Scientists Urge Caution Before Using MGMT to Guide Treatment of Rare Childhood Brain Cancer

A rare and devastating childhood brain tumor has become the focus of an unusually pointed scientific dispute, and the outcome could shape how doctors treat young patients with one of the most challenging diagnoses in pediatric neuro-oncology. In a correspondence published in Acta Neuropathologica, researchers Fang Shen of Melbourne Pathology in Australia and Hansong Sheng of Wenzhou Medical University in China argue that claims about MGMT gene expression in pediatric H3 G34-mutant diffuse hemispheric glioma rest on shaky technical ground, and that those problems must be resolved before the biomarker is used to guide therapy in the clinic.

H3 G34-mutant diffuse hemispheric glioma is a defined molecular subtype of high-grade glioma that arises in the cerebral hemispheres of children and young adults. Like its better-known adult counterpart, glioblastoma, it carries a poor prognosis, and clinicians have long searched for molecular signals that might predict which patients will benefit from temozolomide, the oral chemotherapy drug that is a cornerstone of adult glioma treatment. In adult glioblastoma, methylation of the MGMT promoter, which silences the DNA repair enzyme O6-methylguanine DNA methyltransferase, is a well-established predictor of temozolomide benefit. Whether the same logic applies to pediatric H3 G34-mutant tumors has been an open and clinically urgent question.

That question appeared to move closer to an answer when Tlais and colleagues published a study in Acta Neuropathologica in March 2026 examining clinical, imaging and molecular prognostic factors in this tumor type, including MGMT expression and temozolomide response. The study attracted wide attention because it suggested that MGMT expression might carry prognostic and predictive weight in a disease where treatment options remain desperately limited. For families facing this diagnosis, the prospect of a simple molecular test that could indicate whether an aggressive chemotherapy drug is worth its toxicity was understandably compelling.

Shen and Sheng, however, were not convinced that the data could bear that weight. Rather than generating new experiments, the two researchers did what modern science increasingly allows: they went back to the raw material. Their letter reports a reanalysis of publicly available RNA sequencing feature-count data for twenty samples classified as H3 G34-mutant diffuse hemispheric glioma, obtained from the St. Jude Cloud platform under dataset accessions SJC-DS-1001, SJC-DS-1007 and SJC-DS-1014, together with the supplementary data published alongside the original Tlais study. No new data were generated; the entire exercise was a forensic examination of existing numbers.

The central concern raised in the correspondence is captured in its subtitle: technical covariates. In large-scale gene expression analysis, the measurements scientists care about, such as differences in MGMT expression between groups of tumors, can be contaminated by variables that have nothing to do with biology. Batch effects, the systematic differences introduced when samples are processed, sequenced or analyzed in different runs, are among the most notorious offenders in high-throughput genomics. A decade of methodological literature, including a widely cited 2010 review by Leek and colleagues in Nature Reviews Genetics, has documented how these artifacts can masquerade as biological discoveries if they are not explicitly identified and corrected.

Another technical hazard is RNA quality, particularly in tissue preserved as formalin-fixed paraffin-embedded material, the standard archival format in pathology departments worldwide. Chemical modification and fragmentation of RNA during fixation and storage mean that FFPE-derived samples can yield degraded genetic material whose measured expression profiles differ systematically from those of fresh-frozen tissue. A 2007 study by von Ahlfen and colleagues in PLoS ONE documented these determinants of RNA quality in FFPE samples, and the issue remains a routine consideration in any expression study that mixes preservation methods. If the samples in a comparison differ in how they were preserved or processed, apparent differences in gene expression may reflect the chemistry of preservation rather than the biology of the tumor.

Shen and Sheng’s reanalysis, conducted with standard differential expression tooling of the kind exemplified by the DESeq2 package described by Love, Huber and Anders in Genome Biology in 2014, led them to conclude that such technical covariates genuinely complicate the MGMT expression findings in this dataset. Their argument is not that MGMT is irrelevant in these tumors, but that the evidence base is not yet clean enough to support clinical decision-making. In their framing, the covariates should be addressed before clinical use, a deliberately measured phrase that nonetheless carries significant practical consequences for a field eager to translate molecular findings into treatment guidance.

The stakes of this debate extend well beyond a single gene and a single tumor subtype. Temozolomide is not a benign drug; it causes myelosuppression and other toxicities, and exposing a child to chemotherapy without a reliable expectation of benefit is a real cost. Conversely, withholding a potentially effective therapy because a flawed analysis suggested no benefit would be equally harmful. Biomarker-driven treatment decisions demand that the underlying measurements be robust, and the correspondence is a reminder that the path from a genomic dataset to a clinical test runs through unglamorous but essential quality control.

The episode also illustrates the changing sociology of scientific correction. Reanalyses of public data, once rare and often resented, have become a routine mechanism of accountability in genomics, where repositories such as St. Jude Cloud make raw feature counts available to any researcher with the skills to interrogate them. Shen performed the data reanalysis and drafted the manuscript, while Sheng contributed clinical context and critical revision; both authors reviewed and approved the final version and take responsibility for its integrity. The work received no dedicated funding, and the authors declare no conflicts of interest, details that matter when the subject is a contested biomarker with commercial and clinical implications.

For now, the practical message for clinicians and researchers is one of disciplined patience. The correspondence, published on 14 October 2026 in volume 152 of Acta Neuropathologica as article number 55, does not close the door on MGMT as a biomarker in pediatric H3 G34-mutant diffuse hemispheric glioma; it insists that the door not be opened prematurely. Resolving the question will require analyses that explicitly model and correct for batch effects, preservation-related RNA degradation and other technical covariates, ideally across larger and more uniformly processed cohorts. Until then, the fate of children with this rare tumor should, the authors argue, be decided by evidence that can survive scrutiny, not by expression patterns that may be artifacts of the laboratory rather than signatures of the disease.

Subject of Research: Reanalysis of MGMT gene expression data in pediatric H3 G34-mutant diffuse hemispheric glioma highlighting technical covariates affecting clinical biomarker use

Article Title: MGMT expression in pediatric H3 G34-mutant diffuse hemispheric glioma: technical covariates should be addressed before clinical use

Article References: MGMT expression in pediatric H3 G34-mutant diffuse hemispheric glioma: technical covariates should be addressed before clinical use. (n.d.). https://doi.org/10.1007/s00401-026-03106-2

Image Credits: AI Generated

DOI: 10.1007/s00401-026-03106-2

Keywords: MGMT, pediatric glioma, H3 G34-mutant, diffuse hemispheric glioma, temozolomide, RNA-seq, batch effects, biomarker, Acta Neuropathologica, St. Jude Cloud, FFPE, gene expression