The study, published in Frontiers in Cell and Developmental Biology, addresses a persistent gap in gynecological oncology. Endometrial carcinoma is the most common gynecological malignancy in high-income countries, and roughly eighty percent of patients are diagnosed with FIGO stage I–II disease. About eighty-five percent of these tumors are the endometrioid subtype. Current risk assessment relies on stage, histological grade, depth of myometrial invasion, lymphovascular space invasion, and molecular subtype assignment into POLE-mutant, microsatellite-instability, copy-number-low, and copy-number-high classes. Even with molecular classification, which has measurably improved risk assignment, clinically relevant uncertainty remains. Histological grading suffers from interobserver variability, lymphovascular space invasion is difficult to distinguish from morphological mimics, and molecular subtypes must still be interpreted alongside tumor architecture, spatial heterogeneity, and clinical context.

The new framework, formally named the prototype-resolved interpretable survival model, was built to integrate routine haematoxylin and eosin-stained whole-slide images with RNA-sequencing-derived pathway programmes for progression-free survival prediction. The technical design is distinctive. Diagnostic slides are cut into 256-by-256-pixel patches at an equivalent twentyfold magnification, and a pathology encoder converts each patch into a numerical embedding, producing five hundred image tokens per patient. On the molecular side, gene expression is filtered, log-transformed, standardized, and organized into 236 pathway-level tokens drawn from Hallmark and KEGG gene sets. The critical innovation is an adaptive prototype-learning bottleneck that compresses the hundreds of image tokens into just four learnable morphological prototypes — shared tissue states that recur across patients — before any cross-modal fusion occurs.

This compression matters because directly modelling the interactions between hundreds of image patches and hundreds of molecular pathways creates an enormous parameter space that is difficult to estimate reliably in disease-specific cohorts with limited matched data. By first distilling tissue appearance into four prototypes and then aligning those prototypes with pathway-level representations, PRISM reduces the multimodal fusion problem to a manageable size while retaining traceability back to specific regions of the slide. A pathway–prototype alignment module links transcriptional programmes to morphological states, and the fused representation feeds a Cox proportional-hazards output head that produces a continuous risk score, with higher scores indicating greater probability of progression, recurrence, or death.

The development process was deliberately rigorous. The model was trained on 229 patients from The Cancer Genome Atlas uterine carcinoma cohort, with a held-out internal validation set of 58 patients. Only after the architecture, hyperparameters, preprocessing workflow, and checkpoint-selection rule were completely locked was the model retrained on all 287 TCGA cases and evaluated exactly once, without any tuning, in an independent external cohort of 63 patients from the Clinical Proteomic Tumor Analysis Consortium. The CPTAC data played no role whatsoever in training, feature selection, or threshold optimization — a design feature that lends unusual credibility to the external results in a field where leakage between discovery and validation sets is a common criticism.

Across 350 total patients, PRISM achieved Harrell’s concordance indices of 0.770 in the TCGA training cohort, 0.751 in internal validation, and 0.745 in the external CPTAC cohort, indicating that risk ranking held up when transported to a different patient population, sequencing platform, and slide-scanning environment. At the thirty-six-month horizon, the censoring-adjusted time-dependent area under the curve reached 0.727 externally, compared with 0.685 for a baseline model combining clinicopathological variables and molecular subtype. In a complementary fixed-horizon analysis restricted to patients whose thirty-six-month event status was ascertainable, adding PRISM to the clinicopathological model raised the AUC in the external cohort from 0.609 to 0.719. In the TCGA evaluable subset, the corresponding jump was even larger, from 0.697 to 0.905 when molecular subtype was included. Ablation experiments reinforced the architecture’s logic: a version without the prototype bottleneck and an image-only model both performed worse and varied more across five independently generated data splits, while the image-only model collapsed to a concordance index of 0.584 in external validation.

Perhaps the most striking feature of the study is its interpretability. Because the model’s risk predictions pass through named morphological prototypes, a gynecological pathologist could review the patches assigned to each prototype and attach descriptive labels. Prototype one corresponds predominantly to stromal and fibromuscular tissue, prototype two to tumour epithelial and glandular morphology, and prototype three to hypercellular, pleomorphic tumour regions. A fourth prototype captured low-content and artefact-associated tissue edges and was excluded from biological interpretation. Spatial activation maps showed that the first three prototypes occupied distinct regions within individual slides, meaning the model’s risk estimate could be traced to recognizable tissue compartments. High-risk tumors showed lower stromal prototype burden and higher hypercellular prototype burden — a shift from supportive tissue architecture toward aggressive cellular morphology.

The molecular side of the interpretation proved equally coherent. Tumours classified as high risk by PRISM showed elevated pathway-programme scores for proliferation, cell-cycle activity, glycolysis, hypoxia, focal adhesion, and epithelial–mesenchymal transition, while low-risk tumors scored higher on metabolism- and homeostasis-associated programmes. These are precisely the biological themes that independent studies of endometrial cancer have implicated in aggressive behavior. The pathway–prototype alignment patterns were strongly concordant across the two cohorts, with a pooled Pearson correlation of 0.91, and within-prototype correlations ranged from 0.44 to 0.73. When the TCGA-derived risk cut-point was applied unchanged to the CPTAC cohort, Kaplan–Meier curves separated significantly, and the continuous risk score remained associated with progression-free survival after multivariable adjustment for age, grade, molecular subtype, invasion depth, radiation therapy, and residual tumour status.

The authors are candid about limitations. The external cohort contained only 63 patients and eight progression events, so the confidence intervals around external discrimination gains span zero, and the external calibration analysis showed attenuation and overprediction of absolute risk, making recalibration a necessary step before any clinical transport. The risk cut-point was selected and evaluated within the same TCGA cohort, rendering that analysis descriptive, and each patient was represented by a single diagnostic slide. The multivariable external association is explicitly framed as exploratory rather than definitive evidence of independent prognostic value. The model also requires transcriptomic input, which is not universally available at diagnosis, and the pathway scores and prototype assignments remain model-derived constructs that will need confirmation through spatial transcriptomics, single-cell sequencing, and functional studies.

Nevertheless, the work represents a meaningful proof of concept for prototype-guided morphomolecular risk modelling in a clinically relevant early-stage population. The locked external evaluation, repeated-split sensitivity analysis, and spatially traceable outputs provide a well-defined foundation for the larger, prospective, multi-institutional validation the authors say is warranted before deployment. More broadly, the study raises a question with implications well beyond endometrial cancer: which morphomolecular relationships — the recurring partnerships between how tissue looks and what its cells are doing transcriptionally — remain stable across institutions and tumor types, and which vary with lineage and context? If recurrent adverse programmes such as proliferation, hypoxia, and epithelial–mesenchymal transition are expressed through shared tissue states across cancers, a comparative atlas of prototype–pathway maps could improve model transportability everywhere and clarify, at last, how molecular risk is written into the architecture of tissue itself.

Subject of Research: Interpretable multimodal AI integrating histopathology and transcriptomics for progression-risk stratification in early-stage endometrioid endometrial carcinoma

Article Title: An interpretable multimodal morphomolecular model refines progression-risk stratification in early-stage endometrioid endometrial carcinoma

Article References: Li, J., Tang, Q., Chen, J., Chen, Q., Liu, B., Kong, Q., Ji, J., & Zhao, Q. (2026). An interpretable multimodal morphomolecular model refines progression-risk stratification in early-stage endometrioid endometrial carcinoma. Frontiers in Cell and Developmental Biology, 14, Article 1945984. https://doi.org/10.3389/fcell.2026.1945984

Image Credits: AI Generated

DOI: 10.3389/fcell.2026.1945984

Keywords: endometrial cancer, computational pathology, multimodal AI, risk stratification, transcriptomics, whole-slide imaging, prototype learning, progression-free survival, molecular subtyping, interpretable machine learning, TCGA, CPTAC

For years, molecular pathologists have focused on a dangerous duo. The BRAF V600E mutation, found in about 45 to 60 percent of papillary thyroid cancers, drives tumor growth through the MAPK signaling pathway. Somatic mutations in the promoter of the TERT gene, which encodes telomerase reverse transcriptase, are rarer but potent. When both alterations coexist, the prognosis is grim: landmark work by Xing and colleagues in 2014 showed that patients carrying both mutations had recurrence rates exceeding 68 percent, with a hazard ratio of 8.51 compared to patients without either mutation. The mechanism is elegant. BRAF-driven MAPK activation upregulates FOS, which activates the GABPA-GABPB complex that selectively binds the mutant TERT promoter, creating a feedforward loop of telomerase overexpression and cellular immortalization.

What happens when TERT promoter mutations appear alone, without BRAF? This question has remained surprisingly underexplored, and a team led by Bing Guan at Suzhou BenQ Medical Center in China set out to answer it using publicly available genomic data. Their approach was straightforward. They mined The Cancer Genome Atlas thyroid carcinoma cohort, identified 49 patients with both BRAF and TERT promoter mutation data plus progression-free interval endpoints, and stratified them into four groups: BRAF-negative/TERTp-negative, BRAF-positive/TERTp-negative, BRAF-negative/TERTp-positive, and BRAF-positive/TERTp-positive. They then compared survival outcomes using Kaplan-Meier analysis and Cox proportional hazards regression.

The results were striking. Overall log-rank analysis revealed significant differences in progression-free interval across the four subgroups, with a p-value below 0.001. The BRAF-negative/TERTp-negative group, comprising 33 patients or 67.3 percent of the cohort, showed no progression events at all, with median progression-free interval not reached. The BRAF-positive/TERTp-negative group, with 6 patients, also fared relatively well, with only 2 events and median progression-free interval not reached. But the two TERTp-mutant subgroups told a different story.

The most unexpected finding came from the BRAF-negative/TERTp-positive group. Despite containing only 5 patients, this subgroup showed the worst prognosis of all, with 3 of 5 patients experiencing disease progression, a 60 percent progression rate, and a median progression-free interval of just 5.7 months. In contrast, the BRAF-positive/TERTp-positive group had only 1 event among 5 patients, with a median progression-free interval of 17.5 months. Univariable Cox regression identified BRAF-negative/TERTp-positive status as the strongest adverse predictor, with a hazard ratio of 9.57 and a 95 percent confidence interval spanning 2.02 to 45.25, yielding a p-value of 0.004. This association remained significant after adjustment for age at diagnosis, with an adjusted hazard ratio of 8.38 and a p-value of 0.010.

These findings partially diverge from the prevailing paradigm. The established model holds that the synergistic combination of BRAF and TERT promoter mutations confers the worst prognosis, and the 2014 Xing study, which enrolled 507 patients with 35 dual-mutant cases and 24 months of median follow-up, provided adequate statistical power to detect that synergy. The current analysis, by contrast, found that the isolated TERTp group showed a numerically worse outcome than the dual-mutant group, though the authors are careful to emphasize that this should not be interpreted as contradicting the dual-mutation model. With only 5 patients per TERTp-mutant subgroup and a total of 6 progression events, the Cox models have limited statistical power, and the wide confidence intervals reflect this imprecision directly. The non-significance of the BRAF-positive/TERTp-positive group, with a hazard ratio of 2.88 and a p-value of 0.318, almost certainly reflects type II error rather than a true biological reversal of the established risk hierarchy. A single additional event in that subgroup would substantially alter the hazard ratio estimate.

Recognizing the fragility of small-sample Cox regression, the authors took additional precautions. Because the BRAF-negative/TERTp-negative reference group contained no progression events, standard Cox regression exhibited complete separation and could not produce finite estimates. The team therefore used ridge-penalized Cox regression with a penalizer of 0.1 for primary parameter estimation, and applied Firth’s penalized likelihood correction with a penalizer of 0.05 as a sensitivity analysis to assess small-sample bias. The Firth-corrected estimate yielded a hazard ratio of 16.93 with a 95 percent confidence interval of 2.95 to 97.17 and a p-value of 0.0015, remaining statistically significant and in the same direction as the primary estimate. This suggests that the qualitative conclusion of elevated risk associated with isolated TERT promoter mutation is not solely an artifact of small-sample bias, though the magnitude of the effect remains imprecisely estimated.

To provide biological context, the researchers turned to a second dataset. They analyzed GSE33630, a microarray dataset comprising 58 papillary thyroid carcinoma samples, of which 8 carried BRAF V600E mutations and 34 were BRAF wild-type. Differential expression analysis identified 42 genes with statistically significant differences between the two groups at a false discovery rate below 0.05, comprising 17 upregulated and 25 downregulated transcripts in BRAF-mutant tumors. Among the most notable findings, PCDH20, a protocadherin tumor suppressor that antagonizes Wnt/beta-catenin signaling and suppresses epithelial-mesenchymal transition, was downregulated with a log fold change of -1.54. HES1, a canonical Notch pathway target involved in thyroid differentiation, was also downregulated. TLR2, which plays a key role in innate immune responses, and LGALS3BP were similarly reduced, suggesting impaired immune surveillance in BRAF-mutant tumors. On the other side, DAPK2, a pro-apoptotic serine/threonine kinase, was upregulated with a log fold change of +1.14, possibly as a compensatory response to oncogenic MAPK signaling, though sustained MAPK activation can paradoxically uncouple DAPK2 from its apoptotic function and promote tumor survival.

The authors acknowledge important limitations. The TCGA cohort was not prospectively designed for survival analysis, and selection biases may affect the representation and follow-up of different molecular subgroups. Clinical management differences may also play a role: patients known to carry both BRAF and TERT promoter mutations may have received more aggressive initial therapy, including total thyroidectomy, radioiodine ablation, and external beam radiotherapy, which could partially attenuate observed event rates, while BRAF-negative/TERTp-positive patients may have been under-treated if the absence of BRAF mutation led to lower-risk clinical classification. Furthermore, the progression-free interval endpoint used in this study, standardized by the TCGA Pan-Cancer Clinical Data Resource, differs from the tumor recurrence endpoint used in the 2014 Xing study, and these endpoints are not directly interchangeable. The GSE33630 dataset, collected in 2012 before the identification of TERT promoter mutations, lacks TERTp annotation, preventing TERTp-stratified transcriptomic analysis.

Despite these caveats, the biological plausibility of the finding is compelling. TERT promoter mutations, predominantly the C228T and C250T substitutions, create de novo ETS transcription factor binding sites in the TERT promoter, leading to enhanced telomerase expression and telomere maintenance, which promotes cellular immortalization. In the absence of BRAF-driven MAPK activation, TERTp-mutant tumors may rely on alternative oncogenic cooperators, such as RAS mutations, receptor tyrosine kinase fusions, or epigenetic reprogramming, to drive aggressive behavior. Identifying these cooperating alterations represents an important priority for future research. The authors propose that routine TERT promoter sequencing in papillary thyroid carcinoma may have prognostic value even in the absence of BRAF mutation, and they call for validation in larger prospective cohorts with comprehensive molecular annotation. If confirmed, the finding could reshape how clinicians risk-stratify patients whose tumors lack the more familiar BRAF alteration, potentially catching a hidden high-risk group before disease progression begins.

Subject of Research: Prognostic impact of isolated TERT promoter mutation in papillary thyroid carcinoma

Article Title: Isolated TERT promoter mutation may define a high-risk prognostic subgroup in papillary thyroid carcinoma: an exploratory, hypothesis-generating analysis of TCGA and GSE33630 datasets

Article References: Guan, B., Zhou, R., Zhang, S., Chen, Y., Li, L., & He, P. (2026). Isolated TERT promoter mutation may define a high-risk prognostic subgroup in papillary thyroid carcinoma: an exploratory, hypothesis-generating analysis of TCGA and GSE33630 datasets. Frontiers in Endocrinology, 17, Article 1881969. https://doi.org/10.3389/fendo.2026.1881969

Image Credits: AI Generated

DOI: 10.3389/fendo.2026.1881969

Keywords: papillary thyroid carcinoma, TERT promoter mutation, BRAF V600E, TCGA, progression-free interval, Cox regression, GSE33630, differential gene expression, PCDH20, DAPK2, prognosis, molecular risk stratification

The study behind this map began with a deliberately simple design: twenty-one ovarian cancer cell-line samples and six control cell-line samples, each profiled in parallel for messenger RNA using bulk RNA sequencing and for microRNA using the NanoString platform. Because every sample contributed both types of measurement, the investigators could look for pairs of microRNAs and genes that moved in opposite directions across the same samples, which is the statistical signature expected when a microRNA represses a target transcript. Cell lines, of course, strip away much of the complexity of real tumors, including immune cells, fibroblasts, and endothelial cells, but they also remove confounding variation, making them a reasonable starting point for building a first-pass regulatory network before testing it against patient data.

The scale of the differential expression analysis illustrates how much molecular activity separates cancer cells from their normal counterparts. Nearly eight thousand genes crossed the initial statistical threshold at a false discovery rate below five percent, and after requiring a fold-change of at least two in either direction, almost three thousand genes remained as strict differentially expressed genes. On the microRNA side, forty significant candidates emerged, but the pattern was strikingly lopsided: thirty-five were reduced in the cancer samples while only five were increased. This asymmetry echoes a broader observation in cancer biology, where global attenuation of microRNA expression, sometimes traced to defects in the microRNA-processing machinery itself, is common across many tumor types and can remove multiple brakes on oncogenic programs simultaneously.

To convert these expression patterns into a regulatory network, the team applied conservative filters. Only microRNA–gene pairs showing a negative Spearman correlation stronger than negative 0.4 across matched samples, and surviving Benjamini–Hochberg correction for multiple testing, were retained as candidate edges. This yielded 4,614 inferred connections linking twenty-four microRNAs to 1,236 genes. The next step borrowed from network science: the Leiden community-detection algorithm partitioned the network into five modules, each a cluster of genes and microRNAs more densely connected to one another than to the rest of the graph. The authors are careful to note that the module numbers are arbitrary computational labels rather than rankings of biological importance, a distinction that matters when interpreting downstream results.

Functional annotation of these modules revealed that two of them carried coherent biological themes. One was enriched for genes involved in extracellular matrix organization and developmental processes, which is consistent with the well-known importance of matrix remodeling in ovarian cancer, a disease that spreads primarily by seeding the peritoneal lining rather than through the bloodstream. The other module, centered on miR-15a-5p, was enriched for angiogenesis-related genes. Within this module, miR-15a-5p was connected by inferred edges to 109 of the 119 member genes, an unusually broad reach that marks it as the dominant hub. The angiogenic leading-edge genes attached to this hub included ADGRG1, ANGPTL4, EREG, HTATIP2, ID1, PTK2, RAPGEF3, SYK, and TNFAIP2, several of which have established roles in blood vessel formation, hypoxic signaling, or focal adhesion dynamics.

The directionality of the findings is worth unpacking. In the discovery cell-line dataset, miR-15a-5p was downregulated while its module’s angiogenesis-linked genes were strongly upregulated, a pattern consistent with loss of microRNA-mediated repression unleashing pro-angiogenic transcripts. Yet when the investigators turned to 420 matched primary tumors from The Cancer Genome Atlas ovarian cohort, the Module 5 gene set showed a modest positive association with miR-15a-5p rather than the inverse relationship the cell-line model predicted. The authors interpret this honestly: the bulk-tumor signal likely reflects cellular composition, indirect regulation, or broader angiogenic tumor-state biology rather than simple one-to-one repression. Primary ovarian tumors are mixtures of malignant epithelial cells, stromal fibroblasts, immune infiltrates, and vasculature, and microRNA levels measured in bulk tissue can track the abundance of particular cell populations as much as any direct regulatory relationship. This discrepancy between cell-line inference and tumor-level correlation is a common and instructive feature of network studies, and it underscores why the authors frame their edges as hypotheses for mechanistic follow-up rather than proven interactions.

Beyond the angiogenic hub, the biomarker triage portion of the analysis shifted attention toward epithelial and tumor-associated genes. EPCAM, a canonical epithelial cell adhesion molecule long used as a marker of carcinoma cells; ESRP1, an epithelial splicing regulator; ST14, a protease involved in growth factor activation; TM4SF18, a tetraspanin-family membrane protein; RAB25, a small GTPase frequently implicated in ovarian cancer aggressiveness; and ADRA2C, an adrenergic receptor gene, emerged as priority candidates. Recurring regulatory connections involved three microRNAs: hsa-miR-140-5p, hsa-miR-15a-5p, and hsa-miR-381-3p. The recurrence of miR-15a-5p across both the angiogenic module and the biomarker-linked modules suggests it may sit at a junction between vascular and epithelial programs, though confirming this will require direct experimental perturbation.

Methodologically, the study demonstrates the value of integrating matched mRNA and microRNA measurements from the same samples rather than combining datasets profiled separately. Matched sampling reduces batch-related artifacts and allows correlation-based edge inference within a single biological context. The use of Leiden community detection, a refinement of earlier modularity-optimization algorithms, provides a scalable way to find densely connected subnetworks without requiring the number of modules to be specified in advance. Gene Ontology enrichment then serves as the bridge between abstract graph structure and interpretable biology, and the fact that two of five modules produced significant, coherent themes suggests the partition captured real signal rather than noise.

Several caveats temper the translational implications. The discovery cohort consisted of cell lines, which are typically grown in two-dimensional culture under supraphysiologic oxygen and nutrient conditions, environments that can distort angiogenic signaling pathways. The TCGA validation, while drawn from a large and well-characterized primary tumor cohort, relies on bulk transcriptomics that cannot resolve which cell type expresses a given microRNA or target gene. Single-cell RNA sequencing and spatial profiling of ovarian tumors, combined with microRNA measurement, would be needed to localize the miR-15a-5p program to specific cellular niches. Functional validation, such as restoring miR-15a-5p in ovarian cancer cell lines and measuring changes in the expression of ANGPTL4, EREG, or ID1, or assessing effects on endothelial tube formation in co-culture assays, would test whether the inferred edges carry causal weight. The miR-15 family has prior form in this regard: members of this family are known to regulate cell-cycle genes and have been studied as potential therapeutics in other cancers, so a pro-angiogenic role in ovarian cancer would extend an already rich literature.

The funding and disclosure details of the study are straightforward: the work was supported by the Ryan Foundation, conducted at the Molecular Diagnostic Research Laboratory at Boys Town National Research Hospital in Omaha, and used only commercially obtained cell lines, so no institutional review board approval was required. The authors also disclose that a large language model was used to improve text clarity and correct grammatical errors, with the authors retaining full responsibility for the content, a transparency practice that is becoming more common in scientific publishing.

For clinicians and translational researchers, the practical takeaway is a prioritized shortlist rather than an immediate diagnostic or therapeutic product. The convergence of network centrality, differential expression, and biomarker triage on a small set of microRNAs and epithelial genes provides a rational starting point for validation studies. Whether miR-15a-5p levels in tumor tissue, ascites, or circulating vesicles could serve as a prognostic or predictive marker, and whether restoring its activity would sensitize tumors to anti-angiogenic therapy, are questions the current study raises but cannot answer. What it does provide is a systems-level framework that connects a specific microRNA hub to a coherent angiogenic gene module and to a set of epithelial biomarkers, giving the ovarian cancer research community a structured map for the next round of hypothesis-driven experiments.

Subject of Research: Ovarian cancer miRNA–mRNA network analysis

Article Title: Integrated miRNA–mRNA network analysis identifies a miR-15a-5p–centered angiogenic program and biomarker-linked regulatory modules in ovarian cancer

Article References: Sarmiento, C., Tom, W. A., Jiang, C., Chandel, D. S., Thelagathoti, R. K., Krzyzanowski, G., Olou, A., & Fernando, M. R. (2026). Integrated miRNA–mRNA network analysis identifies a miR-15a-5p–centered angiogenic program and biomarker-linked regulatory modules in ovarian cancer. Journal of Ovarian Research. https://doi.org/10.1186/s13048-026-02275-8

Image Credits: AI Generated

DOI: 10.1186/s13048-026-02275-8

Keywords: ovarian cancer, miR-15a-5p, microRNA, angiogenesis, gene regulatory network, biomarkers, RNA-seq, TCGA, Integrated, miRNA, mRNA, network

The investigation began in the single-cell realm. Rather than grinding whole tumors into an averaged molecular signature, the researchers mined publicly available single-cell RNA sequencing data from 17 papillary thyroid carcinoma samples drawn from three independent public repositories. After rigorous quality control, they used a technique called Harmony to correct for the technical batch effects that arise when samples are processed in different laboratories, then annotated the tumor ecosystems cell by cell. This resolved the expected cast of characters: epithelial cells, fibroblasts, endothelial cells, pericytes, macrophages, T cells, NK cells, B cells, plasma cells, and rarer immune populations. The epithelial compartment, the seedbed of carcinoma itself, was then subdivided into ten subclusters for closer scrutiny.

Here the study deployed one of its most technically elegant moves: copy-number inference. Normal cells carry two copies of every chromosome, but cancer cells accumulate large-scale chromosomal gains and losses. Using the InferCNV pipeline, with macrophages serving as a genomically clean reference population, the team quantified these broad alterations in each individual epithelial cell. One subcluster, designated C6, showed copy-number disruption as severe as any malignant population, confirming its cancerous identity. Critically, this population also expressed high levels of CRABP1 while retaining classical epithelial markers such as CDH1, the gene encoding E-cadherin, and KRT19, a keratin characteristic of thyroid lining cells. In other words, CRABP1 did not mark an infiltrating stromal cell type; it marked a specific, bona fide malignant epithelial state.

Did that state matter clinically? To find out, the researchers turned to The Cancer Genome Atlas thyroid carcinoma cohort, the largest publicly available bulk transcriptomic dataset for the disease. Differential-expression analysis using DESeq2, with shrinkage of fold-change estimates to guard against noise, revealed that tumors with lymph-node metastasis expressed significantly less CRABP1 than node-negative tumors. CRABP1 levels also varied across pathological T categories, a measure of primary tumor size and local extent, and were lower in tumors than in adjacent healthy thyroid tissue. An exploratory gene set enrichment analysis in an external 157-patient cohort added texture: CRABP1-low tumors showed reduced enrichment for focal adhesion and tight junction pathways and elevated oxidative phosphorylation signatures. The authors are careful to frame these as unadjusted observational associations, but the direction is consistent and biologically coherent.

Correlation is cheap, so the team moved into the wet laboratory. Among seven thyroid cancer cell lines screened by quantitative PCR, HTC-C3, an anaplastic-derived line, expressed the most CRABP1, while FTC-238, isolated from a metastatic follicular carcinoma, expressed comparatively little. These became the complementary test beds. Silencing CRABP1 with small interfering RNAs in HTC-C3 cells made them visibly more mobile: in Transwell assays, cells must squeeze through 8-micrometer pores, and with a Matrigel coating mimicking basement membrane, they must also digest their way through. CRABP1-depleted cells crossed more of both barriers. Conversely, forcing CRABP1 overexpression in FTC-238 cells suppressed migration and invasion. The molecular fingerprint accompanied the behavior: CRABP1 knockdown lowered E-cadherin and raised N-cadherin, Vimentin, and Snail, a classic epithelial-mesenchymal transition signature, while overexpression reversed the pattern.

The in vivo evidence followed. The researchers injected control or CRABP1-depleted HTC-C3 cells into the flanks of immunodeficient BALB/c nude mice and tracked tumor growth for 24 days. Tumors lacking CRABP1 grew faster and ended heavier, and their tissue showed higher Ki-67 staining, a marker of cell proliferation, along with the same E-cadherin-low, mesenchymal-high gene expression shift seen in culture. CRABP1 depletion, in short, pushed thyroid cancer cells toward a more proliferative, more invasive, more mesenchymal identity inside a living animal, not just on plastic.

What silences CRABP1 in the first place? The team’s epigenetic analysis points toward DNA methylation, the chemical addition of methyl groups to cytosines that can dampen gene activity. Using the UALCAN portal on TCGA methylation arrays, they found that aggregated methylation of promoter-region probes near the CRABP1 transcription start site was higher in primary tumors than in normal thyroid, in both node-positive and node-negative disease. Probe-level mapping sharpened the picture: one CpG site, cg02276974, sits inside intron 2 of the CRABP1 gene on the south shore of a CpG island, and in 505 matched tumor samples its methylation was weakly but significantly inversely correlated with CRABP1 RNA abundance, with a Pearson correlation of minus 0.2069. The authors are explicit that this single site explains only about four percent of expression variation and cannot be declared a causal regulatory element, but it is a concrete, testable lead.

To probe that lead pharmacologically, they applied decitabine, a demethylating agent that inhibits DNA methyltransferases and is already approved in oncology for myelodysplastic syndromes. In both cell models, decitabine reduced migration, invasion, and Vimentin fluorescence intensity, and the strongest suppression appeared in CRABP1-overexpressing cells treated with the drug. In the xenograft setting, decitabine was associated with lower tumor weight and volume in the CRABP1-depleted background, alongside reduced Ki-67 and shifted EMT marker expression. Because decitabine acts genome-wide, these experiments establish a methylation-associated pharmacological context rather than proving that the drug specifically resurrects CRABP1 or unmethylates that particular intronic site. Notably, the animal design lacked a control-plus-decitabine arm, a limitation the authors candidly acknowledge.

The study also sketches the tumor’s social network. Using CellChat, a computational tool that predicts ligand-receptor crosstalk from transcriptomic data, the team found that CRABP1-positive malignant epithelial cells appeared poised to communicate with macrophage subpopulations through APP-CD74 and PPIA-BSG signaling axes. These predictions are hypotheses, not demonstrations of physical signaling, but they suggest that this malignant state may help sculpt its own inflammatory microenvironment, a question that now invites experimental testing with co-culture and spatial profiling approaches.

The implications ripple outward in several directions. Clinically, CRABP1 joins a growing list of markers that could eventually help separate tumors worth surgery from those suitable for active surveillance, an increasingly urgent distinction given evidence that a large fraction of diagnosed papillary thyroid carcinomas represent overdiagnosis. Scientifically, the work reframes a bulk-level gene finding as a single-cell state question, reminding researchers that a gene’s clinical signal may reflect the changing prevalence of a particular cell population within the tumor rather than uniform behavior across all cells. And pharmacologically, the decitabine results hint that demethylating therapy could blunt invasive phenotypes in methylation-driven contexts, though the bridge from nude mice to patients remains long and untested. The authors themselves draw careful boundaries: associations are associations, drug responses are drug responses, and no locus-specific epigenetic causality has been established. That restraint is precisely what makes the findings credible. What the study does deliver is a sharply defined malignant epithelial state, a plausible epigenetic handle on its regulation, and a concrete experimental agenda for turning CRABP1 from a molecular curiosity into a clinically useful signpost for thyroid cancer patients.

Subject of Research: CRABP1 downregulation and methylation in aggressive papillary thyroid carcinoma

Article Title: Methylation-associated downregulation of CRABP1 marks an aggressive malignant epithelial cell state in thyroid carcinoma

Article References: Fu, T., Li, Y., Huang, G., Hu, X., Luo, G., Wang, C., Wang, Y., Ding, Z., Zhou, J., & Li, Z. (2026). Methylation-associated downregulation of CRABP1 marks an aggressive malignant epithelial cell state in thyroid carcinoma. Frontiers in Cell and Developmental Biology, 14, Article 1974822. https://doi.org/10.3389/fcell.2026.1974822

Image Credits: AI Generated

DOI: 10.3389/fcell.2026.1974822

Keywords: CRABP1, papillary thyroid carcinoma, single-cell RNA sequencing, DNA methylation, decitabine, epithelial-mesenchymal transition, lymph node metastasis, copy number variation, TCGA, tumor microenvironment, retinoic acid signaling, cancer epigenetics

The research, published in Frontiers in Oncology, analysed a predefined 14-gene sulphur-handling panel across The Cancer Genome Atlas colon and rectal adenocarcinoma cohorts (TCGA-COAD and TCGA-READ) and six independent HG-U133 microarray validation cohorts comprising 1,337 tumours in total. The panel was fixed before any expression analysis and was not modified based on results, a design choice that strengthens the credibility of the findings. It included genes from four functional arms: the synthesis arm (CTH, CBS, MPST), the oxidation/clearance arm (SQOR, ETHE1, SUOX, TST), the cysteine/redox arm (SLC7A11, CDO1, CSAD), and the methionine/other arm (GOT1, GOT2, MAT2A, AHCY).

When the researchers compared tumours with adjacent-normal mucosa, they found that colorectal tumours did not show a uniform change in sulphur-handling capacity. Instead, the alterations were branch-specific. The oxidation/clearance arm showed the largest mean decrease, with SQOR, ETHE1, SUOX, and TST all coordinately downregulated. The synthesis arm declined modestly, with CTH reduced and CBS near-neutral and nonsignificant. The cysteine/redox arm showed a small positive mean, driven by a striking upregulation of SLC7A11 (log2 fold change of +2.57) and CSAD, partly offset by a marked reduction in CDO1 (log2 fold change of −2.37). AHCY, belonging to the methionine/other arm, was also upregulated.

This arm-resolved pattern is more consistent with selective metabolic reorganisation of cysteine and sulphur handling during malignant dedifferentiation than with a simple reduction in hydrogen sulphide biosynthetic capacity. The authors emphasise that transcript-level observations do not permit inference of enzyme activity or hydrogen sulphide flux, so all conclusions are framed at the level of the transcriptional programme and direction. The marked reduction in CDO1 expression recapitulates a well-established colorectal-cancer-associated alteration, frequently silenced by promoter methylation, and serves as an internal biological concordance check supporting the validity of the analytical pipeline.

To assess whether the sulphur-handling programme was associated with established colorectal cancer molecular identity, the team calculated a 14-gene programme score as the mean of within-cohort gene-wise Z-scores and compared scores across CMS1–4 tumours classified by CMScaller. Programme scores differed across CMS subtypes in TCGA-COAD (Kruskal–Wallis p = 6.20 × 10−20) and the independent GSE39582 cohort (p = 5.11 × 10−18). In both cohorts, CMS4—the mesenchymal subtype characterised by stromal and TGF-beta-associated biology—had the lowest programme scores, whereas CMS3, which retains stronger epithelial and metabolic features, showed comparatively higher scores.

Reproducibility was then assessed in the six independent HG-U133 cohorts. CMS4 tumours had lower programme scores than non-CMS4 tumours in every cohort. The pooled random-effects meta-analysis yielded a difference of −0.302 programme-score units (95% confidence interval, −0.37 to −0.23; p = 1.62 × 10−17; I² = 63%), confirming that the CMS4-low signal is a consistent feature across seven independent series rather than a cohort-specific artefact.

A critical concern for any programme-by-subtype association derived from bulk RNA sequencing is confounding by tumour purity and stromal or immune admixture, particularly because CMS4 tumours are stromal-rich. To address this, the researchers compared three complementary purity estimates: expression-signature-based ESTIMATE purity, multi-platform consensus purity estimates (CPE), and copy-number-based ABSOLUTE purity. ESTIMATE-based purity and CPE were correlated (Spearman ρ = 0.65), providing evidence of convergence between independent estimation methods. The CMS4-versus-rest programme difference remained negative after clinical adjustment and after adding ESTIMATE (p = 3.65 × 10−4), CPE (p = 2.60 × 10−9), or ABSOLUTE (p = 5.09 × 10−9) purity, indicating that measured purity differences do not fully account for the subtype association.

Single-cell deconvolution using BisqueRNA against a 2,500-cell donor-balanced reference derived from the GSE178341 single-cell colorectal cancer dataset showed compartment-fraction differences across CMS subtypes. CTH showed the largest compartment association with the CAF/stromal fraction (ρ = −0.22), with weaker associations for lymphoid, myeloid, other stromal, and tumour-epithelial fractions. As an independent cross-resource localisation plausibility check, the team processed 2,000 donor-balanced primary tumour cells from the 4.27-million-cell crc.icbi.at atlas, spanning 49 studies assembled independently of the deconvolution reference. The cross-resource Spearman correlation was ρ = 0.974, supporting the plausibility of the compartment-associated expression patterns, although the authors caution that this analysis tests concordance of aggregated localisation patterns between two single-cell resources and does not validate the accuracy of BisqueRNA fractions in individual tumours.

Across three independent survival endpoints—overall survival in TCGA-COAD, recurrence-free survival in GSE39582, and disease-free survival in GSE17536—the sulphur-handling programme score showed no statistically detectable association with outcome. In TCGA-COAD, the base model gave a hazard ratio of 1.17 (95% confidence interval, 0.68–2.01; p = 0.576), and the CMS-plus-purity model gave a hazard ratio of 1.10 (95% confidence interval, 0.54–2.21; p = 0.797). No individual gene survived Benjamini–Hochberg correction across the 14-gene panel. The authors report this as a clean null finding rather than a weak trend, and it does not support positioning the programme as an independent prognostic biomarker.

As a tertiary exploratory objective, the researchers performed a direction-aware transcriptomic-to-drug triage screen, joining TCGA-COAD tumour-versus-normal log2 fold-change values for all 14 panel genes to live DGIdb 5.0 drug–gene interaction records, OpenTargets tractability annotations, and high-confidence STRING v12 protein associations. Only SLC7A11 met all prespecified triage criteria: significant transcript change in a druggable direction, database-supported pharmacological modulation, and colorectal-cancer-specific preclinical evidence. Sulfasalazine and riluzole, as xCT-inhibiting tool compounds, have sensitised colorectal cancer models to cisplatin and radiotherapy through redox and glutathione disruption in preclinical settings, providing directional mechanistic support. However, no biomarker-selected colorectal cancer clinical efficacy data exist for either agent, and both have limited target specificity, so genetic perturbation and orthogonal chemotypes would be required to validate xCT dependence. CBS inhibition and MAT2A inhibition were retained only as conditional, literature-supported hypotheses given their transcript-neutral or untested status in this cohort. Conversely, the marked decreases in CDO1 and CTH argue against inhibitory strategies for those genes. The authors stress that all nominated candidates represent highly prioritised, data-driven candidates for prospective preclinical testing rather than direct clinical treatment recommendations, and that expression direction and network connectivity alone do not establish drug sensitivity, target dependency, or clinical benefit.

Subject of Research: Association between a sulphur-handling transcriptional programme and consensus molecular subtypes in colorectal cancer

Article Title: A predefined sulphur-handling transcriptional programme is associated with colorectal cancer molecular subtypes

Article References: Shafi, S., Hamid, A., Yang, J., Cheng, X., & Sun, M. (2026). A predefined sulphur-handling transcriptional programme is associated with colorectal cancer molecular subtypes. Frontiers in Oncology, 16, Article 1976885. https://doi.org/10.3389/fonc.2026.1976885

Image Credits: AI Generated

DOI: 10.3389/fonc.2026.1976885

Keywords: colorectal cancer, consensus molecular subtypes, sulphur metabolism, SLC7A11, transcriptional programme, tumour purity, TCGA, single-cell deconvolution, ferroptosis, cysteine metabolism, CMS4, bioinformatics

The core insight behind CIL is that three different claims are routinely conflated in cancer genomics: that a gene is associated with disease, that it is causally relevant to the disease process, and that it is a suitable target for therapeutic intervention. Differential expression may reflect reverse causation or shifts in the cellular composition of a tumour sample. A genome-wide association study signal may merely tag a nearby causal locus. An expression quantitative trait locus may operate in a proxy tissue rather than the tumour. And a gene dependency observed in a cell line may reflect general essentiality rather than a cancer-selective vulnerability. CIL forces researchers to state which type of claim each data modality actually supports.

The ladder itself has five rungs. Level I records disease association from expression, clinical, genomic or other omics data — a nomination tier that confers no causal status. Level II evaluates genetically constrained inference, split into inherited Level II-G evidence from germline regulatory variation and acquired Level II-S evidence from somatic mutation and copy-number alteration. Level III is divided into Level III-A, scalable computational functional evidence such as CRISPR dependency screens and pharmacogenomic correlations, and Level III-B, direct experimental perturbation with appropriate controls and rescue. Level IV reserves translational validation for higher-fidelity efficacy models, therapeutic-window assessment and clinical testing. Crucially, viral etiology — in cervical cancer, persistent high-risk HPV infection — is treated as an explicit context modifier rather than folded into the genetic evidence.

To demonstrate the framework, the team applied it to The Cancer Genome Atlas cervical cancer cohort, TCGA-CESC, comprising 306 tumours and only 3 normal samples. Because the control group was so small, the resulting 25-gene panel — 15 upregulated and 10 downregulated genes ranked by effect size — is explicitly exploratory rather than confirmatory. The upregulated genes told a coherent story of epithelial transformation: keratins KRT5, KRT6A and KRT15; cell-cycle drivers SFN, MYBL2, CDKN2A and CDC20; and the inflammatory alarmin pair S100A8 and S100A9. The downregulated genes were dominated by smooth-muscle and stromal cytoskeletal programs, including DES, CNN1, ACTG2, MYH11 and ACTA2, reflecting the loss of normal tissue architecture as tumour invades.

HPV status proved to be a powerful sensitivity axis. Sample-level annotations were available for 291 primary tumours, of which 280 were HPV-positive, 9 HPV-negative and 2 indeterminate. When the analysis was restricted to HPV-positive cases, all 25 Level I candidates retained their original direction of effect — a robustness result that matters in a cancer where the virus is the necessary etiological event. Exploratory comparisons between HPV-positive and HPV-negative tumours identified six genes with panel-level FDR-significant differences, including higher expression of CDKN2A, S100A9 and SERPINB5 in HPV-positive cases, though the authors caution that the tiny, histologically heterogeneous negative group prevents any claim of HPV-specific effects.

The somatic analysis delivered the study’s most concrete molecular findings. Among 295 copy-number-evaluable tumours, ten candidates showed false-discovery-rate-significant associations between copy number and expression, with CDKN2A standing out: one deep deletion, 58 shallow deletions, 52 gains and a single high-level amplification, coupled with a copy-number–expression correlation of ρ = 0.372 (FDR = 6.5 × 10⁻¹⁰). Because CDKN2A encodes the RB-pathway regulator p16, its somatic dosage must be interpreted jointly with the viral E7 oncoprotein’s inactivation of the same pathway. By contrast, recurrent protein-altering mutation was uncommon among the eight prioritized genes, and KRT5 and KRT6A showed no copy-number–expression association at all, marking them as epithelial and HPV-context candidates rather than CNA-driven ones.

The genetic audit was deliberately harsh. Each of the eight upregulated genes carried forward for Level II had documented cis-eQTL records from GTEx v8 and contextual GWAS annotations from the NHGRI-EBI catalog, yet no candidate possessed a complete, source-auditable cervical-cancer colocalization record at PP.H4 ≥ 0.80 plus a valid harmonized causal estimate. Mendelian randomization was deemed ineligible for this implementation because the retained records lacked aligned variant-level effect estimates. All eight genes therefore remain Partial Level II-G — a result the authors present not as failure but as an honest map of what is missing. Absence of germline support, they stress, is not evidence against somatic, viral or functional relevance.

Computational functional evidence fared similarly. CDC20 posted the strongest cervical-line CRISPR dependency, a Chronos score of −2.74, but pan-essentiality and the absence of model-level distributions kept it at Moderate Level III-A, prioritized for therapeutic-window testing rather than labelled a selective vulnerability. KRT6A’s weaker −0.55 score also remained Moderate. The remaining six candidates were classified as Context-only Level III-A: nominal drug-response correlations from GDSC and single-cell compartment annotations from CZ CELLxGENE guide experiment selection but do not constitute functional support. Notably, the myeloid localization of S100A8 and S100A9 reframes them as microenvironmental rather than tumour-cell-autonomous hypotheses, pointing toward tumour–myeloid co-culture and cytokine-blockade experiments instead of tumour-cell viability assays.

The framework’s most provocative demonstration is its reverse-entry workflow. Three clinically established cervical-cancer therapies — bevacizumab targeting VEGFA, validated in the randomized phase III GOG-240 trial; pembrolizumab targeting PD-1, validated in KEYNOTE-826; and tisotumab vedotin directed at tissue factor, validated in innovaTV 301 — were each retained at Level IV for their specific target-treatment relationships, with lower levels reported as not assessed rather than imputed. This preserves a boundary that clinical success often obscures: the efficacy of a checkpoint inhibitor does not establish that PD-1 is a tumour-cell dependency, and an antiangiogenic survival benefit does not prove that VEGFA initiated carcinogenesis. Therapeutic validation and disease causality are different claims, and CIL refuses to let one stand in for the other.

The practical payoff is a prioritized experimental agenda rather than a list of validated targets. For KRT6A, the most informative next study is inducible knockdown across several HPV-positive cervical models with cDNA rescue, epithelial-state markers and non-malignant cervical keratinocyte comparators. For CDC20, partial time-controlled suppression with rescue should probe the therapeutic window rather than whether complete loss impairs proliferation. For CDKN2A, perturbation should be stratified by RB-pathway state and evaluated alongside CDK4/6-inhibitor response. The authors also outline how a machine-readable CIL implementation — structured evidence records, explicit provenance and fixed decision rules exposed through an MCP-compatible service — could let AI clients query evidence and propose the next uncertainty-reducing analysis, while reserving tier promotion for expert review. The design limitations are stated plainly: three normal samples, proxy-tissue eQTLs, precomputed colocalization and no direct perturbation. But the ladder’s logic — that missing evidence is more informative than an assigned label — may prove the framework’s most transferable idea, from HPV-driven cervical cancer to EBV-related lymphomas and beyond.

Subject of Research: A multilevel evidence framework for evaluating therapeutic targets in cervical cancer

Article Title: The causal integration ladder: a multilevel evidence framework for therapeutic target evaluation in cervical cancer

Article References: Singh, S., Srivastava, M., Uppada, P., Ameta, D., Singh, A., Banerjee, M., & Kushwah, A. S. (2026). The causal integration ladder: a multilevel evidence framework for therapeutic target evaluation in cervical cancer. Frontiers in Systems Biology, 6, Article 1874355. https://doi.org/10.3389/fsysb.2026.1874355

Image Credits: AI Generated

DOI: 10.3389/fsysb.2026.1874355

Keywords: cervical cancer, causal inference, HPV, therapeutic targets, genomics, TCGA, systems biology, drug discovery, causal, integration, ladder, multilevel

The team began by mining the Comparative Toxicogenomics Database, a manually curated repository of chemical-gene interactions, to define a rigorous prior. After filtering the database’s acrolein records to direct human evidence and excluding co-exposure entries, they identified 179 unique primary target genes. These were organized into interpretive modules covering oxidative and electrophile stress, apoptosis and DNA damage, inflammation and immune signaling, barrier and epithelial-mesenchymal transition programs, and cell-cycle control. The researchers stress that these groupings are analytic conveniences rather than additional database assertions, preserving the traceability of the underlying literature evidence.

To test whether this curated target set is perturbed in colorectal cancer, the authors reanalyzed bulk RNA-sequencing data from The Cancer Genome Atlas, comprising 624 primary tumor profiles and 51 solid-tissue normal profiles across the COAD and READ projects. A count-aware differential expression analysis using the DESeq2 negative-binomial framework identified 5,594 genes passing a threshold of false discovery rate below 0.05 and an absolute log2 fold change of at least 1. Of the 179 curated acrolein targets, 155 were measurable in the bulk dataset and 43 met the differential-expression threshold, with 23 higher and 20 lower in tumors. The signal was strongly concordant between colon and rectal cancers, with a Spearman correlation of 0.91 across nearly 20,000 genes.

A central metric in the study is the target dysregulation burden, calculated for each sample as the mean absolute z score of measured acrolein targets relative to normal-tissue reference expression. This burden was substantially higher in tumors than in normal tissue, 1.96 versus 0.75, a difference of 1.21 with a Welch p-value of 3.62e-42. All five prespecified target modules showed positive tumor-minus-normal differences, with the largest shifts in barrier and EMT targets and cell-cycle control targets. The authors emphasize that this is an expression-based perturbation proxy, not a measurement of acrolein exposure in patients, since the transcriptomic data cannot reveal the source, dose, or timing of any actual chemical exposure.

The most provocative findings concern the tumor immune microenvironment. Within primary tumors, target burden correlated negatively with T-cell and endothelial expression programs, each with a Spearman rho of -0.27, and with myeloid, fibroblast-extracellular matrix, and antigen-presentation programs, each around -0.20. Only the cell-cycle program correlated positively, at rho of 0.20. When tumors were stratified into low, middle, and high target-burden quartiles, lower-burden tumors tended to show more positive immune-stromal scores than high-burden tumors. The authors are careful to frame these as correlations among standardized expression summaries rather than evidence of immune exclusion or stromal depletion, and they note that fibroblast and immune programs are biologically heterogeneous and cannot be reduced to a single pro- or antitumor phenotype.

To localize the perturbation at cellular resolution, the team turned to single-cell RNA-sequencing data from the public dataset GSE231559, processing six tumor and three normal libraries into 24,128 retained cells after quality control. Cells were assigned to eight marker-supported compartments: epithelial, T cells, NK cells, B and plasma cells, myeloid, fibroblasts, endothelial, and mast cells. Descriptive mean target scores were highest in myeloid cells at 0.455, fibroblasts at 0.424, and endothelial cells at 0.344, while the 154 measurable targets were detectable across all eight compartments rather than confined to a single lineage. Module-level patterns showed barrier and EMT scores particularly high in endothelial, epithelial, and fibroblast compartments, and inflammation and immune-signaling scores relatively high in myeloid and mast cells.

Comparing the bulk and single-cell scales for 154 shared targets yielded a modest but statistically detectable rank concordance of rho equal to 0.31 with a p-value of 6.95e-05. This partial agreement reflects genuine differences in platform, sampling structure, and inferential unit: the bulk analysis represents hundreds of patient-level profiles, whereas the single-cell component contains only nine libraries with unequal cell-type coverage and just three normal controls. The authors explicitly caution that the single-cell results support localization and heterogeneity descriptions, not claims of a unique cellular origin or a demonstrated intercellular signaling mechanism.

To add an experimental anchor, the researchers performed RT-qPCR in HCT 116 colorectal carcinoma cells exposed to 10 micromolar acrolein for six hours, following MIQE reporting guidelines and the comparative Ct method with GAPDH as the reference gene. Three biological replicates with three technical replicates each showed vehicle-normalized relative expression of 2.18-fold for NFKBIA, 1.83-fold for GSTP1, and 1.57-fold for VIM, all increased, while GPX4 decreased to 0.71-fold. This mixed directional response mirrors the broader conclusion that acrolein-linked targets do not behave as a uniformly shifted gene set, and it provides targeted validation for four prespecified genes without establishing a genome-wide or patient-level effect.

The authors are unusually explicit about the boundaries of their claims. The study does not measure acrolein or acrolein-derived protein adducts in the analyzed patient samples, cannot establish causal toxicological effects in human tumors, and does not demonstrate a specific signaling pathway. The target-burden proxy can be influenced by tumor composition and transcriptional state, the single-cell annotation uses broad marker-based labels that may not resolve finer functional subtypes, and the qPCR experiment covers only four genes in one cell line at a single concentration and time point. Dose-response experiments, independent cell models, protein and adduct measurements, spatial validation, and independent cohorts will all be needed before the biological significance of these associations can be determined.

Nevertheless, the study offers a template for how toxicogenomic evidence can be integrated with modern cancer transcriptomics. By anchoring a curated chemical-gene prior in population-scale bulk data, resolving its cellular localization with single-cell sequencing, and validating selected responses in vitro, the researchers have built a hypothesis-generating framework that links an environmental and endogenous carcinogen to the immune-stromal heterogeneity of colorectal cancer. Whether acrolein-driven perturbation ultimately proves to be a driver of immune evasion, a biomarker of tumor state, or a consequence of tumor metabolism, the analytical pipeline established here provides a reproducible path for answering that question in future mechanistic and clinical studies.

Subject of Research: Acrolein target gene perturbation and tumor-immune microenvironment heterogeneity in colorectal cancer

Article Title: Acrolein target-linked toxicological perturbation and tumor-immune microenvironment heterogeneity in colorectal cancer

Article References: Cheng, G., Shan, L., Shi, W., Liu, X., Tao, X., & Han, J. (2026). Acrolein target-linked toxicological perturbation and tumor-immune microenvironment heterogeneity in colorectal cancer. Frontiers in Cardiovascular Medicine, 13, Article 1929802. https://doi.org/10.3389/fcvm.2026.1929802

Image Credits: AI Generated

DOI: 10.3389/fcvm.2026.1929802

Keywords: acrolein, colorectal cancer, toxicogenomics, tumor microenvironment, single-cell RNA-seq, TCGA, Comparative Toxicogenomics Database, RT-qPCR, target-linked, toxicological, perturbation, tumor-immune

The study reanalyzed the landmark GSE131928/SCP393 dataset originally generated by Neftel and colleagues, comprising 24,131 single cells from IDH-wildtype glioblastoma profiled across two sequencing platforms: Smart-seq2, which contributed 7,930 cells across 28 tumor labels, and 10x Genomics, which contributed 16,201 cells across 9 tumor labels. Rather than treating these platforms as interchangeable, the researchers audited every expression matrix on its verified scale, transforming the 10x data from TPM-like linear values to a log2(TPM/10 + 1) scale to match the deposited Smart-seq2 format. This meticulous attention to data provenance matters because cross-platform integration is notoriously fragile, and small errors in normalization can produce spurious biological conclusions that ripple through an entire field.

At the heart of the analysis lies the now-canonical framework of malignant cell states in glioblastoma. Rather than existing as a single fixed subtype, glioblastoma cells occupy four recurrent transcriptional programs: astrocyte-like (AC-like), mesenchymal-like (MES-like), neural-progenitor-like (NPC-like), and oligodendrocyte-progenitor-like (OPC-like). These states coexist within individual tumors and shift dynamically, reflecting the remarkable plasticity that makes the disease so difficult to treat. The research team recomputed these four broad states from the official module definitions, carefully excluding 287 records whose state modules had been corrupted by gene aliasing, leaving 6,576 malignant cells with valid state scores as the denominator for all downstream analyses. It is the kind of unglamorous data hygiene that rarely makes headlines but fundamentally determines whether a computational result can be reproduced by other laboratories.

To bridge the two sequencing platforms, the researchers built a machine-learning label-transfer model using a grouped five-fold logistic regression in which entire tumor labels were held out as groups during training. This design prevents information leakage between tumors, a common pitfall in single-cell machine learning. The model achieved an out-of-fold accuracy of 0.998 with a macro F1 score of approximately 0.985, and only 141 cells, or 0.87 percent, remained unassigned. When the researchers compared malignant-state fractions between the Smart-seq2 and 10x arms across the six tumor labels shared by both platforms, they found Spearman correlations ranging from 0.66 to 0.83, indicating moderate-to-strong concordance. All 11,448 shared-label malignant pseudobulk gene pairs tracked closely along the diagonal, confirming that the two platforms tell a coherent story about the same biological reality.

But the study’s most consequential finding may be what it did not find. A persistent narrative in glioblastoma research holds that macrophages infiltrating the tumor drive malignant cells toward a mesenchymal-like state, creating a self-reinforcing loop of immune suppression and therapeutic resistance. When the researchers directly tested this hypothesis at the tumor-label level, the result was a null: the MES-like fraction showed no significant association with macrophage fraction, yielding a Spearman rho of just 0.171 with a P value of 0.395 across 27 tumor labels. An adult-only sensitivity analysis produced a similarly negligible result. This null finding does not disprove macrophage-tumor interactions, but it does sound a cautionary note for a field that has sometimes promoted bulk-level correlations into directional mechanisms without adequate evidence.

Extending the analysis to bulk transcriptomics, the team turned to the TCGA-GBM cohort, reducing 176 expression aliquots through 158 primary tumors and 143 primary IDH-wildtype libraries to 142 unique patients. Using fixed gene signatures with coverage between 0.91 and 1.00, they computed patient-level state scores and found that the MES score correlated strongly with both immune (Spearman 0.547, P < 0.001) and stromal (0.552, P < 0.001) ESTIMATE scores, while the myeloid program showed an even more striking correlation with immune signatures at 0.954. Conversely, NPC and OPC scores were inversely associated with immune and stromal estimates. These patterns are biologically plausible, but the researchers are careful to note that bulk RNA cannot resolve which cell type produces which transcript, and that tumor purity and compositional variation may partly drive the observed associations.

A score-defined differential expression contrast between patients with high versus low MES and immune program scores yielded 8,620 genes below a false discovery rate of 0.05, with FCGR2A showing the strongest effect at a log2 fold change of 1.417 and a Hedges g of 2.041. Hallmark pathway analysis led by Interferon Gamma Response (normalized enrichment score 3.990, FDR q = 0.001) confirmed the immune character of the high-score group. Yet the researchers explicitly label these analyses as descriptive rather than independent discovery, because the grouping variable is derived from the very expression programs under investigation. This kind of intellectual honesty, distinguishing score geometry from genuine biological discovery, is increasingly rare and increasingly necessary in the era of large-scale computational oncology.

Survival and prediction analyses further tempered any temptation toward clinical overreach. A fixed median split of the combined MES/immune score produced median overall survival of 12.0 versus 14.0 months with a log-rank P value of 0.663, far from statistical significance. Cox model estimates shifted depending on covariate specification, with the hazard ratio for the MES/immune score moving from 1.095 (P = 0.314) in univariable analysis to 1.372 (95% CI 1.019–1.848, P = 0.037) only after adding Karnofsky performance status, a change the authors attribute to specification sensitivity and reduced sample size. The best nested cross-validated model for predicting 12-month mortality, a combined elastic net incorporating clinical and signature features, achieved an out-of-fold ROC-AUC of just 0.587, with permutation importance led by age and the myeloid program. These are workflow benchmarks, not deployable biomarkers.

To ground the computational findings in experimental data, the researchers performed RT-qPCR on four glioma cell lines (U87MG, LN229, U251, and T98G) alongside normal human astrocytes as a non-malignant reference. Four genes representing malignant, mesenchymal/myeloid, and angiogenic programs were all significantly overexpressed in glioma cells relative to normal astrocytes. CHI3L1 showed the most dramatic elevation, reaching up to 12.8-fold in U87MG cells, while SPP1 and VEGFA were upregulated 6- to 9-fold and 5- to 8-fold respectively, and STAT3 showed consistent 4- to 6-fold upregulation. These results align with prior literature implicating STAT3 in tumor signaling, AXL-linked therapy resistance within mesenchymal programs, and SPP1-mediated macrophage-tumor crosstalk, though the authors emphasize that cell-line expression data cannot validate a tumor-associated macrophage-derived mechanism in living tissue.

The study’s ultimate contribution lies not in a single dramatic discovery but in the framework it establishes. By demonstrating that established malignant cell states can be recovered reproducibly across sequencing platforms, that immune-associated expression programs covary with microenvironmental signatures in bulk tissue, and that popular mechanistic narratives require perturbational and spatial evidence before they can be accepted as causal, the researchers have provided a template for how integrated single-cell and bulk transcriptomics should be conducted in the era of reproducible computational biology. Independent cohorts, spatial profiling, paired longitudinal specimens, and functional perturbation experiments will be needed to convert these descriptive associations into actionable biology, but the foundation is now considerably more solid than it was before this analysis began.

Subject of Research: Machine-learning integration of single-cell and bulk transcriptomics to characterize glioblastoma malignant cell states and immune-associated tumor ecosystem programs

Article Title: Machine-learning-assisted integration of single-cell and bulk transcriptomics characterizes glioblastoma ecosystem states and immune-associated programs

Article References: Cao, J., Cheng, H., Wang, Y., Zhao, B., Wang, Y., & Jiang, X. (2026). Machine-learning-assisted integration of single-cell and bulk transcriptomics characterizes glioblastoma ecosystem states and immune-associated programs. Frontiers in Cell and Developmental Biology, 14, Article 1967483. https://doi.org/10.3389/fcell.2026.1967483

Image Credits: AI Generated

DOI: 10.3389/fcell.2026.1967483

Keywords: glioblastoma, single-cell RNA sequencing, machine learning, tumor microenvironment, macrophages, mesenchymal state, TCGA, RT-qPCR, label transfer, immune programs, transcriptomics, IDH-wildtype

The methodological core of the study is a deliberate corrective to one of the most common pitfalls in modern single-cell genomics. Analyses that treat thousands of cells as thousands of independent observations can manufacture convincing-looking differences out of individual quirks, a phenomenon known as pseudoreplication. The team, led by Yalan Mao, Ran Li, Bin Li, and Hongfen Pan, sidestepped this trap by aggregating raw counts into donor-by-cell-type pseudobulk profiles and applying formal statistical tests only at the donor level. GSE241302, with three membranous nephropathy donors and a single internal normal reference, served as a complete-case exploratory discovery set. OEP004186, with eleven patients and seven controls, provided the principal donor-aware support. GSE171458 offered directional replication despite its tiny control group, and GSE200828 supplied tissue-level disease context across 153 glomerular samples. Leave-one-donor-out and leave-one-gene-out sensitivity analyses were used to confirm that the signal did not hinge on a single patient or a single highly weighted gene.

In the discovery dataset, an unbiased pathway screen across non-podocyte compartments repeatedly surfaced cholesterol programs. Every one of the three membranous nephropathy donors showed positive enrichment for cholesterol-efflux regulation and for the NR1H3/NR1H2-regulated cholesterol transport and efflux pathway in tubular epithelial cells. The team then froze a 47-gene program anchored in Liver X receptor biology, a family of sterol-sensing nuclear receptors that heterodimerize with retinoid X receptors and coordinate cholesterol transport, efflux, and inflammatory signaling. The program spans cholesterol exporters ABCA1 and ABCG1, lipid synthesis and remodeling genes such as SREBF1, FASN, and SCD, and LXR/RXR-associated regulatory components including RXRA, RXRB, and several Argonaute-family genes. Crucially, the score behaved as a continuous, distributed state across tubular cells rather than marking a small discrete cell cluster.

The formal donor-aware test in OEP004186 produced the study’s strongest results. Program scores rose in proximal tubule cells with an effect of +0.480 (false-discovery rate 0.0128), in the injury-associated PT-VCAM1 subtype with +0.348 (FDR 0.0165), and in loop-of-Henle and distal epithelial cells with +0.310 (FDR 0.0242). When each of the eleven patients was removed in turn, the effect stayed positive in every single iteration across all three compartments. Collecting duct cells showed a smaller positive trend that failed to reach significance, and the cross-dataset picture confirmed this anatomical gradient: the state was consistently elevated in proximal tubule and loop/TAL-related compartments across all three single-cell resources, but not reproducibly in the distal nephron. The anatomy makes intuitive sense, since proximal tubule cells handle the heaviest filtered-protein load and are metabolically demanding cells dependent on fatty-acid oxidation.

A key control ruled out a deceptive explanation. PT-VCAM1 cells are often read as an injury-associated proximal tubular state, so the observed elevation could conceivably have simply reflected more of these cells being captured in diseased kidneys. The data say otherwise: the PT-VCAM1 fraction was numerically lower in membranous nephropathy, not higher, with a nonsignificant logit-scale effect of -0.688. Meanwhile, the cholesterol-handling program remained significantly elevated within both conventional PT cells and PT-VCAM1 cells themselves. The signal is therefore a transcriptional change superimposed on existing tubular states rather than a compositional shift in cell populations. Sixteen program genes, from ABCA1 and ABCG1 through SREBF1, FASN, SCD, RXRA, and RXRB, showed positive average effects across all three single-cell datasets, reinforcing the distributed, multigene character of the finding.

The authors were careful about one label above all: LXR-linked, not LXR-activated. Transcription-factor activity inferred from the CollecTRI regulatory network showed nominally positive shifts for NR1H3 in proximal tubule and loop/distal compartments, but none of these survived correction across the full transcription-factor testing space. Similarly, the paired increase of both cholesterol-efflux genes and lipid-synthesis genes is more consistent with broad lipid-homeostasis remodeling than with a straightforward boost in net cholesterol export. The study explicitly does not claim direct LXR activation, increased cholesterol-efflux flux, or disease specificity, and the tissue-level analysis drew a sobering boundary: in GSE200828, the program was higher in membranous nephropathy than in living donors (+0.226, FDR 0.0349) but not higher than in focal segmental glomerulosclerosis, minimal change disease, or other glomerular diseases, suggesting a shared response to proteinuric injury rather than a disease-exclusive signature.

The bridge from computational pattern to laboratory experiment came through HK-2 cells, a human proximal tubular epithelial line. When the researchers flooded cultures with fatty-acid-free bovine serum albumin to mimic protein overload, all five representative program genes rose together: ABCA1 increased 1.66-fold, ABCG1 1.42-fold, SREBF1 1.33-fold, FASN 1.71-fold, and SCD 1.37-fold, with Holm-Sidak-adjusted P values between 0.0058 and 0.0170 across three independent experiments. The direction was coherent across both functional arms of the program, providing expression-level concordance between patient data and a controlled protein-stress model. The authors are quick to note the limits: the experiment measured mRNA rather than transporter protein, intracellular cholesterol, lipid droplets, or actual efflux, and fatty-acid-free albumin does not reproduce the full molecular cargo of urinary proteins, which is known to influence the severity of tubular injury.

Then comes the twist that earned the study its unusual title. The five genes experimentally induced by protein overload, ABCA1, ABCG1, SREBF1, FASN, and SCD, were evaluated in 129 paired tumor-adjacent-normal samples from The Cancer Genome Atlas, spanning clear cell renal cell carcinoma (72 pairs), papillary renal cell carcinoma (32 pairs), and chromophobe renal carcinoma (25 pairs). The panel rose in all three histologies, with a restricted-maximum-likelihood pooled effect of +0.967 (95 percent confidence interval 0.843 to 1.092), and 85.3 percent of patients showed simultaneous increases in the cholesterol-efflux arm and the lipid-synthesis/remodeling arm. A common latent factor explained 45.2 percent of the five-gene variance and kept uniformly positive loadings across 2,000 histology-stratified bootstrap samples, while joint five-dimensional shifts were significant in every histology after multiple-testing correction.

The tumor analysis, however, is framed strictly as cross-disease context rather than validation. The paired tumor-normal design cannot establish tumor-cell-intrinsic LXR activation, immune effects, somatic-mutation dependence, or any causal role, and individual gene effects varied by histology. What it does establish is that this experimentally anchored readout behaves as a coordinated multigene lipid-remodeling state across distinct renal pathologies, which may make it a useful molecular readout for future tumor microenvironment and treatment-response studies in renal cancer.

The overall picture that emerges is of a kidney under protein stress mounting a coordinated lipid-regulatory response that no single gene can capture, and of a research team determined to characterize it without overclaiming. Whether this tubular cholesterol-handling state protects injured nephrons, marks an early step toward maladaptive lipid accumulation, or simply tracks the burden of filtered protein remains unresolved, and the authors call for direct measurements of transporter localization, LXR nuclear signaling, cholesterol pools, and efflux capacity, ideally in primary human tubular cells, organoids, or ex vivo tissue. They also emphasize that future disease comparisons must include other proteinuric conditions rather than healthy tissue alone, given the shared nature of the signature. What the study does deliver is a rigorously donor-supported map of where and how tubular lipid biology responds to one of nephrology’s most burdensome symptoms, and a striking demonstration that the same five-gene pattern echoes from inflamed kidney tubules to renal tumors, a convergence that now invites mechanistic follow-up on both fronts.

Subject of Research: A donor-aware multi-cohort single-cell study of tubular cholesterol-handling transcriptional states in membranous nephropathy and renal tumors.

Article Title: Donor-aware multi-cohort single-cell analysis identifies an LXR-linked tubular cholesterol-handling state in membranous nephropathy with a coordinated five-gene lipid-remodeling signature across renal tumors

Article References: Mao, Y., Li, R., Li, B., & Pan, H. (2026). Donor-aware multi-cohort single-cell analysis identifies an LXR-linked tubular cholesterol-handling state in membranous nephropathy with a coordinated five-gene lipid-remodeling signature across renal tumors. Frontiers in Genetics, 17, Article 1916595. https://doi.org/10.3389/fgene.2026.1916595

Image Credits: AI Generated

DOI: 10.3389/fgene.2026.1916595

Keywords: membranous nephropathy, single-cell RNA sequencing, liver X receptors, cholesterol handling, proteinuria, proximal tubule, HK-2 cells, lipid remodeling, renal cell carcinoma, TCGA, ABCA1, pseudobulk analysis

The paper, authored by Manyata Srivastava, Shishir Singh, Diksha Ameta, Aesha Singh, Monisha Banerjee and Atar Singh Kushwah, proposes what the team calls the Immunogenicity Control Axis, or ICA. The central idea is deceptively simple: a tumor’s visibility to the immune system is an emergent property that arises only when three molecular layers are coordinated. Epigenetic regulation determines whether immune-relevant genes can be transcribed at all, DNA damage response pathways shape the mutational substrate from which neoantigens are generated, and the immune axis reflects the downstream cytolytic and interferon-driven state that results. Break the chain at any layer, the framework predicts, and even an antigen-laden tumor can remain immunologically invisible.

The epigenetic layer is the framework’s upstream boundary condition. Histone deacetylases and DNA methyltransferases control chromatin accessibility at promoters and enhancers governing MHC class I and II antigen presentation and interferon-responsive signaling. A repressive chromatin state can silence these programs entirely, which means a tumor may harbor abundant immunogenic mutations yet never display the resulting peptides to cytotoxic T cells. The authors point to converging evidence, including studies showing that epigenetic therapy can reverse immune evasion in lung cancer, that chromatin accessibility precedes transcriptional activation of immune genes, making it a plausible causal starting point rather than a mere correlate.

The DNA damage response axis supplies the raw material. Double-strand breaks can be repaired through homologous recombination, classical non-homologous end joining or alternative end joining, and the balance among these pathways differs in fidelity and genomic consequence. The team hypothesizes that repair-pathway configuration contributes to variation in the genomic substrate available for immune recognition, with error-prone end-joining potentially generating distinct spectra of candidate neoantigens. Crucially, the authors are careful not to overclaim: they treat relationships between expression-defined repair states and mutational signatures such as SBS3, linked to homologous-recombination deficiency, and the APOBEC-associated SBS2 and SBS13 signatures, prominent in cervical cancer, as exploratory associations rather than evidence that a specific repair pathway directly forges any particular immunogenic phenotype.

What elevates the paper above a purely conceptual essay is its insistence on quantitative falsifiability. Each axis is operationalized as a standardized score derived from single-sample gene-set enrichment analysis of prespecified transcriptional programs: permissive chromatin and antigen-presentation features contribute positively to the Epigenetic score, the DDR score captures the relative transcriptional shift toward end-joining versus homologous-recombination programs, and the Immune score aggregates cytolytic activity, checkpoint expression and interferon signatures. The composite ICA Score is then computed as the arithmetic mean of the three standardized components, an equal-weight formulation the authors describe deliberately as a transparent, outcome-independent reference model rather than a claim that the three biological systems contribute equally.

Cervical cancer serves as the initial testing ground, and for good reason. HPV-driven tumors coexist with viral antigens that should provoke immune recognition, yet many remain immunologically cold, a paradox the ICA framework attributes to upstream constraints: epigenetic repression can mute antigen presentation despite antigen availability, while repair-pathway bias may yield suboptimal mutation profiles. Using 304 evaluable TCGA-CESC cases, the team dichotomized patients by median ICA Score and found no statistically significant difference in overall survival between high- and low-scoring groups, a negative result they report transparently as exploratory benchmarking rather than validation. In a separate three-year survival classification analysis, however, the ICA composite model achieved an internally cross-validated AUC of 0.931, outperforming TMB plus PD-L1 (0.890), TMB alone (0.869) and PD-L1 alone (0.823), although the authors stress these figures represent in-cohort cross-validation, not independent clinical validation.

The framework extends into a perturbational layer that will likely attract the most attention. Differential-expression analysis between ICA-high and ICA-low tumors generates a bidirectional query signature, which the team matched against compound-induced transcriptional profiles in the LINCS L1000 Connectivity Map resource. Natural compounds whose signatures transcriptionally oppose the ICA-associated state, and whose curated targets span multiple ICA dimensions, are ranked as candidate multi-axis perturbators. Compounds such as curcumin and resveratrol, previously associated with modulation of DNMT and HDAC activity, exemplify the class of agents the framework is designed to evaluate, since their multi-target profiles may in principle touch the epigenetic, DDR and immune layers simultaneously rather than acting on a single isolated node.

The authors are unusually explicit about the epistemic status of every component. Consensus clustering of ICA features, UMAP visualizations of molecular states and even a synthetic demonstration matrix of 30 genes across 80 simulated profiles are all labeled as computational proof of principle, not evidence of biological subtypes. Network connectivity between a compound and an ICA protein is prioritization evidence, not target engagement. The roadmap to validation is staged: molecular reproducibility in independent multi-omics cohorts first, systems-level interrogation of inter-axis relationships second, and clinical benchmarking against PD-L1, TMB and clinicopathological variables in treatment-annotated cohorts only at the final stage, with failure at any step triggering refinement rather than reinterpretation.

The limitations are real. The ICA Score’s component axes are transcriptional and methylation-based correlates, not direct measurements of chromatin accessibility, repair-pathway flux or immune-cell function, and bulk transcriptomic deconvolution methods such as EPIC, xCell and MCP-counter infer rather than measure immune infiltration. The framework’s relative weights may need recalibration across tumor types; in low-mutational-burden cancers, mutation-derived antigenicity may matter less, while hematologic malignancies may require fundamentally different formulations. Prospective validation in independent clinical cohorts, particularly those with documented immunotherapy exposure and response endpoints, remains the decisive test.

Still, the conceptual contribution is difficult to dismiss. By proposing that tumor immunogenicity follows a Regulatory Convergence Principle, in which epigenetic accessibility, repair-pathway bias and immune signaling must align for effective immune visibility, the ICA framework reframes immunotherapy resistance as a systems-level regulatory problem rather than a failure of any single biomarker. It offers the field something it badly needs: a falsifiable, computationally tractable model that connects biological hypothesis to quantitative evaluation, and that specifies in advance what evidence would support or refute it. Whether coordinated multi-axis information truly captures what PD-L1 and tumor mutational burden miss, the coming validation cohorts will tell.

Subject of Research: A multi-omics framework integrating epigenetic regulation, DNA damage response and immune state to explain tumor immunogenicity and immunotherapy response variation

Article Title: The immunogenicity control axis: a multi-omics framework for integrating epigenetic, DNA damage response, and immune states in cancer

Article References: Srivastava, M., Singh, S., Ameta, D., Singh, A., Banerjee, M., & Kushwah, A. S. (2026). The immunogenicity control axis: a multi-omics framework for integrating epigenetic, DNA damage response, and immune states in cancer. Frontiers in Genetics, 17, Article 1944709. https://doi.org/10.3389/fgene.2026.1944709

Image Credits: AI Generated

DOI: 10.3389/fgene.2026.1944709

Keywords: immunogenicity control axis, tumor immunogenicity, multi-omics, epigenetics, DNA damage response, immune checkpoint blockade, cervical cancer, TCGA, neoantigens, natural compounds, biomarkers, immunotherapy resistance