Lower-grade gliomas occupy an uneasy middle ground in neuro-oncology. They can smolder quietly for years, sometimes over a decade, allowing patients to work, drive and live largely normal lives under watchful surveillance. Yet nearly half of the patients in the new cohort — 77 of 155 — eventually underwent malignant transformation, the biological inflection point in which a slow-growing tumor acquires the aggressive behavior of a high-grade glioma and drives prognosis, adjuvant-therapy decisions and eligibility for clinical trials. Predicting which tumors will take that turn, and when, has remained one of the field’s most stubborn uncertainties, largely because transformation can only be pinned down retrospectively, at the moment a confirmatory MRI or a repeat biopsy reveals it.

The Austrian-led team attacked this problem from several angles simultaneously. Every patient had histologically confirmed adult-type diffuse glioma, diagnostic-quality multiparametric MRI including pre- and post-contrast T1, T2 and FLAIR sequences, and at least one follow-up scan. The time origin for each patient was the first MRI showing the glioma — a deliberately chosen anchor, since 18 patients in the cohort were managed with a wait-and-see strategy for more than a year, with one waiting 17.3 years before histological confirmation, contributing their untreated natural history to the analysis. Because malignant transformation is only ascertainable between examinations, the researchers bracketed each event within an interval bounded by the last documentation of low-grade status and the first examination showing transformation, rather than falsely assigning the event to a single scan date.

That methodological choice mattered enormously. When the team compared conventional dating — which treats the confirmatory scan as the moment of transformation — with proper interval-censored analysis, the conventional approach overstated the hazard of transformation roughly four-fold. The finding exposes a systematic bias running through much of the existing literature, where standard right-censored Cox models equate radiological detection with biological occurrence, an error whose magnitude grows with the length of the surveillance interval. To ensure their conclusions were not artifacts of any single statistical specification, the investigators cross-checked the primary Bayesian interval-censored Weibull model, fitted with a regularized horseshoe prior across 102 candidate predictors, against three frequentist alternatives: a Fine-Gray subdistribution-hazards model accounting for the competing risk of death without transformation, a Cox counting-process model handling surgery as a time-varying covariate with 20-fold multiple imputation, and a native interval-censored Fine-Gray model estimated by sieve maximum likelihood. All four agreed on the direction and significance of the central result.

And that central result was striking. Contrast enhancement visible on the diagnostic scan — even the subtle, sub-threshold kind that does not itself meet radiological criteria for malignancy — was by far the strongest predictor of subsequent malignant transformation, dwarfing both tumor volume and a co-analyzed panel of 99 quantitative radiomic features. In the full cohort, the Bayesian hazard ratio for enhancement reached 22.1, with a 95 percent credible interval of 10.5 to 53.2. Within IDH-mutant disease the association approached separation: all 22 enhancing tumors eventually transformed, compared with only 30 percent of non-enhancing ones, a pattern so extreme that the statisticians had to deploy Firth penalized-likelihood and exact conditional methods to obtain interpretable lower bounds. The same held true across every molecular entity represented — IDH-mutant astrocytomas, 1p/19q-codeleted oligodendrogliomas and IDH-wildtype tumors — with every single enhancing tumor in each subgroup ultimately transforming.

Critically, the team recognized that much of this dramatic unadjusted effect reflects tumors that were already transforming at or near diagnosis, so they performed a prespecified series of landmark analyses that condition on surviving transformation-free for three, six or twelve months and restart the clock there. Even under this stricter framing, a five- to seven-fold elevation in hazard persisted at every landmark: 7.41 at three months, 5.65 at six months and 6.33 at twelve months. The authors note that an unmeasured confounder would need an implausibly strong association with both enhancement and transformation — a risk ratio of at least 5.8 with each — to explain away the six-month estimate. Blinded re-reading of baseline scans by two independent reviewers achieved perfect agreement on enhancement status, and sensitivity analyses excluding patients who would meet the 2021 WHO grade 4 criteria left the effect essentially unchanged.

Perhaps the most surprising discovery concerned what happened after surgery. Because the exposure and the endpoint are both read from post-contrast imaging, one might suspect the enhancement signal simply marks a focus of high-grade disease that a resection removes. The data say otherwise. In a model of time from first surgery restricted to the 128 patients who had not transformed beforehand, baseline enhancement predicted post-surgical malignant transformation with a hazard ratio of 4.28 — and the effect not only persisted within gross-totally resected tumors, where any enhancing tissue present at baseline should have been removed, but strengthened under multivariable adjustment, rising to 7.58 when grade, IDH status and extent of resection were accounted for. Seven of eight enhancing tumors resected gross-totally still transformed, versus only 15 of 60 non-enhancing ones. A signal that outlives the very tissue generating it, the authors argue, points to a property of the tumor as a whole — an intrinsic biology — rather than a discrete focus awaiting excision.

Surgery itself emerged as the only modifiable determinant in the study. Gross-total resection was associated with a roughly two-thirds reduction in the post-surgical hazard of transformation compared with biopsy (hazard ratio 0.32), a benefit also seen within IDH-mutant disease, while subtotal resection showed no significant advantage over biopsy. The finding reinforces the current standard of early onco-functional resection in diffuse glioma, though the authors are careful to note that the timing of surgery in their retrospective cohort was confounded by indication — surgeons often operated precisely because early signs of transformation appeared — and cannot be interpreted causally. What they do propose is that baseline enhancement, currently discarded once histology becomes available, should be carried forward into postoperative risk estimation, where it independently refines prognostication alongside grade, molecular status and resection extent.

The study also delivered a sobering verdict on radiomics. Despite extracting 428 features from co-registered image volumes, reduced to 99 candidates by correlation filtering, not a single quantitative feature survived Bayesian horseshoe shrinkage or native interval-censored analysis. One texture feature retained significance only under the biased midpoint-imputation framework, collapsing to null under proper censoring — a conditional null the authors frame honestly, acknowledging that 77 events cannot credibly support 99 candidate biomarkers, and calling for radiomics discovery to be re-scoped toward predefined panels or much larger cohorts. With a bootstrap-corrected concordance of 0.83 but imperfect absolute-risk calibration, the team explicitly declines to offer their model for individual prediction. Instead, they position the work as a recalibration of the field’s foundations: contrast enhancement on the diagnostic scan deserves a central place in pre- and postoperative risk stratification, gross-total resection materially changes the transformation hazard, and any future estimate of when a lower-grade glioma will turn malignant should respect the interval-censored reality of how that event is actually observed.

Subject of Research: Predictors and timing of malignant transformation in lower-grade glioma using interval-censored survival analysis

Article Title: Malignant transformation of lower-grade glioma: contrast enhancement, extent of resection, and the natural history under interval-censored analysis

Article References: Malignant transformation of lower-grade glioma: contrast enhancement, extent of resection, and the natural history under interval-censored analysis. (n.d.). https://doi.org/10.1007/s11060-026-05803-0

Image Credits: AI Generated

DOI: 10.1007/s11060-026-05803-0

Keywords: lower-grade glioma, malignant transformation, contrast enhancement, MRI, extent of resection, gross-total resection, interval censoring, Bayesian analysis, radiomics, IDH status, WHO classification, brain tumor

Viral mimicry was formally described in 2015, when two landmark studies showed that inhibiting DNA methylation in cancer cells could trigger an interferon response without any actual virus being present. The trigger turned out to be endogenous double-stranded RNA derived from endogenous retroviruses and other repeats that had been derepressed by epigenetic drugs. Since then, a large body of work has mapped how this process works at the molecular level. Many transposable elements, including long interspersed nuclear elements such as LINE1 and short interspersed elements such as Alu, retain vestiges of their viral ancestry, including promoter sequences and the capacity to generate RNA species that form double-stranded structures. Pairs of oppositely oriented Alu elements, known as IR-Alus, can fold into intramolecular double-stranded RNA hairpins that are recognized by innate immune sensors such as MDA5, while PKR and ZBP1 provide additional surveillance routes for endogenous double-stranded RNA and Z-form nucleic acids.

The review emphasizes that viral mimicry is not an accidental by-product confined to drug treatment. Cancer-associated alterations in DNA methylation, histone modifications, splicing and RNA processing routinely generate immunogenic nucleic acids in precancerous cells. Global DNA hypomethylation, a common feature of tumour genomes, relaxes repression at repetitive loci; loss of histone marks such as H3K9me3, mediated by enzymes like SETDB1, or disruption of Polycomb repressive complexes derepresses endogenous retroviruses. Splicing defects, which are pervasive across cancer transcriptomes, can create retained introns and aberrant junctions that form double-stranded RNA or Z-RNA structures sensed by MDA5, PKR and ZBP1. Even mitochondrial double-stranded RNA and cytoplasmic RNA–DNA hybrids derived from R-loops can contribute to the endogenous pool of alarm signals. In this sense, the review argues, the pro-tumorigenic regulatory chaos of transformation is inseparable from the collateral production of viral-like nucleic acids.

Viewed through this lens, viral mimicry emerges as a tumour-suppressive mechanism that shapes tumour evolution from its earliest stages. Cells that activate these antiviral programmes can undergo apoptosis, necroptosis, pyroptosis or translational shutdown driven by PKR, or they can attract immune cells through type I interferon signalling that enhances antigen presentation and cytotoxic lymphocyte activity. Recent work describing viral mimicry as a bottleneck in early cancer evolution, including evidence that the pathway acts as a tumour suppressor in inflammatory contexts such as colitis, supports the idea that most cells attempting transformation are eliminated precisely because they cannot simultaneously disrupt their epigenome and silence the repeats that become exposed. Only clones that acquire effective escape mechanisms survive this selective filter, which is why the review frames viral mimicry escape as a prerequisite rather than an option for malignant cells.

The mechanisms cancer cells use to escape are diverse and, the authors argue, reveal a fundamental dependency. On the transcriptional side, tumours often impose compensatory epigenetic repression on repetitive elements, re-engaging DNA methyltransferases, SETDB1, the RB–EZH2 complex or KDM5B to re-silence retroelements. At the RNA level, the editing enzyme ADAR1 converts adenosine to inosine within double-stranded RNA, destabilizing the structures that immune sensors require and acting as a master regulator of viral mimicry escape; high ADAR1 activity is a hallmark of many tumours and is exploited in leukaemia relapse and other contexts. RNA modifications add another layer: N6-methyladenosine deposited by METTL3 and related machinery can reshape double-stranded RNA to prevent sensor recognition and control the fate of endogenous retroviruses. RNA decay enzymes provide a further escape route, with the exonuclease XRN1 degrading retroelement transcripts, while proteins such as DHX9 and the LINE1 ORF1 protein — which acts like a viral innate immune evasion factor — can chaperone or sequester problematic nucleic acids.

Beyond eliminating the trigger molecules themselves, tumours can dampen the downstream signalling that would translate them into an immune response. Sensor pathways can be epigenetically silenced, as has been described for STING and RIG-I–MAVS components in various cancers; interferon signalling itself can be attenuated through regulators such as USP18 and STAT2, or through negative feedback that blunts the antiviral state. Oncogenic drivers also contribute: mutant KRAS in colorectal cancer impairs DDX60-mediated double-stranded RNA accumulation and viral mimicry, converting immunologically hot tumours into cold ones, while p53 loss has been shown to create chronic viral mimicry pressure that selected clones must overcome. Chemotherapy-resistant breast cancers can switch epigenetic states to evade viral mimicry, illustrating that escape is a dynamic, evolving process throughout tumour progression rather than a one-time event.

The review also connects escape mechanisms to biomarkers and therapeutic vulnerabilities. LINE1 ORF1 protein circulating in the blood has been developed as an ultrasensitive multicancer biomarker, and genome-wide repeat landscapes measurable in cell-free DNA reflect the extent of repeat deregulation in individual tumours. Ratios of stemness to interferon signalling have been proposed as biomarkers of progression in myeloproliferative neoplasms. More importantly, every escape mechanism creates a dependency: tumours that rely on ADAR1, XRN1, DHX9, SETDB1 or m6A machinery to survive their own endogenous viral alarm are theoretically vulnerable to drugs that disable those factors. Pharmacological reactivation of viral mimicry — through DNA methyltransferase inhibitors, EZH2 inhibitors, LSD1 inhibitors, spliceosome-targeted therapies, PRMT inhibition or METTL3 blockade — has been shown in preclinical models to restore immunogenicity, and several such approaches are now entering clinical testing.

Combining viral mimicry induction with immunotherapy is a central translational theme. Because viral mimicry activation enhances antitumour immunity and sensitizes cells to immune checkpoint blockade, epigenetic priming with DNMT inhibitors followed by anti-PD1 therapy has shown promise, including in relapsed or refractory NK/T-cell lymphoma, where one of the first trials demonstrating that triggering viral mimicry could augment checkpoint immunotherapy has now been reported. Similar synergy has been observed with EZH2 inhibition in prostate cancer and glioblastoma models, with ZBP1-driven immunogenicity in HER2-directed combinations, and with strategies that restore cGAS–STING and RIG-I–MAVS signalling. Conversely, the review notes that sustained type I interferon signalling can also mediate resistance to some therapies, underscoring that timing, context and combination design matter when manipulating these pathways in patients.

By gathering this evidence into a single conceptual model, the authors propose a reframing of malignant transformation itself: cancer is not simply uncontrolled growth, but uncontrolled growth that has necessarily survived an internal antiviral insurgency of its own making. This unifying model explains why escape mechanisms are so consistently observed across tumour types, why they map onto established dependencies, and why deliberately reactivating viral mimicry represents a rational strategy to expose tumours to their own genome once again. If the framework holds up under experimental and clinical scrutiny, the ancient viral fossils scattered through human DNA may prove to be one of oncology’s most powerful untapped weapons — an alarm that every successful cancer has had to silence, and that medicine may now learn to ring.

Subject of Research: Viral mimicry escape mechanisms in malignant transformation and cancer immunotherapy

Article Title: Viral mimicry escape as a necessary feature of malignant transformation

Article References: Viral mimicry escape as a necessary feature of malignant transformation. (n.d.). https://doi.org/10.1038/s41568-026-00977-1

Image Credits: AI Generated

DOI: 10.1038/s41568-026-00977-1

Keywords: viral mimicry, transposable elements, malignant transformation, epigenetics, interferon signalling, endogenous retroviruses, ADAR1, immune checkpoint blockade, DNA methylation, cancer immunotherapy, innate immunity, double-stranded RNA