The clinical problem the researchers set out to solve is a timing one. In current practice, the response of a glioblastoma to treatment is assessed with anatomical MRI performed up to three months after the start of radiotherapy and chemotherapy, with lesion sizes measured according to the Response Assessment in Neuro-Oncology, or RANO, criteria. That delay is dangerous in a disease that progresses so quickly, and early interpretation is further complicated by pseudoprogression, an inflammatory phenomenon that can make a responding tumor look as though it is growing. A suspected progression must typically be confirmed by a follow-up scan four to eight weeks later, or by tissue analysis, delaying any definitive judgment. By the time doctors know whether a therapy is working, precious weeks of a patient’s short median survival have already elapsed.

The French team, led by Noémie Moreau and Aurélien Corroyer-Dulmont of the Centre François Baclesse and the University of Caen Normandy, took a different approach: predict the outcome before treatment starts, using only data that already exist in the clinical workflow. They assembled a retrospective cohort of 175 patients treated for glioblastoma between 2018 and 2023, from an initial pool of 385 patients. After excluding clinical trial participants, patients with multiple courses of radiotherapy, those missing pre-treatment MRI scans or clinical data, and patients who were still alive or lost to follow-up, the final cohort comprised 1,575 contrast-enhanced T1-weighted MRI slices from 36,400 acquired images. Importantly, these were routine clinical scans, not the curated, well-organized images from public research datasets that dominate much of the AI literature and may not reflect the messiness of real-world practice.

On the technical side, the researchers fine-tuned a ResNet-51Q convolutional neural network that had been pretrained on ImageNet, a large database of natural images, using transfer learning to adapt it to brain MRI. For each patient, nine central slices through the region of maximum tumor diameter, representing roughly 20 percent of the delineated tumor volume, were selected, and voxel intensities were normalized to compensate for the arbitrary intensity scale of MRI. The network was trained to perform survival classification as a discrete prediction task: first distinguishing short-term from long-term survivors at a threshold of 332 days, the median survival of the training cohort, then tackling three classes defined by terciles, and finally four classes defined by quartiles. Only deceased patients were included, because the classification framework requires exact survival labels that right-censored, still-living patients cannot provide.

To guard against data leakage and inflated performance, the dataset was split at the patient level into training, validation and test sets of roughly 60, 20 and 20 percent, so that no patient’s slices appeared in more than one subset. Hyperparameters were optimized with Optuna, a Bayesian optimization framework, using only the validation set, while the test set was held out until final evaluation. Because the model processes individual slices, the nine per-slice predictions were aggregated into a single patient-level verdict by averaging the probability vectors. Confidence intervals were estimated by patient-level bootstrap resampling with 2,000 iterations, an approach that respects the fact that slices from the same patient are not independent observations. All code was released publicly on GitHub.

The results are striking. In the binary task, the MRI-only deep learning model achieved 81.08 percent accuracy, 94.44 percent specificity and 68.42 percent sensitivity, with an area under the receiver operating characteristic curve of 0.89. The three-class model performed even better at the patient level, reaching 87.18 percent accuracy, 87.18 percent sensitivity and 93.59 percent specificity, with a macro-averaged AUC of 0.948. Even at the more demanding slice level, the three-class model achieved 81.77 percent accuracy. For context, a simple non-AI baseline using surgery type alone, a known prognostic factor, achieved only 69.71 percent accuracy, confirming that the deep network was extracting genuinely predictive information from the images themselves, including features invisible to the human eye such as subtle tissue textures and spatial heterogeneity in vascularization, edema and necrosis.

The team then asked whether adding clinical variables would push performance higher. Using a deep hybrid learning, or DHL, architecture, they concatenated the deep learning model’s averaged probability outputs with structured clinical data: WHO performance status, age, sex and surgery type. This augmented feature vector was fed to a random forest meta-classifier, trained on out-of-fold deep learning predictions generated via five-fold cross-validation to prevent overconfidence bias. The hybrid model reached 84.62 percent accuracy on the three-class task, slightly lower than the 87.18 percent of the MRI-only model, and 76.32 percent on the four-class task versus 71.05 percent for imaging alone. Rigorous statistical testing, including an exact McNemar test and paired bootstrap comparisons, found no significant difference between the two approaches, and feature importance analysis revealed that the imaging contribution dwarfed everything else, accounting for 65 percent of predictive weight in the three-class model and 67.3 percent in the four-class model, with age and surgery type trailing far behind at roughly 13 percent each.

That the imaging signal proved so dominant is scientifically interesting in itself. The authors suggest that surgery type may be partly redundant with imaging, since post-surgical changes such as the resection cavity are visible on the pre-chemoradiotherapy scan and may already be captured by the network. A sensitivity analysis excluding a single patient reclassified under the updated WHO 2021 taxonomy produced only a minor change in accuracy, indicating the result was robust. Still, the team is candid about limitations: the effective independent sample size is 175 patients, the test cohorts for the three- and four-class tasks were small, no external validation on independent multicenter data has yet been performed, and the model was trained on 2D slices rather than full 3D volumes, which could capture more of the tumor’s heterogeneity. The absence of molecular data such as MGMT methylation status, a well-established prognostic marker, also leaves prognostic information unexploited.

Compared with prior work, the Caen study occupies a distinctive niche. Earlier efforts relied on public benchmarks such as the BraTS challenge, used preoperative rather than pre-chemoradiotherapy imaging, or required multimodal data including resting-state functional MRI. One multicenter study predicted survival at eight months after radiotherapy, but used post-treatment MRI, which limits personalization since the therapy has already been given. Others achieved higher reported accuracies, such as 90.6 percent with multimodal preoperative data, but those inputs are not available at the moment of radiotherapy planning, roughly 13 days before treatment begins in the French workflow. By restricting itself to a single contrast-enhanced T1 sequence plus four routine clinical variables, the new model is designed for rapid clinical integration rather than benchmark supremacy.

The researchers emphasize that the tool is not meant to replace the judgment of a radiation oncologist but to provide additional prognostic information at the pre-treatment stage, potentially flagging high-risk patients before any therapy is delivered. Future work includes external validation on multicenter cohorts with harmonized acquisition protocols, federated learning to train more generalizable models without moving sensitive patient data across institutions, deeper multimodal fusion methods, and the integration of radiotherapy dose distributions to capture the effect of the delivered treatment itself. If validated, a system that forecasts survival from the scan already sitting in the patient’s file could transform prognostic stratification for a disease where, for most patients, every week counts.

Subject of Research: Deep learning prediction of overall survival in glioblastoma patients from pre-chemoradiotherapy MRI and clinical data

Article Title: Deep learning prediction of overall survival in glioblastoma patients using pre-chemoradiotherapy MRI and clinical data

Article References: Moreau, N. N., Desmonts, A., Jaudet, C., Leleu, T., Leclercq, A. G., Brunaud, C., Stefan, D., Valable, S., Lechervy, A., & Corroyer-Dulmont, A. (2026). Deep learning prediction of overall survival in glioblastoma patients using pre-chemoradiotherapy MRI and clinical data. Frontiers in Radiology, 6, Article 1963539. https://doi.org/10.3389/fradi.2026.1963539

Image Credits: AI Generated

DOI: 10.3389/fradi.2026.1963539

Keywords: glioblastoma, deep learning, MRI, survival prediction, artificial intelligence, radiotherapy, chemoradiotherapy, deep hybrid learning, neuro-oncology, prognostic stratification, transfer learning, pseudoprogression

The first study, published in Nature Communications, examined a strategy that many oncologists have considered but few have tested rigorously in brain cancer: giving two immune checkpoint inhibitors instead of one, and giving them before surgery rather than after. In a phase 2 clinical trial, the researchers enrolled 63 patients with recurrent glioblastoma whose tumors had returned after initial treatment and who were already scheduled for an operation. Participants received either the combination of nivolumab, which targets the PD-1 checkpoint, and ipilimumab, which blocks CTLA-4, or nivolumab alone, or a placebo, before their tumors were removed. This neoadjuvant design allowed the team to analyze living tumor tissue harvested during surgery and to see, directly, how each regimen reshaped the immune landscape inside the cancer.

The survival difference was striking for a disease in which incremental gains are the norm. Patients who received the dual combination lived a median of about 13.5 months, compared with roughly nine months for those treated with nivolumab alone. More remarkable still, five patients — about 15 percent of those on the combination — became long-term survivors, while no long-term responders emerged among patients receiving nivolumab by itself. In a cancer where median survival after recurrence is typically measured in months, a subgroup of durable responders is a signal that demands explanation, and the researchers found one written in the tumor tissue itself.

When the team examined the surgically removed tumors, the combination therapy had triggered significantly greater immune activity than the single agent, including increased infiltration of cancer-fighting T cells and higher expression of interferon-related immune genes, the molecular signature of an immune system actively engaged with its target. The most important finding was quantitative: patients whose tumors carried the highest density of T cells lived longer, with T-cell density emerging as an independent predictor of overall survival. The long-term survivors also had substantially higher levels of tumor-infiltrating lymphocytes, suggesting that the physical presence of T cells within the tumor may be the critical ingredient that converts checkpoint blockade from a theoretical therapy into a working one.

“These findings give us a clearer picture of what an effective immune response against glioblastoma may look like,” said Lu Sun, adjunct assistant professor of neurosurgery at UCLA and first author of the study. “The long-term survivors showed evidence of a stronger immune response within their tumors, offering important clues for developing treatments that can help more patients benefit from immunotherapy.” The study also identified early immune changes in the blood that were associated with better outcomes, hinting that blood-based biomarkers could one day allow physicians to monitor a patient’s response without repeated surgery. Imaging added another layer of insight: several patients treated with the dual blockade showed early pseudoprogression on MRI scans, an apparent swelling of the tumor that actually reflected immune cells flooding in, and that pattern was associated with longer survival.

“Several patients treated with dual immune checkpoint blockade showed early pseudoprogression on MRI scans following treatment, which was associated with longer survival,” said Benjamin Ellingson, director of the UCLA Brain Tumor Imaging Laboratory and professor of radiological sciences. “These findings suggest that early inflammatory changes may occur on imaging in patients that have an effective immune response.” Timothy Cloughesy, distinguished professor and director of the UCLA Neuro-Oncology Program, framed the result in broader terms: “This study shows that we can activate the immune system against glioblastoma in ways we have not previously seen. The increase in T cells within the tumor and the association with longer survival give us important clues about what an effective antitumor immune response looks like. Those insights are helping guide the next generation of therapies designed to make immunotherapy work for more patients.”

But knowing what a successful response looks like raises an obvious question: why do most glioblastoma tumors never mount one? The second study, published in Cancer Research, a journal of the American Association for Cancer Research, tackles that problem with a new preclinical model built to mirror human disease far more faithfully than the laboratory mice traditionally used to test brain cancer therapies. The model was developed using a technology called MADR — Mosaic Analysis with Dual-Recombinase — which allows scientists to introduce patient-relevant genetic changes into developing brain cells. The engineered tumors carry alterations commonly found in human glioblastoma, including EGFRvIII, a cancer-driving mutation that keeps the EGFR growth signaling pathway permanently switched on, along with the loss of the tumor suppressor genes PTEN and CDKN2A.

“Unlike commonly used glioblastoma models, which often respond to immune checkpoint inhibitors despite their limited benefit in patients, the new model closely mirrored the biology of human disease,” said David Nathanson, professor of molecular and medical pharmacology at the David Geffen School of Medicine at UCLA. The engineered tumors grew diffusely throughout the brain, exhibited the hallmark features of human glioblastoma and developed an immune microenvironment dominated by myeloid cells — suppressive immune cells that crowd out help rather than provide it — with very few cancer-fighting T cells. Critically, the model was resistant to immune checkpoint blockade, reproducing the disappointing results seen in clinical trials rather than the false optimism generated by conventional mouse models. The researchers also found that the tumor cells were inherently resistant to T-cell killing and actively promoted T-cell exhaustion, a dysfunctional state in which immune cells lose their offensive capacity, revealing multiple overlapping mechanisms the cancer uses to evade destruction.

The two studies illuminate each other with unusual precision. In patients, tumors with higher T-cell density were associated with longer survival after dual immunotherapy; the new preclinical model, in contrast, consistently exhibited very low T-cell infiltration, faithfully recreating the so-called immune cold environment that has made glioblastoma so difficult to treat. “The tumors have very low T-cell density, which helps explain why they are resistant to immune checkpoint blockade,” said Robert Prins, professor of neurosurgery and molecular and medical pharmacology at UCLA and senior author of both studies. “Similarly, the MADR model and the clinical study highlight how EGFR alterations are linked to shorter survival and treatment resistance.” For years, Prins noted, researchers have known that immunotherapy works remarkably well in some cancers but has consistently fallen short in glioblastoma; these studies, he said, help explain both sides of that equation, one showing what an effective response looks like in patients and the other providing a model that recreates the biology preventing it.

The findings are already shaping the next phase of research. Working with Nathanson and colleagues, the team is investigating whether novel brain-penetrant EGFR inhibitors can create a window of opportunity that makes tumors more susceptible to immunotherapy, attacking the very genetic driver linked to treatment resistance. In collaboration with the laboratory of Caius Radu, the researchers are also developing next-generation mRNA vaccines designed to increase T-cell infiltration into glioblastoma tumors — directly targeting the deficiency that both studies identified as the decisive factor. The logic is elegant: if T-cell density determines who survives, then therapies that force T cells into the tumor, combined with drugs that strip away the tumor’s defenses, could convert the immune cold majority into the hot minority that responds. For a disease that has defeated nearly every therapeutic advance of the past two decades, the convergence of a human trial showing that the immune system can be activated and a model that finally reproduces why it usually is not represents a genuinely new starting point — and a roadmap for turning a 15 percent long-term survivor rate into something far larger.

Subject of Research: Immune checkpoint inhibitor response and preclinical modeling in glioblastoma

Article Title: Scientists identify clues to making immunotherapy more effective against glioblastoma

Article References: Scientists identify clues to making immunotherapy more effective against glioblastoma. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: glioblastoma, immunotherapy, immune checkpoint inhibitors, nivolumab, ipilimumab, T-cell infiltration, EGFRvIII, MADR model, tumor microenvironment, biomarkers, pseudoprogression, UCLA

To understand why a chemical modification on a stretch of DNA commands such attention, one has to start with the biology. The MGMT gene encodes O6-methylguanine-DNA methyltransferase, a suicide enzyme that patrols the genome and directly reverses a specific type of alkylation damage: the addition of methyl or larger alkyl groups to the O6 position of guanine. This is precisely the lesion created by temozolomide, the backbone chemotherapy drug for glioblastoma, and by nitrosoureas such as lomustine. When the drug damages DNA at O6-methylguanine, MGMT removes the offending adduct and, in the process, consumes itself, sacrificing one protein molecule per repair event. Tumor cells with abundant MGMT are therefore armored against alkylating chemotherapy; they simply undo the damage faster than the drug can inflict it.

The elegant twist is that the MGMT promoter contains a CpG island, a GC-rich region where cytosines can be methylated. When methylation spreads across this promoter, the gene is transcriptionally silenced, and the repair enzyme disappears. Tumor cells lacking MGMT cannot correct O6-alkylguanine lesions, so the damage accumulates, triggers futile cycles of mismatch repair, and culminates in cell death. Methylation of the promoter, in other words, converts temozolomide from a modestly effective drug into a genuinely lethal one for the tumor. The landmark 2005 study by Hegi and colleagues in the New England Journal of Medicine demonstrated this in patients: glioblastoma patients with a methylated MGMT promoter derived clear survival benefit from adding temozolomide to radiotherapy, while those with an unmethylated promoter benefited far less. That finding transformed MGMT promoter methylation from a laboratory curiosity into a decision-making tool used in clinics worldwide.

The predictive logic has since been extended and refined. In elderly patients, where the toxicity of combined chemoradiation is a serious concern, trials such as NOA-08 and the Nordic trial used MGMT methylation to justify temozolomide alone or hypofractionated radiotherapy as alternatives to standard treatment. A 2024 reanalysis of the CE.6 and pooled Nordic/NOA-08 trials by Hegi and colleagues went further, concluding that glioblastomas with a truly unmethylated MGMT promoter derive no benefit from temozolomide at all, a finding that strengthens the case for omitting the drug in that population. Conversely, the CeTeG/NOA-09 trial showed that patients with methylated tumors could safely intensify therapy with a lomustine-temozolomide combination, improving survival. Methylation status also predicts pseudoprogression, the inflammatory radiographic mimic of tumor growth that appears after chemoradiation, and it is now embedded in the RANO 2.0 response assessment criteria that govern how treatment response is judged in trials and clinics.

Yet the review is equally clear that measuring this biomarker is far harder than interpreting it. There is no single gold-standard assay. Methylation-specific PCR, the quantitative real-time version validated by Vlassenbroeck and colleagues, offers speed and sensitivity but interrogates only a handful of CpG sites. Pyrosequencing provides quantitative methylation percentages across selected CpG positions and has met clinical and analytical performance criteria in prospective multicenter validation, but different laboratories target different CpG sites and apply different cutoffs. Genome-wide methylation arrays, now routine in DNA methylation-based tumor classification, can derive MGMT status through algorithms such as MGMT-STP27, allowing methylation calls to be extracted from data already generated for classification purposes. Each method answers a slightly different question about a heterogeneous chemical landscape, and the results do not always agree.

The consequences of this methodological fragmentation are real. An international survey published in Neuro-Oncology Practice asked practitioners whether they truly knew who had a methylated glioma, and the answers revealed widespread uncertainty about assay choice and interpretation. A comprehensive Cochrane-based meta-analysis by Brandner and colleagues confirmed that MGMT testing predicts overall survival in temozolomide-treated glioblastoma patients, but also highlighted how much the accuracy of that prediction depends on the assay used. Tumors with intermediate, so-called gray-zone methylation sit in an interpretive no-man’s-land: partially methylated, they confound binary methylated-or-unmethylated reporting, yet clinical decisions demand a yes-or-no answer. Studies have shown that the number of methylated CpG sites correlates linearly with outcome, suggesting that the dichotomization itself discards meaningful information.

Pre-analytical and biological pitfalls compound the technical ones. Bisulfite conversion, the chemical step underlying most methylation assays, degrades DNA and can lose material unevenly. Intratumoral heterogeneity means a sample from one region of a glioblastoma may not represent the whole. Rare genetic events can masquerade as methylation results: the review’s own authors have reported that homozygous deletion of the MGMT gene, which removes the DNA template entirely, creates a potential pitfall for methylation testing, and other work has described genomic rearrangements of MGMT that contribute to chemotherapy resistance independently of promoter methylation. MGMT expression can also be induced by treatment itself, as xenograft studies demonstrated, meaning a tumor’s initial methylation status may not capture its evolving resistance biology. Immunohistochemistry for the MGMT protein, once hoped to be a simpler surrogate, has been hampered by observer variability and lack of association with survival in validation studies.

The biomarker’s reach now extends well beyond glioblastoma. In IDH-mutant astrocytomas, MGMT methylation contributes to prognostic stratification, although analyses from the CATNON trial program found that the MGMT-STP27 algorithm’s methylation call was not predictive of temozolomide benefit in IDH-mutant anaplastic astrocytomas, complicating the simple predictive story. In 1p/19q-codeleted oligodendrogliomas and in 1p/19q-intact lower-grade gliomas, methylation status associates with survival after alkylating chemotherapy. Emerging data address rare entities such as H3 G34-mutant diffuse hemispheric glioma. A uniform analysis of more than 15,000 nervous system tumors from the Mayo group, cited in the review, is helping map where methylation matters diagnostically and where it is merely passenger biology. The picture that emerges is of a biomarker whose meaning shifts with tumor type, IDH status, and treatment context, demanding increasingly sophisticated interpretation.

Looking forward, the review identifies several converging trajectories. Nanopore sequencing, which reads native DNA and detects methylation directly without bisulfite conversion, has been shown by Halldorsson and colleagues to deliver accurate and comprehensive evaluation of MGMT promoter methylation, potentially eliminating an entire class of pre-analytical artifacts. Droplet digital PCR assays for MGMT methylation are undergoing analytical and clinical validation as more quantitative and robust alternatives to conventional qPCR. Artificial intelligence applied to multiparametric MRI is being trained to predict methylation status noninvasively, using radiomic features from intratumoral and peritumoral imaging, with machine learning models combining imaging and clinical data to distinguish pseudoprogression from true progression. CRISPR-based epigenetic tools, including CRISPRoff and dCas9-mediated silencing, are being used experimentally to manipulate MGMT expression and sensitize tumors to temozolomide, hinting at a future where the biomarker becomes a therapeutic target rather than merely a prognostic readout.

What makes this review resonate beyond the specialist literature is the tension it exposes at the heart of precision oncology. MGMT promoter methylation is simultaneously a triumph, a two-decade-old biomarker that still determines whether thousands of patients receive a drug that can only help them, and a cautionary tale, a test with no standardized method, contested cutoffs, and a gray zone that defies the binary answers clinicians need. Platform trials such as GBM AGILE and the Individualized Screening Trial of Innovative Glioblastoma Therapy now use methylation status to stratify patients toward experimental arms, raising the stakes of every ambiguous result. The Mayo Clinic authors’ verdict is measured but firm: the technical and interpretative challenges are ongoing, yet the biomarker’s clinical value is not in doubt. The task ahead is not to replace MGMT methylation but to tame it, with harmonized assays, quantitative reporting, and technologies that measure the epigenome as it actually exists rather than as a convenient yes-or-no label.

Subject of Research: MGMT promoter methylation as a molecular biomarker for diagnosis and treatment decisions in central nervous system tumors

Article Title: MGMT promoter methylation as a molecular biomarker in neuro-oncology: current practices, challenges, and future directions

Article References: Do, T. J., Shafqat, S., Kizilbash, S. H., & Ida, C. M. (2026). MGMT promoter methylation as a molecular biomarker in neuro-oncology: current practices, challenges, and future directions. Journal of Neuro-Oncology, 180(1), Article 16. https://doi.org/10.1007/s11060-026-05821-y

Image Credits: AI Generated

DOI: 10.1007/s11060-026-05821-y

Keywords: MGMT promoter methylation, glioblastoma, temozolomide, DNA methylation, neuro-oncology, biomarker testing, pyrosequencing, methylation-specific PCR, nanopore sequencing, epigenetics, IDH-mutant glioma, pseudoprogression

The study, published in the Journal of Neuro-Oncology, tackles a problem that has grown more pressing as whole-brain radiation therapy has fallen out of favor. Brain metastases, the spread of cancer to the brain from tumors elsewhere in the body, are estimated to occur in a substantial fraction of patients with advanced cancer, and as systemic therapies extend survival, more patients live long enough to face complications in the brain itself. Modern guidelines from ASTRO, ASCO, and the Society for Neuro-Oncology favor stereotactic radiosurgery, which concentrates high radiation doses on individual tumors while sparing healthy tissue. But large lesions, defined in this study as those with a volume of at least 4 cubic centimeters, pose a special challenge. A single massive dose of radiation risks injuring surrounding brain tissue, so clinicians must choose between single-session treatment, splitting the dose into staged sessions weeks apart, or delivering it in a handful of larger-than-conventional fractions known as hypofractionation. Which strategy works best, and at what dose, has remained genuinely uncertain.

The research team, led by Yuta Koketsu and corresponding author Toshinori Hasegawa, took advantage of a decade of experience at a single Gamma Knife center. Between 2016 and 2024, they treated patients with large brain metastases using all three approaches, individualizing both the treatment schedule and the radiation dose to each patient’s anatomy and clinical situation. For the analysis, they focused on one dominant lesion per patient and tracked it with serial imaging. Radiographic progression was defined conservatively as an increase of at least 20 percent in the longest tumor diameter from its smallest recorded size, the nadir. But the investigators went a crucial step further than most radiographic studies. They distinguished true progression, meaning enlargement that persisted or forced the team to intervene with additional treatment, from pseudoprogression, in which a lesion that initially appeared to grow later stabilized or shrank on its own without any salvage therapy. This distinction matters enormously, because treating pseudoprogression with more radiation or surgery exposes patients to harm for a problem that would have resolved by itself.

The headline result concerns dose. Using the concept of biologically effective dose, a radiobiological formula that converts different dose-and-fractionation schedules into a single number reflecting the expected biological effect on tumor cells, the team found a clear threshold. Lesions that received a BED10 of at least 57.6 Gy had a significantly lower risk of true progression, with a subdistribution hazard ratio of 0.46, meaning the hazard was cut by more than half, and the difference was statistically significant at p equals 0.02. In practical terms, the BED10 framework rewards clinicians for thinking beyond the raw dose number on a treatment plan. A modest dose delivered in a single session and a larger total dose spread over several sessions can be biologically equivalent to tumor cells, and the 57.6 Gy threshold gives clinicians a common currency for comparing schedules. Notably, the analysis found that true progression rates did not differ significantly by treatment modality, whether single-session, staged, or hypofractionated, or by tumor volume, suggesting that what matters is the total biological punch delivered rather than the specific way it is packaged.

The cumulative incidence figures tell the story of what patients face over time. At six months after Gamma Knife surgery, 17.4 percent of lesions had undergone true progression. By twelve months, that figure had climbed to 31.5 percent, and by twenty-four months it had essentially plateaued at 33.1 percent. That plateau is clinically meaningful: it suggests that if a large treated metastasis is going to truly regrow, it almost always does so within the first year, and a patient who reaches the two-year mark with local control is unlikely to lose it. The median overall survival of the entire cohort was 15.3 months, which means that for a large share of these patients, the question of local tumor control is not academic; it unfolds within their actual lifespan and directly affects their quality of life and neurological function.

Of the 143 lesions followed, 47 eventually showed radiographic progression, but the breakdown reveals how treacherous imaging interpretation can be. Forty-two turned out to be true progression, four were pseudoprogression, and one was attributed to intratumoral hemorrhage. In other words, roughly one in ten radiographic failures was not a failure at all. The pseudoprogression phenomenon reflects radiation necrosis, a delayed injury to brain tissue in and around the treated lesion that can cause the enhancing area on MRI to enlarge, mimic tumor growth, produce mass effect and edema, and even bring back a patient’s original neurological symptoms. Biopsy series from other centers have repeatedly shown that enlarging lesions after radiosurgery frequently contain necrosis rather than viable tumor. The Komaki team illustrated the point vividly in a supplementary case: a man in his seventies with a 66 cubic centimeter occipital metastasis causing hemianopsia and Gerstmann syndrome received two-stage Gamma Knife treatment with 12 Gy per stage. His symptoms resolved and the tumor shrank dramatically, but a year later the lesion enlarged again with worsening edema and recurrent vision loss, apparently demanding salvage therapy. Instead, it stabilized on its own at eighteen months and the edema improved by twenty-two months, consistent with radiation necrosis running its natural course.

How, then, can clinicians tell the impostor from the real thing? The study’s most actionable finding is a timing signature. Recurrent peritumoral edema, the reappearance or worsening of fluid accumulation in brain tissue around the lesion, developed far earlier in true progression than in pseudoprogression, with a median of 173.5 days versus 564 days, a difference that was highly significant at p equals 0.001. When tumor enlargement and renewed edema arrive together within roughly the first year after treatment, the pattern strongly indicates genuine tumor regrowth, and the authors conclude that such a finding should prompt consideration of early salvage intervention rather than a prolonged wait-and-see approach. Conversely, enlargement with edema emerging well beyond the one-year mark is more consistent with late radiation necrosis, where observation may spare the patient unnecessary surgery or reirradiation. This simple temporal rule requires no advanced imaging, no perfusion sequences, no spectroscopy, and no biopsy, only careful attention to the calendar alongside the MRI.

The study also speaks to what happens when a lesion does fail and clinicians reach for a second round of Gamma Knife treatment. Among patients who underwent repeat radiosurgery, a lower marginal dose at the second session was associated with subsequent true progression, with a hazard ratio of 0.82 per dose increment, significant at p equals 0.02. The interpretation is sobering: underdosing at salvage treatment appears to invite another failure, yet pushing the dose higher raises the specter of radiation necrosis, particularly since the surrounding tissue has already been irradiated. Repeat stereotactic radiosurgery for recurrent brain metastases is well established in the literature, with systematic reviews and meta-analyses documenting both its efficacy and its necrosis risk, and the new data add a quantitative argument for dosing salvage sessions deliberately rather than timidly. The team notes that their institution has also explored combined approaches, including reirradiation followed by resection, for select recurrent lesions.

Why does dose matter so much biologically? The answer lies in classic radiation biology. Tumor cells vary in their oxygenation, and cells at intermediate oxygen levels can substantially influence the response of a tumor to fractionated radiotherapy, a relationship explored in foundational work by Wouters and Brown. Hypoxic cells require more radiation to kill, and fractionation schedules that allow tumor reoxygenation between sessions can partially compensate, which is precisely why the BED10 framework, which encodes the effect of fractionation, is a better predictor of control than the physical dose alone. Large metastases are also more likely than small ones to contain poorly oxygenated cores and to press against eloquent brain structures that force dose reductions, creating a tension between tumor control and toxicity that the 57.6 Gy threshold helps clinicians navigate quantitatively.

The study’s limitations are those inherent to its design: it was retrospective, conducted at a single institution, and treatment decisions were individualized rather than randomized, so selection biases cannot be fully excluded. Datasets are available from the corresponding author on reasonable request, and the work received no specific external funding. Still, the practical takeaways are hard to ignore. For patients with large brain metastases, the total biologically effective dose delivered to the tumor edge should be planned to reach at least 57.6 Gy BED10, using whatever fractionation scheme best protects the surrounding brain. And in follow-up, the pairing of tumor enlargement with recurrent peritumoral edema within a year of Gamma Knife surgery should be read not as an ambiguous radiological curiosity but as a call to act, because in this disease, waiting for certainty can cost patients the window in which salvage treatment works best.

Subject of Research: Dose selection and progression assessment after Gamma Knife radiosurgery for large brain metastases

Article Title: Dose selection and clinical decision-making after Gamma Knife surgery for large brain metastases: implications for early identification of true progression

Article References: Koketsu, Y., Hasegawa, T., Kato, T., Naito, T., Mizuno, A., Ando, Y., Kosaka, N., & Saito, R. (2026). Dose selection and clinical decision-making after Gamma Knife surgery for large brain metastases: implications for early identification of true progression. Journal of Neuro-Oncology, 180(1), Article 3. https://doi.org/10.1007/s11060-026-05792-0

Image Credits: AI Generated

DOI: 10.1007/s11060-026-05792-0

Keywords: Gamma Knife surgery, stereotactic radiosurgery, brain metastases, biologically effective dose, pseudoprogression, radiation necrosis, peritumoral edema, salvage therapy, hypofractionation, tumor progression, radiation oncology, neuro-oncology

The research, conducted at Mayo Clinic in Rochester, Minnesota, describes the institution’s early experience deploying a research-based cerebrospinal fluid (CSF) cell-free DNA profiling program across its neurology, oncology, and neurosurgery departments. Beginning in August 2021 and continuing through December 2024, any clinician performing a clinically indicated lumbar puncture could request that the fluid be analyzed for tumor-derived genetic material. Seventy-six consecutive patients were enrolled under an institutional review board-approved protocol, with informed consent obtained for the brain tumor biomarkers study and for storage in the Mayo Clinic Neuro-Oncology biorepository. The result is a portrait of liquid biopsy testing as it actually happens in a busy hospital — messy, heterogeneous, and full of both promise and practical constraints.

The technical pipeline behind the study is worth understanding, because it illustrates the engineering challenge of detecting vanishingly small amounts of tumor DNA. After collection, CSF samples were kept on ice and centrifuged within one to two hours, then frozen and shipped to Predicine, Inc., where cell-free DNA was extracted using a silica-membrane kit and quantified with a fluorometer and fragment analyzer. The median yield was just 0.27 nanograms per milliliter — a tiny quantity compared with what is typically recovered from blood plasma. Depending on the amount and quality of DNA recovered, samples then underwent either targeted next-generation sequencing of a 152-gene cancer panel on an Illumina NovaSeq 6000, low-pass whole-genome sequencing to generate genome-wide copy-number profiles, or both. Error suppression was achieved by merging sequencing reads from the same DNA molecules into consensus sequences, and a variant was only called as positive if it was supported by at least three independent DNA fragments, including at least one duplex fragment, and met a stringent log-odds threshold.

Patients fell into three broad clinical categories, each representing a classic diagnostic dilemma in neuro-oncology. Forty-six percent had a brain lesion of unknown origin seen on MRI that had not yet been biopsied. Twenty-one percent had a known glioma or other tumor, and their clinicians were trying to distinguish true tumor progression from treatment-related changes such as radiation necrosis or pseudoprogression — a distinction that conventional MRI handles poorly, with sensitivity and specificity ranging from only 30 to 70 percent. The remaining 26 percent had known systemic cancer with new neurological findings, and the question was whether leptomeningeal disease — the spread of cancer into the membranes lining the brain and spinal cord — was present. Cytology, the traditional test for that condition, is notoriously insensitive: in this cohort it was negative in 16 of the 18 patients in whom it was obtained.

Across all three indications, the assay produced what the researchers call a positive tumor call — detection of a cancer-associated variant allele or copy-number alteration — in 20 of the 76 patients, or 26 percent. Another 29 patients, 38 percent, had adequate DNA but no detectable tumor signal, while 27 patients, 36 percent, had insufficient cfDNA yield or quality to permit analysis. Notably, the distribution of positive, negative, and uninformative results was remarkably similar across the three clinical questions, suggesting that the assay behaves consistently regardless of why it is ordered. Only two of the sequencing runs failed quality control, a low failure rate that the authors attribute to careful sample handling and the robustness of the extraction workflow.

One of the study’s most striking findings concerns what predicts success. Increasing the volume of CSF collected did not correlate with the amount of DNA recovered — a counterintuitive result that underscores how heterogeneous tumor DNA shedding is among patients. Instead, positive tumor calls were strongly associated with higher cfDNA concentration itself: samples with detectable tumor signals had a median yield of 0.91 nanograms per milliliter, compared with 0.30 for negative samples and essentially zero for the low-yield group. Tumor proximity to the CSF space is thought to be a major driver of this variability, since fully embedded parenchymal lesions with little contact with CSF are unlikely to release detectable DNA into the fluid. This means that a negative result must always be interpreted in light of the lesion’s anatomy, not as definitive proof of absence of disease.

When tumor DNA was detected, it told clinically meaningful stories. Low-pass whole-genome sequencing uncovered the 1p/19q co-deletion characteristic of oligodendroglioma in one patient and the chromosome 7 gain and chromosome 10 loss typical of glioblastoma in others. Targeted sequencing revealed mutations in genes including TP53, TERT, IDH1, BRAF, EGFR, and MYD88 — the latter a disease-defining alteration in primary central nervous system lymphoma. In the three patients who had also undergone clinical tissue sequencing, CSF and tissue shared a substantial fraction of mutations, but each patient also carried mutations found only in the spinal fluid, a phenomenon the authors attribute to spatial heterogeneity within tumors or to genomic evolution between the times of tissue and fluid sampling.

The study also identified a quantitative biomarker with prognostic power. By summing genome-wide copy-number deviations into a single copy-number burden score, the team found that a threshold of 7.42 cleanly separated samples with true tumor-specific copy-number alterations from those without, achieving an area under the receiver operating curve of 0.998. Patients whose CSF exceeded that threshold had significantly worse overall survival, with a hazard ratio of 3.4 after controlling for the clinical indication. More broadly, patients with any positive tumor call survived significantly less long than those with negative calls or low DNA yield — a difference driven largely by the unknown-diagnosis group, with a similar trend among patients being evaluated for leptomeningeal disease. In other words, even a negative result carries information: abundant tumor DNA in spinal fluid is a bad sign, while its absence, in the right clinical context, is reassuring.

Three case vignettes bring the statistics to life. A 72-year-old man with chronic lymphocytic leukemia presented with confusion and diffuse brain abnormalities on MRI; his CSF revealed chromosome 12 amplification consistent with trisomy 12 CLL, plus KRAS and BRAF mutations, helping confirm rare central nervous system involvement of his leukemia. A 38-year-old man with metastatic gastroesophageal adenocarcinoma showed an extensive copy-number burden and TP53, NTRK1, and EGFR mutations in his spinal fluid, confirming leptomeningeal spread and guiding treatment decisions. And a 51-year-old woman with a surveillance question about her grade 3 oligodendroglioma had her tumor’s molecular signature — the 1p/19q co-deletion and IDH1 R132H mutation — detected in CSF months before imaging confirmed ventricular progression, ultimately informing her enrollment in an IDH inhibitor trial.

The authors are candid about the limitations. The copy-number burden threshold was derived post hoc without an independent validation cohort, tissue sequencing was available for only a handful of patients, and cytology’s poor sensitivity made it an imperfect gold standard for leptomeningeal disease. For progression-versus-pseudoprogression questions, the mixture of recurrent tumor and treatment effects likely explains why survival did not differ by tumor-call status in that subgroup. Still, the study demonstrates that CSF cfDNA sequencing can be deployed across the full breadth of a real neuro-oncology practice with acceptable failure rates, and that requests for testing have grown substantially since the program began. As low-input sequencing platforms mature, the authors argue, spinal fluid liquid biopsy is poised to become a routine complement — and in some cases a safer alternative — to the neurosurgeon’s needle.

Subject of Research: Cerebrospinal fluid cell-free DNA sequencing as a liquid biopsy for diagnosing and monitoring brain tumors and leptomeningeal disease in neuro-oncology practice

Article Title: Early experience with cerebrospinal fluid cell-free DNA molecular profiling in a neuro-oncology practice

Article References: Riviere-Cazaux, C., Kumar, R., Rechberger, J. S., Obiri-Yeboah, D., Dai, C., Li, J., Huang, Y., Warrington, A. E., Sharif, E. H., Palmer, E. A., Wang, X., Lachance, D. H., Kizilbash, S. H., Ruff, M. W., Carabenciov, I. D., Fortin Ensign, S. P., Zadeh, G., Neth, B. J., Sener, U., … Burns, T. C. (2026). Early experience with cerebrospinal fluid cell-free DNA molecular profiling in a neuro-oncology practice. Journal of Neuro-Oncology, 180(1), Article 2. https://doi.org/10.1007/s11060-026-05799-7

Image Credits: AI Generated

DOI: 10.1007/s11060-026-05799-7

Keywords: cerebrospinal fluid, cell-free DNA, liquid biopsy, neuro-oncology, brain tumors, glioma, leptomeningeal disease, next-generation sequencing, copy-number alterations, pseudoprogression, Mayo Clinic, biomarkers

The rationale behind the trial was grounded in a genuine biological puzzle. In an earlier phase 1 study, inhaled recombinant human interleukin-15, or rhIL-15, proved safe in dogs with metastatic osteosarcoma and melanoma and produced durable responses in a subset of animals. Intriguingly, dogs that entered the trial with lower absolute lymphocyte counts were statistically more likely to benefit. The Davis team hypothesized that reducing lymphocytes before treatment might remove so-called cytokine sinks, circulating immune cells that soak up interleukin-15 before it can reach the natural killer cells and CD8-positive memory T cells the drug is meant to energize.

Lymphodepletion is a well-established strategy in human adoptive cell therapy, where chemotherapy is routinely used before infusing engineered T cells to improve their expansion and persistence. Doxorubicin, a mainstay chemotherapy agent in veterinary medicine, had previously been shown to reduce circulating T lymphocytes in tumor-bearing dogs. The researchers therefore designed a trial in which dogs received a standard dose of doxorubicin, 30 milligrams per square meter for larger animals or 1 milligram per kilogram for smaller ones, followed one week later by two weeks of inhaled rhIL-15 at 50 micrograms twice daily, delivered by nebulizer at home by trained owners.

The study population consisted of ten client-owned dogs with histologically or cytologically confirmed melanoma or bone tumors, all carrying at least one lung lesion of a centimeter or larger on thoracic radiographs. Five dogs had osteosarcoma, four had melanoma, and one had a multilobular tumor of bone. Eligibility criteria were strict: dogs needed adequate organ function, no prior rhIL-15 exposure, no recent chemotherapy or immunotherapy, and screening echocardiograms to rule out cardiac disease that could be worsened by doxorubicin. The trial was approved by the UC Davis institutional review boards, and all owners provided informed consent.

When the results came in, the numbers told a sobering story. Two dogs progressed before the Day 35 response assessment, leaving eight evaluable patients. One dog achieved a complete response that has now lasted more than two years, two had stable disease, and five progressed. That translates to an overall response rate of 10 percent and a clinical benefit rate of 30 percent, figures essentially indistinguishable from the phase 1 trial of inhaled rhIL-15 alone, which reported a response rate of 11 percent and a clinical benefit rate of 39 percent. The trial had been powered to detect a jump in response rate from 11 to 50 percent, so the null result is unlikely to be a statistical fluke of modest effect size.

The hematological data confirmed the team did achieve the biological precondition they set out to create. Seven days after doxorubicin, mean absolute lymphocyte counts fell from 1,452 to 1,023 cells per microliter, a statistically significant drop that brought the dogs to lymphocyte levels comparable to those seen in responding patients from the earlier trial. White blood cells, monocytes, and neutrophils also declined significantly, and one dog developed grade 4 neutropenia with fever that resolved within two days of hospitalization and supportive care. Crucially, adding doxorubicin did not increase the toxicity of the inhaled cytokine therapy itself, and no allergic reactions or pulmonary adverse events were attributed to rhIL-15.

Transcriptomic analysis added an unexpected layer of insight. Using the NanoString nCounter canine immuno-oncology panel covering 780 genes, the researchers profiled peripheral blood mononuclear cells collected before doxorubicin, after doxorubicin, and after two weeks of inhaled rhIL-15. While overall lymphocyte counts fell, the relative proportions of T cells, CD8-positive T cells, exhausted CD8-positive T cells, natural killer cells, and Th1 cells did not change significantly, indicating that doxorubicin depleted lymphocytes indiscriminately rather than selectively removing suppressive populations. The one clear casualty was the B cell compartment: B cell scores and B cell function genes, including CD19, CD79B, MS4A1, and immunoglobulin genes, dropped significantly and had not recovered by Day 21, consistent with the rapid proliferation rate of B cells making them especially vulnerable to chemotherapy.

Perhaps the most tantalizing clue came from the single dog that achieved a complete response. That patient initially appeared to progress at Day 35, with new lung lesions appearing on radiographs, but after receiving palliative radiation to its primary mandibular melanoma, every lung lesion resolved by week 15. The researchers suggest this may represent pseudoprogression, a well-recognized phenomenon in human immunotherapy where apparent tumor growth reflects immune cell infiltration rather than true malignancy, though an abscopal effect triggered by the radiation cannot be excluded. Transcriptomically, this dog trended toward higher T cell, CD8-positive T cell, and cytotoxic cell scores, and started with a lower exhausted CD8 T cell score than non-responders, hinting that baseline immune fitness may predict who benefits from inhaled rhIL-15.

The study carries important limitations that the authors acknowledge candidly. Ten patients is a small cohort, and two dogs did not survive to formal response assessment. The therapy used human rather than canine interleukin-15, which likely provoked neutralizing antibodies, although the two-week treatment window was chosen to minimize that risk. Target lesions smaller than the standard 2-centimeter RECIST threshold were permitted, matching the phase 1 design but introducing radiographic measurement imprecision. Tumor type may also have influenced outcomes, since one of the stable disease patients had a multilobular tumor of bone, an inherently slower-growing cancer.

Even so, the trial delivers a valuable double message. First, it definitively rules out standard-dose doxorubicin preconditioning as a way to enhance inhaled rhIL-15 efficacy, redirecting the field toward other combination strategies such as radiation, which has already shown promise with cytokine therapy in canine sarcoma and melanoma. Second, it confirms that durable, multi-year complete responses are achievable even in dogs with diffuse pulmonary metastatic disease, a population whose prognosis otherwise hovers below 80 days. For the dogs that respond, and for the human osteosarcoma and melanoma patients their disease so closely mirrors, that signal is worth chasing with better-designed combinations.

Subject of Research: Doxorubicin preconditioning and inhaled recombinant human IL-15 immunotherapy in dogs with pulmonary metastasis

Article Title: Evaluating the impact of doxorubicin preconditioning on the efficacy of inhaled recombinant human IL-15 immunotherapy in dogs with pulmonary metastasis

Article References: Luker, M. E., York, D., Al-Nadaf, S., Johnson, E. G., Withers, S. S., Cruz, S. M., Lejeune, A., Skorupski, K. A., Young, J., Cohen-Davidyan, T., Sparger, E. E., Culp, W. T., Murphy, W. J., Kent, M. S., Canter, R. J., & Rebhun, R. B. (2025). Evaluating the impact of doxorubicin preconditioning on the efficacy of inhaled recombinant human IL-15 immunotherapy in dogs with pulmonary metastasis. Veterinary Oncology, 2(1), Article 23. https://doi.org/10.1186/s44356-025-00040-5

Image Credits: AI Generated

DOI: 10.1186/s44356-025-00040-5

Keywords: immunotherapy, interleukin-15, doxorubicin, canine osteosarcoma, melanoma, lung metastasis, lymphodepletion, veterinary oncology, natural killer cells, CD8 T cells, pseudoprogression, transcriptomics