The biological signal is striking. GSTP1 encodes the pi-class glutathione S-transferase, which participates in glutathione conjugation and protection against oxidative and genotoxic stress. The gene is strongly expressed in normal basal prostate cells, can be induced in proliferative inflammatory atrophy, and is frequently silenced in high-grade prostatic intraepithelial neoplasia and prostate adenocarcinoma. Experiments in prostate cancer cells showed that reversing promoter methylation restored both GSTP1 transcription and protein expression, while methylation of promoter constructs reduced promoter activity. This supports a direct relationship between the epigenetic alteration and transcriptional silencing, although restoring GSTP1 did not reverse the malignant growth of already transformed cells.

More recent tissue analysis indicates that methylation accumulates as prostate lesions evolve. In a study of normal epithelium, proliferative inflammatory atrophy, high-grade prostatic intraepithelial neoplasia and cancer from 32 prostatectomy specimens, 0.52% and 1.3% of evaluated CpG sites were methylated in normal and atrophic epithelium, compared with 24% in high-grade intraepithelial neoplasia and 51% in invasive carcinoma. Methylation clustered near the transcription start site and binding regions for AP1 and SP1 transcription factors. Separate studies have detected GSTP1 methylation in 91.1% of malignant tissues versus 5.4% of adjacent healthy tissue and in 97% of radical-prostatectomy tumors, establishing a powerful malignant–benign contrast in specimens rich in tumor DNA.

That strength comes with important qualifications. GSTP1 methylation appears early in carcinogenesis and is highly prevalent, meaning it may identify malignant prostate epithelium without reliably distinguishing consequential from indolent tumors. In a 2024 analysis, higher tumor methylation was associated with biochemical recurrence and metastatic relapse in univariable models, but the association did not remain independent after adjustment for established clinicopathological factors. The marker also exists in heterogeneous forms: partially methylated promoter alleles and different patterns in separate tumor foci can coexist. Which CpG sites an assay interrogates may therefore change not only whether a sample appears positive, but also the amount of signal measured. The review treats GSTP1 principally as a marker of malignant tissue, not a stand-alone measure of tumor aggressiveness.

Measurement conditions are equally consequential. Tissue contains the greatest proportion of tumor-derived DNA and consequently gives the most reproducible signal. In urine, prostate cells and nucleic acids can be recovered non-invasively, but collection method, prostate manipulation, preservatives, processing delays, temperature, pH, storage and DNA extraction all affect recovery. Urinary cell-free DNA is especially vulnerable to degradation and loss of short fragments. Clinical findings in 2000 already illustrated the gap between tissue and urine: one study detected methylated GSTP1 in 94% of tumors but only 36% of urine samples, with higher detection after prostate massage. Urine is therefore promising, but “urinary GSTP1 test” does not describe a single standardized measurement.

A recent 120-patient study suggests that a carefully controlled workflow can improve urinary performance. Among biopsy-referred patients, including 59 with cancer and 61 with benign prostatic hyperplasia, standardized post-examination first-void urine analysis reported 81.4% sensitivity, 83.6% specificity, 82.5% accuracy and an area under the receiver operating characteristic curve of 0.87. In the PSA grey zone of 4–10 nanograms per milliliter, the corresponding values were 87.5%, 86.2%, 86.7% and 0.90. These findings were encouraging but came from a retrospective, single-center cohort already selected for biopsy, and the endpoint was any biopsy-confirmed cancer rather than clinically significant disease defined as Grade Group 2 or higher. Comparison with PSA alone also did not establish that GSTP1 adds useful information beyond PSA density, risk calculators, multiparametric MRI and other biomarkers.

Blood-based results are more limited because circulating tumor DNA may account for about 1% or less of total cell-free DNA in localized disease. A 2023 meta-analysis of cell-free-DNA studies pooled 37% sensitivity and 97% specificity, with substantial heterogeneity, indicating that a positive result may be informative while a negative result cannot reliably exclude a small localized tumor. A smaller 2024 study illustrates the variation: GSTP1 methylation was detected in only 3 of 13 cancers, producing 23% sensitivity and 32% specificity among 39 biopsy-received men and 15 healthy controls. Adding RASSF1 improved both measures, whereas requiring GSTP1, RASSF1 and RASSF2 raised specificity to 83% but reduced sensitivity to 8%. Multigene testing therefore does not automatically produce simultaneous gains in sensitivity and specificity.

Assay technology has advanced from qualitative methylation-specific polymerase chain reaction to quantitative, digital and sequencing approaches, but the review cautions that greater molecular resolution has not yet yielded a uniformly validated stand-alone test. Quantitative assays can be affected by unmethylated CpG sites located within primer and probe sequences, especially in urine containing partially methylated epialleles. Bisulfite conversion remains practical for PCR assays but can damage scarce DNA; enzymatic conversion and third-generation sequencing can preserve molecular integrity while introducing greater cost and bioinformatics demands. A small single-molecule plasma study found that combinations of methylation patterns at GSTP1 and two other loci could improve discrimination, but its 20 cancer cases, 17 patients with benign prostatic hyperplasia and 18 healthy donors raised concerns about model overfitting. Prespecified targets, thresholds and independent validation are therefore essential.

The clearest clinical niche is the repeat-biopsy setting, where ConfirmMDx measures GSTP1, APC and RASSF1 methylation in histologically negative biopsy tissue to identify an epigenetic field effect associated with potentially unsampled cancer. The American Urological Association and Society of Urologic Oncology lists the assay among biomarkers available for this setting. Population screening, pre-MRI triage, assessment after equivocal imaging and risk refinement in the PSA grey zone remain proposed research contexts rather than established indications. For pre-biopsy studies, Grade Group 2 or higher cancer should be the principal endpoint, and prospective multicentre testing should compare GSTP1-assisted care with contemporary pathways that already include risk models, established biomarkers and multiparametric MRI. The decisive measure would be whether the assay reduces unnecessary biopsies or scans without increasing missed clinically significant cancers, rather than whether it merely improves stand-alone sensitivity, specificity or area under the curve.

Subject of Research: Review of GSTP1 promoter methylation as a prostate cancer detection biomarker and its clinical translation

Article Title: GSTP1 promoter hypermethylation in prostate cancer detection: current evidence, clinical positioning, and translational gaps

Article References: Joo, J.-L., Wong, R. S.-Y., Goh, B. H., & Ng, N. C.-S. (2026). GSTP1 promoter hypermethylation in prostate cancer detection: current evidence, clinical positioning, and translational gaps. Molecular Biology Reports, 53(1), Article 1777. https://doi.org/10.1007/s11033-026-13011-2

Image Credits: AI Generated

DOI: 10.1007/s11033-026-13011-2

Keywords: GSTP1 methylation, prostate cancer, DNA methylation, liquid biopsy, urine biomarkers, cell-free DNA, ConfirmMDx, prostate cancer screening, multiparametric MRI, clinical validation, GSTP1, promoter

Ovarian cancer accounted for 324,398 new cases and 206,839 deaths globally in GLOBOCAN 2022, a mortality burden disproportionately high relative to incidence. Outcomes are more favorable when disease is detected early, but large screening efforts combining CA125 and transvaginal ultrasonography have not demonstrated sufficient survival benefit to support population screening. CA125 also has limited sensitivity in early-stage disease and can be elevated in benign and non-ovarian malignant conditions. Modelling cited by the review indicates that an early-stage screening test would need sensitivity above 75% and specificity of at least 99.6% to achieve a positive predictive value of 10% or greater.

DNA methylation offers a possible signal because aberrant methylation may arise early in tumourigenesis and can be measured in circulating cell-free DNA. However, plasma cell-free DNA contains material from many tissues, and only a fraction may be tumour-derived. Delays in blood processing can increase contamination with high-molecular-weight DNA, while methylation conversion, extraction and threshold selection can further affect measured signal. These issues are particularly consequential when early tumours release relatively few tumour-derived molecules.

Single-gene studies produced variable results. RASSF1A sensitivities of 25.4%, 37.7% and 85.7% were reported across three blood-based studies. OPCML showed the strongest overall performance among commonly assessed single-gene markers, with reported sensitivities of approximately 80.0% to 97.8%, but it has not been reproduced in large independent prospective cohorts. In a North Indian case-control study, serum HOXA9 and HIC1 methylation had sensitivities of 62.2% and 71.1%, respectively, at 100% specificity, although the stage I–II subset contained only 10 patients and controls were healthy rather than women with benign pelvic disease.

Multi-gene panels often improved discrimination within their source cohorts. A HOXA9–HIC1 combination tested in 45 epithelial ovarian cancer cases and 25 healthy controls achieved 88.9% sensitivity, 100% specificity and an area under the curve of 0.95, and was positive in all 10 stage I–II serum cases. Separately, an 11-marker random-forest model achieved 79% sensitivity at 96% specificity in 91 ovarian cancer cases and 91 controls and identified all five stage I–II high-grade serous cancers included in its plasma cohort. These small early-stage denominators and retrospective case-control designs do not establish screening performance.

Pre-diagnostic testing exposed a substantial gap between diagnostic discrimination and early detection. The three-region WID-cfOC score achieved 97.6% specificity and 66.7% sensitivity across 27 cancer cases in a diagnostic cohort that included women with benign pelvic pathology. Applied to prediagnostic samples from 29 women who subsequently developed ovarian cancer and 29 risk-matched controls, however, sensitivity was 27.3% for high-risk cancers with relatively lower genomic-DNA contamination and 33.3% in samples collected within one year of diagnosis, despite 100% specificity. Neither methylation nor CA125 detected stage I cancers in the diagnostic cohort, and combining the tests did not improve sensitivity beyond CA125 alone in the pre-diagnostic subset.

Machine-learning classifiers produced encouraging but threshold-dependent results. A transformer-based MethylBERT model evaluated 493 cfDNA methylation markers in 754 ovarian cancer cases, including 205 early-stage cases, and 1,118 healthy women. In its held-out internal validation set, the model achieved 89.24% sensitivity at 94.39% specificity and 79.45% sensitivity among 73 early-stage cases. Raising specificity from about 95% to 99.2% reduced early-stage sensitivity to 63.01%. Combining the classifier with CA125 increased early-stage sensitivity from 82.51% to 89.62% but reduced specificity from 97.36% to 93.55%, illustrating the trade-off between sensitivity and false-positive classification.

Validation quality remains a central limitation. A tissue-based high-grade serous classifier developed with more than 850,000 probes achieved an area under the curve of 0.84 (95% CI 0.76–0.93) in external validation after reduction to nine probes, compared with 0.98 for a much larger model. Its tissue samples, predominantly advanced-stage disease and a largely homogeneous population limit extrapolation to blood-based early detection. Across the wider literature, retrospective case-control designs, small FIGO stage I–II subsets, predominance of serous disease and healthy-control groups limit generalisability. Histotype-specific MethylBERT estimates also rested on only 43 serous, 11 endometrioid, eight mucinous and six clear-cell early-stage cases.

Assay chemistry and pre-analytical procedures add further uncertainty. Targeted PCR methods are comparatively practical for predefined loci, but bisulfite conversion can cause DNA fragmentation and loss, with reported recovery rates ranging from approximately 22% to 66% across conversion kits. Enzymatic and enrichment methods can reduce some conversion losses and work with low DNA input, but they have not established ovarian cancer screening performance. In 29 reviewed blood-based studies, only six reported the interval between collection and processing, only four used double centrifugation, and only 11 used extraction kits designed specifically for cell-free DNA. The review’s bounded conclusion is that methylation assays remain promising research components rather than clinically validated population-screening tests.

Subject of Research: Evaluation of DNA methylation biomarkers for early detection of ovarian cancer

Article Title: DNA methylation biomarkers for early detection of ovarian cancer

Article References: Leow, J. Y.-S., Ng, N. C.-S., Wong, R. S.-Y., Ming, L. C., Yap, T. L., & Goh, B. H. (2026). DNA methylation biomarkers for early detection of ovarian cancer. Molecular Biology Reports, 53(1), Article 1562. https://doi.org/10.1007/s11033-026-12750-6

Image Credits: AI Generated

DOI: 10.1007/s11033-026-12750-6

Keywords: ovarian cancer, DNA methylation, cell-free DNA, early detection, liquid biopsy, biomarkers, machine learning, CA125, diagnostic accuracy, methylation, early, detection

The research team, led by Guangqi Li and colleagues at the General Hospital of Ningxia Medical University, conducted a three-part investigation. First, they performed a one-year retrospective epidemiological analysis of 16,970 blood culture specimens processed between January and December 2024. Second, they assembled a retrospective clinical cohort of 295 patients with culture-confirmed BSI, extending the study period from January 2022 to December 2024 to capture enough polymicrobial cases, which are relatively rare. Third, they carried out a prospective molecular comparison in 20 patients with suspected BSI, running three detection methods in parallel: conventional blood culture, 16S rDNA PCR library construction, and direct 16S rDNA amplicon high-throughput sequencing.

The epidemiological survey revealed that 1,321 of the 16,970 blood culture specimens—7.8 percent—were positive for microbial growth. The overwhelming majority, 96.6 percent, were classified as monomicrobial infections, while polymicrobial BSIs accounted for just 3.4 percent. Escherichia coli emerged as the most frequently identified pathogen, followed by coagulase-negative staphylococci and Brucella species. The prominence of Brucella is notable: the Ningxia Hui Autonomous Region, where the study was conducted, is an area where human brucellosis remains a significant public health concern linked to extensive livestock farming and pastoral communities, with Brucella melitensis the predominant circulating species. The authors caution that this regional epidemiological background shapes the pathogen spectrum and may not generalize to other healthcare settings.

To understand what distinguishes polymicrobial infections clinically, the researchers compared inflammatory markers between the monomicrobial and polymicrobial groups in the 295-patient cohort. The differences were striking. Polymicrobial BSI was associated with significantly higher white blood cell (WBC) counts, neutrophil percentage, absolute neutrophil count, neutrophil-to-lymphocyte ratio, neutrophil-to-platelet ratio, systemic immune-inflammation index, systemic inflammatory response index, and procalcitonin (PCT) levels—all with P values below 0.05. Median WBC counts reached 15.18 × 10⁹/L in the polymicrobial group versus 10.13 × 10⁹/L in the monomicrobial group, while median PCT levels were 14.00 ng/mL compared with 1.40 ng/mL. Moreover, within polymicrobial infections, those involving mixed Gram types—both Gram-positive and Gram-negative organisms—showed even higher inflammatory burdens than single Gram-type infections, with median PCT soaring to 31.00 ng/mL in mixed cases.

Multivariable logistic regression confirmed that WBC count and PCT were independent predictors of polymicrobial BSI. Elevated WBC carried an odds ratio of 1.723 (95 percent confidence interval 1.472–2.017, P < 0.001), while PCT carried an odds ratio of 1.028 (95 percent confidence interval 1.003–1.054, P = 0.031). For mixed Gram-type infections specifically, WBC (OR = 1.486) and PCT (OR = 1.042) were again independently associated. Receiver operating characteristic (ROC) analysis quantified the discriminatory power of these markers. WBC alone achieved an area under the curve (AUC) of 0.890, with 77.1 percent sensitivity and 90.0 percent specificity at a cutoff of 12.54 × 10⁹/L. PCT alone achieved an AUC of 0.910, with 88.0 percent sensitivity and 80.0 percent specificity at 3.20 ng/mL. Combining the two markers improved performance further, yielding an AUC of 0.927 with balanced sensitivity of 84.0 percent and specificity of 91.8 percent. For identifying mixed Gram-type infections, the combined markers performed even better, reaching an AUC of 0.950 with 94.3 percent sensitivity and 89.8 percent specificity.

These biomarker findings carry practical significance. Because polymicrobial BSIs are associated with higher mortality, increased treatment failure, and greater clinical complexity than monomicrobial infections, identifying patients at elevated risk early in their clinical course could guide decisions about escalating empiric antimicrobial coverage and pursuing additional molecular testing. The authors suggest that when WBC and PCT monitoring suggests a possible polymicrobial infection, 16S rDNA molecular diagnosis can be deployed as a supplementary method to achieve more rapid identification of the full pathogen spectrum.

The molecular comparison cohort of 20 patients provided the study’s most technically revealing results. For the clone-based workflow, cell-free DNA (cfDNA) extracted from plasma was subjected to broad-range PCR amplification of bacterial 16S rDNA using universal primers 16F and 1492R, producing amplicons of approximately 1,500 base pairs—consistent with the near-full-length 16S rRNA gene. The purified amplicons were ligated into T5 cloning vectors and transformed into Escherichia coli DH5α competent cells. Positive clones were screened by colony PCR with M13 primers and then subjected to amplified ribosomal DNA restriction analysis (ARDRA) using the restriction enzymes HhaI and RsaI. Clones displaying identical restriction fragment patterns were clustered into operational taxonomic units (OTUs), and representative clones from each OTU were sequenced by Sanger sequencing, with taxonomic classification determined by BLAST alignment against the NCBI GenBank database. This cloning step physically separates individual rDNA sequences before sequencing, which is the method’s defining feature.

The three methods produced distinctly different detection profiles. Blood culture detected 0 to 1 microbial taxon per patient, identifying one pathogen in 17 patients and returning negative results in 3 patients. The clone-based 16S rDNA PCR library method detected a broader spectrum, identifying 0 to 5 taxa per patient, and uncovered 16 bacterial species overall—including Escherichia coli, Klebsiella pneumoniae, and Chryseobacterium chrysogenum as the most frequently detected taxa—compared with only 10 species detected by culture. Direct 16S rDNA amplicon high-throughput sequencing showed the broadest detection range of all, with more than five taxa detected in 9 of the 20 patients. At the genus level, the most abundant taxa in the sequencing data included Escherichia (23.3 percent), Streptococcus (18.6 percent), and Staphylococcus (14.0 percent). Mixed bacterial signals—defined as detection of two bacterial species in a single sample by the 16S method—were observed in all 20 samples analyzed by the library approach.

When blood culture was used as the reference standard, the clone-based method demonstrated an overall percent agreement of 80.0 percent with a sensitivity of 88.2 percent, whereas high-throughput sequencing showed a concordance rate of 70.0 percent and a sensitivity of 82.4 percent. Concordance analysis between blood culture and the library method classified results as completely consistent in 6 cases (30 percent), partially consistent in 10 cases (50 percent), and inconsistent in 4 cases (20 percent). The researchers interpret the higher agreement of the clone-based approach cautiously, noting that it may partly reflect the narrower taxonomic yield compared with high-throughput sequencing. However, because the library method detected more taxa than culture, its higher concordance cannot be explained simply by an inability to detect additional organisms. Instead, the cloning step appears to provide an additional layer of signal separation, reducing redundant sequencing signals from highly abundant organisms and allowing individual sequence reads to be traced back to representative clones—physical material that is retained for downstream confirmatory testing, an advantage high-throughput sequencing lacks.

The study’s authors position the 16S rDNA PCR library approach not as a replacement for modern next-generation sequencing platforms but as a practical, comparatively lower-cost adjunctive strategy for settings where broad molecular detection is needed but standardized NGS interpretation remains challenging. High-throughput sequencing, despite its superior depth, suffers from the absence of clinically validated diagnostic thresholds, making it difficult to distinguish true pathogens from commensal microbiota, environmental contaminants, and low-abundance background signals. A multilaboratory assessment involving 90 metagenomic NGS laboratories previously documented substantial interlaboratory variability in microbial identification, particularly for low-abundance organisms. The study also illustrates the library method’s real-world value through case examples: in an immunocompromised patient with lung adenocarcinoma whose blood cultures were negative, the 16S library successfully detected Streptococcus constellatus, cross-validated by a parallel wound secretion culture; in a case of hollow viscus perforation where empirical antibiotics rendered routine cultures sterile, the library identified Klebsiella pneumoniae and Escherichia coli, matching the anatomical source of gastrointestinal leakage and validating the clinician’s escalated antibiotic strategy.

Limitations remain. The single-center design and modest sample size constrain generalizability, the exclusion of viral and fungal co-infections leaves the true polymicrobial burden incompletely assessed, and the 30-to-36-hour turnaround time for library construction—although optimized with automated instruments and simultaneous multi-sample processing—remains impractical for acute clinical decisions. Universal primers also introduce amplification bias against highly divergent 16S sequences such as Mycoplasma, and fungal pathogens, which are critical in polymicrobial BSI among neutropenic patients, are not captured by the current bacterial workflow. Cost is another barrier, with per-sample molecular testing expenses often an order of magnitude higher than traditional culture. Nevertheless, the study demonstrates that WBC and PCT can serve as accessible biomarkers to triage patients for molecular testing, and that the cloning-based 16S rDNA PCR library offers a balanced middle ground—broader than culture, more focused than direct amplicon sequencing, and traceable at the sequence level—for unraveling the microbial complexity of polymicrobial bloodstream infections.

Subject of Research: Molecular detection of polymicrobial bloodstream infections using 16S rDNA PCR library amplicons compared with blood culture and high-throughput sequencing

Article Title: Enhanced detection of polymicrobial bloodstream infections with 16S rDNA PCR library amplicons and clinical relevance

Article References: Li, G., Wu, G., Yin, H., McGraw, K. A., Hou, J., & Zhao, Z. (2026). Enhanced detection of polymicrobial bloodstream infections with 16S rDNA PCR library amplicons and clinical relevance. Frontiers in Microbiology, 17, Article 1916658. https://doi.org/10.3389/fmicb.2026.1916658

Image Credits: AI Generated

DOI: 10.3389/fmicb.2026.1916658

Keywords: bloodstream infection, polymicrobial infection, 16S rDNA, PCR library, blood culture, procalcitonin, white blood cell count, high-throughput sequencing, sepsis, molecular diagnostics, Brucella, cell-free DNA

DNA methylation is chemically simple: a methyl group is attached to cytosine bases, most often at cytosine-guanine dinucleotides known as CpG sites, catalysed by enzymes called DNA methyltransferases. Yet the consequences are profound. Methylated promoters can silence genes, methylation patterns define chromatin states, and the overall methylation landscape helps keep the genome structurally stable. As the editors of the Research Topic, Pietro Salvatore Carollo of the University of Enna Kore, Andrea Fuso of Sapienza University of Rome and Viviana Barra of the University of Palermo, frame it, DNA methylation acts as the critical software that executes precise cellular programs within the genome’s hardware. When that software is corrupted, the result can be inappropriate silencing or activation of genes, altered chromatin, genomic instability or the disruption of developmental programs, which is why methylation patterns are increasingly treated as predictors of health rather than mere byproducts of it.

Crucially, methylation is not a static feature of the genome. It responds to developmental environment and cellular cues, adapting dynamically and profoundly affecting cellular function and identity. That dynamism cuts both ways: depending on the biological context, a change in methylation may be the initiator of a pathological cascade or simply the consequence of a cell that has already gone awry. Disentangling cause from effect is one of the central challenges of modern epigenetics, and it is precisely the question that unites the six contributions gathered in this collection, which the editors organise into three strands: the enzymology of active demethylation, the methylation-based measurement of biological age, and methylation signatures in inflammatory and chronic disease.

The first strand centres on TET2, a member of the Ten-Eleven Translocation family of enzymes that mediate active DNA demethylation. While methyltransferases write the mark, TET enzymes erase it, and the process is far more intricate than simple reversal. TET2 catalyses the first step of the cascade, oxidising 5-methylcytosine, the canonical methylated base, into 5-hydroxymethylcytosine, an intermediate that carries its own regulatory meaning and can be further oxidised or passively diluted through cell division. In a review by Tang and colleagues, the authors survey how wild-type TET2 maintains cellular physiology and genetic stability, while somatic mutations in the gene are implicated in a striking range of pathologies, from inflammatory disorders to cancers. Because demethylation involves complex multi-step enzymatic cascades whose fine-tuning is still being worked out, understanding TET2’s behaviour in pathological contexts is viewed as a promising route to new therapeutic strategies, exploiting the fact that methylation, unlike the underlying DNA sequence, is chemically reversible.

The second strand turns to time. In 2013, Steve Horvath introduced the concept of the epigenetic clock, showing that variations in DNA methylation patterns across human tissues and cell types correlate with ageing phenotypes. Since then, a family of epigenetic clocks has been developed, and they can now be applied not only to individual tissues but to populations and to specific pathological contexts. Chen and colleagues review the mechanistic basis of these methylation-based age estimates and their expanding implications for ageing research, health monitoring and disease prediction. The idea that a blood sample can yield a molecular readout of biological age, one that may diverge from chronological age and predict disease risk, has moved epigenetics from the laboratory bench into preventive medicine, and the review charts both the technical progress and the open questions in the field.

The third strand delivers some of the collection’s most concrete clinical findings. In rheumatoid arthritis, an autoimmune disease characterised by chronic joint inflammation and erosion, Delgado-Cruzata and colleagues examined cell-free DNA, fragments of genetic material circulating in plasma, from 27 patients. Using pyrosequencing, a quantitative method that reads methylation levels at individual CpG sites, they found that patients displayed lower methylation of the DUSP22 gene promoter at one specific site compared with healthy controls. More strikingly, decreased methylation levels in circulating DNA were associated with increased joint space narrowing, a radiographic marker of disease progression. The result points toward cell-free DNA methylation as a new, non-invasive biomarker for rheumatoid arthritis, potentially allowing clinicians to track erosive disease from a simple blood draw.

Hypertension tells a similar story of methylation as a disease mirror. Zhu and colleagues used Illumina Infinium Methylation EPIC BeadChips, arrays capable of interrogating hundreds of thousands of CpG sites across the genome, to profile methylation in peripheral leukocytes from 179 patients with high blood pressure. Their epigenome-wide analysis uncovered associations between altered methylation patterns and inflammation biomarkers, reinforcing the emerging view that hypertension is not solely a vascular or haemodynamic disorder but one with a measurable inflammatory and epigenetic dimension. Patterns like these could eventually help stratify patients who respond differently to treatment, or flag individuals at risk before blood pressure becomes chronically elevated.

Inflammation also takes centre stage in neonatal medicine. Sankar and colleagues studied newborns with late-onset neonatal sepsis, a devastating bacterial infection that strikes after the first days of life. Infants with the condition displayed a distinctive methylation signature: pro-inflammatory genes were hypomethylated, a state generally permissive of gene expression, while anti-inflammatory genes were hypermethylated, relative to healthy newborns. The expression of key cytokines, including TNF-α, IFN-γ, IL-10 and TGF-β, was accordingly shaped by methylation status. Because neonatal sepsis is notoriously difficult to diagnose quickly, a methylation-based signature could improve both diagnostics and therapy, identifying which infants mount a runaway inflammatory response and guiding interventions before organ damage occurs.

The final study demonstrates how methylation profiles can capture the interplay of disease, sex and geography. Miranda-Morales and colleagues performed bisulfite pyrosequencing on leukocyte DNA from Parkinson’s disease patients and age- and sex-matched healthy individuals in the Mexican Mestizo population, targeting CpG sites within two genes central to the disease: MAPT, which encodes the tau protein, and SNCA, which encodes alpha-synuclein, the main component of the protein aggregates that define the disorder. Patients showed reduced methylation in the MAPT promoter region and in intron one of SNCA. Intriguingly, the methylation differences were sex-specific: men displayed significant differences in MAPT methylation, while women showed significant differences at SNCA. Patterns also varied with regional location within the population, establishing that sex- and geography-specific methylation signatures exist in Parkinson’s disease, with implications for how epigenetic risk is assessed across diverse populations.

Taken together, the six studies converge on a common theme that is reshaping biomedical research: DNA methylation is a dynamic, responsive layer of genome regulation whose disturbances both reflect and drive human disease. By combining multi-omic and clinical data, the field is evolving from the identification of individual methylation loci toward an integrated view of disease biology, one that clarifies the mechanisms underlying disease heterogeneity and uncovers molecular signatures with value for diagnostics, prognosis and therapy. And because methylation is reversible in a way that genetic mutations are not, aberrant patterns represent an inviting therapeutic target. From the demethylase activity of TET2 to the ticking of epigenetic clocks and the methylation signatures circulating in a drop of blood, the software of the genome is proving to be not just a record of disease, but a place where disease might be rewritten.

Subject of Research: The role of DNA methylation in human diseases, including TET2-mediated demethylation, epigenetic clocks, and methylation biomarkers in inflammatory and neurodegenerative conditions.

Article Title: Editorial: Discussing the implication of DNA methylation in human diseases

Article References: Carollo, P. S., Fuso, A., & Barra, V. (2026). Editorial: Discussing the implication of DNA methylation in human diseases. Frontiers in Molecular Biosciences, 13, Article 2015661. https://doi.org/10.3389/fmolb.2026.2015661

Image Credits: AI Generated

DOI: 10.3389/fmolb.2026.2015661

Keywords: DNA methylation, epigenetics, TET2, epigenetic clock, cell-free DNA, rheumatoid arthritis, hypertension, neonatal sepsis, Parkinson's disease, biomarkers, ageing, gene expression

A new perspective published in Transplant International, launching an educational series called Decoding AI for Transplantation, challenges the way clinicians, journalists and patients talk about medical AI. The authors, a team spanning clinical transplantation, biostatistics and surgery in Paris, Madrid, Leuven and Vienna, argue that the pervasive habit of evaluating artificial intelligence as a single monolithic entity, a uniform black box expected to revolutionize care overnight, is fundamentally misleading. In reality, there is no singular all-encompassing AI tool for transplantation. Instead, the field consists of many highly narrow algorithms, each engineered for a specific purpose and each reading a particular kind of data. The essential clinical skill, they contend, is not the capacity to critique AI in the abstract but the practical ability to tell these disparate technologies apart.

The framework the authors propose rests on two deceptively simple questions. The first concerns modality: what kind of data does the tool actually read? Some tools consume structured clinical data such as creatinine values, proteinuria measurements, HLA mismatch counts and donor-specific antibodies. Others read images, typically digitized biopsy slides. Some read molecular data, whether the gene-expression signature of a tissue sample or donor-derived cell-free DNA circulating in the blood. Still others follow longitudinal data, tracking how a value moves over months rather than its level on a single day, and a growing class of tools now reads free text, including clinical notes, referral letters and the scientific literature, using large language models. Each modality is, in effect, a different language, and most tools speak only one of them.

The second question concerns the task: what clinical job is the tool built to perform? A tool may be designed to diagnose what is happening in the graft right now, to screen or monitor for something developing before it becomes clinically obvious, to predict long-term risk, or to guide a treatment decision. The same modality can serve different jobs, and the same job can be approached through entirely different modalities, which is precisely why the word AI on its own conveys so little. Placing real transplant tools on these two axes makes the point concrete. Digital-pathology models read biopsy images to help diagnose rejection. Transcriptomic classifiers, such as those built on the Banff Human Organ Transplant gene panel and the MMDx molecular phenotyping project, read molecular data to answer the same diagnostic question by a completely different route. Donor-derived cell-free DNA assays read the blood to monitor for subclinical rejection before a biopsy is ever taken.

One tool in the landscape already stands out for combining sources rather than relying on a single stream. The iBox risk score, developed through an international derivation and validation study, reads clinical, histological and immunological data together to predict long-term graft survival after kidney transplantation. Four tools, four distinct positions on the modality-by-task map, and only by naming each tool’s data input and clinical objective can a clinician judge what it genuinely offers and, just as importantly, what it cannot. This categorization, the authors argue, should become the reflexive first move whenever a new AI product or paper arrives in the transplant community.

Categorizing a tool this way exposes a critical limitation that a single headline metric can easily hide: most algorithms capture only a narrow slice of the total clinical picture. The biopsy-reading model has never seen the cell-free DNA result. The purely clinical risk score knows nothing of the histology. Each model can be excellent within its specific diagnostic domain and is, by design, blind to everything outside it. That is the exact reverse of what a clinician does at the bedside, where the value of the judgement comes not from any one number but from holding individual results together, the rising creatinine made meaningful by the proteinuria, the antibodies reinterpreted in the light of the biopsy findings.

A narrow answer delivered with confidence, the authors warn, is easy to mistake for a complete one. A tool that returns one confident result from one source is doing something a clinician cannot, answering a narrower question often very well, but a narrower one nonetheless. To be fair, some tools already reach beyond a single modality, and the honest description of the field is a spectrum: many single-modality tools, a growing few that combine two or three sources, and none that yet sees the whole patient. The authors are careful to note that neither does the clinician alone, and that multimodal integration is precisely the path toward seeing more than either the human eye or any single tool can on its own.

The frontier of the field, and the reason many promising tools stumble, lies in making that integration work. Different modalities live on different scales and timeframes: a histology slide, a single blood value and a years-long trajectory of lab results are not naturally comparable, and forcing them into one model confronts researchers with a genuine technical problem rather than a formality. Sources are often missing in clinical practice, with no biopsy performed at a given visit or an anti-HLA donor-specific antibody result not yet back from the laboratory, and a model trained on complete data can behave unpredictably when inputs are absent. Unlike traditional statistics, which offers a mature and flexible arsenal of methods for handling missing data, AI methodology has not yet reached that stage. Moreover, the more sources a model blends, the harder it becomes to see why it reached its conclusion, exactly at the moment when clinicians most want to know.

This is where foundation models and large language models enter the picture, and the authors are careful about their role. These general-purpose systems, often trained on vast text corpora, are increasingly proposed not as oracles that replace clinical judgement but as connective tissue, able to take in very different kinds of input and help bind them into a coherent whole. It is a useful job and a difficult one. The stated ambition, which the authors call the holy grail, would be to train computational models to replicate what clinicians already perform instinctively: evaluating the patient holistically rather than viewing an isolated organ system or a singular assay. Yet the authors argue the goal should not be to develop machine learning methods that replicate what clinicians do, but rather to provide better tools that help clinicians make better decisions.

For clinicians and researchers confronting the growing flood of AI claims, the authors distill the reframing into five questions that turn an intimidating field into a set of answerable ones. What kind of data does the tool read? What is its clinical job? Does it read one modality or combine several, and if it integrates, how does it cope when a source is missing? Was it tested in patients and centres other than the ones it was built on, the crucial test of external validation? And does it actually change a decision or improve an outcome, or does it only describe a risk the clinician already sensed? A tool that reads one modality for one task can be excellent and still be only part of the picture. A tool that claims to integrate many should be held to a higher standard, not a lower one, because its reach is wider and its failures quieter. The question, the authors conclude, is not whether AI can replace clinical judgement but how it can be used to make the best possible judgement to help patients.

Subject of Research: A framework for classifying and critically appraising artificial intelligence tools in organ transplantation by data modality and clinical task

Article Title: Decoding AI for transplantation: many models, one patient: why there is no single “AI for transplantation” – and what to ask instead

Article References: Aubert, O., Cortes Garcia, E., Pineda, S., Neyens, T., & Pilat, N. (2026). Decoding AI for transplantation: many models, one patient: why there is no single “AI for transplantation” – and what to ask instead. Transplant International, 39, Article 17554. https://doi.org/10.3389/ti.2026.17554

Image Credits: AI Generated

DOI: 10.3389/ti.2026.17554

Keywords: artificial intelligence, transplantation, multimodal AI, kidney allograft, rejection diagnosis, cell-free DNA, digital pathology, foundation models, external validation, clinical decision support, machine learning, iBox risk score

The research, led by Yunze Wang, Yuying Hou, Xinyu Liu, Wen Zhang, and Xionghui Zhou of Huazhong Agricultural University together with Qingjiao Li of Sun Yat-Sen University’s Eighth Affiliated Hospital, tackles a practical problem that has quietly limited the field. Whole-genome methylation sequencing of plasma cfDNA is attractive because a single assay can capture both DNA methylation and fragment-level features. But nobody had rigorously tested whether the chemical steps used to read methylation actually preserve the fragmentomic signals that other methods rely on.

The team systematically compared two methylation library preparation protocols, whole-genome bisulfite sequencing (WGBS) and enzymatic methyl sequencing (EM-seq), against conventional whole-genome sequencing (WGS). Their verdict was nuanced. Canonical cfDNA fragmentomic patterns, such as characteristic fragment size distributions and end motifs, were partially preserved across the methylation protocols, but the degree of preservation varied from feature to feature. In other words, methylation sequencing is not a perfect stand-in for WGS when it comes to fragmentation, and researchers combining both modalities from one assay need to know exactly which signals survive the chemistry.

Beyond the technical audit, the study probed the biology linking the two modalities. DNA methylation levels turned out to be associated with characteristic fragment length and end-motif patterns, confirming that how a DNA molecule is chemically decorated influences how it is cut, protected, and degraded in the blood. This coupling is not a nuisance; it is information. Tumor cells, with their altered chromatin and epigenetic landscapes, release DNA that carries coordinated epigenetic and physical fingerprints, and the study set out to organize those fingerprints into a framework a computer model can actually use.

The organizing principle the researchers chose is the methylation haplotype block, or MHB. Rather than treating each CpG site or each fragment independently, the team delineated MHBs directly from plasma cfDNA. These blocks, stretches of the genome where methylation states tend to be inherited together across reads, exhibited tissue-associated regulatory features and distinct patterns of chromatin accessibility, repeat enrichment, and methylation heterogeneity. That makes them biologically interpretable units, anchored in real regulatory architecture rather than arbitrary genomic windows.

Within these MHB regions, methylation and fragmentomic signals showed coordinated yet complementary patterns. Methylation tells the model which tissue the DNA came from; fragmentation tells it how that DNA was processed and cleared. Neither signal alone captures the full picture, but together they form a richer portrait of what is happening in a patient’s body. This insight guided the construction of FAME, the Fragmentomics And Methylation Ensemble, an interpretable multimodal diagnostic model that integrates MHB-centered methylation and fragmentomic features for cancer detection and tissue-of-origin prediction.

The researchers did not stop at a compact model. They extended FAME with complementary genome-wide methylation and fragmentomic features, producing FAME-GW, a genome-wide-augmented version. Across cross-validation, independent validation cohorts, and head-to-head comparisons with established methods including THEMIS (THorough Epigenetic Marker Integration Solution), SPOT-MAS (screening for the presence of tumor by methylation and size), and DELFI (DNA evaluation of fragments for early interception), the FAME-based models achieved robust diagnostic performance. FAME delivered a compact and interpretable representation, while FAME-GW delivered the strongest overall predictive performance.

What makes this work stand out in the crowded liquid biopsy field is its insistence on interpretability. Many high-performing cancer detection models are black boxes, blending thousands of features in ways clinicians cannot inspect. By centering the analysis on MHBs, units with known regulatory and chromatin properties, the FAME framework lets researchers trace a prediction back to biologically meaningful regions. A clinician or scientist can ask which blocks drove a cancer call and what their methylation and fragmentation patterns look like, rather than trusting an opaque score.

The study also carries a practical message for anyone designing cfDNA assays. Because methylation library preparation partially preserves fragmentomic signals, a single whole-genome methylation sequencing run can serve double duty, capturing both epigenetic and fragment-level information from one precious plasma sample. That is a meaningful efficiency gain for multi-cancer early detection programs, where sample volume is limited and every read counts. But the feature-by-feature variation in preservation means assay developers must validate each fragmentomic signal they intend to use under their specific protocol.

The work, published on 15 October 2026 as an open-access article in Genome Medicine, was supported by the Fundamental Research Funds for the Central Universities and the Huazhong Agricultural University Scientific and Technological Self-innovation Foundation. The in-house cohort drew healthy controls from The Eighth Affiliated Hospital of Sun Yat-sen University and gastric cancer samples from Fujian Medical University Union Hospital, with ethics approval and written informed consent obtained for all participants. As methylation haplotype blocks move from a research framework toward clinical assay design, the study’s central conclusion is likely to shape the next generation of cfDNA diagnostics: the epigenetic and fragmentomic stories written in circulating DNA are not separate chapters but a single, coupled narrative, and models that read them together, in interpretable units, are the ones most likely to reach the clinic.

Subject of Research: Integrating coupled methylation and fragmentomic signals in plasma cell-free DNA for interpretable cancer diagnosis

Article Title: Decoding coupled epigenetic-fragmentomic signals in circulating cell-free DNA for interpretable cancer diagnosis

Article References: Wang, Y., Hou, Y., Liu, X., Li, Q., Zhang, W., & Zhou, X. (2026). Decoding coupled epigenetic-fragmentomic signals in circulating cell-free DNA for interpretable cancer diagnosis. Genome Medicine. https://doi.org/10.1186/s13073-026-01793-w

Image Credits: AI Generated

DOI: 10.1186/s13073-026-01793-w

Keywords: cell-free DNA, DNA methylation, fragmentomics, methylation haplotype blocks, liquid biopsy, cancer detection, whole-genome bisulfite sequencing, EM-seq, FAME model, tissue of origin, Genome Medicine, interpretable machine learning

Osimertinib is a third-generation EGFR tyrosine kinase inhibitor (TKI), engineered to block the epidermal growth factor receptor even when it carries the T790M mutation — the classic escape route tumors use to shrug off first- and second-generation TKIs like gefitinib and erlotinib. In the clinic, osimertinib transformed outcomes for these patients, but resistance is inevitable. Tumors eventually find new molecular workarounds, and knowing which workaround is operating in a given patient can determine the next treatment decision. The problem has always been access: getting that information traditionally requires a tissue biopsy, an invasive procedure that is often impractical in patients who are frail or whose tumors sit in hard-to-reach locations.

The Thai team, led by Piyakarn Watcharenwong and Narumol Trachu, who contributed equally to the work, set out to answer whether blood-based testing could capture the same resistance intelligence. Between January 2016 and December 2019, they prospectively enrolled patients with advanced EGFR T790M-positive NSCLC at Ramathibodi Hospital. Every participant had progressed on a first- or second-generation EGFR TKI before starting osimertinib. The researchers drew paired plasma samples — one before osimertinib treatment began and a second at the moment of disease progression — and analyzed the cell-free DNA (cfDNA) circulating in the blood using a targeted next-generation sequencing panel covering 52 cancer-relevant genes.

The technical logic behind this approach is elegant. Dying tumor cells shed fragments of their DNA into the bloodstream, and those fragments carry the same mutations driving the cancer. Sequencing cfDNA therefore offers a molecular snapshot of the tumor without a needle in the tumor itself. The 52-gene panel used in this study was broad enough to catch not just EGFR mutations but also alterations in other canonical resistance genes, including TP53, PIK3CA, BRAF, and amplification of MET, a receptor tyrosine kinase gene that tumors frequently overproduce to bypass EGFR blockade entirely.

Fifty patients had usable paired plasma samples, and what the sequencing revealed was strikingly heterogeneous. Nineteen patients — 38 percent of the cohort — carried more than one resistance-associated alteration at progression, confirming that resistance rarely arrives through a single door. At baseline, plasma T790M was detectable in 32 patients. By the time osimertinib failed, 16 of those 32 — exactly half, representing 32 percent of the overall cohort — had lost detectable T790M altogether. This loss matters clinically: T790M-negative tumors generally do not respond to earlier-generation TKIs, so a patient whose T790M vanishes has lost one of the few re-treatment options available to them.

The genomic alterations detected at progression painted a detailed resistance map. TP53 mutations appeared in 22 percent of patients, PIK3CA mutations in 14 percent, the EGFR C797S mutation — a direct on-target escape that alters the very amino acid osimertinib binds to — in 10 percent, BRAF mutations in 8 percent, and MET amplification in 2 percent. Each of these represents a distinct evolutionary strategy: C797S changes the drug’s binding pocket, MET amplification floods the cell with an alternative growth pathway, and PIK3CA and BRAF mutations activate downstream signaling cascades that no longer depend on EGFR at all.

Clinically, the cohort fared as expected for this stage of disease: median time to treatment failure (TTF) on osimertinib was 9.3 months and median overall survival (OS) was 16.4 months. But the molecular subgroups told a more nuanced story. Patients whose plasma T790M disappeared at progression had a numerically shorter TTF — 6.0 months versus 10.1 months for those who retained T790M — though the difference did not reach statistical significance (P = 0.21). Among patients with T790M loss, those who also carried co-occurring resistance alterations fared even worse, with a median TTF of 4.1 months compared with 10.6 months for those without co-alterations (P = 0.06), a trend that hovered just at the edge of significance in this small cohort.

Perhaps the most intriguing findings were the associations between specific resistance mechanisms and patterns of metastatic spread. In univariable analyses, the presence of brain metastasis at baseline was significantly associated with subsequent plasma T790M loss (P = 0.04). Metastasis to uncommon sites — locations outside the typical lung, bone, liver, and adrenal distribution — was significantly linked to the emergence of EGFR C797S (P < 0.01). And progression specifically in the brain during osimertinib treatment was significantly associated with BRAF mutations (P = 0.01). These correlations suggest that the anatomical environment a tumor spreads to may exert selective pressure shaping which resistance mechanism evolves, a hypothesis that could eventually guide which molecular tests clinicians order based on a patient’s metastatic pattern.

The study’s authors are careful to frame these as exploratory findings that require validation in larger cohorts with tissue-based analyses, and that caution is warranted. With fifty patients, the statistical power to detect modest differences is limited, and plasma-based cfDNA testing has known blind spots — alterations present at low allele fractions, or in tumor regions that shed little DNA, can go undetected. Brain metastases in particular are notoriously underrepresented in plasma cfDNA because the blood-brain barrier limits the exchange of circulating nucleic acids, which means some of the brain-associated findings may reflect biology that plasma testing only partially captures.

Even so, the study delivers a practical message with real clinical weight: a blood draw, repeated at the moment of progression, can reveal the molecular landscape of resistance in T790M-positive NSCLC and flag patients on faster trajectories of decline. As osimertinib continues to move into earlier lines of treatment worldwide, understanding — and monitoring — the mechanisms by which tumors escape it becomes ever more urgent. This Thai prospective cohort adds a valuable piece to that puzzle, demonstrating that liquid biopsy is not merely a convenience but a window into the evolutionary battle between drug and cancer, fought in real time and readable from a vial of blood.

Subject of Research: Liquid biopsy detection of acquired resistance mechanisms to osimertinib in EGFR T790M-positive advanced non-small cell lung cancer

Article Title: Non-invasive molecular testing of osimertinib acquired resistance in T790M-positive advanced non-small cell lung cancer

Article References: Watcharenwong, P., Trachu, N., Iemwimangsa, N., Munthum, D., Monnamo, N., Jiaranai, P., Charoenyingwattana, A., Nitiwarangkul, C., Chamroonrat, W., Chansriwong, P., Ativitavas, T., Dejthevaporn, T., Chantratita, W., & Reungwetwattana, T. (2026). Non-invasive molecular testing of osimertinib acquired resistance in T790M-positive advanced non-small cell lung cancer. BMC Cancer. https://doi.org/10.1186/s12885-026-17148-y

Image Credits: AI Generated

DOI: 10.1186/s12885-026-17148-y

Keywords: osimertinib, EGFR T790M, non-small cell lung cancer, liquid biopsy, cell-free DNA, acquired resistance, next-generation sequencing, EGFR C797S, MET amplification, BRAF mutation, brain metastasis, targeted therapy

The research, led by Hailing Ou and colleagues, took advantage of a fundamental feature of cancer biology. Tumor cells leave behind a distinctive epigenetic signature, and fragments of their DNA released into the bloodstream carry traces of that signature. DNA methylation, the addition of chemical tags to cytosine bases that can silence genes without altering the genetic code, is one of the most stable and informative of these marks. Because methylation patterns are preserved in cell-free DNA fragments, they can serve as a molecular fingerprint of tumor presence detectable through a simple blood draw.

To identify such fingerprints, the team first performed whole-genome bisulfite sequencing on 29 pairs of gastric tumor tissue and adjacent normal tissue. Bisulfite treatment converts unmethylated cytosines into uracil while leaving methylated cytosines intact, allowing researchers to map methylation across the entire genome at single-base resolution. This analysis revealed two hallmark features of the gastric cancer epigenome: widespread global hypomethylation, in which large stretches of the genome lose their methylation marks, and focal hypermethylation at CpG islands, the gene-regulatory regions where dense clusters of cytosine-guanine dinucleotides reside. These opposing patterns are well-documented in cancer and can disrupt gene expression in ways that promote tumor growth.

The critical innovation of the study lay in connecting tissue-level findings to what can actually be measured in blood. The researchers mapped the differentially methylated regions discovered in tumor tissue onto methylation data from plasma cell-free DNA obtained from 61 gastric cancer patients and 29 healthy controls. This tissue-plasma concordance analysis identified 32 regions whose methylation changes were consistent in both compartments, suggesting that the tumor’s epigenetic alterations are faithfully reflected in the circulating DNA fragments that reach the bloodstream.

From these 32 candidate regions, the team applied two complementary feature-selection algorithms, LASSO regression and Random Forest, to narrow the panel to eight core diagnostic markers. LASSO, or least absolute shrinkage and selection operator, shrinks the coefficients of less informative features toward zero, effectively eliminating them, while Random Forest ranks features by how much they improve the purity of decision trees built on random subsets of the data. The convergence of both methods on the same eight markers strengthened confidence that these regions carry genuine diagnostic signal rather than statistical noise. One of the markers maps to the ATP8A2 gene, which encodes a phospholipid-transporting ATPase involved in membrane lipid trafficking.

With the marker panel established, the researchers built diagnostic classifiers using four machine learning algorithms: Support Vector Machine, Random Forest, K-Nearest Neighbor, and Naïve Bayes. Each algorithm approaches classification differently. Support Vector Machines find the hyperplane that best separates cancer from control samples in high-dimensional feature space; K-Nearest Neighbor classifies samples by the majority label of their closest training examples; Naïve Bayes applies probabilistic reasoning under an assumption of feature independence; and Random Forest aggregates hundreds of decision trees trained on bootstrapped samples. When the models were evaluated on training and independent test sets, the Random Forest classifier built on the eight cfDNA methylation markers performed best.

The performance figures are striking. The Random Forest model achieved an area under the receiver operating characteristic curve, or AUC, of 0.978 in the training set and 0.938 in the independent test set. An AUC of 1.0 represents perfect discrimination, so values approaching unity indicate that the model separates cancer patients from healthy controls with very few errors. In the test set, the model reached an accuracy of 88.46 percent, a sensitivity of 94.44 percent, and a specificity of 75.00 percent. Sensitivity measures how reliably the test catches true cancer cases, a crucial metric for screening, while specificity reflects how well it avoids false alarms. Because age can influence methylation patterns independently of disease, the researchers also adjusted for age and found the model retained an AUC of 0.924 on the test set, confirming that its discriminatory power is not merely an artifact of demographic differences between groups.

Beyond diagnosis, the study explored what the eight markers might reveal about tumor biology. Functional enrichment analysis linked the differentially methylated regions to biological pathways including phagocytosis, the process by which cells engulf and digest particles, and axon guidance, the signaling system that steers growing nerve fibers. The markers also showed significant correlation with immune cell infiltration patterns in the tumor microenvironment, hinting that methylation changes at these loci may reflect or influence the interplay between gastric tumors and the immune system. Such connections could eventually inform not only detection but also prognostic stratification and immunotherapy decisions.

Perhaps the most consequential technical finding concerns the limitations of existing methylation screening platforms. The Illumina methylation arrays known as 450K and EPIC v2, which are widely used in epigenetic research, do not cover the probes corresponding to any of the eight core DMRs identified in this study. In other words, researchers relying solely on array-based methylation profiling would never have found these biomarkers. This underscores a key advantage of whole-genome bisulfite sequencing: by surveying the entire methylome without probe constraints, it can uncover regulatory regions that fall outside the curated content of commercial arrays, opening discovery space that standardized platforms systematically miss.

The study’s design, matching tissue and plasma methylation profiles from the same patients before applying machine learning, offers a template for biomarker development in other cancers where non-invasive detection remains an unmet need. Liquid biopsy approaches based on circulating tumor DNA have already transformed management of several malignancies, and methylation-based cfDNA analysis is emerging as a particularly informative modality because methylation changes occur early in carcinogenesis and are tissue-specific. For gastric cancer, where endoscopic screening is impractical at scale in many regions and symptoms often appear late, a blood test achieving over 90 percent sensitivity could meaningfully shift the detection curve toward earlier, more treatable stages. The authors conclude that their tissue-plasma concordance strategy has produced a panel of plasma cfDNA methylation biomarkers with strong diagnostic potential, and that the resulting Random Forest model represents a promising non-invasive strategy for gastric cancer screening. Larger validation cohorts across diverse populations and clinical settings will be the necessary next step before such a test can move from research bench to clinical practice, but the evidence presented here demonstrates that the epigenetic whispers of a stomach tumor can indeed be heard in a tube of blood.

Subject of Research: Genome-wide cfDNA methylation biomarkers for non-invasive gastric cancer diagnosis

Article Title: Genome-Wide DNA methylation profiling of tissue-plasma pairs identifies Novel cfDNA biomarkers for the diagnosis of gastric cancer

Article References: Ou, H., Miao, J., Ling, F., Gao, J., Wang, T., Ye, Z., Zhang, C., Huang, M., & Qin, Y. (2026). Genome-Wide DNA methylation profiling of tissue-plasma pairs identifies Novel cfDNA biomarkers for the diagnosis of gastric cancer. PLOS Genetics, 22(10), e1012322. https://doi.org/10.1371/journal.pgen.1012322

Image Credits: AI Generated

DOI: 10.1371/journal.pgen.1012322

Keywords: gastric cancer, cell-free DNA, DNA methylation, liquid biopsy, machine learning, Random Forest, whole-genome bisulfite sequencing, biomarkers, epigenetics, cancer screening, PLOS Genetics, differentially methylated regions

The clinical stakes are considerable. Hepatocellular carcinoma is a leading cause of cancer-related mortality worldwide, and its prognosis depends heavily on whether the tumor is caught while it is still small and potentially curable. Current surveillance for people at high risk, particularly those with cirrhosis or chronic hepatitis B infection, relies on abdominal ultrasonography combined with the blood protein alpha-fetoprotein. Yet these tools have well-documented limitations. Meta-analyses cited in the review show that surveillance imaging and alpha-fetoprotein together miss a substantial fraction of early tumors, and even refined protein markers such as AFP-L3 have struggled to reliably distinguish early-stage disease from benign cirrhotic liver tissue. This diagnostic gap has pushed researchers toward liquid biopsy approaches that interrogate molecular signals released by tumors into the circulation.

Cell-free DNA refers to small fragments of genetic material that circulate in blood plasma, shed by dying cells throughout the body. In people with cancer, a portion of this DNA originates from tumor cells and is known as circulating tumor DNA. What makes methylation analysis particularly attractive is that DNA methylation, the addition of methyl groups to cytosine bases in CpG dinucleotides, is an epigenetic modification that does not alter the underlying genetic sequence but profoundly changes gene activity. Aberrant methylation is a hallmark of hepatocarcinogenesis: hypermethylation of promoter CpG islands silences tumor suppressor genes, while broader patterns of methylation dysregulation reshape the epigenetic landscape of the malignant cell. Because these methylation changes are cancer-specific and occur at defined genomic loci, they can be detected in the small quantity of tumor-derived DNA present in plasma.

The review places particular emphasis on the interplay between hepatitis B virus and the host epigenome. The viral X protein, a multifunctional regulator encoded by HBV, can promote methylation dysregulation through direct interactions with DNA methyltransferases, the enzymes that establish and maintain methylation patterns. Studies summarized in the review show that HBx upregulates DNA methyltransferases DNMT1, DNMT3A and DNMT3B, leading to hypermethylation and silencing of genes such as SOCS-1 and RASSF1A, and that the viral protein can induce epigenetic repression of E-cadherin through its interaction with DNMT3A. The virus genome itself also undergoes methylation during infection, and methylation of the HBV genome has been associated with the development of hepatocellular carcinoma and occult infection. This virus-driven epigenetic rewiring means that methylation signatures in HBV-related liver cancer carry information about both the tumor and the viral history that shaped it.

On the clinical performance side, the review assembles a striking set of results from recent studies. Multi-feature cfDNA models that integrate methylation data with additional fragmentomic features, such as fragment size and end-motif patterns, have achieved sensitivity of 95.42 percent and specificity of 97.67 percent, with an area under the receiver operating characteristic curve of 0.997, for differentiating hepatocellular carcinoma from liver cirrhosis. Critically, these models maintained a sensitivity of 90.0 percent for early-stage disease, the setting in which detection matters most. In specific cohorts, methylation-based approaches complemented alpha-fetoprotein by detecting nearly 75 percent of early malignancies that the protein marker alone missed, and adding methylation data to a combined predictive model raised its AUC from 0.87 to 0.90. Combined mutation-methylation assays, which simultaneously profile genetic mutations and hypermethylation patterns in circulating DNA, have reached validation AUC values as high as 0.93.

Several individual methylation markers have emerged as recurring candidates across studies. The RASSF1A promoter, frequently silenced in hepatocellular carcinoma, has been detected as methylated cell-free DNA in plasma, and a study of patients with hepatitis B-related cirrhosis found that cell-free methylation of RASSF1A and CDKN2AIP helped distinguish malignant from non-malignant disease. Genome-wide methylation profiling of plasma cell-free DNA across the stages of HBV-related hepatocarcinogenesis has identified potential biomarkers for early detection and risk monitoring, and more recent plasma cfDNA methylome profiling has pointed to GJA4 as a candidate biomarker specific to HBV-related disease. Whole-genome bisulfite sequencing approaches have further refined the search for blood-based methylation biomarkers, while digital droplet PCR has been applied to quantify circulating tumor DNA methylation with high sensitivity.

The technical foundations of these assays deserve attention because they explain both the promise and the difficulty of methylation-based liquid biopsy. Detecting a tumor-specific methylation signal requires distinguishing a small tumor-derived fraction from a large background of DNA released by normal cells, particularly in patients with cirrhosis, where inflammation and cell turnover elevate background cfDNA levels. Bisulfite conversion, which chemically converts unmethylated cytosines to uracil while leaving methylated cytosines intact, remains the workhorse of methylation analysis, but it degrades DNA and complicates quantification. Newer approaches, including methylation-aware sequencing and single-molecule assays, aim to preserve more information from each fragment. Computational methods for cell-type decomposition of circulating cell-free DNA, such as the CelFiE framework, help researchers account for the tissue origins of cfDNA and isolate the tumor-specific component of the methylation signal.

The review is careful to note that these impressive numbers come largely from retrospective, case-control studies in specific cohorts, and that the field has not yet demonstrated clinical utility in the way guidelines demand. Prospective validation and direct head-to-head comparisons with existing surveillance strategies, namely ultrasonography and alpha-fetoprotein, are still required before methylation-based liquid biopsy can be recommended for routine screening of high-risk populations. Issues of patient selection bias, standardization of assay protocols, and the cost and accessibility of sequencing-based platforms in the regions where hepatitis B is most prevalent all remain practical obstacles. The authors, writing from Burkina Faso, where chronic hepatitis B infection is endemic, underscore that any diagnostic advance must ultimately be deliverable in the health systems that carry the greatest burden of disease.

Even with those caveats, the trajectory of the field is clear. Methylation markers are stable, cancer-specific, and detectable in plasma at levels that modern assays can measure, and their performance in early-stage disease exceeds what protein biomarkers alone have achieved. As multi-feature models combine methylation with fragmentomics and mutation data, and as prospective trials test these panels against current standards of care, cell-free DNA methylation analysis may move from a research tool to a clinical mainstay in the surveillance of hepatitis B-related hepatocellular carcinoma, offering patients the possibility of catching a deadly cancer while it is still curable.

Subject of Research: Cell-free DNA methylation biomarkers for early diagnosis of hepatitis B virus-related hepatocellular carcinoma

Article Title: Cell-free DNA methylation markers in the diagnosis of hepatitis B virus-related- hepatocellular carcinoma: a review

Article References: Soubeiga, S. T., Ouattara, A. K., Ilboudo, C., Djigma, W. F., & Simpore, J. (2026). Cell-free DNA methylation markers in the diagnosis of hepatitis B virus-related- hepatocellular carcinoma: a review. Molecular Biology Reports, 53(1), Article 1722. https://doi.org/10.1007/s11033-026-12942-0

Image Credits: AI Generated

DOI: 10.1007/s11033-026-12942-0

Keywords: hepatocellular carcinoma, hepatitis B virus, cell-free DNA, DNA methylation, liquid biopsy, biomarkers, epigenetics, cirrhosis, alpha-fetoprotein, early detection, tumor suppressor genes, HBx protein

The clinical motivation behind the tool centers on minimal residual disease, the microscopic burden of cancer cells that persists after treatment and drives relapse. In germinal center B-cell derived lymphomas, a family of blood cancers that includes some of the most common lymphoma subtypes, the malignant cells carry a distinctive legacy of their origin. During their development in germinal centers, these cells activate a molecular machinery known as somatic hypermutation, which deliberately scrambles DNA in antibody genes to generate immune diversity. The machinery is not perfectly targeted, however, and it also leaves clusters of simultaneous mutations scattered across the genome, particularly within super-enhancer regions, stretches of DNA that control the activity of key genes in B cells. These clustered mutations, called multi-nucleotide variants, or MNVs, act as patient-specific fingerprints that are extraordinarily unlikely to arise from sequencing noise.

MNVista works by searching for co-occurring single-nucleotide variants and evaluating candidate pairs with a Bayesian statistical model that incorporates read-level metrics from the sequencing data. Rather than requiring specialized file formats, the tool accepts standard VCF and indexed BAM files, the common currencies of next-generation sequencing pipelines, which means laboratories can slot it into existing workflows with minimal friction. The software can operate with or without matched normal control samples, a flexibility that matters in clinical settings where a healthy tissue reference is not always available. By scoring how likely it is that two or more nearby mutations appeared together on the same set of sequencing reads by genuine biology rather than by independent errors, the Bayesian framework filters out noise while preserving true clustered variants.

The performance gains reported by the team are striking. In publicly available dilution series, in which DNA from lymphoma patients was diluted into healthy background DNA to simulate minimal residual disease, MNVista detected patient-specific multi-nucleotide variants down to a variant allele fraction of 0.0025 percent, equivalent to 25 parts per million. The widely used standard caller Mutect2, by comparison, reached a detection limit of 1,783 parts per million on the same data. That gap spans nearly two orders of magnitude, and it translates directly into clinical potential: the lower the fraction of tumor DNA that can be reliably detected, the earlier a relapse can be flagged and the more confidently a patient can be declared disease-free after therapy.

The mathematics behind this sensitivity advantage are worth unpacking. Sequencing errors strike individual bases essentially at random, so the probability that two specific errors occur side by side on the same DNA fragment is the product of two independent error rates, an astronomically small number. When somatic hypermutation produces genuinely linked mutations, however, those bases are altered together in the tumor genome and therefore appear together in every descendant DNA fragment circulating in the blood. By demanding co-occurrence rather than trusting any single variant call, MNVista effectively multiplies its power to distinguish signal from noise. The researchers showed that the tool recovered complex multi-nucleotide variants composed of up to nine co-occurring single-nucleotide changes, a combinatorial specificity so extreme that it implies a theoretical detection limit below the 1 part per million threshold that the National Comprehensive Cancer Network has cited as a target for clinical minimal residual disease detection.

Validation did not rest on patient dilution data alone. The team also ran synthetic benchmark experiments in which they spiked artificial multi-nucleotide variants into real sequencing data and compared the results against Mutect2. On clean backgrounds, MNVista delivered near-perfect performance, recovering the spiked-in variants with essentially no false positives. More revealing, however, was the stress test: when the researchers synthetically introduced background single-nucleotide variants into genuine patient data, including both real somatic variants and sequencing artifacts, MNVista’s accuracy remained robust while Mutect2’s performance degraded. This resilience to noisy backgrounds is critical, because real clinical samples are never as pristine as benchmark datasets, and a caller that collapses under background variation offers little value at the bedside.

The design choices embedded in MNVista reflect a deliberate effort to bridge the gap between research algorithms and clinical practice. Many specialized ultra-sensitive detection methods demand bespoke wet-lab protocols, custom capture panels, or proprietary analysis chains, all of which slow adoption and add cost. MNVista instead operates entirely downstream of standard sequencing, consuming ordinary VCF and BAM files that any clinical genomics facility already produces. This architecture means the tool does not require new laboratory infrastructure or revalidation of sequencing chemistry; it simply extracts more information from data that already exist. For hospitals running lymphoma surveillance programs, the path from sample to result could therefore remain largely unchanged apart from an additional analysis step.

The implications for B-cell lymphoma care could be substantial. Minimal residual disease monitoring using cell-free DNA is increasingly recognized as a way to personalize treatment intensity, identifying patients who need additional therapy before overt relapse and sparing others from unnecessary toxicity. In lymphomas driven by germinal center programs, the abundance of super-enhancer-associated multi-nucleotide variants provides a rich reservoir of markers that are unique to each patient’s tumor, sidestepping the need to hunt for recurrent mutations that may not exist in every case. Because these clustered signatures arise from the tumor’s own mutational machinery, they can be catalogued at diagnosis from tumor tissue and then tracked in blood over time, giving oncologists a quantitative, dynamic measure of disease burden measured in parts per million.

Limitations and open questions remain, as with any new method. The tool’s sensitivity figures derive from publicly available dilution series and synthetic benchmarks, and prospective clinical studies will be needed to confirm that ultra-sensitive MNV detection translates into earlier relapse detection and improved outcomes in real patient cohorts. The approach is also specifically tailored to cancers shaped by somatic hypermutation and related error-prone processes, so its applicability to solid tumors with different mutational signatures remains to be established. Nevertheless, the software is released as open source, which invites the broader bioinformatics community to test, adapt, and extend it, and the authors acknowledge support from the Anita Veldman Foundation and the high-performance computing resources of the University of Groningen in bringing the project to fruition.

What makes this study resonate beyond lymphoma is its demonstration that biological insight can outperform brute-force engineering in the quest for analytical sensitivity. Rather than pushing sequencing depth ever higher, the Groningen team harnessed a distinctive mutational process to create signatures whose statistical improbability alone defeats sequencing noise. If the tool’s performance holds in clinical validation, patients with germinal center B-cell lymphomas may one day benefit from relapse surveillance capable of spotting one cancer-derived DNA fragment among a hundred thousand healthy ones, a detection limit that until recently belonged to the realm of theory rather than routine laboratory practice.

Subject of Research: Computational detection of multi-nucleotide variants in cell-free DNA for minimal residual disease monitoring in B-cell lymphoma

Article Title: MNVista: accurate identification of multi-nucleotide variants from next generation sequencing data

Article References: Silljé, L., Lunter, G., Nijland, M., Diepstra, A., Dalenberg, J. R., Zhong, Y., Kluiver, J. L., Plattel, W. J., Berg, A. V. D., & Veltmaat, N. (2026). MNVista: accurate identification of multi-nucleotide variants from next generation sequencing data. BMC Bioinformatics. https://doi.org/10.1186/s12859-026-06702-x

Image Credits: AI Generated

DOI: 10.1186/s12859-026-06702-x

Keywords: MNVista, multi-nucleotide variants, cell-free DNA, minimal residual disease, B-cell lymphoma, circulating tumor DNA, next-generation sequencing, somatic hypermutation, super-enhancer regions, Bayesian model, variant calling, bioinformatics