Introduction
Diffuse intrinsic pontine glioma (DIPG) is an aggressive pediatric brain tumor with a median survival of less than 1 year, despite current multimodal therapies [
1,
2]. Midline, non-brainstem high-grade gliomas (mHGGs) in children share clinical and biological features with DIPG and have a similarly dismal prognosis [
3‐
5]. Historically, reluctance to biopsy these precariously located tumors to obtain tissue has impeded the understanding of their biology. Recently, greater acceptance of the safety of biopsy [
6‐
9], development of autopsy-based protocols [
10‐
12], and advancement of high-throughput sequencing technology have enabled unprecedented insight into the molecular underpinnings of DIPG and mHGG. Genomic studies have detailed recurrent aberrations in many canonical cancer pathways and mutations in novel oncogenes, such as highly recurrent histone mutations (
H3F3A or
HIST1H3B/C/I) and
ACVR1 [
5,
10,
13,
14].
Despite remarkable genomic discoveries, therapeutic progress for DIPG and mHGG has remained static. The standard treatment, focal radiotherapy, provides only transient local control and fails to address the recently reported metastatic potential of these highly infiltrative tumors [
1,
15,
16]. Clinical trials of adjuvant chemotherapy and targeted therapy, including those targeting critical pathways, such as platelet-derived growth factor receptor alpha (PDGFRA) [
7‐
9,
17], have not improved outcome over many decades [
11,
12,
18]. Intra-tumoral genomic heterogeneity and clonal evolution are well-recognized in the pathogenesis and therapeutic resistance of adult HGG [
19‐
21]. We previously reported intra-tumoral histopathological variation in DIPGs from autopsy [
1]. However, evaluation of spatial genomic heterogeneity, which carries important implications for determining the generalizability of molecular profiles derived from small diagnostic biopsies and scientifically-sound integration of molecularly-targeted therapies, has not been reported in DIPG or mHGG. We used whole exome sequencing (WES) to comprehensively evaluate spatial intra-tumoral genomic heterogeneity in eight children with DIPG or mHGG.
Discussion
Recent large-scale genomic studies have established the inter-patient molecular complexity of pediatric HGGs [
5,
10,
13,
14,
24]; however, limited understanding of spatial subclonal architecture has impeded development of effective therapies. To our knowledge, we are the first to report multi-regional genetic analyses of matched primary, contiguous, and metastatic sites from DIPG and mHGG. Disease-defining somatic mutations in
H3F3A,
HIST1H3B, and
ACVR1 were conserved across all tumor locations, as were mutations affecting the (RTK)-PI3K-MAPK pathway. However, other targetable molecular aberrations, such as
PDGFRA, were spatially heterogeneous. Findings from this study reiterate the distinct biology between pediatric and adult HGG and are critical for establishing the role of diagnostic biopsy and informing therapeutic strategy for these lethal pediatric tumors.
Our findings confirm the propensity of DIPG and mHGG for aggressive local and distant metastatic spread (Fig.
1) and are consistent with our prior study in which extra-brainstem and leptomeningeal disease was found in 25 and 39 % of DIPGs from autopsy, respectively [
1]. Caretti et al. similarly reported leptomeningeal involvement in a quarter of DIPGs from autopsy [
26]. Importantly, we found histological evidence of metastatic disease by post-mortem MRI in five cases in which tumor was not seen on pre-mortem imaging or grossly at autopsy. Hence, disease dissemination is prevalent, but in most cases appears late in the disease course.
Spatial histological heterogeneity, frequently reported in adult GBM [
27,
28], was observed in all but one patient in our cohort (Fig.
1). Tissue from two DIPGs demonstrated intra-pontine histologic variation, including one with evidence of WHO grade II, III, and IV astrocytoma. We previously reported substantial histologic heterogeneity at various levels of the brainstem in DIPGs from autopsy [
1]; other autopsy studies have reported limited histologic evaluation of one pontine site, most of which were high-grade [
12]. Metastatic lesions in our cohort tended to be of lower histologic grade, a finding consistent with our prior report [
1]. While variation in metastatic regions may represent sampling bias in areas of lower tumor content, our finding of histologic heterogeneity within the densely tumor-packed pons in DIPG reiterates the poor reliability of histopathological grading for clinical stratification [
1].
Significant temporal and spatial genetic and epigenetic heterogeneity has been reported in adult HGG, including regional variation of known driver mutations
EGFR,
MET,
PTEN, and
CDK6 [
19], as well as heterogeneity of
MGMT promoter methylation [
29]. We are the first to undertake comprehensive evaluation of the spatial genomic landscape of DIPG and mHGG in children. The most significant genomic discovery in DIPG and mHGG, to date, has been that of recurrent mutations in evolutionarily highly-conserved histone genes
H3F3A and
HIST1H3B, resulting in replacement of lysine 27 by methionine (K27M); these mutually exclusive somatic alterations occur in approximately 80 % of DIPGs [
14] and 50 % of pediatric thalamic GBMs [
4,
5]. Unlike the remarkable intra-tumoral heterogeneity of driver mutations in adult HGG, we confirmed the presence of heterozygous K27M mutations in
H3F3A or
HIST1H3B in 100 % of tumor cells across all disease compartments in four and three DIPGs, respectively. This finding builds on prior reports demonstrating consistently high allelic frequency of histone mutations in all primary DIPG tumor cells using deep sequencing [
1,
30] and Kambhampati et al. finding of conservation of H3.3-K27M by IHC between the pons and ventricular tumor extension in one DIPG patient from autopsy [
11]. Activating point mutations of
ACVR1, present in 20–30 % of DIPGs [
5,
14] and potentially targetable, were also conserved across all tumor compartments.
ACVR1 mutations were strongly associated with H3.1-K27M, as previously described [
14]. Furthermore, spatial conservation of activating mutations of
FGFR1,
PIK3CA, and
MET supports the therapeutic relevance of targeting the (RTK)-PI3K-MAPK pathway in DIPG. Spatial homogeneity of
PIK3CA mutations, in particular, supports their role as “founder mutations” in pediatric HGG, as suggested by Wu et al. who discovered a common
PIK3CA mutation (Q546K) in multiple tumor subclones derived from matched mHGG samples from diagnosis and recurrence, suggesting not only spatial but also longitudinal conservation [
14]. Although
TP53 aberrations were present across all tumor compartments, the type and location of aberration varied suggesting that they are critical, but likely secondary hits.
Similar to adult HGG, we observed subclonal variation of
PDGFRA aberrations in our cohort. Spatial heterogeneity was particularly evident for Patient 3 in whom amplification was observed in three of six tumor locations (Fig.
2). Fontebasso et al. similarly reported amplification in one of five primary tumor sites in a treatment-naive mHGG [
5]. Interestingly, in Patient 3, one pontine site bearing high-level
PDGFRA amplification also harbored a missense
PDGFRA mutation (E229K); other sites with
PDGFRA gain/amplification were wild-type with sufficient read coverage, suggesting later acquisition of this activating mutation in
PDGFRA-amplified clones. Our findings are similar to Puget et al. who reported co-occurrence of
PDGFRA mutation and amplification in three treatment-naïve DIPGs [
31] but differ from a report by Paugh et al. who found mutual exclusivity of amplification and mutation in 26 and 5 % of 43 DIPGs, respectively [
32]. While it is apparent that
PDGFRA mutations and copy number changes arise in subclonal populations, further studies are needed to determine their temporal order and functional consequence.
Our description of the spatial genomic landscape of DIPG and mHGG provides critical insight into the utility of diagnostic biopsy and the biologic rationale behind selection of therapeutic targets. Since the 1980s, imaging, rather than tissue, diagnosis has been the standard for DIPG and most mHGGs. More recently, pre-treatment biopsy has gained wider acceptance given the procedure’s low morbidity [
7,
8] and discovery of potentially actionable genetic alterations that may inform therapy [
5,
6,
14]. Despite a notable shift in perspective in the pediatric neuro-oncology community, ongoing concerns about intrinsic heterogeneity, poor ability to decipher driver from bystander mutations, and sampling bias from the small tissue yield of stereotactic biopsy have impeded uniform implementation of pre-treatment biopsy to guide therapy. Our study offers several important insights in favor of diagnostic biopsy for DIPG and mHGG, including definitive demonstration that disease-driving histone mutations are intra-tumorally conserved. Several large clinico-genomic studies have demonstrated the prognostic relevance of H3 mutations [
33,
34]. Ongoing acquisition of pre-treatment tissue will allow refinement of histone mutation-based risk groups and molecular signatures, which have potential to elucidate oncogenic mechanisms and resistance pathways. Our findings also demonstrate that stereotactic biopsy of the safest intracranial disease location offers
bona fide representation of certain molecular aberrations, abrogating the need for multiple biopsies at different sites of primary or metastatic tumor to elucidate the molecular signature from which therapy may be informed. Furthermore, the finding from this and other reports [
1,
32] that primary and metastatic tumor sites may demonstrate variable histopathological grades of astrocytoma (grades II-IV) without affecting prognosis or defining risk groups also supports the recommendation to limit the number of biopsy locations. Stereotactic biopsy of DIPG and mHGG should only be performed at specialized medical centers with skilled pediatric neurosurgeons trained in such techniques to minimize risk.
Development of adjuvant therapies should be focused on targeting highly conserved genetic aberrations, which likely represent true disease drivers. Promising preclinical data demonstrating potent ability of demethylase inhibitors (e.g. GSKJ4) to reverse the broad epigenetic dysregulation and transcriptional signature induced by H3 mutations in DIPG support ongoing efforts for clinical translation of such agents [
3,
7,
8]. Grasso et al. also reported compelling in vivo data demonstrating the therapeutic efficacy of multi-histone deacetylase inhibitor panobinostat in H3-mutant DIPG xenografts, including synergistic effect with GSKJ4 [
5,
6,
14]. Ongoing work with ALK2 inhibitors for
ACVR1-mutant DIPG is also encouraging [
10,
33,
34]. Therapies directed against subclonal variants in DIPG and mHGG, such as
PDGFRA, should not be abandoned but rather pursued in combination instead of monotherapy. Therapeutic implications of our findings should be formally tested in clinical trials implementing biopsy-directed targeted therapies. Indeed, target-based stratification based on biopsy results is already underway in several ongoing clinical trials in newly-diagnosed DIPGs (NCT01182350, NCT02233049).
Competing interests
The authors have no disclaimers or conflicts of interest.
Authors’ contributions
LMH, MD, SR, PB, JL, RD, CEH, and MF contributed to writing the manuscript. LMH, MD, SR, PB, PD, CEH, and MF designed the study. All authors contributed to data collection. LMH, MD, SR, PB, JL, LM, AR, MB, SSK, RD, QRL, BJ, DW, BA, CEH, and MF contributed to data analysis and interpretation. All authors critically reviewed the manuscript and agreed to submit for publication.
Lindsey M. Hoffman, Mariko DeWire, Scott Ryall are co-first authors.
Cynthia E. Hawkins, Maryam Fouladi are co-senior authors.