Genomic and epigenomic re-categorization of congenital glioblastoma and desmoplastic infantile ganglioglioma

The recently updated World Health Organization classification of central nervous system (CNS) tumors, 5th edition, (CNS5) reclassifies pediatric tumors according to their distinct molecular drivers, recognizing a new entity—infant-type hemispheric glioma (IHG). Defined by its unique epigenetic signature, and/or genomic fusions in ALK, ROS1, NTRK, or MET gene, IHG subsumes many cases previously classified as congenital glioblastoma (cGBM). Histologic features of IHG are still poorly defined with known overlap with a clinic radiologically similar entity-desmoplastic infantile ganglioglioma/astrocytoma (DIG). We revisited our cohort of cGBMs and DIGs, now reclassifying them according to CNS5 and compared the clinical, radiologic, molecular and histologic features between the two. 3/6 cases of cGBM that underwent targeted NGS fusion mutation panel were positive for ALK fusions (involving MAP4, MZT2Bex2, and EML4 genes as fusion partners), and 1/6 showed GOPC:ROS1 fusion. Interestingly, GOPC:ROS1 fusion was also shared by 1/5 cases of histologically defined DIG. DNA methylation profiling using the Heidelberg classifier (v12.3) recategorized 2/5 DIG cases as IHG (including the case with ROS1 alteration). In conclusion, histology alone is insufficient to distinguish IHG from DIG, necessitating epigenomic and genomic testing for the diagnosis of early-life gliomas.


Introduction
Prior to 2021, most pediatricians, pediatric neuro-oncologists, and pediatric neurosurgeons recognized that high-grade gliomas in children were distinct from those in adults, although there were no histologically distinguishable features, especially in congenital glioblastomas (cGBMs) as we [1] and others [2] have shown. Nevertheless, cGBMs often had a generally more favorable prognosis [1], although still highly variable [1,2], making it highly desirable to further subclassify these tumors.
The recent update to the World Health Organization classification of Central Nervous System (CNS) tumors, 5th edition (CNS5), published in late 2021, now divides pediatric high-grade gliomas into 4 types, with what is beginning to be understood as displaying very different molecular drivers and clinical outcomes [3,4]. These 4 new types eliminate "cGBM" nomenclature and are designated: diffuse midline glioma, H3 K27-altered, diffuse hemispheric glioma, H3 G34mutant, pediatric-type high-grade glioma, H3-wildtype and IDH-wildtype, and infant-type hemispheric glioma (IHG), all of which are assigned a CNS WHO grade of 4 except for IHG, which appears to show a better prognosis and remains officially not graded [3]. Of note, the first two categories are defined by alterations in the histone genes and the latter two by molecular alterations other than histone modifications [3,4].
CNS5 incorporates emerging data from 2 seminal articles published in 2017 [5] and 2019 [6]. The first [5] utilized a relatively new technique known as DNA methylation, which has been shown to robustly separate CNS tumors, based on their epigenetic features [7] and the second [6] employed more readily-available molecular testing by next generation sequencing (NGS). These techniques successfully separate the 2 non-histone altered pediatric-type high-grade tumors beyond their histological features into pediatric-type high-grade glioma (pHGG), H3-wildtype and IDH-wildtype, and infant-type high-grade gliomas (IHGs).
In contrast, NGS allows identification of IHG as having receptor tyrosine kinase (RTK) fusions including those in the NTRK family or in ROS1, ALK, or MET, all of which are definitional for the entity and targetable with current agents such as crizotinib [10], larotrectinib [10,11], lorlatinib [12], and theoretically with other agents such as ceritinib and ensartinib [10]. It was suspected that at least some cGBMs would now fall into this more prognosticallyfavorable IHG group [13], since the reported median age at presentation for IHGs is 2.8 months (range: 0-12 months) [6]. While most IHGs manifest classic histological features of glioblastoma, a complicating feature for identifying IHG on a strictly morphological basis alone is that there acknowledged histological overlap between molecularlydefined IHG and tumors formerly histologically diagnosed as anaplastic gangliogliomas (GG), desmoplastic infantile ganglioglioma/astrocytoma (DIG), ependymomas, and CNS primitive neuroectodermal tumors [4,14]. It is unclear whether the use of CNS5 criteria and molecular techniques, such as NGS and DNA-methylation profiling, leads to the reclassification of tumors formerly believed to be low-grade tumors, such as the DIGs that were formerly WHO grade 1. DIG and IHG share infantile onset and cerebral hemispheric location, and although IHGs usually lack the desmoplasia of DIG and lack their sharply demarcated growth patterns, whether this histological distinction is always completely distinguishing is unclear.
In this study, we re-interrogate our published experience with cGBM [1,13] and DIG [15] to determine how these might be reclassified in CNS WHO 2021. We included cases originally published as cGBM or DIG, as well as cases from our tertiary care pediatric hospital diagnosed since our first publication in 2012 [1], to determine whether molecular characterization by NGS (a genomic technique) and DNA methylation (an epigenomic technique), which might be superior to histological criteria in separating out these 2 entities.

Methods
Originally published cases, including the 7/8 hemispheric cases from our recent cGBM series [13] and 6 from our DIG series [15], were available for study, supplemented by one case of cGBM at our pediatric hospital since that time.
Histology methods had been previously described [13,15]. Details of NGS methodology were also previously published, including a detailed description of the NGS panels [13,16] and a listing of genes that are reported on one of our in-house NGS panels [13].
For DNA methylation, the Illumina 850 K whole genome assay and the Heidelberg classifier 12.3 were employed, with details of the methodology as previously described for our laboratory [17].

Results
The cGBM cohort included 3 males and 5 females, ranging in age from 1 week-3 months in age (see Table 1). The 6 DIGs included 4 males and 2 females, all 1-12 months in age (see Table 1).
The 8 tumors histologically diagnosed as cGBMs demonstrated large size and hemispheric location. They manifested a predominantly astrocytic histological phenotype with hypercellularity and palisading necrosis and variable pleomorphism, mitotic activity, and microvascular proliferation that did not allow distinction from adult GBM counterparts, as previously shown [1,2,13].
The 6 DIA/DIGs, while also showing large size and hemispheric location, more often displayed a mixed astrocytic and neuronal phenotype and desmoplasia that could be best recognized by numerous interspersed reticulinpositive fibers [15]. Although several DIA/DIGs in our series contained areas with more undifferentiated appearing cells, i.e., high-grade like histological features, they generally lacked necrosis or hemorrhage, unlike several of our cGBMs [1,13,15].
Using NGS or DNA-methylation testing, tumors originally diagnosed as cGBMs, including 5/7 of our original cGBM series for which testing could be performed, as well as the 1 newly accrued case, fell into the IHG category, based on definitional molecular features including ROS, ALK, or MET gene alterations or on DNA-methylation profiling (Table 1). Specifically, 4/7 showed an ALK fusion and 1 demonstrated a ROS1 fusion, namely, GOPC::ROS1 fusion. The fusion partners with ALK included MAP4, MZT2Bex2, EML4, and PCM1 genes. DNA methylation on the 3/4 cases for which sufficient material was available showed that 3 fell into the Heidelberg category of IHG with high confidence scores of 0.999, while the 4th case could only be poorly categorized by DNA methylation. Indeed, even though an ALK fusion (EML4 ex2:: ALK ex20) was identified in this case by NGS (currently definitional of IHG in the correct clinical-histologic context), the closest match to DNA methylation class was that of high-grade neuroepithelial tumor (HGNET), albeit with a low confidence score of 0.128 (see Table 1). The 3 tumors that matched IHG on DNA-methylation had no distinguishing age at presentation, clinical, or histological features from the case that did not match IHG (see Table 1). On NGS testing of our 6 DIGs, 1/6 had insufficient material for further assessment; of the remaining 5 cases, 3/5 had no identifiable mutations or fusions despite the 500 + gene coverage of our panel and 1 had a BRAF V600D mutation, as previously reported [15]. Surprisingly, 1/5 DIGs showed the identical GOPC::ROS1 fusion we also found in our histologically-defined cGBM group (Table 1). By DNA methylation analysis, there was even more diversity in DNA methylation class than for our cGBM cases. Namely, 2/5 histologically defined DIGs which underwent DNA methylation testing fell into the IHG class with high confidence scores of 0.99 and 0.92 and 1/5 fell into the DNA methylation class of DIG with high confidence score of 0.94. The remaining 2 assessable cases were "no match" on DNA methylation studies (with low confidence scores) (see Table 1). Cases showed no distinguishing age at presentation, clinical, or histological features (see Table 1).
Representative radiologic and histologic features of one of these cases, that were histologically DIG/ DIA cases but matched IHG on molecular testing, are shown in Fig. 1D-F. In comparison, radiologic and histologic features of a case that was classified as DIG on methylation are also shown ( Fig. 1A-C). Both cases showed similar features with marked cystic appearance ( Fig. 1A and D) and extensive desmoplasia ( Fig. 1C and F) with one of the cases (case 3) also showing ganglion-like cells (Fig. 1E). These features are consistent with the original histologic descriptions of DIA/DIG and did not allow differentiation of cases of DIA/DIG that matched the methylation class IHG.
T-distributed stochastic neighbor embedding (tSNE) plots of the 5 histologically defined DIA/ DIG cases similarly showed incomplete separation from IHG cases (Fig. 2) as these cases clustered with DIG and IHG cases.
None of our cases originally diagnosed as cGBM or DIG fell into the DNA methylation classes encompassed in diffuse pediatric-type high-grade glioma, H3-wildtype, and IDH-wildtype, a not surprising feature given the absence of prior radiation therapy or known congenital mismatch repair deficits in any of our patients.

Discussion
This study of relatively uncommon, but important, subtypes of pediatric gliomas, namely, IHGs and DIGs shows that despite recent CNS WHO 2021 categorization as pediatric high-grade gliomas and pediatric glioneuronal tumors, respectively, there is still much to be learned after re-interrogation with DNA methylation. Our first conclusion from this study is that histology alone may be insufficient to distinguish IHG from DIGs. This is concordant with the observation of others. Namely, although the presence of necrosis and classic "glioblastoma" features is most common in IHGs, and while lower-grade features and desmoplasia usually predominate in DIGs, overlap can occur. High-grade areas can be seen in some DIGs and a few IHGs can show ganglion cells, desmoplasia, or even lower-grade histological features, especially those with ALK fusion [6,10,18]. Our second conclusion is that histologically defined DIG appears to be a heterogeneous group when assessed by genomic or epigenomic criteria. Indeed, NGS showed BRAF V600D mutation in 1/5 assessable cases and 1 of our histologically-classic DIGs surprisingly manifested a genomic finding that is definitional for IHG, namely, a ROS fusion, specifically a GOPC::ROS1 fusion. Interestingly, the same fusion was also found in a case in our histologically defined cGBM cohort (Table 1). For DIGs, DNA methylation yielded even more diverse results, with only 1 of 5 assessable cases falling into the DIG DNA methylation class, 2 into the IHG DNA methylation class, all 3 with high confidence scores, and the remaining 2 into other DNA methylation classes. Although these 2 had low confidence scores; conversely, they clearly did not fit into either IHG or DIG categories. Our result that histologically-classic DIGs can fall into diverse DNA methylation classes despite similar morphology is concordant with the recent study presented in abstract format by Chiang et al. that DIGs may or may not fall into IHG categories [19].
Our third conclusion is that tumors formerly histologically diagnosed as cGBM [1] usually do fall into the new IHG category, albeit not invariably. Specifically, while 4/7 showed an ALK fusion and 1 demonstrated a ROS1 fusion, namely, GOPC::ROS1, the one with an ALK fusionwhich should be definitional for IHG by genomic findings-was not confirmed as IHG on DNA methylation testing. Instead, it fell into the DNA methylation class of HGNET. While this tumor had a low confidence score, it clearly did not match IHG or DIG. Thus, genomic and epigenomic testing for this case, 1 of 7 cGBMs, yielded different CNS5 WHO classification.
Our fourth conclusion builds on the results of this study; namely, we cannot make generalizations from our study whether histology, genomic, or epigenomic categorization should be the "gold standard" /final definitive result for categorization of tumors. However, we can conclude from our study that there may be discordances between NGS and DNA methylation classification results for individual cases. While it is clear that currently DNA methylation is a costly and complicated test modality that is not widely available, it was utilized in CNS WHO5 for categorizing tumors and is rapidly becoming standard of care for at least complicated pediatric tumors in a few areas of the world.
These new CNS WHO5 classification schemas have therapeutic and prognostic implications for clinicians, with possibly better prognosis for IHGs than the terminology of "high-grade" would imply. As noted above, IHGs by definition have receptor tyrosine kinase (RTK) fusions including those in the NTRK family or in ROS1, ALK, or MET, all of which are definitional for the entity and targetable with current agents such as RTK inhibitors larotrectinib [10], crizotinib [10], and lorlatinib [12]. Although we did not identify other DNA methylation classes within the new pediatric high-grade glioma group in our study of either cGBMs or DIGs, it should be mentioned that the 3 subtypes of pediatric-type high-grade glioma, H3-wildtype, and IDHwildtype also can have differing prognoses. Namely, DNA methylation allows grouping into 3 subtypes, with the RTK2 tumors showing a significantly longer survival time (median OS 44 months), the MYCN subtype with extremely poor LGG_GG, low-grade glioma -ganglioglioma; MB_SHH, medulloblastoma, sonic hedgehog activated subtype outcomes (median OS 14 months), and the RTK1 tumors manifesting an intermediate prognosis [5].
One acknowledged potential limitation of our study, as well as all DNA methylation studies, is erroneously low confidence scores in some cases due to low DNA/tumor content [20,21]. However, low DNA content can also be a limitation of NGS studies and sampling errors in histological tumor diagnosis on small biopsies or subtotal resection specimens are well-known occurrences as well. In some cases, the use of histology plus either NGS and DNA methylation, or both, may better classify a tumor.
Since this was a retrospective study, decisions for use of adjunctive post-resection therapy were made by the clinical team at the time of initial diagnoses of cGBM or DIG, using the histological criteria available at the time. However, as noted in the Table 1, occasional patients even with original diagnosis of DIG required adjunctive therapy on a case-bycase basis. We acknowledge that rendering a diagnosis of GBM-or for that matter even infant-type "high-grade glioma"/IHG-might prompt some clinicians to automatically administer adjunctive post resection therapy. As this is a small cohort, we can make no specific recommendations regarding treatment based on DNA methylation class, but we can provide caution that histology alone may not be the best way to categorize these tumors.
Finally, we have shown that sometimes there may be mismatch between genomic and epigenomic results. Our most obvious example is 1 of our cGBMs in which an ALK fusion was identified by NGS (a finding currently definitional of IHG), yet the closest match to DNA methylation class was that of high-grade neuroepithelial tumor (HGNET), not IHG. Two other possible examples of NGS versus DNA methylation "mismatch" might be for 2 of our DIGs. Namely, 1 had a BRAF V600D mutation, yet classified as germ cell tumor/teratoma, albeit with low confidence score, and conversely, our 1 histologicallydiagnosed DIG that fell into the DNA methylation class of GG did not have a BRAF mutation or any other detectable mutation or fusion typical for ganglioglioma [22]. While our study does not allow us to say whether NGS or DNA methylation is a superior test modality for pediatric tumors, we do observe that NGS currently identifies fusions or mutations that might lead to targeted therapies, whatever the DNA methylation class.
Author contribution AG and BKD contributed to acquisition, analysis, and interpretation of data for the work; drafting, revising, and critical review. ZS performed the DNA-methylation analysis and contributed to preparing the figures. EK and JK helped with review of clinical data.
Data Availability IDAT files from the DNA methylation analysis are available upon request.