The motivation behind the study lies in a practical bottleneck of modern oncology. Molecular profiling, which guides targeted therapies and immunotherapy decisions, currently depends on genomic or transcriptomic assays such as next generation sequencing. These tests are expensive, require specialized infrastructure, and can add days or weeks to the diagnostic timeline, a delay that matters enormously for patients with aggressive disease. If a machine could reliably estimate molecular features from the slide that is already being examined for diagnosis, the turnaround time for biomarker information could shrink dramatically and the cost per test could fall to nearly nothing beyond the computational expense. That is the promise the authors set out to test on a pan cancer scale rather than in a single tumor type.

The system, whose name stands for Rapid and Intelligent Detector for Genetic Estimation, is built on weakly supervised learning, a strategy that sidesteps one of the most stubborn obstacles in computational pathology. Whole slide images are gigantic, often exceeding one hundred thousand by one hundred thousand pixels, and labeling specific regions of interest by hand is prohibitively laborious. RIDGE instead learns from slide level labels only, using a multiple instance learning formulation in which the slide is treated as a bag of smaller tissue patches and only the overall slide carries a molecular label. The architecture employs fully convolutional networks to aggregate patch level information, allowing the model to identify which morphological patterns are associated with a given biomarker without ever being told where to look.

Several technical choices distinguish the framework. The authors incorporate clustering guided contrastive learning, a self supervised pretraining approach that teaches the network to group visually and biologically similar tissue patterns together before any biomarker prediction begins. Attention mechanisms, including a focused linear attention module, allow the model to weigh the contribution of thousands of patches efficiently, while depthwise convolutions and a mixture of experts design help the network specialize across the heterogeneous landscape of tumor types. Multi task learning enables a single model to predict multiple biomarkers and gene expression signatures simultaneously, sharing learned representations across related prediction problems. The result is a general purpose system intended to work across cancers rather than a bespoke model for each disease.

The training data were substantial. The team developed and validated RIDGE using 4,983 whole slide images from 4,680 patients spanning 12 solid tumor types in The Cancer Genome Atlas, including breast, lung, colon, rectal, stomach, liver, pancreatic, cervical, head and neck, and three kidney cancer cohorts. Performance was measured with the area under the receiver operating characteristic curve, the standard metric for binary classification tasks in medicine. Across all 12 cancer types, RIDGE achieved an overall AUC of 0.763, with a 95 percent confidence interval of 0.724 to 0.802. In a field where individual biomarker prediction models often hover in a similar range, a single framework reaching this level across such a diverse set of tumors and molecular targets is a meaningful benchmark.

Perhaps the most important result concerns generalization beyond the training data. Machine learning models in medicine frequently fail when moved to new datasets, a phenomenon driven by differences in staining protocols, scanners, and patient populations. To test reproducibility, the researchers applied RIDGE to an external validation cohort of 221 whole slide images from 105 colorectal cancer patients in the Clinical Proteomic Tumor Analysis Consortium, a completely independent resource, asking the model to predict microsatellite instability status. MSI is a critical biomarker because microsatellite unstable tumors respond well to immune checkpoint inhibitors. RIDGE achieved an AUC of 0.769, with a confidence interval of 0.686 to 0.841, closely matching its internal performance and suggesting the learned features reflect genuine biology rather than dataset specific artifacts.

Beyond raw accuracy, the authors emphasize interpretability, a persistent concern for clinicians asked to trust black box algorithms. The study reports that the model captures morphological visual characteristics that make gene expression signatures detectable directly from the slides, and supplementary analyses include explainability experiments examining how molecular features manifest in gastric cancer tissue. In effect, the network learns to associate particular cellular and tissue architectures, such as the appearance of tumor infiltrating immune cells, gland formation patterns, or nuclear features, with underlying molecular states. This capacity to quantify genotype phenotype relationships from images could itself become a research tool, allowing investigators to map molecular biology onto tissue morphology at a scale that manual review could never achieve.

The clinical implications, if the approach matures, are considerable. Because H&E stained slides are universally produced, an AI layer on top of standard pathology workflows could provide preliminary biomarker estimates within minutes of slide scanning, flagging patients who should receive priority for confirmatory molecular testing. In hospitals without access to sequencing facilities, such predictions could guide referral decisions and broaden equitable access to precision oncology. The authors argue that RIDGE could significantly expedite cancer screening and personalized therapy, and the pan cancer design means a single deployment could serve pathology departments handling many tumor types rather than requiring separate pipelines for each indication.

Caution is nonetheless warranted before such systems reach the clinic. An AUC of roughly 0.76, while respectable, indicates imperfect discrimination, meaning the model would need to function as a triage and prioritization tool rather than a replacement for definitive molecular assays. The study relied on publicly available, de identified data from TCGA and CPTAC, which, although multi institutional, may not capture the full heterogeneity of staining practices and patient demographics seen in routine global practice. Prospective clinical validation, regulatory review, and demonstration of impact on patient outcomes remain necessary steps. The published work was exempt from additional ethics approval because it used existing consented datasets, but real world deployment would face a new and more demanding evaluation landscape.

Even with those caveats, the study adds to a rapidly growing body of evidence that the humble pathology slide is an information rich object whose molecular content can be unlocked computationally. By demonstrating a single weakly supervised framework that predicts multiple biomarkers across a dozen cancer types and reproduces its performance on external data, the RIDGE team has moved the field closer to a future in which every diagnostic slide yields both a morphological diagnosis and a molecular profile. For patients, that could mean faster answers at lower cost; for researchers, a powerful new lens on the relationship between how a tumor looks and what it is, at the level of its genes.

Subject of Research: Deep learning prediction of pan-cancer molecular biomarkers from H&E-stained whole slide pathology images

Article Title: Deep learning-based large-scale pan-cancer multiple biomarkers prediction using RIDGE with pathological images

Article References: Xi, H., Feng, X., Lu, Y., Li, G., Zhang, Y., Li, J., Wang, Y., Xu, J., Zhang, Y., Sha, C., & He, M. (2026). Deep learning-based large-scale pan-cancer multiple biomarkers prediction using RIDGE with pathological images. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02695-4

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02695-4

Keywords: computational pathology, deep learning, whole slide imaging, biomarker prediction, weakly supervised learning, multiple instance learning, microsatellite instability, TCGA, CPTAC, precision oncology, gene expression, pan-cancer

Alternative splicing is the process by which a single gene produces multiple distinct RNA molecules. After a gene is transcribed into precursor messenger RNA, segments called introns are removed and the remaining exons are stitched together. Depending on which exons are retained, skipped, or joined at alternative boundaries, the same gene can yield proteins with very different functions. Seven major categories of these events are recognized, including exon skipping, intron retention, alternate promoter and terminator usage, and mutually exclusive exon selection. When this editing machinery goes awry in cancer, the consequences can be profound: aberrant splicing has been linked to suppressed apoptosis, enhanced drug resistance, altered DNA damage repair, and uncontrolled cell cycle progression.

The ABC transporter family is among the largest and oldest gene families in the human genome, organized into subfamilies A through G. These proteins act as efflux pumps, using the energy of ATP hydrolysis to move molecules across cell membranes. Their duties range from nutrient uptake and cellular detoxification to lipid balance, signal transduction, antiviral defense, and antigen presentation. In oncology, they are infamous for a darker role: overactive transporters such as ABCB1, ABCC1, and ABCG2 can pump chemotherapy drugs out of tumor cells, producing the multidrug resistance that frustrates treatment. A nonsense mutation in ABCB1 that truncates the P-glycoprotein pump, for example, renders tumors newly sensitive to anticancer drugs, underscoring how central these molecules are to therapeutic response.

To probe how splicing reshapes this family, the team, led by Yidan Zhang and supervised by Xiao Zhu of Guangdong Medical University, first compiled a curated set of 114 ABC transporter genes from the MSigDB, DGIdb, and GeneCards databases. They then mined the TCGA SpliceSeq resource, identifying a staggering 10,617 alternative splicing events across 33 cancer types, drawn from 736 normal control samples and 9,881 patient samples. Using univariate Cox regression, they found that 44 splicing signals from 29 genes were potentially tied to overall survival. LASSO regression, a technique that shrinks weak predictors to zero, narrowed the field, and multivariate Cox analysis ultimately pinned down 14 splicing signals from 11 genes—including PSMA4, PSMD7, UBA52, and PSMF1—that were significantly associated with pan-cancer prognosis.

The prognostic power of these signals was substantial. When patients were stratified by a risk score built from the model, the high-risk group showed dramatically lower overall survival than the low-risk group. The researchers then integrated clinical variables from 1,299 TCGA patients with complete records, confirming that age, tumor grade, and TNM staging were independent prognostic factors alongside the splicing-derived risk score. A nomogram combining these variables predicted one-, three-, five-, and ten-year survival with areas under the curve of 0.695, 0.746, 0.756, and 0.756 respectively—moderate accuracy that the authors themselves caution requires external validation in independent cohorts before clinical deployment.

Perhaps the most striking findings emerged when the splicing signature was overlaid on the tumor immune microenvironment. Using the ESTIMATE algorithm and single-sample gene set enrichment analysis, the team showed that patients in the high-risk group carried heavier immune infiltration, with twelve immune phenotypes positively correlated with risk scores. Checkpoint genes central to modern immunotherapy—CTLA4, CD274 (the gene encoding PD-L1), PDCD1, HAVCR2, and LAG3—were all expressed at higher levels in high-risk patients. Two genes stood out in deeper analysis: PSMA4 and PSMD7. PSMD7, a component of the 19S proteasome, showed that lower expression triggers the activation of potent antitumor immune cells, while higher PSMA4 expression correlated with increased naive CD4 T cells and follicular helper T cells. TIMER database analysis revealed that PSMA4 expression correlated with CD8-positive T cell infiltration in 20 tumor types, and both genes’ expression significantly co-varied with CD274 and CTLA4 across most cancers—hinting that splicing of these transporter-linked genes could help predict which patients will respond to checkpoint inhibitors.

Behind every splicing event stands a splicing factor, the protein that decides where the RNA scissors cut. By mapping the expression of 390 known splicing factors against the prognostic splicing events, the researchers constructed a regulatory network in Cytoscape. It revealed 17 significant splicing events—11 upregulated and 6 downregulated—each controlled by one or more of 50 splicing factors. Notably, some factors exerted dual regulatory effects, with the same factor influencing multiple events and single events being governed by several factors simultaneously. These hub nodes represent potential intervention points: if a splicing factor drives a harmful splice variant, inhibiting that factor could theoretically restore normal transcript production.

Correlation, however, is not causation, and this is where the study makes its boldest methodological move. Mendelian randomization uses genetic variants as natural experiments: because alleles segregate randomly at conception, SNPs associated with a gene’s expression can serve as instruments to test whether that gene genuinely influences disease risk, largely free of confounding. Drawing on eQTL data from the IEU Open GWAS database and outcome data covering 17,254 samples across 33 cancers, the team applied inverse variance weighting, weighted median, and MR-Egger methods, with rigorous sensitivity checks including heterogeneity testing and leave-one-out analysis. The verdict: UBA52 emerged as a protective factor, with its SNPs (rs10414427, rs9908158, rs139767434) associated with reduced risk across all 33 cancers, while ABCB4 acted as a risk factor, its SNPs (rs80351204, rs45493392, rs9275406) linked to increased disease occurrence.

To harden these conclusions, the researchers turned to Bayesian Weighted Mendelian randomization, a technique designed to handle weak polygenic effects and pleiotropy—the situation where a genetic variant influences multiple traits—by downweighting outlier instruments through Bayesian inference. The BWMR results corroborated the standard analysis, with UBA52 (P = 0.038) confirmed as protective and ABCB4 (P = 0.002) confirmed as a risk factor for pan-cancer development. The evidence-of-convergence plots showed rapid stabilization, and posterior weight analysis flagged one observation as potentially confounded, which the Bayesian framework appropriately discounted. Together, the two causal inference approaches elevate UBA52 and ABCB4 from statistical associations to genetically supported candidates for therapeutic targeting.

The authors are candid about limitations. Pan-cancer analyses can mask cancer-type-specific heterogeneity, and the protective or harmful effects of individual splicing events may vary in magnitude or even direction across tumor types. The TCGA and GWAS datasets are dominated by populations not representative of global genetic diversity, and the team calls for multi-ethnic validation in Asian and African cohorts, as well as proteomic confirmation of the functional consequences of the identified splice variants. They also note that PSI values—the quantitative measure of splicing—carry inherent measurement error, and that alternative splicing is shaped by environmental and epigenetic factors no genetic instrument can capture. Still, the convergence of transcriptomic modeling, immune profiling, network biology, and dual causal inference methods marks this study as a template for how splicing biology might finally earn its place in the oncology clinic. If validated, a simple readout of RNA editing patterns in drug transporter genes could one day tell oncologists not only how long a patient is likely to survive, but which immunotherapies their tumor’s molecular editing has primed them to receive.

Subject of Research: Alternative splicing of ABC transporter genes as a pan-cancer prognostic marker and therapeutic target

Article Title: Multi-omics and Mendelian randomization reveal ABC transporter alternative splicing as a pan-cancer prognostic marker and therapeutic target

Article References: Zhang, Y., Wu, J., Lin, Y., Diao, Z., Zhang, X., Yu, L., Cao, Z., & Zhu, X. (2025). Multi-omics and Mendelian randomization reveal ABC transporter alternative splicing as a pan-cancer prognostic marker and therapeutic target. Clinical Cancer Bulletin, 4(1), Article 20. https://doi.org/10.1007/s44272-025-00049-9

Image Credits: AI Generated

DOI: 10.1007/s44272-025-00049-9

Keywords: ABC transporters, alternative splicing, pan-cancer, Mendelian randomization, Bayesian weighted Mendelian randomization, TCGA, tumor microenvironment, immune checkpoints, UBA52, ABCB4, splicing factors, prognostic biomarkers

The numbers are striking. Across 28 studies included in the meta-analysis, patients whose tumors expressed higher levels of miR-221 faced a 75 percent greater risk of death at any given time compared with those expressing lower levels, reflected in a hazard ratio of 1.75 for overall survival. The association was even stronger for disease-free survival, at a hazard ratio of 1.85, and strongest of all for progression-free survival, where the hazard ratio climbed to 2.11, meaning high-miR-221 patients were more than twice as likely to see their disease advance. In oncology statistics, hazard ratios above 1.5 are considered clinically meaningful; values above 2 raise the prospect of a genuinely actionable biomarker.

Certain cancers stood out. In glioma, the devastating family of brain tumors that includes glioblastoma, high miR-221 expression carried a hazard ratio of 2.13, among the strongest signals in the entire analysis. Colorectal cancer followed closely at 1.91. The pattern was not universal, however: in breast cancer, the analysis found no significant prognostic role for the microRNA, a reminder that even broadly oncogenic molecules operate within the specific biology of each tissue. That context-dependence, the authors argue, is not a weakness of the finding but a clue to how miR-221 actually works, which is by latching onto different target networks depending on the cellular environment.

MicroRNA-221 belongs to a class of molecules that biologists have come to regard as master regulators of gene expression after transcription. These tiny RNAs, roughly 21 to 23 nucleotides long, do not encode proteins. Instead, they guide a cellular complex called the RNA-induced silencing complex to specific messenger RNA molecules, binding through partial sequence complementarity and either degrading those messages or blocking their translation into protein. A single microRNA can thereby dampen hundreds of different genes simultaneously, and when the microRNA in question is miR-221, the genes it silences tend to be the very ones cells rely on to keep division in check, to trigger apoptosis when things go wrong, and to suppress tumor growth.

Among the best-characterized casualties of miR-221 overexpression are the cyclin-dependent kinase inhibitors CDKN1B, also known as p27, and CDKN1C, known as p57. These proteins act as molecular brakes on the cell cycle, holding cells at the checkpoint between the G1 and S phases where DNA replication begins. By suppressing them, miR-221 releases the brakes and pushes cells into uncontrolled proliferation. The molecule also undermines apoptosis through targets such as PUMA and BIM, both mediators of programmed cell death, and interferes with tumor suppressors like PTEN, the phosphatase that restrains the PI3K/AKT/mTOR signaling axis, one of the most frequently hijacked growth pathways in human cancer. Additional targets include TIMP3 and RECK, which restrain invasion and metastasis, effectively equipping tumor cells with tools to break free of their surroundings.

To move beyond individual gene lists, the Italian team turned to pan-cancer protein-level analysis using the TCGA-LinkedOmics database, cross-checked for concordance with cBioPortal, drawing on six independent datasets from The Cancer Genome Atlas. The most consistent signal across tumor types was an inverse relationship between miR-221 and tumor-suppressive proteins: wherever the microRNA ran high, the protective proteins ran low. The association was strongest in lower-grade glioma and liver hepatocellular carcinoma, two cancers in which the clinical survival data had already pointed to miR-221 as a driver of poor outcomes. Concordant protein-level associations included SRC, a non-receptor tyrosine kinase central to growth signaling; VHL, the von Hippel-Lindau tumor suppressor; EEF2, a translation elongation factor; IRS1, a node in insulin and growth factor signaling; AXL, a receptor tyrosine kinase implicated in immune escape and therapy resistance; and STAT5A, a transcription factor in the JAK-STAT pathway. In colorectal cancer, TP53BP1, a binding partner of the legendary tumor suppressor p53, showed the strongest correlation, hinting that miR-221 may erode the cell’s DNA damage response in that setting.

The methodological rigor of the analysis deserves attention. The authors followed PRISMA criteria, the international standard for systematic reviews, searching PubMed, Google Scholar, and the Dimensions database, and pooled hazard ratios using random-effects models that account for statistical heterogeneity between studies. One of the most revealing subgroup findings concerned the type of specimen used to measure miR-221. When the microRNA was quantified in plasma or blood serum, the so-called liquid biopsy approach, the estimates showed essentially no heterogeneity, with an I-squared statistic of 0 percent, meaning the results were remarkably consistent across studies. Tissue-based measurements, by contrast, showed substantial heterogeneity at 89.4 percent, reflecting differences in tumor composition, assay platforms, and cutoff definitions. For clinicians hoping to deploy miR-221 as a blood-based prognostic marker, that consistency in circulating specimens is encouraging news, because liquid biopsies are far easier to obtain and repeat than tumor biopsies.

The therapeutic implications are already taking shape. Several of the study’s authors hold patents on LNA-i-miR-221, a locked nucleic acid inhibitor of the microRNA whose rights are owned by Magna Graecia University. Locked nucleic acids are chemically modified oligonucleotides whose sugar rings are locked in a rigid conformation, dramatically increasing their binding affinity and stability, which makes them well suited to neutralizing short RNA targets inside cells. By binding miR-221 directly, such inhibitors would free the suppressed tumor suppressor genes to resume their normal function, restoring the cell’s own defenses rather than attacking cancer through a single downstream pathway. The strategy has been explored most extensively in multiple myeloma and liver cancer, and the new pan-cancer evidence broadens the rationale for testing it elsewhere.

What makes this study resonate beyond its immediate findings is the way it illustrates the changing architecture of cancer research. Rather than asking whether a molecule matters in one tumor type, the researchers integrated clinical survival data with protein-network correlations across the entire TCGA compendium, letting the biology declare itself at scale. The result is a portrait of miR-221 as a context-dependent conductor of oncogenic programs: a molecule that suppresses cell cycle brakes in one cancer, undermines p53 signaling in another, and dampens angiogenesis inhibitors in a third, all while leaving a consistent statistical fingerprint of shortened survival. The authors caution that their work supports further evaluation rather than immediate clinical deployment, and the heterogeneity in tissue-based measurements means standardized measurement protocols will be essential before miR-221 enters routine prognostic panels. But the convergence of meta-analytic survival evidence, multi-omics protein correlations, and an existing therapeutic pipeline built around a locked nucleic acid inhibitor makes oncomiR-221 one of the more compelling microRNA stories in translational oncology, and a reminder that some of cancer’s most influential players are among the smallest molecules in the cell.

Subject of Research: The prognostic and therapeutic role of the oncomiR microRNA-221 across human cancers

Article Title: Deciphering the clinical and epigenetic impact of oncomiR-221 through multi-omics integration: evidence from a pan-cancer systematic review and meta-analysis

Article References: Munir, M., Grillone, K., Vocaturo, M., Di Martino, M. T., Staropoli, N., Tagliaferri, P., Caracciolo, D., & Tassone, P. (2026). Deciphering the clinical and epigenetic impact of oncomiR-221 through multi-omics integration: evidence from a pan-cancer systematic review and meta-analysis. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08996-0

Image Credits: AI Generated

DOI: 10.1186/s12967-026-08996-0

Keywords: miR-221, oncomiR, meta-analysis, pan-cancer, microRNA, prognosis, hazard ratio, TCGA, glioma, colorectal cancer, liquid biopsy, locked nucleic acid inhibitor

The research, whose first authors include Mohammed Inayatullah and Engin Demirdizen, integrated an unusually broad array of data modalities to address a question that has lingered in the field: whether conserved molecular programs underpin therapy resistance regardless of cancer type. The team combined single-cell RNA sequencing, spatial transcriptomics, regulatory network modeling, transcription factor binding data, and pharmacologic perturbation experiments. By layering these approaches across several cancer types, they were able to define what they describe as a pan-cancer, proliferative chemoresistant tumor state—a shared cellular identity that resistant cells from different organs appear to converge upon under the selective pressure of chemotherapy.

The technical findings are striking in their departure from expectation. Rather than exhibiting the low proliferative activity associated with quiescence, resistant tumor cells in the study displayed elevated G2/M and S-phase cell cycle signatures, indicating that they were actively progressing through DNA replication and mitosis at the time of or following treatment. Gene set analyses revealed enriched expression of E2F and MYC target genes, two transcriptional programs classically associated with driving cell cycle entry and growth. Alongside these, the resistant cells showed activation of DNA repair pathways, consistent with an enhanced capacity to mend the damage inflicted by genotoxic chemotherapy, and of PI3K/AKT signaling, a pro-survival axis that supports metabolism and continued proliferation even under stress.

Central to this resistant state, the researchers identified the transcription factor MYC as its master regulator. MYC, one of the most frequently dysregulated oncogenes in human cancer, showed progressive activation along the resistance trajectory—in other words, as cells moved toward a chemoresistant phenotype, MYC activity climbed steadily. Spatial transcriptomics added a clinically meaningful dimension: MYC expression was concentrated in focal pockets within resistant epithelial niches, suggesting that resistant cells do not emerge randomly throughout a tumor but cluster in specific microenvironments where the MYC-driven program is sustained. This spatial organization may explain why resistant clones can dominate recurrences so rapidly once first-line therapy fails.

Perhaps the most consequential discovery from the study concerns a previously underappreciated MYC target gene: SRM, which encodes spermidine synthase, an enzyme central to the biosynthesis of polyamines. The team found that SRM acts as a conserved effector of chemoresistance, promoting polyamine production that the resistant cells require for chromatin stability and metabolic resilience. Polyamines—putrescine, spermidine, and spermine—are small, positively charged molecules that bind nucleic acids and support numerous aspects of cell growth, but their specific role in maintaining the chromatin architecture of drug-resistant cells had not been defined in this context. The new data position SRM not merely as a downstream passenger but as a functional pillar of the resistant phenotype.

The clinical implications of the MYC–SRM axis were reinforced by survival analyses. SRM expression in patient tumors correlated with MYC binding at its regulatory regions and predicted poor patient survival, marking the enzyme as a potential biomarker of therapeutic failure as well as a target in its own right. When the authors examined whether this axis was merely correlative, the answer was emphatically no. Functional validation experiments spanning cancer cell lines, patient-derived organoids, and mouse models demonstrated that pharmacologic inhibition of MYC, of SRM, or of WNT signaling restored chemotherapy sensitivity, suppressed resistance-associated pathways, and reactivated apoptosis—the programmed cell death that chemotherapy is designed to trigger but that resistant cells evade.

What makes this work particularly compelling is the convergence of validation across model systems. Cell lines allow precise mechanistic dissection, but they can diverge from human disease. Patient-derived organoids, which retain much of the cellular heterogeneity and drug response of the original tumors, and in vivo mouse models provide stronger translational evidence. Demonstrating that disrupting the MYC–SRM module resensitizes tumors across all three platforms, and that spatial and survival analyses in human tissues confirm the axis’s clinical relevance, argues that this is a druggable vulnerability rather than a laboratory artifact. The authors go so far as to establish the MYC–SRM axis as a tractable module in treatment-refractory cancers, a claim supported by the pharmacologic tools already available against components of the polyamine biosynthesis pathway.

The conceptual reframing is equally significant. If chemoresistance is not principally a matter of cells going dormant but of cells rewiring themselves into an aggressive, MYC-driven proliferative and repair-competent state, then therapeutic strategies must be recalibrated. Approaches that simply target quiescent or mesenchymal phenotypes may miss the dominant biology of resistance. Conversely, combination regimens pairing conventional chemotherapy with inhibitors of MYC activity, spermidine synthase, or WNT signaling could in principle prevent the emergence or persistence of resistant clones by striking at the very engine of their survival. The finding that PI3K/AKT signaling and DNA repair programs are co-activated in the resistant state further suggests additional combination nodes for drug development.

There are, of course, well-known challenges ahead. MYC has long been considered notoriously difficult to drug directly, though indirect strategies—such as targeting MYC-dependent metabolic enzymes like SRM, or exploiting synthetic lethal interactions—have gained traction in recent years. The identification of SRM as a conserved, druggable effector downstream of MYC offers exactly the kind of actionable node that the field has sought: inhibiting a metabolic enzyme is far more tractable pharmacologically than inhibiting a transcription factor. Whether SRM inhibitors can be advanced safely into clinical combination trials, and whether the pan-cancer signature holds uniformly across all epithelial malignancies, will require prospective clinical validation. Nonetheless, the breadth of evidence assembled in this study—from single-cell and spatial genomics to organoids and animal models—makes a strong case that the MYC–SRM axis represents a genuine Achilles’ heel of chemotherapy-resistant tumors.

The study, funded by the Neye Foundation, the Novo Nordisk Foundation, the Danish National Research Foundation, the Danish Cancer Society, and ICURe grants, is among the first comprehensive efforts to redefine chemoresistance at a pan-cancer level using integrated single-cell and spatial technologies. By demonstrating that a single, conserved, MYC-orchestrated proliferative program underlies treatment failure across diverse cancer types, and by pinpointing spermidine synthase as a druggable linchpin of that program, the work offers oncologists a new conceptual map of resistance and a concrete therapeutic entry point. For patients whose tumors have exhausted standard options, the prospect of restoring chemotherapy sensitivity by dismantling this shared molecular machinery is a development worth watching closely in the years ahead.

Subject of Research: Pan-cancer molecular signatures of chemotherapy resistance and the MYC–SRM axis

Article Title: Uncovering pan-cancer signatures of chemoresistance

Article References: Inayatullah, M., Demirdizen, E., Keepers, Z., Correia, C. M., Hashemi, S. M., Tripathi, K., Sadhukhan, S., Bardhan, I., Mariappan, A., Rassool, F. V., Terp, M. G., Shukla, H. D., & Tiwari, V. K. (2026). Uncovering pan-cancer signatures of chemoresistance. Genome Medicine. https://doi.org/10.1186/s13073-026-01763-2

Image Credits: AI Generated

DOI: 10.1186/s13073-026-01763-2

Keywords: chemoresistance, MYC, SRM, spermidine synthase, single-cell RNA sequencing, spatial transcriptomics, pan-cancer, drug resistance, polyamine biosynthesis, E2F, PI3K/AKT signaling, Genome Medicine

The central premise of the study is that macrophages and monocytes inside tumors exist in a continuum of functional states that cannot be reduced to a single binary. To quantify this continuum, the researchers built the MDRi around three core programs: an injury program capturing stress responses such as iron and heme handling, hypoxia, inflammatory chemokines and danger-sensing pathways; a resolution program reflecting tissue repair, efferocytosis, lipid processing and resident-like macrophage biology; and a combined antigen-presentation and interferon program, abbreviated APC/IFN, that brings together major histocompatibility complex activity and interferon-stimulated genes. From the injury and resolution programs they also derived an injury-resolution axis, a single metric that places each cell along a spectrum from active damage to active healing.

Constructing and validating a framework of this kind required an unusually broad evidence base. The team established MDRi in an immune checkpoint blockade-treated multi-cancer atlas containing 47,750 myeloid cells drawn from 192 samples spanning eight cancer types. They then tested the framework separately in an independent multi-cancer myeloid dataset to confirm that the identified programs were not artifacts of a particular cohort or sequencing batch. Because the antigen-presentation and interferon-response gene sets were deliberately constructed to be non-overlapping, agreement between these two signatures provided gene-independent support for the validity of the APC/IFN dimension, an important safeguard against circular reasoning in signature-based immunology.

One of the more technically interesting aspects of the study concerns the geometry of myeloid cell states. Using trajectory inference, the researchers found that macrophage and monocyte states organized along related but non-identical functional dimensions, and root-sensitivity analyses revealed a subtle but important caveat: the connectivity of the states was stable regardless of where the trajectory was anchored, but the inferred directionality of pseudotime depended on the choice of root. The authors interpret this as evidence that MDRi captures genuine transcriptional topology rather than a universal developmental sequence, a deliberately cautious conclusion that resists over-interpreting pseudotime as a maturation timeline in tumor myeloid biology.

The framework was then put to the test against clinical data. In exploratory analyses of checkpoint blockade-treated patients, the researchers observed that post-treatment non-responders showed concurrent elevation of injury, resolution and APC/IFN scores, suggesting a globally activated but functionally ambivalent myeloid compartment. However, when the analyses were adjusted for cancer type and treatment regimen, the data did not support an independent predictive effect of MDRi scores on response. Complementary analyses of T and natural killer cells indicated that immune differences associated with treatment response were actually more evident in pretreatment samples, hinting that the pretreatment immune landscape, rather than treatment-induced myeloid changes, may carry the stronger predictive signal.

Prognostic analysis in bulk tumor data added another layer of context dependence. The researchers projected MDRi programs onto patient-level cohorts from The Cancer Genome Atlas and fitted joint multivariable Cox models containing the injury, resolution and APC/IFN scores. The derived injury-resolution axis was excluded from these joint models because it is mathematically dependent on its two components, a statistically transparent choice. The results showed MDR injury acting as an adverse factor in selected cancers, while resolution and APC/IFN displayed associations that varied by cancer type and clinical endpoint. In other words, the same myeloid program can be associated with better outcomes in one tumor type and worse outcomes in another, which is precisely the kind of context dependence the framework was designed to expose.

Because single-cell atlases are far less common than bulk transcriptomic cohorts, a framework is only as useful as its portability. The researchers benchmarked MDRi against established immune-deconvolution methods such as MCP-counter, TIMER and xCell, as well as against published tumor-associated macrophage signatures. The comparison demonstrated partial but non-uniform overlap, meaning that MDRi captures myeloid-state information that is related to, but not fully explained by, conventional deconvolution scores. This positions the index as a complementary rather than redundant measurement, adding functional granularity that abundance-based methods cannot provide.

The spatial dimension of myeloid biology received particular attention. Using Visium spatial transcriptomics in nasopharyngeal carcinoma, the team mapped MDRi programs onto hematoxylin and eosin stained tissue sections, revealing that injury, resolution and APC/IFN programs occupy focal tissue niches rather than being uniformly distributed. A parallel multi-cancer Xenium analysis at single-cell spatial resolution across cervical cancer, glioblastoma, lung cancer and melanoma further revealed platform- and cancer-dependent spatial distributions of the MDRi-related programs, including cancer-specific estimates of how APC/IFN-high myeloid cells position themselves near tumor cells and how checkpoint interactions vary in their vicinity. These findings suggest that the functional identity of myeloid cells is shaped not only by cancer type but by precise anatomical microenvironment.

To make the framework accessible, the researchers released MDRi Explorer, an open-source Shiny application that implements scoring, reference comparison, survival analysis and benchmark visualization. Users can apply MDRi to their own transcriptomic data, compare their results against built-in references, and explore cancer-specific score distributions and survival associations from TCGA. The authors are careful to frame the contribution appropriately: MDRi is presented as a reusable, hypothesis-generating framework for investigating context-dependent myeloid functional organization, not as a universal prognostic signature or a clinically validated predictor. That restraint is notable in a field where signature-based tools are often oversold.

The broader significance of the work lies in its refusal to flatten myeloid biology into a single number. By decomposing the tumor myeloid compartment into injury, resolution and antigen-presentation/interferon dimensions, and by documenting honestly where those dimensions matter, where they do not, and how their meaning shifts across cancers, platforms and treatment settings, the study provides the immunology community with a map that is as much about uncertainty as about discovery. For researchers designing myeloid-targeted therapies, the message is that the same cell population may be a friend in one tumor and an enemy in another, and that any intervention must be interpreted against the local context of tissue damage, repair and antigen presentation.

Subject of Research: A single-cell-derived myeloid damage response index quantifying context-dependent myeloid functional states across cancers

Article Title: A single-cell-informed framework maps context-dependent myeloid damage-response states across cancers

Article References: Ding, R., Zheng, W., Long, Z., Cao, Z., Liang, J., & Quan, Q. (2026). A single-cell-informed framework maps context-dependent myeloid damage-response states across cancers. Cancer Immunology, Immunotherapy. https://doi.org/10.1007/s00262-026-04543-4

Image Credits: AI Generated

DOI: 10.1007/s00262-026-04543-4

Keywords: tumor-associated macrophages, myeloid cells, single-cell RNA sequencing, spatial transcriptomics, pan-cancer, immune checkpoint blockade, immune deconvolution, antigen presentation, interferon signaling, TCGA, myeloid damage response index, tumor microenvironment