Exome sequencing identifies BRAF mutations in papillary craniopharyngiomas

Craniopharyngiomas are epithelial tumors that typically arise in the suprasellar region of the brain¹. Patients experience substantial clinical sequelae both from extension of the tumors and from therapeutic interventions which damage the optic chiasm, the pituitary stalk, and the hypothalamic area^2–4. Using whole exome sequencing we identified mutations in beta-catenin (CTNNB1) in nearly all adamantinomatous craniopharyngiomas (11/12; 92%) and recurrent mutations in BRAF (V600E) in all papillary craniopharyngiomas (3/3; 100%). Targeted genotyping revealed BRAF V600E in 95% of papillary craniopharyngiomas (36 of 39 tumors) and CTNNB1 mutation in 96% of adamantinomatous craniopharyngiomas (51 of 53 tumors). The CTNNB1 and BRAF mutations were clonal in each tumor subtype and no other recurrent mutations or genomic aberrations were detected in either subtype. Adamantinomatous and papillary craniopharyngiomas harbor mutations that are mutually exclusive and clonal. These findings have important implications for the diagnosis and treatment of these neoplasms.

Craniopharyngiomas occur at an average age-adjusted incidence rate of 0.18 per 100,000⁵. There are two main subtypes of craniopharyngiomas – the adamantinomatous form that is more common in children and the papillary form that predominantly occurs in adults. Located in or above the sella turcica, craniopharyngiomas grow adjacent to the optic chiasm and often extend to involve the hypothalamus, cranial nerves, the ventricular system, visual pathways and major blood vessels at the base of the brain. Curative surgery is exceedingly difficult⁶ and resection can contribute to complications. The spectrum of complications include visual defects, severe headaches, pan-hypopituitarism, impaired intellectual function and wide-ranging hypothalamic dysfunction leading to sleep disorders⁷, abnormal thermo-regulation, and diabetes insipidus as well as to hyperphagia and uncontrollable obesity^2–4. Aptly, Harvey Cushing, the father of neurosurgery, who introduced the term “craniopharyngioma” declared them “the most formidable of intracranial tumors^8,9.”

Knowledge of the molecular mechanisms that drive craniopharyngiomas remains limited, and this has hampered the development of systemic therapies for this tumor. Mutations in exon 3 of beta-catenin (CTNNB1), which encodes a degradation targeting motif, have been described in adamantinomatous craniopharyngiomas, occurring in 60–75% of patients in most published series^10–12. However, mutations that drive the growth of papillary craniopharyngiomas have not been identified. To that effect, we performed massively parallel sequencing of 15 craniopharyngiomas from both subtypes and targeted genotyping in samples from 95 additional patients.

We first used whole exome sequencing to analyze the DNA from a discovery cohort of adamantinomatous (n=12) and papillary craniopharyngiomas (n=3) (Supplementary Table 1). As craniopharyngiomas are often heterogeneous tumors with nests of tumor cells interspersed between large amounts of reactive tissue and stromal cells, we expected mutations to have low allelic fractions (i.e. the mutations might only be found in a small subset of the sequenced cells due to substantial contamination by normal cells). Thus, we analyzed the sequencing data with a recently described method – MuTect – that uses a Bayesian classifier to identify somatic mutations with very low allelic fractions coupled with filters that provide high specificity¹³.

In consonance with the benign histology of these tumors, we identified only a relatively small number of nonsynonymous somatic mutations in both craniopharyngioma subtypes when compared to large cohorts of other tumor types¹⁴ (Fig. 1). The nonsynonymous mutation rate of 0.9/MB in the craniopharyngioma samples was similar to that found in a number of pediatric tumors as well as low grade tumors in adults such as WHO grade I meningioma¹⁵ (Fig. 1). 58% of the mutations were cytosine to thymidine at CpG dinucleotides which is consistent with spontaneous deamination and not a carcinogen-induced process (Fig. 1)¹⁴. As anticipated, the allelic fraction for many of the mutations was very low (median 3%; range 0.96–48%).

Figure 1.

Plot of the number of non-synonymous mutations per megabase in craniopharyngiomas in comparison to a broad range of pediatric and adult tumors. Data for all other tumor types, as well as the figure design were taken from Lawrence et al¹⁴. Each dot in this plot corresponds to a matched tumor–normal pair. The vertical position indicates the frequency of somatic mutations in that exome. Tumor types are ordered based on their median non-synonymous frequency and within each tumor type tumor-normal pairs are ordered from lowest to highest frequency. The relative proportions of six different possible base-pair substitutions are indicated in the bottom panel. Craniopharyngioma data are derived from whole exome sequencing of 15 tumor-normal pairs including 12 adamantinomatous and 3 papillary craniopharyngioma and marked in red. AML – Acute myelogenous leukemia, Cranio – Craniopharyngioma, DLBCL – Diffuse large B-cell lymphoma, LUAD – Lung adenocarcinoma; LUSC – Lung squamous cell carcinoma.

The most frequently mutated gene was CTNNB1, present in 11 of 12 adamantinomatous craniopharyngiomas (Fig. 2; Supplementary Tables 2–4). These events overlapped with mutations previously described in this tumor type and were found exclusively in exon 3^10,12. In addition, adamantinomatous craniopharyngiomas harbored isolated mutations in genes that have previously been implicated in cancer (Fig. 2). We identified mutations in genes listed in the Cancer Gene Census¹⁶ that are involved in transcriptional regulation (BCOR, CRTC3, MITF and PRDM1) and epigenetic regulation (KDM5A and SMARCA4) as well as DNA repair (BRCA2) (Fig. 2). Additional mutations were detected in genes not listed in the Cancer Gene Census but with described roles in cancer including genes involved in chromatin remodeling and regulation of transcription (including ASCC2, KAT5, PIWIL1, SIN3A, SMC1A), cell cycle and DNA repair (including ATR, BARD1, CDC25B, HBP1, RBBP8, RTEL1) and cell adhesion (ITGA3, PKD1, PKD1L1, PODN, ROCK1, RPTN)¹⁶. Many of these, however, may be ‘passenger’ mutations that do not contribute to tumor initiation or progression.

Figure 2.

Mutations in adamantinomatous and papillary craniopharyngiomas. Panel a depicts the number of mutations per megabase in each of 15 craniopharyngioma whole exome sequencing data sets (12 adamantinomatous and 3 papillary samples). Specific genes are listed bearing non-synonymous somatic mutations in genes listed in Cancer Gene Census¹⁶. At the right we show the false discovery rate q-values providing the significance of mutations in each listed gene. Cancer relevant genes that are not listed in the Cancer Gene Census are not shown. The q value is an evaluation of whether a gene is significantly mutated above the expected basal rate of mutation. The q values for BRAF and CTNNB1 are <0.00001. The q values of other genes in the plot are equal to 1. A full list of mutated genes is provided in Supplementary Table 2. Colors indicate the type of genetic change identified. Only one mutation is indicated even if multiple mutations were found in a particular gene. The cancer cell fraction (CCF, 0–1.0) for each indicated mutation is shown in white type in the corresponding box. The allelic fraction for mutations in each sample is shown in box plots with the median indicated by a red line (the allelic fraction for each mutation is listed in Supplementary Table 2. For each sample, the relative frequencies of six different possible base-pair substitutions are presented in the bottom panel. In Panel b, schematics for CTNNB1 and BRAF indicating the location of identified mutations (11 in CTNNB1 and 3 in BRAF) are shown.

All three of the papillary craniopharyngiomas in our discovery set harbored mutations (c.1799T>A) in the well-established oncogene BRAF (V600E) that have been shown to constitutively activate this serine-threonine kinase that regulates MAP kinase/ERK signaling and affects cell division and differentiation (Fig. 2). None of the three papillary craniopharyngiomas had mutations in CTNNB1. Although two of these samples were resected from adults (31 and 52 years old), one of the samples was from a rare papillary craniopharyngioma that occurred in a child (9 years old) suggesting a similar pathogenesis between pediatric and adult papillary craniopharyngiomas. In these tumors, we did not identify mutations in other genes listed in the Cancer Gene Census but did note isolated mutations in other genes with potential roles in cancer including ones encoding chromatin remodeling factors (CHD5, CHD6)¹⁷ and cell adhesion molecules (CDH26, PTPRT)¹⁸, and one (KIAA1549) that is fused to BRAF in most cases of pilocytic astrocytoma^19,20. These mutations, like some of those found in adamantinomatous craniopharyngiomas, may also be ‘passenger’ mutations.

MutSig¹⁴ was used to analyze the list of mutations identified in our discovery cohort of 15 cases to identify genes that are significantly mutated. Although our cohort is small compared to those from other genomic studies¹⁴ (Fig. 1), the prevalence of mutations in CTNNB1 and BRAF was so high that these mutations were readily detected as statistically significant (Fig. 2).

To validate our findings, we used targeted genotyping approaches to analyze an additional 98 craniopharyngioma samples (from 95 different patients) for mutations in the most commonly mutated genes in our discovery cohort – BRAF and CTNNB1. We also performed immunohistochemistry (IHC) using an antibody (VE1) that selectively recognizes the BRAF V600E mutant epitope and not the wild-type epitope from BRAF²¹. In addition, we evaluated the activation status of beta-catenin using an antibody that allowed us to detect nuclear (activated) and membranous (inactivated) beta-catenin¹². In total, our validation cohort consisted of 39 papillary craniopharyngioma tumors from 36 different patients and 59 adamantinomatous craniopharyngioma tumors, each from different patients (Supplementary Table 1). While CTNNB1 mutations were detected in the adamantinomatous craniopharyngioma samples (51 of 53 samples, 96%), none of these samples harbored BRAF mutations (neither V600E, D, L, M nor K). Cytoplasmic and nuclear beta-catenin was identified by IHC in all of the adamantinomatous samples tested but beta-catenin was exclusively localized to the cytoplasmic membrane in all papillary craniopharyngioma samples tested (Supplementary Table 1, Fig. 3). Remarkably, by targeted genotyping and IHC we detected V600E mutations in 34 of the 36 patients with papillary craniopharyngiomas (94.4%) (Supplementary Table 1).

Figure 3.

Beta-catenin localization is different in adamantinomatous and papillary craniopharyngiomas. Immunohistochemistry for beta-catenin was performed. In Panel a, beta-catenin is localized to the cytoplasm and the nucleus in an adamantinomatous craniopharyngioma. In Panel b, beta catenin is localized to the cell membrane in a papillary craniopharyngioma. Scale bars, 50 µm. Beta-catenin localization in many samples from the discovery and validation cohort is reported in Supplementary Table 1.

The allelic fraction of mutations (including those in BRAF and CTNNB1) in our whole exome data was low (Fig. 2a, middle panel, Supplementary Table 2). This raised the possibility of genetic heterogeneity in our tumor samples. To assess this, we used a recently developed computational method (see Methods section) that corrects for tumor purity and estimates the fraction of cancer cell nuclei that harbor a particular mutation (cancer cell fraction; CCF). Although these methods demonstrated that most of the somatic mutations identified in our samples were indeed subclonal, we found that the BRAF and CTNNB1 mutations were clonal (i.e. present in all tumor cells) in the analyzed samples (Fig. 2a upper panel, Fig. 4a and Supplementary Table 2-column BB).

Figure 4.

BRAF and CTNNB1 mutations are clonal in craniopharyngiomas. In Panel a (left), a violin plot shows the cancer cell fraction (CCF) for BRAF (orange) and CTNNB1 (pink) mutations in each tumor analyzed with whole exome sequencing. The median CCF of all non-synonymous somatic mutations for each sample is represented by a black dot. The bar graph (Panel a, right) shows the computed purity for each sample (see Methods sections); error bars represent standard error of the mean. In Panel b, we show hematoxylin and eosin (H&E) staining of adamantinomatous and papillary craniopharyngiomas. Immunohistochemistry (IHC) shows that adamantinomatous craniopharyngiomas are negative for BRAF V600E but that there is a diffuse distribution of BRAF V600E mutant protein in the neoplastic epithelium of papillary craniopharyngiomas. Stromal elements in the fibrovascular cores of the papillary tumors are negative for the BRAF V600E mutant protein. Scale bars, 100 µm.

To further validate our analysis of intra-tumor heterogeneity, we reviewed the pattern and distribution of BRAF staining in the craniopharyngioma samples (Supplementary Table 1). Using the BRAF V600E selective antibody, we did not detect BRAF V600E in the adamantinomatous craniopharyngioma samples (Supplementary Table 1) nor in two of the papillary craniopharyngioma samples with wild type BRAF that we tested (Fig. 4b). In the genetically confirmed BRAF V600E mutant samples, however, we observed widespread immunoreactivity across the tumor cell population. In these samples, lymphocytes and stromal cells of the fibrovascular core of the tumors were not immunoreactive, but the squamopapillary tumor epithelium stained diffusely for BRAF V600E. This observation supports the conclusion that BRAF mutations are present uniformly throughout the neoplastic epithelial cells.

Our whole exome sequencing data of craniopharyngiomas demonstrates that the adamantinomatous and papillary subtypes have distinct molecular underpinnings, each principally driven by mutations in a single well-established oncogene – CTNNB1 (beta-catenin) in the adamantinomatous form and BRAF in the papillary form, independent of age.

These mutations appear to be critical events in the pathogenesis of these tumors for several reasons. First, the high prevalence of mutations in CTNNB1 and BRAF in craniopharyngiomas occurs amongst an overall paucity of additional mutations, supporting that these are likely instrumental to the growth of these tumors. Second, the mutations are clonal and perfectly segregate with their respective histologic subtypes indicating that these are defining genetic aberrations. Moreover, the frequency of CTNNB1 mutations and BRAF mutations are much higher in craniopharyngiomas than in most other tumor types that bear these mutations.

Given the mutual exclusivity of BRAF and CTNNB1 mutations, immunohistochemistry for BRAF V600E and beta-catenin could be used to routinely distinguish papillary from adamantinomatous craniopharyngiomas, and thereby direct patients to appropriate clinical trials. While agents that target WNT signaling remain in development^22,23, the availability of BRAF inhibitors such as vemurafenib^24–29 and dabrafenib^30,31 suggests that patients with papillary craniopharyngiomas could immediately benefit from such targeted therapeutics. These agents have shown a robust clinical response against BRAF V600E mutant melanomas²⁵ and hairy cell leukemias³² as well as brain tumors such as pleomorphic xanthoastrocytoma^33,34 and ganglioglioma^34,35. Trials of these therapeutics for papillary craniopharyngiomas should be explored in patients with either residual or recurrent tumor following surgical resection. Also, BRAF inhibitors could be evaluated as first-line therapy, prior to surgical resection or radiation therapy.

In summary, our discovery of frequent and clonal mutations in adamantinomatous and papillary craniopharyngioma in CTNNB1 and BRAF, respectively, offers valuable opportunities for testing molecularly guided therapeutics for the treatment of these formidable brain tumors.

ONLINE METHODS

Sample Selection and Preparation

The study was reviewed and approved by the human subjects institutional review boards of the Dana-Farber Cancer Institute, Brigham and Women’s Hospital, Broad Institute of Harvard and MIT, Boston Children’s Hospital, Johns Hopkins Hospital, Children's Cancer Hospital Egypt, and Children’s Hospital of Philadelphia. Written informed consent was obtained from all participants whose tumor samples were subjected to whole-exome sequencing. Histologic diagnosis was re-confirmed on all samples by a board certified neuropathologist (S.S.) and representative fresh-frozen and paraffin embedded blocks with estimated purity of ≥10% were selected. DNA was extracted from tissue shavings of frozen tissue or 1 mm core punch biopsies (Miltex, cat# 33-31AA-P/25) from FFPE tissue and from buffy coat preparations of paired blood using standard techniques (QIAGEN, Valencia CA). The DNA was then quantified using PicoGreen® dye (InVitrogen, Carlsbad CA). Mass spectrometric genotyping with a well-established 48-SNP panel was used to confirm the identity of tumor-normal pairs (Sequenom, San Diego CA)³⁶.

Whole Exome Sequencing and Analyses

Whole exome sequencing was performed as previously described¹⁵. In brief, DNA was fragmented by sonication (Covaris Inc., Woburn, MA) to 150 bp and further purified using Agencourt AMPure XP beads. 50 ng size selected DNA was then ligated to specific adaptors during library preparation (Illumina TruSeq, Illumina Inc, San Diego, CA). Each library was made with sample specific barcodes, quantified by QPCR (Kapa Biosystems, Inc, Woburn, MA) and 2 libraries were pooled to a total of 500 ng for exome enrichment using the Agilent SureSelect hybrid capture kit (Whole Exome_v2 and Whole Exome_v4, 44 Mb; Agilent Technologies, Santa Clara, CA). Several captures were pooled further and sequenced in one or more lanes to a final equivalent of 2 exomes per lane on a Hiseq 2500 (Illumina Inc, San Diego, CA).

Read pairs were aligned to the hg19 reference sequence using the Burrows-Wheeler Aligner³⁷ and sample reads were de-multiplexed using Picard tools. Data were sorted and duplicate-marked using Samtools and Picard. Bias in base quality score assignments due to flowcell, lane, dinucleotide context, and machine cycle were analyzed and recalibrated, and local realignment around insertions/deletions was achieved using the Genome Analysis Toolkit (GATK)^38,39. Somatic variant calling was performed within the Firehose environment⁴⁰ using MuTect¹³. All sample pairs passed a QC pipeline to test for any tumor/normal and inter-individual mix-ups by comparing insert-size distribution and copy-number profile as described previously⁴¹.

Somatic mutations and short insertions/deletions were called and post-filtered using MuTect¹³ and IndelLocator^40,41. These were annotated to genes and compared to events in the Catalogue of Somatic Mutations in Cancer (COSMIC) using Oncotator, and spurious calls caused by mis-mapping and other previously identified systematic errors were removed using an established list of known problematic sites^15,40,42,43. MutSig¹⁴ was used to determine the significance of mutated genes. MutSig compares observed mutations against expected sequence specific, context-specific, tumor-specific and gene specific background mutation frequencies. Additionally MutSig prioritizes mutations that are positionally clustered or occur in highly conserved regions^40,42,43. We analyzed mutational spectra and rate across multiple tumor types using previously described methods¹⁴.

Mutation clonality analysis

To assess whether mutations are clonal (i.e. present in all cancer cells), we assessed the cancer cell fraction (CCF) of each mutation, as described in Carter et al⁴⁴. Mutations for which the CCF is close to 1 are considered clonal. Those mutations with lower probable CCFs are considered subclonal. To determine the CCF we first calculated the sample purity (i.e. the percentage of tumor cells in our sample) by constructing the probability density function for the allele fractions of mutations. Using that information we then identified a clonal heterozygous peak at or below 0.5 allele fraction:

$$\mathit{\text{pdf}}(af)=\sum _m\mathit{\text{betapdf}}(af,\mathit{\text{alt}}(m)+1,\mathit{\text{ref}}(m)+1)$$

where af is the true allele fraction between 0 and 1, alt is the alternate allele counts, ref is the reference allele counts, m is the index for mutation count⁴⁴. We assumed local copy number at each mutation is two for craniopharyngioma and that each tumor has heterozygous clonal mutations that are sufficient to form a peak^45,46. Given these assumptions purity is peak af * 2.

Once we had estimated tumor purity we then estimated the cancer cell fraction (CCF) for each mutation. The cancer cell fraction is the percentage of tumor cells harboring a given mutation. Clonal mutations have a true cancer cell fraction of one and subclonal mutations have a true cancer cell fraction of less than one. The observed allele counts correspond to a probability density of the CCF and this can be estimated with the following equation:⁴⁴

$$\mathit{\text{pdf}}(CCF,m)=\mathit{\text{betapdf}}(\frac{CCF}{CN(m)}*\mathit{\text{purity}},\mathit{\text{alt}}(m)+1,\mathit{\text{ref}}(m)+1)$$

where CN(m) is the local copy number at the given mutation m, and CCF ranges from 0 to 1. The probability for any allele fraction greater than the maximum CCF allowed was folded back to a CCF of 1. We assumed again a copy number of two^26,27. We also considered the possibility that the site is homozygous in both copies of the chromosome (loss of heterozygosity) by multiplying the CCF by 2 in the equation above and chose the solution with the highest peak probability. The CCF formulation is approximately equivalent to ABSOLUTE⁴⁴. The pdfs also provided a direct estimate of the confidence intervals on CCF. Each mutation was classified as clonal or subclonal based on the probability that the CCF exceeded 0.80. A probability threshold of 0.5 was used throughout for this classification⁴⁷.

Validation of Candidate Mutations by Sequenom Genotyping and Snapshot

We validated all of the common mutations in BRAF and CTNNB1 by mass spectrometric genotyping based on the Sequenom MassARRAY^® technology (Sequenom Inc, San Diego, CA) using a multi-base homogenous Mass-Extend (hME) as previously described^35,48. The complete hME assay list is provided in the supplementary appendix. A second targeted sequencing platform (SNaPshot genotyping) was performed as previously described to validate the BRAF V600E (c.1799T>A) mutation in all papillary craniopharyngiomas^33,49,50.

Immunohistochemistry

We performed immunohistochemical studies on five-micrometer-thick whole tissue sections of formalin-fixed, paraffin-embedded tissue in a Bond 3 automated immunostainer (Leica Microsystems, Bannockburn, IL, USA) using a primary antibody against BRAF V600E (clone: VE1, 1:100, Spring Bioscience, Pleasanton, CA) and a primary antibody against beta-catenin. For BRAF V600E, we deparaffinized the sections on the Leica Bond using Bond Dewax solution and performed antigen retrieval with an EDTA-based solution (Leica) at pH 9 and Leica Polymer Refine Kit for DAB (Diaminobenzidine) staining. Appropriate positive and negative controls were included. Positive staining was characterized by diffuse and moderate cytoplasmic staining of the tumor cells. We considered isolated nuclear staining, weak staining of occasional cells, or faint diffuse staining as negative staining. For beta-catenin (BD pharmigen, cat# 610154, mouse-monoclonal, clone: 14), antigen retrieval was performed in a pressure cooker in citrate buffer (pH=6.0, 1:1000 dilution) with a 45 minute incubation followed by Dako anti-mouse-HRP for 30 minutes at room temperature. Cases with nuclear staining (which ranged from low level to high level) were scored as positive and membranous staining were scored as negative.