 inhibition as a lineage-remodeling strategy to restore androgen receptor signaling and antiandrogen sensitivity in castration-resistant prostate cancer

Article Title: KIF11 inhibition remodels lineage state and sensitizes castration-resistant prostate cancer to antiandrogen therapy via enhancer–promoter looping

Article References: Deng, Z., He, L., Li, J., She, S., Luo, X., Lin, J., Qi, Z., Zhuang, Y., Gao, X., Chi, Z., Xu, Z., Ji, Y., Liu, J., Lin, S. L., & Cui, Y. (2026). KIF11 inhibition remodels lineage state and sensitizes castration-resistant prostate cancer to antiandrogen therapy via enhancer–promoter looping. Molecular Cancer. https://doi.org/10.1186/s12943-026-02792-6

Image Credits: AI Generated

DOI: 10.1186/s12943-026-02792-6

Keywords: KIF11, castration-resistant prostate cancer, androgen receptor, lineage plasticity, enhancer-promoter looping, CTCF, epigenetics, enzalutamide, apalutamide, FOXA1, patient-derived organoids, drug resistance