Beyond variant detection: Mismatch repair gene alterations in plasma cell-free DNA from 1,226 glioma samples, and plasma-confirmed functional deficiency in an unbiopsied pontine glioma

Abstract

Background

Mismatch repair (MMR) gene alterations are uncommon in glioma but can produce hypermutation and sensitivity to checkpoint inhibition. Profiling is often unavailable in diffuse intrinsic pontine glioma (DIPG), where biopsy carries substantial risk. We assessed whether plasma cell-free DNA (cfDNA) sequencing detects these alterations in gliomas and report an unbiopsied DIPG case in which plasma findings were functionally corroborated.

Methods

Retrospective analysis of an index DIPG case profiled in plasma, together with sample-level analysis of 1,226 glioma plasma cfDNA profiles from a clinical digital-sequencing assay and 5,369 glioma tissue samples from public datasets. We assessed MSH6, MSH2, MLH1, and PMS2 alterations, tumor mutational burden (TMB), and microsatellite instability (MSI). Plasma–tissue comparisons used Fisher's exact tests, Bonferroni threshold p < 0.0125.

Results

In the index case, plasma cfDNA showed a truncating MSH6 variant at 100% variant allele fraction, MSI-high status, and TMB of 132.3 mut/Mb in a child never given an alkylating agent; all other somatic-range variants lay below 1.7%, indicating constitutional origin. Pembrolizumab was given across 21 doses with radiographic stability, with progression at approximately 19 months; at 24 months the patient was alive, performance status 3 to 4. In the cohort, 680 of 1,226 samples (55.5%) had at least one non-synonymous alteration. Within these, MSH6 was altered in 8 (1.2%), MSH2 and MLH1 in 6 each (0.9%) and PMS2 in 2 (0.3%). MSH6 was the most altered MMR gene in both plasma and tissue, at 1.2% each, and no gene differed significantly after correction. Median TMB was higher in MMR-altered samples in both cohorts.

Conclusions

Plasma cfDNA detected MMR gene alterations at frequencies indistinguishable from tissue, and in one unbiopsied case gave functional evidence of deficiency through concurrent MSI-high status and hypermutation. Where tissue is unobtainable a positive result identifies a subgroup with a specific therapeutic hypothesis, whereas a microsatellite-stable result is weak evidence against deficiency. Plasma complements rather than replaces tissue testing.

Highlights

• •Biopsy is often unsafe in DIPG, leaving immunotherapy decisions without molecular data.
• •In the index case plasma cfDNA showed MSH6 loss at 100% VAF and TMB 132.3 mut/Mb.
• •Hypermutation without alkylator exposure supports functional MMR deficiency.
• •Pembrolizumab: 15 months radiographic stability, progression at 19 months.
• •Of 1,226 plasma samples 55.5% had detectable ctDNA; MSH6 led at 1.2%, as in tissue.

1. Introduction

Gliomas represent a biologically heterogeneous group of tumors whose diagnosis, classification, and treatment increasingly rely on molecular characterization [[1], [2], [3]]. Contemporary clinical practice guidelines recommend testing specific molecular alterations that define glioma subtypes, inform prognosis, and guide therapy [4], while broader genomic profiling has also identified uncommon but clinically actionable biomarkers in selected patients including IDH1, H3K27M, EGFR, BRAF, NTRK [5]. Within this evolving molecular landscape, MMR deficiency, although rare, has important biological and therapeutic implications. MMR deficiency promotes a hypermutator phenotype characterized by high TMB, increased neoantigen load, and enhanced tumor immunogenicity, features that may confer sensitivity to immune checkpoint inhibition [6,7]. MMR-deficient gliomas may arise sporadically, through germline pathways, or following alkylating therapy [6,7]. These routes are not therapeutically equivalent. Hypermutation arising after temozolomide exposure has been associated with a low rate of response to PD-1 blockade, whereas primary, predominantly germline-driven MMR deficiency defines the glioma subgroup in which checkpoint blockade has been associated with a survival advantage [8,9].

Despite the growing importance of molecular profiling, comprehensive genomic evaluation in gliomas is often constrained by tumor location and biopsy-associated risk. This limitation is particularly pronounced in diffuse midline gliomas, including DIPG, where surgical sampling is frequently unsafe or infeasible [10]. As a result, clinically relevant molecular alterations may remain unidentified in a subset of patients. Liquid biopsy using plasma cfDNA has transformed molecular testing across many solid tumors, but its adoption in gliomas has been limited by concerns regarding low cfDNA passage across the blood–brain barrier [[11], [12], [13]]. That view derives from an early series of 27 glioma patients in which an earlier-generation assay detected ctDNA in fewer than 10% [14]; higher-sensitivity digital-sequencing assays have since reported detection in approximately half of primary brain tumor samples [15,16], so yield is assay-, panel- and timing-dependent rather than precluded by the barrier. Multiple studies have demonstrated that cfDNA is detectable in a subset of patients with primary brain tumors, even in the presence of an intact blood–brain barrier [13,17].

Emerging evidence indicates that plasma cfDNA analysis can capture actionable genomic alterations in glioma subsets, including hypermutation, oncogenic driver mutations, and MSI-high status [18,19]. The analytical performance of plasma-based MSI detection has been validated in a large pan-cancer cohort, supporting its use in tumors with relatively low DNA shedding [20]. In pediatric diffuse midline gliomas, including DIPG, ctDNA from cerebrospinal fluid and plasma has been shown to detect hallmark alterations such as H3K27M mutations, underscoring the feasibility of this approach when tissue testing is limited [21].

Here, we report a pediatric DIPG case in which plasma cfDNA NGS identified MSI-high status associated with an MSH6 truncating mutation, informing treatment with immune checkpoint inhibition, followed by a period of clinical and radiographic stability and subsequent progression at approximately 19 months from diagnosis. We integrate this illustrative case with large-scale plasma cfDNA and tissue sequencing datasets to evaluate the feasibility and clinical relevance of plasma-based detection of MMR deficiency in gliomas.

2. Materials and methods

2.1. Study overview

This retrospective study integrates data from three sources: publicly accessible glioma tissue sequencing datasets retrieved from cBioPortal [22], plasma cfDNA NGS results from glioma patients profiled with Guardant360®, and a pediatric DIPG case evaluated using the Guardant360® plasma cfDNA assay. The objective was to assess whether MMR gene alterations and associated hypermutation patterns observed in the index case were consistent with findings across larger tissue-based and plasma-based cohorts. All analyses were exploratory and descriptive, with no prespecified statistical hypotheses.

2.2. Case description

A 9-year-old boy presented in January 2024 with diplopia and impaired extraocular movements. MRI showed an expansile pontine lesion with extension into the midbrain, thalami, medulla and cerebellar peduncles; diffuse intrinsic pontine glioma was diagnosed on clinical and radiological grounds at a multidisciplinary neuro-oncology board, and biopsy was not undertaken. A ventriculoperitoneal shunt was placed and radiotherapy (54 Gy in 30 fractions) was completed in March 2024. No temozolomide or other alkylating chemotherapy was given at any point, so the hypermutant phenotype described below cannot be treatment-induced. Six weeks after radiotherapy, imaging showed stability and plasma cfDNA testing was performed; no corticosteroids were being given at collection, and dexamethasone was started only later, at clinical progression. Pembrolizumab monotherapy was then started on a three-weekly schedule. Radiological response was assessed using iRANO criteria [23] by a single unblinded neuroradiologist without central review and toxicity by CTCAE v5.0. The full case narrative, including imaging timepoints and perfusion values, is provided in the Supplementary Material. The index case was not among the 1,226 plasma samples analyzed in the cohort.

2.3. Tissue genomic dataset

Glioma sequencing datasets available through cBioPortal were queried for pathogenic or likely pathogenic alterations in the MMR genes MSH6, MSH2, MLH1, and PMS2. Microsatellite instability status was not available in the tissue datasets. Tumor mutational burden values were extracted when reported. The cBioPortal platform aggregates genomic data from multiple large-scale sequencing initiatives, including TCGA and MSK-IMPACT [24,25].

2.4. Plasma liquid biopsy dataset

Guardant360® plasma sequencing data (versions 2.11–2.12) from glioma patients were reviewed for mismatch repair (MMR) gene alterations, plasma tumor mutational burden (TMB), and microsatellite instability (MSI). Guardant360® is a hybrid-capture–based next-generation sequencing assay that analyzes cell-free DNA from plasma and interrogates single-nucleotide variants, insertions/deletions, copy number alterations, and select gene fusions across a multigene panel [26]. Analytical validation of variant and MSI calling is described in the following subsection. Of 1,226 glioma plasma samples submitted for routine clinical testing through December 2025, 719 (58.6%) had at least one alteration of any class detected; in 39 (3.2%) the only findings were synonymous variants, and the remaining 680 (55.5% (95% CI 52.6–58.3)) had at least one non-synonymous alteration and formed the analysis set. Samples were run on assay version 2.11, which interrogates 74 genes, or version 2.12, which interrogates up to 498 genes and reports 83. Microsatellite instability was assessed on both versions; tumor mutational burden was reported only on version 2.12, so TMB analyses are restricted to the 272 of 680 samples (40.0%) that were TMB-evaluable. All analyses are at the sample level. Samples were not deduplicated to patients in either cohort: the plasma cohort comprises 1,226 samples from 901 patients and the tissue cohort 5,369 samples for which patient-level deduplication was not undertaken. Observations are therefore not fully independent (Supplementary Methods). Samples came from the same de-identified database as a previously reported primary brain tumor cohort and a small proportion may overlap; that study addressed overall ctDNA yield and targetable alterations, not MMR genes, TMB by MMR status, or MSI [15]. A sample was counted as informative if at least one non-synonymous single-nucleotide variant, insertion or deletion was reported. MSI and TMB were reported only for samples meeting the respective evaluability criteria (673 and 272 samples, respectively).

2.5. cfDNA-based MSI and TMB determination

Microsatellite instability was assessed using the validated MSI module of the Guardant360® assay [20]. Cell-free DNA is enriched by hybrid capture and sequenced using a molecular barcoding (digital sequencing) approach that permits error correction of individual DNA molecules. Candidate microsatellite loci comprising mononucleotide repeats of seven or more units were selected for susceptibility to instability across tumor types and for reliable capture, mapping and coverage in short cell-free DNA fragments; BAT-25 and BAT-26, from the original Bethesda panel, and NR-21, from the pentaplex mononucleotide panel, are included, whereas NR-24 and MONO-27 were excluded because of poor mappability. A sample is classified as MSI-high when the number of unstable loci exceeds a predefined proprietary threshold; samples with insufficient tumor content for classification are reported as not determined rather than as microsatellite stable. Tumor mutational burden was calculated across a footprint of 2.0 Mb of genomic and 1.0 Mb of exonic coverage, and is derived from the full interrogated region rather than from the reported gene set, so the alterations listed for the index case represent only those within the reported genes. The pan-cancer clinical validation of this module, and the loci and model detail, are summarized in the Supplementary Methods and Supplementary Fig. S1. The reported positive predictive value applies at the MSI-high prevalence of that cohort, and the module has not been separately validated in glioma.

An analytical workflow schematic is provided in Supplementary Fig. S1.

Sensitivity for cfDNA MSI determination depends on tumor shedding, so a microsatellite-stable result does not exclude MSI; and isolated loss of MSH6 or MSH2 generates less instability than loss of MLH1 or PMS2, so an MSI-high call in an MSH6-driven case is informative but is not the expected finding in MMR-deficient glioma.

2.6. Statistical analysis

Proportions are reported with their numerators, denominators and exact (Clopper–Pearson) 95% confidence intervals. Plasma and tissue prevalence were compared gene-wise using Fisher exact tests, with a Bonferroni-corrected threshold of p < 0.0125 for four comparisons; these comparisons are exploratory and are not powered to establish equivalence. Tumor mutational burden is summarized as the median and the number of evaluable samples in each group. Because the tissue and plasma cohorts are neither matched nor drawn from a common sampling frame, and because tumor mutational burden was derived using different panels and bioinformatic definitions, the two sets of estimates are presented in parallel and were not compared inferentially. All analyses are hypothesis-generating.

2.7. Ethics statement

The study was conducted in compliance with the Declaration of Helsinki. Institutional Review Board approval was not required for the tissue genomic data, as it originated from publicly available cBioPortal datasets (which have been appropriately referenced). The protocol for the analysis of plasma cfDNA results was reviewed and approved by Advarra Institutional Review Board Services, 6100 Merriweather Dr., Suite 600, Columbia, MD 21044, USA, Approval No. Pro00034566/CR00 218935. Furthermore, the Advarra Institutional Review Board waived the need for informed consent to analyze de-identified data. Written informed consent for genomic testing and for publication of this case, including anonymized imaging, was obtained from the patient's parent or legal guardian.

3. Results

3.1. Index case: plasma findings and clinical course

Plasma cfDNA NGS identified a truncating MSH6 E544fs variant at 100% variant allele fraction, an IDH1 R132C variant at 0.5%, a tumor mutational burden of 132.3 mut/Mb, and MSI-high status. Additional reportable alterations were a TP53 splice-site variant (0.8% VAF), KRAS G60D (0.4%), and NOTCH1 L1678P (0.1%), with 13 variants of uncertain significance at 0.2–1.7% and 9 synonymous alterations at 0.2–1.0%. The predicted consequence of the frameshift and the variant allele fractions of all detected alterations are shown in Fig. 3. H3K27M was not covered by the panel. Every somatic-range variant in this sample lay between 0.1% and 1.7%, indicating a low tumor fraction. A somatic variant cannot reach complete allelic fraction in such a sample, so the MSH6 fraction of 100% indicates constitutional rather than somatic origin. Confirmatory germline testing on peripheral blood identified the same MSH6 variant in the homozygous state, establishing constitutional mismatch repair deficiency (CMMRD). There was no relevant family history of Lynch-spectrum malignancy and no prior malignancy in the patient; the family was referred for genetic counseling and cascade testing.

Fig. 3.

Predicted Consequence of the Plasma-Detected MSH6 Frameshift in the Index Case

(A) Schematic of MSH6 (1360 amino acids; UniProt P52701, NP_000170.1) showing the position of the frameshift variant detected in plasma cell-free DNA at codon 544. The variant is predicted to truncate the protein, removing 817 residues (60% of the coding sequence) including the region containing the conserved Walker-A adenine nucleotide-binding motif required for the ATPase-dependent conformational switch of the MutS alpha heterodimer. The N-terminal PWWP domain, which binds trimethylated Lys-36 of histone H3, is retained. On this basis the variant is classified as pathogenic. In this case the predicted loss of function is corroborated functionally rather than assumed: microsatellite instability-high status and a tumor mutational burden of 132.3 mutations per megabase were present in a patient who never received an alkylating agent, so the hypermutant phenotype cannot be treatment-induced. Microsatellite instability is a phenotypic readout of failed mismatch repair rather than a genotype, and a variant allele fraction of 100% indicates the absence of a wild-type allele; functional mismatch repair deficiency is therefore supported for the index case and not merely predicted. This is distinct from the cohort analysis, in which pathogenic MMR gene alterations were identified without accompanying functional data, and for which the term gene alteration rather than deficiency is used throughout. Germline testing on peripheral blood confirmed homozygosity for the same MSH6 variant. Residual uncertainties are that the microsatellite instability call itself was made on plasma without tissue confirmation, that the module has not been separately validated in glioma, and that POLE and POLD1 are not reported by either assay version, so a polymerase proofreading defect — the principal differential for ultra-hypermutation at this level — could not be excluded. (B) Variant allele fractions of all non-synonymous alterations detected in the same plasma sample, on a logarithmic scale, with the fraction expected for a heterozygous germline variant (approximately 50%) and the assay limit of detection indicated. Reportable alterations were MSH6 p.Glu544fs (100.0%), the TP53 splice-site variant (0.8%), IDH1 R132C (0.5%), KRAS G60D (0.4%) and NOTCH1 L1678P (0.1%). Variants of uncertain significance in genes recurrently involved in clonal hematopoiesis are shown separately: MPL (0.4%), JAK3 (0.4%) and ATM (0.3%).

On pembrolizumab the pontine lesion was radiographically stable for approximately 15 months across 21 three-weekly doses, with maximum rCBV falling to 1.0–1.3 by February 2025 and the patient remaining ambulatory and in school. Approximately 19 months after diagnosis, in August 2025, new gait disturbance and facial palsy accompanied interval enlargement of the enhancing component with rCBV rising above 4.0. Pembrolizumab was stopped and ivosidenib with bevacizumab begun; imaging at three months showed interval increase within the irradiated field, so treatment-related change could not be distinguished from progression. At last follow-up in January 2026, 24 months after diagnosis, the patient was alive with performance status 3 to 4 and dependent for daily activities. The clinical course, treatments given and imaging timepoints are summarized in Fig. 1. Serial imaging is shown in Fig. 2 and detailed in the Supplementary Material.

Fig. 1.

Clinical, Treatment and Imaging Timeline of the Index Case

Timeline of the index case from presentation in January 2024 to last follow-up in January 2026, expressed in months from diagnosis. Bars indicate treatment periods: radiotherapy (54 Gy in 30 fractions, completed March 2024), pembrolizumab (21 doses on a three-weekly schedule), bevacizumab (four cycles) and ivosidenib (250 mg once daily), which was continuing at last follow-up. Markers above the axis indicate clinical events and the timing of plasma cell-free DNA testing, performed approximately six weeks after completion of radiotherapy. Markers below the axis indicate MRI studies; filled markers denote the three studies shown in Fig. 2. Bars indicate the periods over which each agent was given and no inference about treatment effect is implied.

Fig. 2.

MRI of the Index Case at Three Timepoints

Representative axial MR images documenting the appearance of the lesion at three timepoints, one per row. Rows, top to bottom: 11 April 2024, six weeks after completion of radiotherapy; 8 February 2025, during pembrolizumab; and 10 August 2025, at clinical progression. Columns, left to right: T2-weighted, diffusion-weighted, apparent diffusion coefficient (ADC) map, T1-weighted, susceptibility-weighted, FLAIR, and post-contrast T1-weighted.

Each row shows an expansile T2-and FLAIR-hyperintense mass centered in the pons with extension into the midbrain, thalami, medulla and cerebellar peduncles. The section level differs between rows, so the structures visible and the apparent extent of the lesion differ accordingly. Sections were not acquired at matched anatomical levels across timepoints and volumetric measurements were not performed. These images therefore illustrate the appearance and multiparametric characteristics of the lesion at each timepoint and do not support comparison of lesion extent or size between rows. The serial changes described in the Results, including the interval enlargement of the enhancing component at the third timepoint, are taken from the corresponding radiology reports rather than from the sections displayed here. Perfusion maps are not shown, and relative cerebral blood volume is not a validated response endpoint in glioma.

3.2. Cohort characteristics

Of 1,226 glioma plasma samples tested, 680 (55.5%; 95% CI 52.6–58.3) had at least one non-synonymous alteration detected; 719 (58.6%) had an alteration of any class. This is consistent with previously reported digital-sequencing cfDNA series in primary brain tumors, which detected alterations in roughly half of samples, and in 55% of newly diagnosed glioblastoma sampled before resection [16,27]. Direct numerical comparison is limited because assay composition differed between series, the largest prior cohort spanning four panel versions of 54 to 73 genes accrued between 2014 and 2017, so these figures are offered as context rather than as a controlled comparison. The overall detection rate is shown in Fig. 4. Detection rate by histological subtype and grade could not be determined, as the subtype and grade breakdown of all 1,226 tested samples was not retrievable. The diagnostic distribution of glioma subtypes in the analysis set (n = 680) and the tissue cohort (n = 5,369) is summarized in Table 1. Glioblastoma comprised the majority of cases in both cohorts. In the plasma cohort, glioblastoma (n = 607) was followed by glioma not otherwise specified (NOS; n = 37) and oligodendroglioma (n = 22), with only three diffuse midline gliomas. In the tissue cohort, glioblastoma (n = 3,043) was followed by diffuse midline glioma (n = 1,634), oligodendroglioma (n = 418) and glioma NOS (n = 206). The two cohorts therefore differ substantially in subtype composition, which should be borne in mind when the prevalence estimates below are compared.

Fig. 4.

Plasma ctDNA Detection Rate in Glioma

Proportion of glioma plasma samples with at least one non-synonymous alteration detected. Left, the present cohort (680 of 1,226 samples; 55.5%). Center and right, published comparators profiled on the same assay platform: 302 of 665 samples (45.4%) in a primary brain tumor cohort, using a definition that also counted synonymous variants and variants of uncertain significance, and 11 of 20 patients (55%) in a prospective series of newly diagnosed glioblastoma sampled before resection. Error bars are exact (Clopper–Pearson) 95% confidence intervals. The present estimate is obtained under a stricter variant definition than the 45.4% comparator.

Table 1.

Diagnostic distribution of glioma subtypes in plasma and tissue cohorts.

Diagnosis	Liquid Samples, n (%)	Tissue Samples, n (%)
Glioblastoma	607 (89.3)	3,043 (56.7)
Glioma (NOS)	37 (5.4)	206 (3.8)
Oligodendroglioma	22 (3.2)	418 (7.8)
Diffuse Midline Glioma	3 (0.4)	1,634 (30.4)
Anaplastic Oligodendroglioma	4 (0.6)	61 (1.1)
Brainstem Glioma	1 (0.1)	0 (0)
High-Grade Glioma	5 (0.7)	6 (0.1)
Low-Grade Glioma	1 (0.1)	1 (0.0)
Total	680 (100)	5,369 (100)

Table 1: Distribution of glioma diagnoses among samples in the plasma analysis set (n = 680 informative samples, of 1,226 tested) and the tissue sequencing cohort (n = 5,369). Glioblastoma constituted the majority of cases in both datasets, followed, in the plasma cohort, by glioma not otherwise specified (NOS) and oligodendroglioma, and, in the tissue cohort, by diffuse midline glioma and oligodendroglioma. Values are numbers of samples with column percentages in parentheses; percentages are calculated using the cohort totals (n = 680 for plasma and n = 5,369 for tissue) as denominators and may not sum to 100 because of rounding. Diagnostic labels are reproduced as recorded in the source datasets; some, including anaplastic oligodendroglioma and brainstem glioma, predate the 2021 WHO classification and are retained here for fidelity to the source data rather than as current diagnostic entities.

3.3. Prevalence of MMR gene alterations in tissue cohorts

Among the tissue sequencing datasets (n = 5,369), MMR gene alterations were detected across all four genes evaluated. MSH6 was the most frequently altered gene, present in 1.2% of samples, followed by PMS2 (0.6%) and MLH1 (0.5%). MSH2 alterations were least frequent, occurring in 0.4% of tumors. The gene-wise distribution of MMR alterations in the tissue cohort is shown in Fig. 5.

Fig. 5.

Prevalence of Mismatch Repair Gene Alterations in Plasma and Tissue Cohorts

Gene-wise prevalence of pathogenic or likely pathogenic alterations in MSH6, MSH2, MLH1 and PMS2 in plasma (n = 680 samples with at least one non-synonymous alteration detected) and tissue (n = 5,369 sequenced samples), with exact (Clopper–Pearson) 95% confidence intervals. MSH6 was the most frequently altered gene in both compartments and at the same rate (1.2% in plasma, 1.2% in tissue). Confidence intervals overlap for all four genes, and no gene differed significantly after Bonferroni correction for four comparisons (MSH6 p = 1.000, MSH2 p = 0.114, MLH1 p = 0.260, PMS2 p = 0.424; threshold p < 0.0125). Numbers within bars are event counts. Referred to all 1,226 samples tested, the corresponding rates were 0.7% for MSH6, 0.5% for MSH2, 0.5% for MLH1 and 0.2% for PMS2. All plasma estimates rest on fewer than a dozen events per gene, and the two cohorts differ substantially in subtype composition (Table 1); the comparison is descriptive.

3.4. Prevalence of MMR gene alterations in plasma cohorts

Among the 680 informative plasma samples, MMR gene alterations were detected across all four genes: MSH6 in 8 samples (1.2%; 95% CI 0.5–2.3), MSH2 in 6 (0.9%; 0.3–1.9), MLH1 in 6 (0.9%; 0.3–1.9) and PMS2 in 2 (0.3%; 0.0–1.1). MSH6 was the most frequently altered mismatch repair gene in plasma, at the same rate as in tissue (1.2% in each). The rank order of the remaining genes differed between compartments: MSH2 and MLH1 were equally frequent in plasma (0.9% each) and PMS2 least frequent (0.3%), whereas in tissue PMS2 ranked second (0.6%), followed by MLH1 (0.5%) and MSH2 (0.4%). Gene-level rates are reported among the 680 informative samples as the primary denominator, following the convention in circulating tumor DNA studies; the corresponding rates referred to all 1,226 samples tested are given as a secondary estimate. All gene-wise comparisons with tissue below use the primary denominator. Gene-wise prevalence in plasma and tissue with exact 95% confidence intervals is shown in Fig. 5. On the primary denominator no gene differed significantly between plasma and tissue after correction for four comparisons (MSH6 p = 1.000, MSH2 p = 0.114, MLH1 p = 0.260, PMS2 p = 0.424; threshold p < 0.0125).

3.5. Tumor mutational burden and MSI

Tumors harboring MMR gene alterations showed higher tumor mutational burden than those without in both datasets (Table 2). The plasma comparison rests on eight MMR-altered samples and is reported descriptively; no formal testing was performed, and differences in panel size and bioinformatic definition preclude quantitative comparison between compartments. In plasma samples, the median tumor mutational burden was 12.5 mut/Mb in MMR-altered tumors compared with 6.8 mut/Mb in non–MMR-altered tumors. In tissue samples, corresponding tumor mutational burden values were 6.9 mut/Mb and 1.5 mut/Mb, respectively.

Table 2.

Tumor mutational burden by mismatch repair status in plasma and tissue cohorts.

Group	Plasma TMB (mut/Mb); n	Tissue TMB (mut/Mb); n
MMR-Altered	12.5; n = 8	6.9; n = 144
Non–MMR-altered	6.8; n = 264	1.5; n = 5,225

Table 2: Comparison of median tumor mutational burden (TMB) between mismatch repair (MMR)–altered and non–MMR-altered tumors in plasma ctDNA and tissue sequencing datasets. MMR-altered tumors demonstrated consistently higher TMB across both modalities, consistent with a hypermutator phenotype. MSI data were not available for the tissue datasets. Values are median tumor mutational burden; n is the number of TMB-evaluable samples in each group; tumor mutational burden was reported only on assay version 2.12, so these denominators are the version 2.12 subset rather than the full cohort. Tumor mutational burden was reported only for samples meeting the assay tumor-fraction requirement, so n is smaller than the corresponding cohort total. Panel size and bioinformatic definition differ between the plasma and tissue platforms, so the two columns are not quantitatively comparable.

MSI-high status was detected in 4 of 673 MSI-evaluable plasma samples (0.6%); seven samples were classified as not determined because of insufficient tumor content and are not counted as microsatellite stable. Referred to all 1,226 samples tested, MSI-high status was detected in 0.3%. Microsatellite instability data were unavailable in the tissue datasets.

4. Discussion

This study shows that plasma-based liquid biopsy can detect MMR gene alterations in gliomas, with no significant difference from tissue detected after correction for multiple comparisons, and with associated elevation in TMB. cfDNA was informative in 55.5% of samples tested, consistent with prior series and with detection being concentrated in high-grade tumors, extensive disease and hypermutation-driven genomic instability [[28], [29], [30]]. These observations support an expanding role for plasma-based genomic profiling in selected patients with central nervous system tumors.

The index case illustrates plasma cfDNA identifying an MMR gene alteration and extreme hypermutation where tissue characterization was not feasible. Checkpoint-inhibitor responses in DIPG are uncommon overall, but MMR-deficient gliomas appear to be a biologically distinct subgroup with increased sensitivity to PD-1 blockade [7,9,31,32]. Prior reports have documented durable responses to checkpoint inhibition in hypermutated gliomas driven by MMR deficiency, supporting the biological plausibility of the clinical course observed in this case [7]. Identification of a concurrent IDH1 alteration also informed subsequent treatment selection, underscoring the potential of plasma-based profiling to reveal clinically actionable information beyond a single therapeutic decision [18].

No significant difference in gene-wise prevalence between plasma and tissue was detected for any of the four genes after correction for multiple comparisons. Absence of a detected difference is not evidence of equivalence: the study was not powered for an equivalence test, and the risk ratio for MSH6 was 0.99 (95% CI 0.48–2.05), so a two-fold difference in either direction cannot be excluded. Point estimates were higher in plasma than tissue for MSH2 and MLH1, lower for PMS2, and near-identical for MSH6. Two mechanisms likely contribute. The tissue datasets are matched tumor–normal and report somatic variants after germline subtraction, whereas tumor-only plasma sequencing retains germline variants, which sit near 50% variant fraction irrespective of tumor content and so remain detectable in minimally shedding samples, selectively inflating apparent plasma prevalence; the index case, with an MSH6 frameshift at complete allelic fraction, illustrates this. Separately, because a sample becomes informative as a function of shedding rather than at random, the informative subset is biologically selected in a way the tissue cohort is not.

A notable limitation in neuro-oncology liquid biopsy remains the low abundance of circulating cfDNA, which constrains assay sensitivity in many patients. Low-input cfDNA workflows, including low-coverage whole-genome sequencing adapted for picogram-level inputs, have enabled genome profiling from all pediatric CNS tumor liquid biopsy samples in a cohort of 56 patients, with cfDNA detectable in a subset of serum samples and a higher proportion of CSF samples [33]. Such approaches highlight potential technical strategies to improve cfDNA yield and tumor signal detection in settings where traditional input requirements and low cfDNA shedding limit assay performance.

MSH6 was the most frequently altered mismatch repair gene in both compartments, at 1.2% in each, which is consistent with prior molecular profiling studies and with the established association of MSH6 loss with glioma-associated hypermutation. That the gene most strongly linked to glioma mismatch repair deficiency is also the gene most readily detected in plasma, at a rate for which no significant difference from tissue was detected, is the principal cohort-level observation of this study. All plasma estimates rest on fewer than a dozen events each and should not be over-interpreted. A caveat applies to the framing of this analysis: what is detected is a pathogenic or likely pathogenic MMR gene alteration, which is not equivalent to functional mismatch repair deficiency. Establishing deficiency requires biallelic inactivation, loss of protein expression, MLH1 promoter hypermethylation or a deficient mutational signature, none obtainable from a targeted plasma panel. Similarly, the observation of higher tumor mutational burden among MMR-altered tumors in both datasets aligns with the known hypermutator biology of this subgroup. Detection of MSI-high status in a small proportion of plasma samples further supports emerging evidence that microsatellite instability assessment in central nervous system tumors is feasible using cfDNA-based approaches [20,34].

4.1. Clinical relevance of TMB and MSI in glioma

The relationship between tumor mutational burden and benefit from checkpoint blockade in glioma is not uniform and depends on how hypermutation arose [8]. In a comprehensive analysis of gliomas with hypermutation, the phenotype emerged either de novo, in association with constitutional mismatch repair defects, or after temozolomide exposure [8], and these routes were not biologically equivalent: post-treatment hypermutated gliomas showed sparse T-cell infiltration, extensive intratumoral heterogeneity and a low rate of response to PD-1 blockade [8]. A high TMB in glioma therefore cannot be read as a tumor-agnostic predictor of benefit in the way it can in several extracranial cancers, and checkpoint-inhibitor studies in biomarker-unselected diffuse midline glioma and other pediatric central nervous system tumors have reported objective response rates of only a few per cent [35].

The evidence points the other way for a defined subgroup. In the largest reported series of primary mismatch repair deficient glioma in children, adolescents and young adults, checkpoint blockade was associated with improved survival [9], and durable responses have been documented in hypermutant glioblastoma arising from biallelic mismatch repair deficiency [7]. Primary MMR deficiency is also more frequent in pediatric high-grade glioma than historically appreciated [9] and is enriched in TP53-mutant and IDH-mutant subgroups, which corresponds to the alteration profile observed in the present case. The combination of an IDH1 mutation with mismatch repair deficiency in a child also raises primary mismatch repair deficient IDH-mutant astrocytoma as an alternative to diffuse midline glioma; the diagnosis here rests on clinical and radiological grounds without tissue or histone H3 status, so neither entity can be confirmed [36]. Microsatellite instability itself is rare in glioma when assessed by conventional tissue methods, with reported frequencies well below 1% in unselected series, which is one reason MMR deficiency in this tumor type is more reliably approached through MMR gene sequencing, protein expression and mutational-signature analysis than through microsatellite testing alone.

4.2. Clinical implications

Liquid biopsy may complement tissue-based profiling in glioma, particularly where biopsy is unsafe or infeasible, as in diffuse midline glioma and DIPG, and where tissue is limited, nonrepresentative or compromised by necrosis.

The expected yield of plasma cfDNA testing in glioma is modest and should be understood before the test is ordered. MMR gene alterations were detected in approximately 1% or fewer of plasma samples per gene, and MSI-high status in fewer than 1% overall, in a population already enriched for advanced disease. A negative or microsatellite-stable plasma result is therefore weak evidence against MMR deficiency and should not withhold further evaluation; where tissue is obtainable, MMR immunohistochemistry and tissue sequencing remain the reference standard. The value lies in the positive result: where tissue is unobtainable, an MMR gene alteration with high TMB identifies a subgroup for which a specific therapeutic hypothesis exists and can prompt germline evaluation.

5. Limitations

This study has several limitations. First, it is retrospective and integrates heterogeneous sources generated on different platforms and pipelines, so tissue–plasma comparisons should be read cautiously.

Second, matched tissue and plasma samples from the same patients were not available, so concordance could not be assessed at the individual patient level. For the index case no tumor tissue exists, biopsy was not undertaken at diagnosis for the reasons that motivate this report, and re-biopsy of a diffuse pontine lesion at progression was not clinically justifiable. Germline testing confirmed biallelic MSH6 loss, but the plasma MSI-high call itself was not confirmed against tissue.

Third, cfDNA shedding in gliomas varies with tumor grade, burden, location and treatment status, and the selection bias this introduces is discussed above.

Fourth, microsatellite instability status was not available in the tissue datasets, precluding direct tissue–plasma concordance analyses for MSI. In addition, the number of MMR-altered tumors in both cohorts was small, limiting statistical power and precluding formal hypothesis testing or subgroup analyses.

Reported concordance between liquid biopsy and tissue genotyping in glioma varies by compartment rather than being uniformly low. In blood plasma, a prospective paired series detected no alteration by both platforms, and a larger treated-glioma cohort matched characterized alterations in only a small minority of paired cases; cerebrospinal fluid has reproduced tumor-tissue alterations in the large majority of cases [37,38] and tumor in situ fluid in roughly a third [39]. Plasma occupies the accessible, low-sensitivity end of this gradient, which is why it is positioned here as complementary to, not a substitute for, tissue characterization.

Clonal hematopoiesis-type abnormalities are prevalent in cfDNA from treated gliomas [40], and this patient was sampled six weeks after cranial radiotherapy. The report contains candidate variants in genes recurrently involved in clonal hematopoiesis: a reportable TP53 splice-site variant at 0.8% and variants of uncertain significance in MPL (0.4%), JAK3 (0.4%) and ATM (0.3%). Without paired leukocyte sequencing a hematopoietic origin cannot be excluded for these or for the low-fraction IDH1 R132C, KRAS and NOTCH1 variants, and the IDH1 result that informed subsequent therapy should be read with that caveat. Clonal hematopoiesis cannot produce a variant at complete allelic fraction, so the MSH6 finding is unaffected.

Fifth, the pediatric DIPG case presented is illustrative and hypothesis-generating. Clinical outcomes observed in a single patient should not be generalized, and no conclusions regarding treatment efficacy can be drawn from this case alone.

Sixth, the radiological assessment has specific limitations. Relative cerebral blood volume is not a validated response endpoint; iRANO [23] assessment was by a single unblinded reader without central review; the reduction in enhancement six weeks after starting bevacizumab is consistent with pseudoresponse, since bevacizumab reduces enhancement and perfusion independently of tumor control; stability in the months after radiotherapy has alternative explanations; and the later changes lay within the irradiated field. The new parieto-occipital focus is of uncertain significance. Survival beyond two years is reported in approximately 10% of children with this diagnosis, so this course is presented as consistent with, rather than demonstrating, benefit.

Finally, a single commercial assay was used, so results may be influenced by gene coverage, input requirements and bioinformatic thresholds. Such studies should incorporate paired plasma and tissue and, where obtainable, cerebrospinal fluid sampling in biologically selected patients, with germline testing, serial cfDNA collection at predefined intervals, and MMR immunohistochemistry and mutational-signature analysis as reference comparators, with response assessed according to iRANO criteria [23].

6. Conclusions

MMR gene alterations can be identified using plasma cfDNA–based NGS in gliomas, including diffuse midline gliomas such as DIPG. In clinical settings where tissue sampling is unsafe or infeasible, cfDNA analysis may provide a noninvasive approach to detecting hypermutation and MSI-high status, with potential implications for immunotherapy selection. The case shows how plasma-based genomic profiling can inform treatment decisions when conventional molecular characterization is limited. Collectively, these findings support the role of plasma cfDNA–based NGS as a complementary tool in glioma molecular evaluation and highlight the need for prospective studies to further define its clinical utility.

Author contribution

Suman Karanth: Conceptualization, Methodology, Investigation, Formal analysis, Data curation, Resources, Project administration, Writing – original draft, Writing – review & editing.

Ankur Bahl: Conceptualization, Methodology, Investigation, Writing – review & editing.

Nitesh Rohatgi: Conceptualization, Methodology, Investigation, Formal analysis, Data curation, Writing – review & editing.

Rohit Bharat: Conceptualization, Methodology, Formal analysis, Data curation, Resources, Project administration, Writing – original draft, Writing – review & editing.

Anil Kumar Anand: Investigation, Writing – review & editing.

Rakesh Kumar Gupta: Investigation, Formal analysis, Data curation, Writing – review & editing.

Rana Patir: Investigation, Formal analysis, Data curation, Writing – review & editing.

Nippun Sandhir: Formal analysis, Data curation, Writing – review & editing.

Ethics declaration

Written informed consent for genomic testing and for publication of the case, including anonymized imaging, was obtained from the index patient's parent or legal guardian. For the de-identified cohort data, the Advarra Institutional Review Board waived the requirement for informed consent. The privacy rights of participants have been observed. (Approval No. Pro00034566/CR00 218935.) This research follows the CARE guidelines and the CARE checklist.

Data availability

Publicly available tissue genomic datasets used in this study are available through cBioPortal. The plasma cfDNA dataset contains de-identified patient-level genomic and clinical data generated under institutional and privacy restrictions and is not publicly available. De-identified aggregate data supporting the findings may be requested from the corresponding author and will be shared where permissible under applicable approvals, data-use agreements, and privacy regulations.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work the authors used AI-assisted tools for language editing and for drafting assistance. The authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

Funding

No external grant funding was received for this work. Guardant Health provided access to de-identified plasma cfDNA data and supported the analysis through contributions of NS and RB, who are Guardant Health employees. No additional funding was reported. Guardant Health co-authors contributed to study design, data access and analysis, and to drafting of the manuscript. They had no role in the clinical management of the index case, and no author employed by the sponsor participated in treatment decisions for that patient. The sponsor did not restrict the reporting of any result or the decision to submit for publication.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Rohit Bharat reports administrative support and article publishing charges were provided by Guardant Health Asia Middle East & Africa Inc. Rohit Bharat reports a relationship with Guardant Health Asia Middle East & Africa Inc that includes: employment and equity or stocks. Nippun Sandhir reports a relationship with Guardant Health Asia Middle East & Africa Inc that includes: employment and equity or stocks. Nitesh Rohatgi reports a relationship with Guardant Health Asia Middle East & Africa Inc that includes: consulting or advisory and travel reimbursement. Ankur Bahl reports a relationship with Guardant Health Asia Middle East & Africa Inc that includes: consulting or advisory. Nitesh Rohatgi reports a relationship with Datar Cancer Genetics Ltd that includes: equity or stocks. Nitesh Rohatgi reports a relationship with Roche Products (India) Pvt. Ltd that includes: speaking and lecture fees. Nitesh Rohatgi reports a relationship with Eli Lilly and Company India Pvt Ltd that includes: speaking and lecture fees.The remaining authors declare no competing interests.

Appendix A. Supplementary data

The following is the Supplementary data to this article: