Introduction
Diffuse midline gliomas are high-grade glial neoplasms of the thalamus or brainstem (including diffuse intrinsic pontine glioma, DIPG), and occur almost exclusively in young children. These tumors are not surgically resectable due to their anatomic location, which limits tissue available for diagnosis and molecular study. However, recent studies have revealed molecular characteristics of diffuse midline gliomas that are distinct from hemispheric pediatric and adult gliomas. Specifically, recurrent somatic mutations in genes encoding the replication-independent histone H3 isoform, H3.3 (
H3F3A), and the replication–dependent isoform, H3.1 (
HIST1H3B), are reported in a majority of pediatric midline and high-grade gliomas [
10,
20,
33,
38]. These mutations result in lysine 27 to methionine substitution (H3.3 or H3.1 K27M) or glycine 34 to valine or arginine substitution (H3.3 G34V/R) The K27M mutation is observed in up to 80% of diffuse midline gliomas, and G34V/R mutations occurs in up to 30% of hemispheric pediatric gliomas [
10,
20,
33,
38]. Because K27 and G34 are located in the N-terminal tail of the Histone H3 protein, these amino acid residues are critical sites for post-translational histone modification [
28]. As a result, H3K27M and H3G34V mutations have a significant impact on regulation of gene transcription and DNA methylation [
2,
27,
31,
32]. Because patients with H3 mutations demonstrate a more aggressive clinical course and poorer overall response to therapy [
3,
16,
20], the biological effects H3 mutations are thought to contribute to the lack of clinical response to treatments that are more effective in H3 wild type gliomas [
16].
Given the biological and clinical implications of histone H3 mutation in diffuse midline glioma, mutation detection is of great interest for advancing understanding of tumor biology and improving patient treatment. However, biopsy of these tumors for genetic analysis is not without clinical risk [
30]. In contrast, cerebrospinal fluid (CSF) is more easily obtained than midline brain tumor tissue, and tumor-specific genetic alterations may be detected in CSF due to direct contact with brain tumor tissue [
5,
14,
26,
37]. CSF from midline glioma patients may therefore serve as a reasonable alternative for detection of these mutations without the risk of tissue biopsy. Therefore, we set to detect Histone H3 mutation in archival CSF collected from pediatric patients with diffuse midline glioma, including DIPG, and to validate these findings in patient-derived tumor tissue. This approach could serve as a safe and robust method of “liquid biopsy” for histone H3 mutation detection in children with diffuse midline glioma, to potentially facilitate clinical stratification to targeted therapies and measure response to treatment.
Discussion
We present the first report of Histone H3 mutation detection in CSF from children with diffuse midline glioma. Given the high frequency and significant biological implications of histone H3 mutation in these tumors, “liquid biopsy” via CSF analysis may serve as an important approach for H3 mutation detection to impact patient treatment. Indeed, pre-clinical evaluation of agents aimed at the downstream effects of H3K27M in diffuse midline glioma demonstrate efficacy [
11,
12,
27], and biopsy-based clinical trials for patient stratification to molecularly targeted treatments based on H3 mutation status are now underway [
15,
17,
36]. However, while recent advances in neurosurgical and imaging techniques have made tumor biopsy for genetic analysis technically feasible, tissue acquisition from the brainstem or thalamus is not without risk, and brainstem glioma biopsy is not yet routinely performed. In contrast, CSF is more safely accessible than midline brain tumor tissue, and may provide a more accurate representation of mutation status than small tissue specimens. For example, tumor-specific mutations can be detected and quantified in CSF-derived tumor DNA from patients with primary and metastatic brain tumors as a correlate of tumor burden for clinical diagnosis and measuring tumor response to therapy [
19,
25,
26,
34,
37]. Our results demonstrate the feasibility and specificity of H3K27M detection in DNA from CSF from children with diffuse midline glioma, and suggest the potential clinical utility of CSF analysis for determining H3 mutation status in these patients.
We selected available archival CSF specimens collected from a cohort of pediatric brain tumor patients (n = 11) for Histone H3 mutation detection. An additional CSF specimen from a child with congenital hydrocephalus and no history of brain tumor was analyzed as a negative control. Specimen selection was based on tumor histopathologic diagnosis, location, grade, site and time of CSF acquisition, and specimen availability. Given previous reports we anticipated a low yield of DNA from CSF specimens. Therefore, in order to maximize the likelihood of H3 mutation detection, CSF specimens from children with intraventricular tumors (n = 2) and CSF diversion devices in close proximity to tumor tissue (n = 5) were preferentially selected for study. Diffuse midline glial tumors (PID 1–6) were hypothesized to have a higher likelihood of H3K27M mutation, while the remaining specimens in our cohort studied were expected to be H3.3G34V (PID 8) or H3 wild type (PID 7, 9–12). H3K27M status was evaluated matched tumor tissue when available (n = 8) to validate the sensitivity and specificity of CSF analysis for mutation detection.
In order to develop a robust, reliable method for H3 mutation detection in CSF, we first sought to identify the most suitable precipitation carrier for nucleic acid extraction. In an RNA analysis workflow, both the extracted target mRNA and carrier RNA is subjected to reverse transcription and second-strand synthesis, which can confound downstream analysis and library construction for RNA-sequencing. To our best knowledge, there is no effective way to isolate carrier RNAs from target mRNAs. While the Illumina Truseq RNA preparation workflow can be used to purify poly-A containing mRNA molecules using poly-T oligo-attached magnetic beads, this approach is not effective for isolating carrier RNAs, as these also contain poly-A tails. Size selection also cannot be used to isolate carrier RNA (yRNA), as the carrier is often several orders of magnitude longer than extracted nucleic acids of interest. We therefore compared linear polyacrylamide (LPA) as an alternative to carrier RNA [
1,
9,
13], and demonstrate that LPA is as effective as carrier RNA for nucleic acid precipitation. Given our intent to investigate CSF-derived RNA, we used LPA for all subsequent CSF DNA extractions.
In order to determine the source of DNA isolated from CSF specimens in our cohort (genomic tumor DNA or cell-free ctDNA), we evaluated extracted DNA fragment size. Our data demonstrate that centrifugation at 1000
g × 10 min is sufficient to isolate 150 bp DNA fragments, consistent with cell-free circulating tumor DNA (ctDNA). Our results also suggest CSF specimens in the present cohort contain a mixture of both genomic tumor DNA and ctDNA (Additional file
2: Figure S3). However, if quantifying changes in H3 mutation frequency, for diagnosis or monitoring disease progression or response to treatment, it is necessary to distinguish the source of DNA isolated in CSF specimens [
29]. Further studies to preferentially isolate ctDNA from CSF specimens submitted for H3 mutation analysis are therefore warranted, and currently underway.
Overall, DNA was isolated from all CSF specimens studied (
n = 12). In 8/12 specimens, DNA yield was sufficient for sequencing of amplified
H3F3A gene product for c.83A > T and c.104G > T transversion. In two
H3F3A wild type cases with sufficient DNA for further testing,
HIST1H3B sequencing for c.83A > T transversion was also performed. Sanger sequencing can detect histone H3 point mutations with precision, without the need for negative controls [
39], but does require a threshold quantity and quality of gene fragments to ensure the predominant wild type allele does not mask the mutant signal to yield a false-negative result. In our study, the DNA yield from 4/12 CSF specimens was below this threshold (<10.5 ng). Rather than applying multiple rounds of PCR amplification to these specimens, a nested-PCR strategy was employed for selective amplification of H3.3K27M mutant
H3F3A alleles from a total pool of
H3F3A in order to prevent amplification bias of smaller-sized DNA fragments [
6]. For this approach, a forward H3.3K27M mutation-specific primer was designed with the 3′-end anchoring to the variant-base of the mutant allele (Fig.
1d), to ensure that only the allele containing the missense mutation would be elongated and amplified, minimizing the likelihood of a false negative result. Three reverse primers, all complementary to the wild type sequence, were also designed to amplify 150 bp of the
H3F3A gene. Using this approach, we detected H3.3K27M in two additional CSF specimens from patients with midline glioma. This strategy is therefore an effective alternative sequencing approach when working with a very low amount of starting nucleic acid and for detecting mutations with low starting allelic frequency, as the primer is highly mutation-specific. In three DIPG cases with sufficient amount of DNA isolated, both H3K27M detection strategies were performed and the results were found to be 100% concordant.
The described method for H3K27M detection in CSF-derived DNA effectively circumvents a major challenge in detecting an oncogenic mutation: low relative concentration of mutant DNA in source material. Because oncogenic mutations occur in only 0.1–1% of all DNA molecules for a given genomic locus, deep sequencing coverage is required to achieve sufficient sensitivity for detection [
26]. Our cohort included CSF specimens from seven children harboring diffuse midline gliomas (Table
1). We detected the H3 mutation in 66.7% (4/6) of CSF-derived DNA analyzed, including three DIPGs (H3.3K27M) and one thalamic anaplastic astrocytoma. Analysis of one additional CSF specimen from diffuse midline glioma (PID 3) did not reveal H3.3 or H3.1 K27M mutation, while sequencing of CSF from PID 6 was not possible due to the low quantity of starting DNA (<0.5 ng). As expected, all CSF specimens from children with tumors located outside midline were negative for H3K27M, and H3.3G34V was detected in CSF from the patient with supratentorial glioblastoma (PID 8). H3K27M status was validated in tumor tissue (
n = 8) via IHC staining (
n = 7) and/or genetic sequencing (
n = 4). Tissue analysis results were 100% concordant with DNA analysis results of H3.3 K27M status (Table
1) for cases in which both analyses were possible (
n = 6). The lack of available tumor tissue for analysis from three patients in our cohort (PID 1, 3 and 7) underscores the need for an alternative technique for H3 mutation detection in children harboring midline glioma. Finally, since global loss of H3K27 trimethylation is associated with H3K27M mutation, tumor tissue specimens were also stained H3K27me3. Results were consistent with expected relative patterns of K27 post-translational modification in H3 mutant and wild type tumors, providing additional validation of our CSF mutation analysis results.
We postulate that timing, technique and location of CSF collection may influence test sensitivity, and must therefore be considered when interpreting CSF DNA results. Importantly, a higher concentration of CSF-derived DNA was isolated from patients with intraventricular tumor extension and/or from CSF collected from ventricles in close anatomic proximity to tumor tissue (PID 5–8, 10–11; mean = 1.0 ng/μL CSF), compared to CSF collected from the lateral ventricle in patients with posterior fossa or brainstem tumors (PID 1–4, 9; mean = 0.16 ng/μL CSF, Additional file
5: Figure S2). Of note, the CSF specimens in our cohort were archival and hence not preserved in nuclease-free tubes. Given that low starting DNA can limit mutation detection, we have subsequently collected tumor tissue and CSF specimens from adult and pediatric brain tumor patients using nuclease-free tubes for increased mutation detection sensitivity using a recently published next generation sequencing technique [
24].
Importantly, the H3K27M mutation allelic frequency in midline glioma tissue is thought to be between 15 and 58% [
8,
23]. Likewise, CSF-derived DNA has a lower proportion of wild type DNA relative to the mutant form. However, the described technique of DNA amplification and sequencing achieves high specificity for H3K27M mutant allele detection, without the need for costly next-generation sequencing. Given the emerging evidence of spatial and temporal heterogeneity of diffuse midline glioma tissue, coupled with the small volume of tissue typically acquired via stereotactic needle tumor biopsy, selective amplification of tumor DNA from CSF specimens may serve as a superior method for clinical detection of H3 mutation in this patient population. Indeed, detection of somatic mutations in CSF collected by lumbar puncture from brain tumor patients via droplet digital PCR (ddPCR) and targeted amplicon sequencing, with utility for monitoring tumor burden and response to therapy, was recently reported [
25,
26]. Comparison of perilesional cisternal, ventricular and lumbar CSF collected from patients with diffuse midline glioma, using mutation-specific primers and a ddPCR technique, could therefore be of value for determining the threshold DNA concentration and optimal source of tumor DNA for H3 mutation detection. In turn, CSF acquisition for longitudinal quantitative H3 mutation detection, via serial reservoir taps or lumbar puncture in lieu of tumor tissue sampling, could be of utility for measuring tumor burden and response to molecularly-directed treatments in these patients.
Overall, our results demonstrate the feasibility of histone H3 mutation detection in DNA derived from archival CSF from children with brain tumors. In our patient cohort, the described technique was 100% specific for H3K27M detection in patients with diffuse midline glioma, with 87.5% sensitivity (in specimens for which matched tissue was available for validation). These data suggest that CSF collection for H3 mutation analysis may be considered for patients with diffuse midline glioma, including DIPG, for diagnosis and to potentially facilitate stratification to molecularly-directed cancer therapy without the need for tumor tissue.
Acknowledgements
The authors would like to express their sincere gratitude toward all the patients and their families whose generous gift of biological specimens for research made this work possible. Thanks also to Marc Morgan, PhD (Department of Biochemistry and Molecular Genetics, Northwestern University Feinberg School of Medicine), for providing custom Histone H3 PCR primers; Rintaro Hashizume MD PhD (Department of Neurological Surgery, Northwestern University Feinberg School of Medicine) for providing pediatric brain tumor cell lines; Claudia Rivetta MS and Katy McCortney BA (Nervous System Tumor Bank, Department of Neurological Surgery, Northwestern University Feinberg School of Medicine,) for tissue processing and immunohistochemical staining; and Nitin Wadhwani MD (Department of Pathology, Ann & Robert H. Lurie Children’s Hospital of Chicago) for providing archival pathology specimens.