Molecular Diagnosis of Diffuse Gliomas through Sequencing of Cell-Free Circulating Tumor DNA from Cerebrospinal Fluid

This article is featured in Highlights of This Issue, p. 2711

Diffuse gliomas are the most frequent primary malignant tumors of the central nervous system (CNS) and include IDH-wild-type glioblastoma multiforme (GBM), IDH-mutant GBM, diffuse astrocytomas, anaplastic astrocytomas, oligodendrogliomas, anaplastic oligodendroglioma, and diffuse midline gliomas (1, 2). Diffuse gliomas exhibit a wide range of prognosis depending on the grade, from 1 to 15 years median overall survival. Grade IV glioma, GBM, is one of the most aggressive tumors with less than 10% of patients surviving beyond 5 years (3). Magnetic resonance imaging (MRI) is the principal imaging modality for patients with suspected brain lesions but for a definitive diagnosis, tumor tissue (from biopsy or surgical resection) is required (4–6).

Genomic characterization of tumors is crucial for optimal diagnosis and treatment. However, characterization of cancer is challenged by evolving intratumor heterogeneity, which requires thorough and continuous analysis of genomic complexity over time. This is particularly relevant in brain malignancies where the genomic landscape changes in response to treatment or during relapse (7, 8). However, availability of glioma samples for characterization and correct diagnosis can be challenging. The anatomical location of gliomas can complicate tumor access, incurring high-risk surgical procedures and stereotactic biopsies. Moreover, specimens may be small and not representative, hampering correct diagnosis or even necessitating multiple surgical samplings to clarify final pathologic diagnosis. In addition, the surgical intervention strategy and assessment of the surgical risk–benefit balance depend on the glioma subtype (1, 9–11). This implies that an intraoperative histologic diagnosis may be required, possibly delaying the surgical procedure. Repeat surgical interventions may be needed to clarify tumor pseudoprogression on treatment or confirm early relapses. Taken together, a relatively noninvasive method for brain tumor characterization to facilitate molecular diagnosis would greatly assist the management of diffuse gliomas, minimizing complex and high-risk surgical procedures.

During the past decade, large-scale DNA sequencing efforts have identified key genomic alterations across glioma subtypes, facilitating classification (12–15). Recently, the 2016 CNS WHO proposed the integration of histology and genetic analyses to define glioma entities. Diffuse gliomas can be subdivided by the presence of IDH mutations, and within the subgroup of IDH-mutant gliomas the presence of codeletions of the 1p/19q chromosome arms as well as mutations of the TERT gene promoter ATRX and/or TP53 distinguish between diffuse astrocytomas and oligodendrogliomas (12, 13, 16). Moreover, diffuse midline gliomas are characterized by mutations in the H3F3A and HIST1H3B genes (17). Importantly, these tumors are usually located in the brain stem, thalamus, and spinal cord, making surgery to obtain diagnostic tissues extremely difficult.

We and others have reported the presence of cell-free circulating tumor DNA (ctDNA) in the cerebrospinal fluid (CSF) of patients with brain tumors (18–20). While absent or low levels of ctDNA are found in the plasma of these patients, ctDNA is frequently present in the CSF and can be used to characterize brain tumors (18). Here, we postulated that in diffuse glioma patients, CSF ctDNA could complement and facilitate molecular diagnosis. Based on recent knowledge about the repertoire of genomic alterations in diffuse gliomas, we argue that the identification of a set of gene mutations in CSF ctDNA could facilitate classification of this type of tumors. We therefore developed a platform to simultaneously and rapidly genotype mutations in IDH1, IDH2, TP53, TERT gene promoter, ATRX, H3F3A, and HIST1H3B by targeted sequencing and droplet digital PCR (ddPCR).

Tumor tissue, CSF (2 mL) were obtained from 20 diffuse glioma patients. Tumors were identified by MRI (Table 1; Supplementary Fig. S1) and diagnosed by histologic analysis. All patients underwent surgical resection, and CSF was collected prior to surgery by lumbar puncture, except for two samples obtained from the cisterna magna during a warm autopsy procedure, and one sample from the brain ventricle through a cerebral shunt. The study was approved by the local ethics committee, and written informed consent was obtained from all patients.

DNA from fresh tumor tissue samples was extracted using the QIAamp DNA micro kit (Qiagen), and germline DNA from peripheral blood lymphocytes was extracted using the QIAamp DNA blood mini kit. For formalin-fixed paraffin-embedded tumor samples, five 10-μm sections were obtained from the previously selected tumoral area and processed using the QIAamp DNA FFPE tissue kit. CSF-derived ctDNA was extracted using the QIAamp Circulating Nucleic Acids kit and quantified using a Qubit Fluorometer.

DNA from matched glioma tumor tissue, peripheral blood, and CSF samples underwent custom amplicon sequencing, targeting all exons of IDH1, IDH2, ATRX, and TP53. In brief, DNA libraries were prepared using the Nebnext Library Prep kit for Illumina, and paired-end 100-bp reads were generated on the Illumina Hiseq2500. Reads were aligned to the reference human genome hg19 using the Burrows-Wheeler Aligner v.07.12. Somatic variants were called using VarScan2, and only mutations with VarScan2 P < 0.05, total coverage ≥10 reads, variant coverage ≥7 reads, and with allelic frequencies >5%, were considered. All candidate mutations were reviewed manually using the Integrative Genomics Viewer.

Genomic DNA (10 ng) from tumor tissues, and germline DNA from peripheral blood lymphocytes, and CSF DNA (2–5 ng) from the same patient were used for digital PCR analysis, using the QX200 Droplet Digital PCR system per the manufacturer's protocols and the literature (21). Custom Taqman SNP genotyping assays for ddPCR were designed to detect IDH1, IDH2, TP53, H3F3A K27M, and HIST1H3B K27M point mutations and the corresponding wild-type alleles. Two ddPCR-specific assays (dHsaEXD72405942 and dHsaEXD46675715, respectively) were used to detect TERT promoter C228T and C250T mutations.

Immunohistochemistry was performed on 5-μm formalin-fixed paraffin-embedded tissue sections using a BenchMark Ultra immunostainer. Sections were stained with antibodies specific for IDH1-R132H–mutant protein H09 (Master Diagnostica), ATRX HPA001906 (Sigma-Aldrich), and p53 (mouse monoclonal DO-7; Ventana). After deparaffinization, sections were incubated with primary antibody (1:200) for 2 hours. Standard Ventana signal amplification was used. Sections from known mutation-positive and immunoreactive GBM tumors were used as positive controls.

1p/19q deletions were detected by fluorescence in situ hybridization (FISH) on 5-μm paraffin-embedded sections using Vysis 1p36/1q25 and 19q13/19p13 dual-colur probes. Signals were scored with at least 100 nonoverlapping intact nuclei. 1p/19q codeletion was defined as the signal ratio of 1p/1q <0.80.

In silico genomic analyses were performed using tumor mutational data and RNA-seq of two tumor cohorts from The Cancer Genome Atlas (TCGA), with Firebrowse (01 2016 version; N = 648). Tumors were considered as harboring a mutation in the TERT promoter as described previously (12), when expression levels are above a certain threshold, determined through the study of whole genome sequenced tumors. Tumors were considered mutated for IDH1, IDH2, ATRX, or TP53 when any protein affecting mutation (PAM) in the gene's canonical transcript was present, as identified by Variant Effect Predictor (Ensembl release 90). IDH-mutated was defined as a tumor bearing any PAM in either IDH1 or IDH2. In silico clinical analyses were performed using TCGA clinical data from the same source as the genomic data. Python lifelines library was used to compute the Kaplan–Meier curve, and its statistical significance between tumor genomic subgroups was assessed through a log-rank test. Survival COX regression was computed by using the same library and was adjusted by age.

We focused our search for relevant genomic alterations on gene missense or nonsense mutations, which can be detected with a higher sensitivity in ctDNA than genomic deletions or amplifications. Our panel included targeted sequencing of four genes (IDH1, IDH2, TP53, and ATRX) and ddPCR probes for the most common mutations in the TERT gene promoter (C228T and C250T), H3F3A (K27M), and HIST1H3B (K27M). In some cases, we also used ddPCR for mutations in IDH1, IDH2, TP53, and ATRX to increase the sensitivity of the method or validate the results from the targeted sequencing. Importantly, the methodology allowed rapid analysis of genomic alterations, with a 7- to 10-day turnaround time. Based on previous reports and the WHO 2016, the mutational status of these genes can facilitate the molecular diagnosis of diffuse gliomas according to three tumor entities, IDH-wild-type and TERT-mutant tumors suggestive of IDH-wild-type GBM; IDH-mutant and ATRX and/or TP53-mutant tumors suggestive of IDH-mutant GBM or diffuse astrocytomas; and IDH-mutant and TERT-mutant tumors suggestive of oligodendrogliomas. Although we did not analyze 1p/19q codeletions to determine oligodendrogliomas in TERT-mutant tumors, we took advantage of the fact that these codeletions tend not to coincide with TP53 and/or ATRX mutations (12, 13). Analysis of TERT, ATRX, and TP53 thus facilitated the diagnosis of oligodendroglioma.

We first determined how our proposed analysis could facilitate the subtyping of diffuse gliomas. To this end, we analyzed the tumor mutational and RNA sequencing data of two tumor cohorts produced by TCGA that contain 648 diffuse gliomas and evaluated how the combination of mutations in the IDH1, IDH2, TP53, TERT, and ATRX genes was distributed (Fig. 1A). We observed that 25% of tumors were IDH-wild-type and TERT-mutant, suggestive of IDH-wild-type GBM; 27% of tumors were IDH-mutant and ATRX and/or TP53-mutant, suggestive of IDH-mutant GBM or diffuse astrocytomas; and 27% of tumors were IDH-mutant and TERT-mutant, suggestive of oligodendrogliomas (Fig. 1A and B). Importantly, analysis of these five genes allowed 79% of gliomas to be attributed to three subgroups (IDH-wild-type GBM, IDH-mutant GBM/diffuse astrocytoma, and oligodendroglioma). As expected, the vast majority of tumors with ATRX mutations and/or TP53-mutations did not exhibit 1p/19q deletions (0.4% showed ATRX mutations and 1p/19q deletions, and 1% showed TP53 mutations and 1p/19q deletions), confirming this tumor group as suggestive of astrocytomas (Fig. 1C). Only one tumor had a K27M mutation in H3F3A and none in HIST1H3B as in this cohort very few pediatric tumors were included.

Importantly, the identified subgroups showed quite different prognosis. Median overall survival was 1.1 years for IDH-wild-type and TERT-mutant cases, 6.7 years for IDH-mutant and ATRX and/or TP53-mutant cases and 11.2 years for IDH-mutant and TERT-mutant cases (Fig. 1D). Thus, the molecular classification based on our gene set provided valuable information about prognosis, crucial for the surgical and clinical managing of the patients.

We then tested our sequencing gene panel in a cohort of 20 diffuse glioma cases representing several subtypes, as a proof-of-concept study to compare tumor and CSF ctDNA. CSF ctDNA and tumor DNA were analyzed using the described panel of targeted sequencing and ddPCR probes (Table 1). Gene mutations identified in tumor samples were found in the corresponding CSF ctDNA except for three cases (Fig. 2A). Moreover, the mutational analysis of the CSF ctDNA coincided with the histologic diagnosis. TERT mutations were observed in IDH-wild-type GBM; IDH, TP53, and ATRX mutations were found in IDH-mutant GBM and astrocytomas; and IDH and TERT mutations in the absence of ATRX and/or TP53 mutations were found in oligodendrogliomas where the 1p/19q status was confirmed by FISH (Fig. 2B).

The three cases in which the ctDNA did not recapitulate the results of the tumor analysis, corresponded to a diffuse astrocytoma and two oligodendrogliomas. All were grade II tumor subtypes with diverse tumor volumes, and the distance between the tumors and the brain cortex or ventricles varied from 0 to 1 cm (Fig. 3). Although the number of cases was too small to perform statistical analyses, our results suggest that tumor aggressiveness may be a relevant factor in determining the presence of CSF ctDNA (Fig. 3).

In our cohort, we had three cases of midline gliomas and H3K27 mutations were present in the CSF ctDNA of all of them (Fig. 2A). This is highly relevant because the identification of H3K27 mutations determines the diagnosis of diffuse midline glioma, a type of tumor that is minimally surgically accessible. Our results indicate that sequencing CSF ctDNA can facilitate the diagnosis of this glioma subtype in the absence of a tumor specimen. The three cases had different anatomical locations (thalamus, pons, and bulbous-medullar) and interestingly the allelic frequencies of H3K27 mutations was similar in all three (Table 1; Fig. 4).

Our work shows that analysis of mutations in IDH1, IDH2, TP53, TERT, ATRX, H3F3A, and HIST1H3B in CSF ctDNA facilitates the molecular diagnosis of diffuse gliomas.

Diagnosis of diffuse gliomas requires access to tumor specimens. Moreover, the evolution of the genomic characteristics of the tumor during progression and relapse, as well as the need to differentiate between pseudoprogressions or true progressions, or clarify early relapse, might imply multiple surgical samplings over time to adequately manage these tumors. In an attempt to overcome the heavy clinical burden associated with the surgical management of diffuse glioma patients, we developed a methodology to allow brain tumor characterization and molecular diagnosis in a relatively noninvasive manner.

The 2016 update of the WHO classification incorporates well-established molecular parameters into the classification of diffuse gliomas, with specific genomic alterations guiding diagnosis in combination with histologic analysis. Studies over recent years have shown that the analysis of IDH, ATRX, TP53, TERT, and 1p/19q status facilitates diagnosis of diffuse gliomas (12, 13). Interestingly, almost all TERT-mutant tumors that exhibit 1p/19q codeletion do not show mutations in ATRX and/or TP53, allowing 1p/19q status to be inferred from analysis of TERT, ATRX, and TP53 (12, 13). Thus, identification of IDH1 and IDH2, ATRX, TP53, and TERT status facilitates the classification of diffuse gliomas into IDH-wild-type GBM, IDH-mutant GBM/diffuse astrocytoma, and oligodendroglioma and can have prognostic value. Importantly, our results indicate that the analysis of the mentioned genes will be relevant for the subtyping of astrocytic tumors.

Moreover, mutations in H3F3A and HIST1H3B determine the diagnosis of diffuse midline gliomas (17). This is of great importance because the anatomical location of this type of tumors can prevent the possibility of obtaining surgical specimens.

The in silico analysis of a cohort of 648 diffuse gliomas indicated that 79% of them can be subtyped by analyzing IDH, ATRX, TP53, and TERT and thus can be detected by our approach. Furthermore, we could identify ctDNA in a large proportion of patients with diffuse glioma (17 of 20 patients). A lack of ctDNA in the CSF may depend on the aggressiveness of the tumor or the anatomical location (18–20). In our cohort, the three patients without CSF ctDNA (one diffuse astrocytoma and two oligodendrogliomas) were all grade II gliomas. The cohort was not large enough to conclude any tendencies; however, the tumor grade may contribute to predicting the presence of CSF ctDNA. Our results indicate that the analysis of CSF ctDNA in low-grade glioma might have limitations challenging its clinical utility in this context. The inability to detect ctDNA in these three patients in addition to the fact that we could not subtype 21% of diffuse gliomas provides evidence that not all patients will benefit from this type of analysis and that in several cases analysis of CSF ctDNA will not be informative, especially in low-grade gliomas. Nonetheless, for patients in whom this analysis is informative, it could be of crucial relevance in clinical practice, and all the more so as technologic advances improve sequencing sensitivity, thus reducing the number of noninformative cases.

In addition to the possibility of using CSF ctDNA to complement diagnosis and circumvent complex and challenging surgical procedures, CSF ctDNA can play an important role in the context of genomic characterization of tumors, identify potential actionable mutations, and facilitate the monitoring of tumor progression (18). Importantly, serial sampling of CSF after tumor resection might facilitate the characterization of tumor recurrence. Moreover, studies to determine the role of CSF ctDNA in identifying pseudoprogression to treatment as well as the possibility that it could facilitate early diagnosis even before the identification of a relapsed lesion by imaging are still needed.

Our platform can be improved by assessing more genomic alterations. In this manner, we could improve sensitivity by including gene mutations (such as H3F3A G34A). We could also incorporate targetable gene mutations (such as BRAF mutations) or other genomic alterations to improve prognostic stratification. However, the balance between complexity and speed of the assays and cost should be taken into consideration.

This pivotal study warrants future studies with a larger number of patients, with the demonstration that ctDNA in the CSF can be used as a liquid biopsy to help diagnose brain tumors and avoid risky surgical procedures. This opens a novel avenue in the field of noninvasive methods for molecular diagnosis aimed at better management of CNS malignancies.

J. Tabernero is a consultant/advisory board member for Bayer, Boehringer Ingelheim, Genentech/Roche, Lilly, MSD, Merck Serono, Merrimack, Novartis, Peptomyc, Roche, Sanofi, Symphogen, and Taiho. No potential conflicts of interest were disclosed by the other authors.

Conception and design: F. Martínez-Ricarte, R. Mayor, E. Martínez-Sáez, J. Tabernero, J. Sahuquillo, J. Seoane

Development of methodology: F. Martínez-Ricarte, R. Mayor, E. Martínez-Sáez, J. Tabernero, J. Sahuquillo, J. Seoane

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Martínez-Ricarte, R. Mayor, E. Martínez-Sáez, E. Pineda, E. Cordero, M. Cicuéndez, M.A. Poca, S. Ramon y Cajal, J. Carles, J. Tabernero, A. Vivancos, S. Gallego, F. Graus, J. Sahuquillo, J. Seoane

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Martínez-Ricarte, R. Mayor, E. Martínez-Sáez, C. Rubio-Pérez, N. López-Bigas, M. Vieito, J. Carles, J. Tabernero, F. Graus, J. Sahuquillo, J. Seoane

Writing, review, and/or revision of the manuscript: F. Martínez-Ricarte, R. Mayor, E. Martínez-Sáez, E. Pineda, E. Cordero, M.A. Poca, M. Vieito, J. Carles, J. Tabernero, S. Gallego, F. Graus, J. Seoane

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Martínez-Ricarte, R. Mayor, E. Martínez-Sáez, E. Pineda, M. Cicuéndez, S. Ramon y Cajal

Study supervision: F. Martínez-Ricarte, E. Pineda, N. López-Bigas, J. Seoane

This study was funded by the Spanish public grants from Asociación Española contra el Cáncer (AECC), the FIS PI16/01278, and the FERO and Cellex foundations. We thank all the patients and their families who participated in this study.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.