Acylation is a diverse class of post-translational modification in which metabolites act as chemical donors that alter proteins. Beyond the canonical form of acetylation, emerging acylation marks are dynamically controlled by the availability of acyl donors and by “writer,” “eraser” and “reader” systems. The authors describe these systems as regulatory networks that can reshape chromatin, metabolism, protein stability and localization, cytoskeletal dynamics and immune responses.

The review integrates the chemistry and enzymatic control of major protein acylations with their roles across metabolic, epigenetic, signaling and tumor-immune networks. In this framework, acylation is not presented as one isolated pathway. Instead, it connects changing metabolic states with protein function and signaling, providing a possible basis for the adaptability and plasticity seen in cancer cells under stress.

The scope is a synthesis of existing evidence rather than a report of a new clinical trial or a demonstration of a precision treatment. Translational applications remain an important part of the review, but they are not presented as established clinical outcomes. The authors examine whether acylation patterns could support biomarkers or patient stratification and discuss pharmacological targeting and mechanism-guided combination strategies.

Major obstacles include the specificity of protein modifications and the need for clinical validation.

Subject of Research: Protein acylation in cancer metabolism, tumor plasticity and therapeutic response

Article Title: Protein acylation in cancer: from regulatory networks to precision therapeutics

Article References: Wei, Z., Ma, C., Gao, J., Zhang, X., Xu, Y., Ma, D., Liu, W., & Xing, X. (2026). Protein acylation in cancer: from regulatory networks to precision therapeutics. Journal of Experimental & Clinical Cancer Research. https://doi.org/10.1186/s13046-026-03857-6

Image Credits: AI Generated

DOI: 10.1186/s13046-026-03857-6

Keywords: protein acylation, cancer metabolism, post-translational modification, tumor plasticity, therapeutic resistance, biomarkers, precision oncology, epigenetic regulation, tumor immunity, drug targeting, Protein, acylation

The stakes could hardly be higher. Diffuse midline glioma, including diffuse intrinsic pontine glioma, accounts for roughly 10 to 20 percent of pediatric central nervous system tumors and is the leading cause of cancer-related death in children. Median overall survival is a mere 11 months, and fewer than 10 percent of patients survive beyond two years. For decades, radiotherapy has been the only standard-of-care option, offering transient benefit at best. In August 2025, the US Food and Drug Administration granted accelerated approval to dordaviprone, also known as ONC201, for previously treated patients aged one year and older with H3K27M-mutant disease, and preclinical work supporting combination with the PI3K/mTOR inhibitor paxalisib provided the rationale for the adaptive Phase II PNOC022 trial. Yet responses remain variable and often fleeting, raising the question that sits at the heart of the review: which molecular programs allow these tumors to adapt, evolve and resist?

The answer begins with the tumor’s origin. Diffuse midline glioma is defined by alterations in histone H3 lysine 27, most commonly the H3K27M mutation, or by overexpression of EZHIP. Both events converge on disruption of polycomb repressive complex 2, the cellular machinery that maintains repressive chromatin marks, producing a global collapse in H3K27 trimethylation and a profoundly dysregulated epigenetic landscape. The consequence is aberrant reactivation of developmental programs that should have been silenced after cells committed to a lineage. Single-cell studies show the tumor arises within a developmentally restricted glial progenitor context and recapitulates an oligodendroglial differentiation hierarchy, dominated by proliferative oligodendrocyte precursor cell-like states. This extensive heterogeneity and plasticity means therapy can select for, or actively induce, resistant cellular states that remain poorly characterized.

To bring order to this complexity, the authors organize tumor adaptation into four axes: compositional shifts in cellular populations, functional state remodeling, malignant cell plasticity, and intercellular communication networks. Each axis represents a different way a tumor can respond to treatment, and each demands its own computational approach. The framework matters because resistance is not a single phenomenon. A tumor may expand a pre-existing resistant population, reprogram its transcriptional programs without changing cellular identity, transition cells between differentiation states, or rewire the signaling conversations between malignant cells and their microenvironment. Disentangling these possibilities from high-dimensional single-cell and spatial transcriptomic data requires sophisticated mathematics and machine learning.

On the first axis, the review highlights methods that model differential abundance while accounting for the compositional and technical pitfalls that plague single-cell data. Because cell populations are measured as relative proportions, depletion of one group can masquerade as enrichment of another, and conventional clustering can force cells into discrete groups even when states exist along a continuum. Tools such as Milo represent cellular states as neighborhoods within a k-nearest neighbors graph, enabling detection of responsive intermediate states, while newer machine learning approaches like Cellograph use graph neural networks to identify treatment-responsive and resistant states across conditions. Spatially informed methods such as NicheCompass extend this to tissue architecture, relevant because in diffuse midline glioma, proliferating OPC-like and OC-like cells cluster within mitotic niches, while mesenchymal-like cells colocalize with vascular cells and SPP1-positive, GPNMB-positive myeloid cells in an immunosuppressive niche.

Functional state remodeling captures a subtler phenomenon: a cell population may persist after treatment while undergoing profound reprogramming. The H3K27M mutation drives extensive metabolic rewiring, creating both vulnerabilities and escape routes. Inhibiting oxidative phosphorylation with phenformin extends survival in preclinical models, but tumors adapt by activating alternative pathways including cholesterol synthesis. Similarly, these tumors depend heavily on purine metabolism, and after radiotherapy they upregulate the purine salvage pathway; targeting hypoxanthine-guanine phosphoribosyltransferase alongside radiation produces durable responses in mouse models. Computational tools such as RECODR can detect shifts in pathway activity and identify the transcription factors driving adaptation, while TransMarker and CancerTrace reconstruct dynamic regulatory networks. Spatial methods including C-SIDE and SPATA2 then correlate these programs with anatomical position, from tumor core to infiltrative margin.

Plasticity, the third axis, may be the defining feature of this disease. H3K27 alterations elevate α-ketoglutarate, a cofactor for histone demethylases, reinforcing the hypomethylated state. Dordaviprone disrupts mitochondrial homeostasis and alters α-ketoglutarate availability, partially restoring H3K27 methylation, proof that epigenetic states are therapeutically modifiable. To track how cells move between states, the review describes trajectory inference methods such as Monocle and Slingshot, which have revealed multiple OPC-like states in diffuse midline glioma and a resistance-associated trajectory in DIPG models, with resilient mesenchymal-like and mitotic populations emerging after triple therapy. Because pseudotime from static snapshots cannot establish directionality, RNA velocity approaches and deep generative models such as PRESCIENT, GRNvelo and CytoTRACE2 add temporal and potency information. Optimal transport methods including STIFT, Moscot and STORIES reconstruct cellular trajectories while preserving spatial context, potentially revealing whether resistant states pre-exist or are induced by therapy.

The fourth axis acknowledges that tumor cells do not adapt in isolation. The microenvironment of diffuse midline glioma is dominated by tumor-associated myeloid cells with sparse lymphocyte infiltration, and neuron-glioma connectivity is increasingly implicated in tumor genesis. Ligand-receptor inference tools such as CellChat and CellPhoneDB have exposed extensive tumor-myeloid communication, including the IGSF11-VISTA immune checkpoint axis. Newer frameworks push further: scComm uses supervised contrastive learning to reconstruct communication at single-cell resolution, CELLetter integrates large language models to predict novel ligand-receptor pairs, Tensor-cell2cell models interactions across multiple conditions simultaneously, and COMMOT and CellNEST exploit spatial proximity and graph attention to map signaling between neighboring cells. DualCellChat even infers directionality, identifying senders and receivers. These networks highlight actionable pathways that could be disrupted to restore antitumor immunity or enhance treatment efficacy.

Translating these models into the clinic remains formidable. True longitudinal sampling is largely unfeasible in patients, with matched specimens typically limited to diagnostic biopsies and post-mortem autopsies, confounded by cumulative therapy effects. Deep learning demands patient-level replication, careful separation of training and test data, and robust generalization across institutions. The authors point to preclinical models, organoids, patient-derived xenografts and liquid biopsies of blood or cerebrospinal fluid as bridges toward prospective modeling, and note that biomarkers such as TP53 mutation status, linked to radioresistance and reduced dordaviprone response, could stratify patients in adaptive trials. Foundation models like NicheFormer, pretrained on large single-cell atlases, have yet to be applied to resistance trajectories but represent a promising direction. If the field succeeds, the authors conclude, artificial intelligence integrated with scientific and ethical rigor could make diffuse midline glioma care predictive, personalized and proactive, a future in which resistance is anticipated before it emerges rather than discovered after treatment fails.

Subject of Research: Machine learning and computational modeling of therapeutic response and resistance in diffuse midline glioma using single-cell and spatial transcriptomics

Article Title: Machine learning and computational approaches to model therapeutic response and resistance in diffuse midline glioma

Article References: Dubois, O. A., Loughnan, O. E. L., Jiang, Y., Petraello, B., Douglas, A. M., McLachlan, T., Findlay, I. J., Vo, T., Savary, C., & Dun, M. D. (2026). Machine learning and computational approaches to model therapeutic response and resistance in diffuse midline glioma. Journal of Neuro-Oncology, 179(3), Article 104. https://doi.org/10.1007/s11060-026-05817-8

Image Credits: AI Generated

DOI: 10.1007/s11060-026-05817-8

Keywords: diffuse midline glioma, DIPG, machine learning, single-cell RNA sequencing, spatial transcriptomics, therapeutic resistance, H3K27M, tumor plasticity, tumor microenvironment, dordaviprone, computational biology, pediatric oncology

The phenomenon is called lineage infidelity, and it describes a form of cellular plasticity in which small-cell lung cancer cells transition between molecularly distinct states to evade both physiological pressures and therapeutic attack. Historically, the disease was viewed as a single, uniform neuroendocrine tumor. Modern genomic and transcriptomic profiling has dismantled that view entirely, revealing at least four major molecular subtypes defined by the dominant activity of key transcription factors: ASCL1 in the SCLC-A subtype, NEUROD1 in SCLC-N, POU2F3 in SCLC-P, and YAP1 marking a mesenchymal-like, non-neuroendocrine phenotype often called SCLC-Y. The status of YAP1 as a true lineage-defining factor remains debated among researchers, but its value as a marker of therapeutic escape is widely accepted.

Crucially, these states are not fixed. They represent fluid points along a phenotypic spectrum, allowing tumors to shift identities as conditions change. A tumor that begins as a highly neuroendocrine SCLC-A malignancy may transition toward a non-neuroendocrine or inflamed state when treatment targets its original vulnerabilities, effectively changing molecular camouflage mid-course. Single-cell RNA sequencing and genetically engineered mouse models have been instrumental in mapping these transitions, showing that they are governed less by new mutations and more by an epigenetic rheostat—a dynamic, reversible control system of gene expression built on chromatin remodeling and lineage-specific enhancers.

The mechanistic details are striking. The best-characterized switch occurs between the SCLC-A and SCLC-N states, both neuroendocrine but occupying distinct regulatory territories. Activation of NOTCH signaling induces the expression of HES1, which directly represses ASCL1, facilitating movement toward NEUROD1-positive or non-neuroendocrine phenotypes. In SCLC-A cells, ASCL1 acts as a pioneer factor keeping chromatin accessible at its target genes; when epigenetic co-factors such as LSD1 are perturbed, methylation marks at H3K4 and H3K9 change, effectively erasing the cell’s transcriptional memory and opening the door to alternative identities. MYC amplification can drive further progression, pushing cells from ASCL1-dominant states through NEUROD1-high states and ultimately toward YAP1-driven mesenchymal phenotypes.

Perhaps the most clinically consequential observation concerns the SCLC-I, or inflamed, subtype, which recent evidence suggests may be a common evolutionary endpoint for tumors treated with chemotherapy. This transition involves a global decrease in DNA methylation that activates endogenous retroviruses and triggers interferon signaling, supporting the shift toward an inflamed phenotype. Meanwhile, the tumor microenvironment supplies external cues that flip internal switches: hypoxia within the tumor core stabilizes HIF-1α, which downregulates neuroendocrine markers and promotes a migratory, mesenchymal-like state. Chemotherapy itself, beyond killing sensitive clones, induces stress responses that can push surviving cells into quiescent or variant states—raising the uncomfortable possibility that standard care may inadvertently steer tumors toward more recalcitrant identities.

But the review’s authors argue that this plasticity is not merely a survival strategy; it is also a weapon clinicians can turn against the tumor. Because every lineage transition opens a new window of vulnerability, they propose a ‘Push-and-Pull’ strategy of evolutionary steering. In the ‘Push’ phase, epigenetic modifiers such as LSD1 inhibitors destabilize the dominant lineage, inducing ASCL1-high neuroendocrine cells to lose their identity and transition toward a more inflamed or non-neuroendocrine state. In the ‘Pull’ phase, a second agent lethally targets the newly acquired state. If the Push produces an inflamed SCLC-I phenotype, the Pull would be immune checkpoint blockade, which is significantly more effective in that molecular context.

Subtype-specific vulnerabilities provide the ammunition for the Pull. Tumors transitioning from SCLC-A/N states toward SCLC-P develop a profound dependence on the DNA damage response, driven by replication stress from MYC or POU2F3 activity, making them hypersensitive to PARP and ATR inhibitors. Cells adopting a YAP1-high mesenchymal phenotype often upregulate the surface marker TROP2, and clinical trials are now investigating TROP2-targeted antibody-drug conjugates for patients whose tumors have undergone this transition—effectively converting a resistance mechanism into a delivery system for chemotherapy. Epigenetic agents such as HDAC or EZH2 inhibitors may even reset the chromatin landscape of resistant non-neuroendocrine cells, reverting them to a neuroendocrine state in which they regain sensitivity to original platinum regimens, a concept known as chemo-resensitization.

Implementing these strategies in the clinic is far from straightforward. Small-cell lung cancer rarely exists as a single pure subtype; high-resolution single-cell mapping shows that most clinical samples are composite tumors containing multiple subtypes in varying proportions. Treating the dominant clone often triggers explosive expansion of a pre-existing minor subclone, and distinguishing clonal selection from true transdifferentiation is essential, because the two demand different interventions. Real-time monitoring is therefore central to the entire framework: longitudinal analysis of circulating tumor DNA and circulating tumor cells can detect molecular signatures of a subtype switch—such as rising POU2F3 or YAP1 fragments, or shifts in subtype-specific DNA methylation patterns—months before imaging reveals tumor growth, allowing preemptive therapeutic pivots while disease burden remains low.

Pharmacological hurdles compound the challenge. HDAC inhibitors like vorinostat and EZH2 inhibitors like tazemetostat are FDA-approved for other malignancies, but their efficacy in small-cell lung cancer remains confined to early-phase trials. LSD1 inhibitors such as iadademstat have shown early promise but are not yet standard of care. Concurrent administration of epigenetic Push agents and cytotoxic Pull agents often produces prohibitive toxicities, including severe myelosuppression and gastrointestinal distress, because chromatin modifiers exert broad transcriptional effects. Sequential targeting mitigates toxicity but demands precise, validated liquid biopsy biomarkers that do not yet fully exist in clinical form. Overcoming these barriers, the authors suggest, will require more selective epigenetic tools such as PROTACs and dynamic trial designs that use ctDNA to trigger therapy switches before relapse becomes visible.

What emerges from this synthesis is a fundamentally new vision for treating one of medicine’s deadliest cancers. Rather than reacting to resistance after it appears, oncologists of the future may proactively direct tumor trajectories toward therapeutic dead ends—luring cancer cells into states where their acquired vulnerabilities become lethal traps. Preclinical platforms including genetically engineered mouse models, patient-derived xenografts, and organoids are already providing the testing grounds for such interventions, and precision immunotherapy approaches aim to extend the benefits of immune checkpoint inhibitors to cold tumors through epigenetic priming. If the capacity for change is the tumor’s greatest strength, the reviewers contend, it may also prove to be its ultimate undoing—transforming small-cell lung cancer from a recalcitrant malignancy into a disease that is manageable, and perhaps one day curable.

Subject of Research: Lineage infidelity and subtype plasticity driving chemoresistance in small-cell lung cancer

Article Title: Lineage infidelity in small-cell lung cancer: driving subtype transitions and acquired therapeutic vulnerabilities

Article References: Ajeh, I. J., Ikukpla’si, O. S. I., Bisoye, D. A., & Danraka, A. (2026). Lineage infidelity in small-cell lung cancer: driving subtype transitions and acquired therapeutic vulnerabilities. Clinical Cancer Bulletin, 5(1), Article 9. https://doi.org/10.1007/s44272-026-00061-7

Image Credits: AI Generated

DOI: 10.1007/s44272-026-00061-7

Keywords: small-cell lung cancer, lineage infidelity, ASCL1, NEUROD1, POU2F3, YAP1, epigenetic reprogramming, tumor plasticity, chemoresistance, Push-and-Pull strategy, liquid biopsy, immune checkpoint blockade

The research team, led by investigators at the First Affiliated Hospital of Xinjiang Medical University, combined single-cell RNA sequencing, spatial transcriptomics, CRISPR-based gene editing, pharmacological pathway modulation, and xenograft modeling to trace SOX9’s role across both time and tissue space. Using a publicly available single-cell dataset, they reconstructed a pseudotime trajectory with Monocle3, mapping the continuous transition from astrocytes to malignant glioma cells. The analysis revealed substantial cellular heterogeneity within glioma samples, encompassing malignant cells alongside astrocytes, macrophages, T cells, B cells, monocytes, neurons, fibroblasts, and endothelial cells.

The most consequential finding to emerge from the trajectory analysis was SOX9’s temporal behavior. Its expression was concentrated almost exclusively in astrocyte and malignant cell populations, and along the inferred developmental arc it peaked early and then declined steadily as cells matured into malignant states. Correlation analysis confirmed this downward trend, with Pearson and Spearman coefficients both strongly negative and highly significant. At the level of trajectory nodes, SOX9 dominated the early, astrocyte-rich segments and faded in the terminal malignant zones, a pattern inconsistent with the conventional view of SOX9 as a constitutively active stemness factor.

To probe what SOX9 might be doing during that early window, the researchers scored the activity of the Hippo pathway, a master regulator of organ size and stem cell fate whose downstream effectors YAP and TAZ are established drivers of glioblastoma plasticity. Using Gene Set Variation Analysis, they quantified an upstream kinase module, including MST1/2, LATS1/2, SAV1, MOB1A/B, NF2, and WWC1, and a canonical YAP/TAZ target module containing genes such as CTGF, CYR61, ANKRD1, AXL, and BIRC5. Both modules showed inverse relationships with SOX9 expression, and generalized additive modeling revealed that pseudotime and SOX9 each contributed independent, nonlinear effects on pathway activity. In other words, the SOX9–Hippo relationship was phase-dependent, strongest during early-to-intermediate stages of the transition and not a simple monotonic association.

Spatial transcriptomics added a second, geographic dimension to the story. Analyzing four anatomically distinct regions of glioblastoma tissue—the tumor–normal interface, the pure tumor core, the perivascular compartment, and the tumor–necrosis interface—the team mapped SOX9 expression against the probability that each spatial spot contained malignant cells. The coupling between SOX9 and malignancy was weak or unstable at the tumor edges and the necrotic margin, modest in the tumor core, and most robust in the perivascular niche. There, SOX9 expression was markedly elevated, correlations with malignant cell probability were strongest, and neighborhood enrichment and spatial autocorrelation statistics all confirmed significant co-localization.

This regional specificity is biologically meaningful. The perivascular niche has long been recognized as a reservoir for stem-like glioblastoma cells, bathed in vascular, hypoxic, and paracrine signals that nurture cellular plasticity. The findings suggest that SOX9-dependent reprogramming is not only time-restricted but niche-conditioned, with perivascular regions providing the most permissive anatomical context for effective SOX9–Hippo–YAP/TAZ coupling. Elsewhere in the tumor, downstream malignant programs may be sustained through alternative inputs, weakening the spatial coherence of the axis.

Functional experiments brought the correlation studies into the laboratory. Using lentiviral vectors, the team generated U87 glioma cells stably overexpressing SOX9 and used CRISPR/Cas9 to knock out the gene in U251 cells, validating the edits by Sanger sequencing and confirming a frameshift-inducing deletion in the knockout clone. SOX9 overexpression modestly increased proliferation, migration, and invasion while reducing apoptosis, whereas SOX9 knockout produced the opposite phenotype across wound-healing, Transwell invasion, and flow-cytometric apoptosis assays. Critically, treatment with XMU-MP-1, an inhibitor of the upstream Hippo kinases MST1/2, partially rescued the defects caused by SOX9 loss, linking SOX9 function experimentally to Hippo pathway state.

Phosphorylation-level Western blotting sharpened the mechanistic picture. SOX9 overexpression raised the ratio of phosphorylated to total YAP and lowered the phosphorylated-to-total MOB1 ratio, while SOX9 knockout produced the reciprocal pattern. XMU-MP-1 shifted both readouts toward the SOX9-overexpression signature, and total MOB1 remained unchanged across groups, indicating that the pathway rewiring was phosphorylation-dependent rather than a simple change in protein abundance. YAP and TAZ mRNA and protein levels rose with SOX9 gain and fell with SOX9 loss, and drug treatment partially restored them in SOX9-deficient cells, extending the transcriptomic associations to protein-level pathway readouts.

In vivo, the story held. Subcutaneous xenografts in nude mice showed that SOX9 overexpression significantly accelerated U87-derived tumor growth from day 14 onward, while SOX9 knockout markedly suppressed U251-derived tumors. XMU-MP-1 treatment further enlarged SOX9-overexpressing tumors and partially reversed the growth inhibition caused by SOX9 loss. Histopathology mirrored these dynamics: SOX9-overexpressing tumors displayed increased necrosis and nuclear atypia, whereas knockout tumors showed milder pathology, and the drug partially reversed both patterns. Ki67 immunohistochemistry confirmed the corresponding changes in proliferative activity, and CD68 staining revealed that myeloid and macrophage-like cell accumulation also shifted with SOX9 status, adding an immune dimension to the tumor microenvironmental effects.

Taken together, the study proposes what the authors call an early priming–late decoupling model. SOX9 acts early, at the moment of astrocyte-to-malignant conversion, to initiate Hippo–YAP/TAZ-linked malignant reprogramming. Once downstream transcriptional networks consolidate, the tumor becomes progressively less dependent on sustained SOX9 expression, which explains why stemness programs can persist in advanced disease even as SOX9 levels fall. The translational implication is pointed: therapies aimed at SOX9 may work best before malignant programs fully consolidate, whereas in later-stage tumors, blocking SOX9 alone may prove insufficient. The results argue for stage-specific and niche-aware strategies targeting the SOX9/Hippo/YAP–TAZ axis, particularly in the perivascular compartment where the axis is most strongly engaged. Limitations remain, including the reliance on public cohorts of limited size, established cell lines, and subcutaneous rather than orthotopic models, and the absence of YAP/TAZ nuclear localization data or a second pathway inhibitor. Even so, the study repositions SOX9 from a static stemness marker to a dynamic state-switch regulator, and it provides a conceptual framework for timing future interventions against one of medicine’s most stubborn cancers.

Subject of Research: The temporal role of SOX9 in priming Hippo–YAP/TAZ signaling during glioblastoma stemness acquisition

Article Title: Early SOX9 Activation Primes Hippo–YAP/TAZ Rewiring During Glioblastoma Stemness Acquisition

Article References: Mijiti, M., Li, Y., Maimaiti, A., Zhang, G., Lian, M., Song, C., Zhang, Y., Wang, J., Wang, Y., Chen, X., Kong, X., Xiaokaiti, X., Wusiman, N., Zhu, G., Geng, D., & Shi, H. (2026). Early SOX9 Activation Primes Hippo– YAP / TAZ Rewiring During Glioblastoma Stemness Acquisition. Journal of Cellular and Molecular Medicine, 30(17), Article e71341. https://doi.org/10.1111/jcmm.71341

Image Credits: AI Generated

DOI: 10.1111/jcmm.71341

Keywords: glioblastoma, SOX9, Hippo pathway, YAP/TAZ, glioma stem cells, pseudotime analysis, spatial transcriptomics, perivascular niche, XMU-MP-1, CRISPR, tumor plasticity, xenograft