Prognostic impact of distinct genetic entities in pediatric diffuse glioma WHO‐grade II—Report from the German/Swiss SIOP‐LGG 2004 cohort

Reports on pediatric low‐grade diffuse glioma WHO‐grade II (DG2) suggest an impaired survival rate, but lack conclusive results for genetically defined DG2‐entities. We analyzed the natural history, treatment and prognosis of DG2 and investigated which genetically defined sub‐entities proved unfavorable for survival. Within the prospectively registered, population‐based German/Swiss SIOP‐LGG 2004 cohort 100 patients (age 0.8‐17.8 years, 4% neurofibromatosis [NF1]) were diagnosed with a DG2. Following biopsy (41%) or variable extent of resection (59%), 65 patients received no adjuvant treatment. Radiologic progression or severe neurologic symptoms prompted chemotherapy (n = 18) or radiotherapy (n = 17). Multiple lines of salvage treatment were necessary for 19/35 patients. Five years event‐free survival dropped to 0.44, while 5 years overall survival was 0.90 (median observation time 8.3 years). Extensive genetic profiling of 65/100 DG2 identified Histone3‐K27M‐mutation in 4, IDH1‐mutation in 11, BRAF‐V600‐mutation in 12, KIAA1549‐BRAF‐fusions in 6 patients, while the remaining 32 tumor tissues did not show alterations of these genes. Progression to malignant glioma occurred in 12 cases of all genetically defined subgroups within a range of 0.5 to 10.8 years, except for tumors carrying KIAA1549‐BRAF‐fusions. Histone3‐K27M‐mutant tumors proved uniformly fatal within 0.6 to 2.4 years. The current LGG treatment strategy seems appropriate for all DG2‐entities, with the exemption of Histone3‐K27M‐mutant tumors that require a HGG‐related treatment strategy. Our data confirm the importance to genetically define pediatric low‐grade diffuse gliomas for proper treatment decisions and risk assessment.


| INTRODUCTION
Low-grade gliomas (LGG) are a heterogeneous group of slow-growing glial or glioneuronal brain tumors usually assigned to WHO-grades I or II, which may arise in all regions of the CNS. 1,2 Circumscribed pilocytic astrocytomas (PA; WHO-grade I) represent the best characterized and largest subgroup in children and adolescents with a share of 70% to 80% in most pediatric series. [3][4][5] According to the 2016 WHO-classification of tumors of the central nervous system 2  usually IDH1/2-wild-type. 6,8,10 While no constant mutations of IDH1/2 or ATRX were found in DA, pediatric ODG typically lack 1p/19qcodeletion. 8 Whole genome analyses documented that pediatric LGG may harbor several different alterations of BRAF-, FGFR1-, MYB-or MYBL1-genes which are usually mutually exclusive. 9,10 Moreover, pediatric DG2 comprise just 6% to 10% in LGG-cohorts (<18 years), 4,5,[11][12][13] and even recent series of molecularly investigated tumors report only up to 36 cases preventing general conclusions about the prognostic significance of their molecular genetic profile. 10,[14][15][16][17] While long-term overall survival (OS) is generally excellent in pediatric LGG, early series indicated shorter OS for diffuse astrocytoma as compared to PA, 12,18 and even in the more recent reports of Stokland et al 4 11 OS remained reduced for DG2. After chemotherapy DG2 had a worse clinical course and shorter OS compared to PA in the multivariable analysis of the international SIOP-LGG 2004 trial, although progressionfree survival (PFS) was not impaired. 19 This result implied an unfavorable response to salvage treatment for progressive tumors beyond first-line therapy and raised the question whether pediatric DG2 may consist of prognostically diverse subentities. The combined information of pathological and genetic features improved risk assignment and treatment stratification for adult LGG. [20][21][22] This approach was attempted for pediatric LGG as well, but the reported cohorts were small, collected retrospectively or over extended periods of time and

What's new?
Pediatric low-grade diffuse gliomas histologically resemble their adult counterparts, but they differ greatly in terms of genetics, clinical behavior, and prognosis. Here, the authors investigated genetic mutations in 65 pediatric grade 2 diffuse gliomas (DG2), looking for a correlation with long-term outcome. All 4 tumors carrying the K27M mutation in the Histone3 gene were fatal. Conversely, none of the 6 tumors carrying a particular duplication of the BRAF gene, called the KIAA1549-BRAF fusion, progressed to malignant glioma. The authors characterized the tumor genetics with respect to prognosis, age at onset, and response to treatment. not treated within a comprehensive treatment strategy. 14,15,17,23,24 Thus, no general conclusion could be drawn so far. Therefore, we investigated the natural history, response to treatment and long-term outcome of 100 pediatric patients with centrally confirmed DG2 who were included in the prospectively registered, population-based German/Swiss cohort of the SIOP-LGG 2004 study. We focused on the frequency of genetically defined DG2-entities and their prognostic impact.

| Eligibility
The prospective, multinational/multicenter SIOP-LGG 2004 study registered patients with LGG of all CNS-sites from 2004 to 2012.
Inclusion criteria comprised age <18 years, histologic diagnosis of LGG according to the effective WHO-classification (2000,2007), without prior nonsurgical therapy. 19 Central review for radiology and neuropathology was recommended; central neuropathological review was mandatory for DG2 included in this series.

| Treatment strategy
All patients with DG2 followed the study strategy: at diagnosis, best safe resection of the primary tumor was recommended. Patients with complete resection were to be observed, as well as patients following incomplete resection, biopsy or radiological diagnosis if no threatening neurological symptoms were present. Severe initial symptoms or clinical/ radiological progression during observation were an indication for the start of nonsurgical/adjuvant treatment, if resection remained infeasible.
Children <8 years and all children with neurofibromatosis (NF1) were allowed to receive primary chemotherapy. Older children ≥8 years without NF1 were allowed to receive either primary RT or chemotherapy.
Chemotherapy was scheduled for 18 months with a 24-week induction (7 courses at 3-4 weeks intervals) and 60 weeks consolidation (10 courses every 6 weeks). Details for standard (vincristine/carboplatin) and intensified (additional etoposide) induction and consolidation were reported by Gnekow et al. 19 Focal radiotherapy (RT) was scheduled with a total dose of 54 Gy (1.8 Gy/fraction).
Brachytherapy with 125-Iodine-seeds was applied for suitable tumors. 25 Treatment for progression after primary RT or chemotherapy was not standardized, but included all modalities after discussion in local and reference tumor boards.
For radiological response assessment, contrast-enhanced MRI was performed at defined intervals at week 24, 54 and 85 after the start of nonsurgical therapy. Complete, partial and objective responses (regression) as well as stable disease (SD) were considered positive responses. 19,26

| Materials and neuropathological evaluation
Formalin-fixed paraffin-embedded (FFPE) tissue specimens of DG2 were retrieved from the archive of the German DGNN Brain Tumor Reference Center, Bonn, Germany.
All DG2-tumors were re-reviewed by two neuropathologists (M. G., T. P.) and classified according to the WHO-classification of tumors of the CNS 2 using standard histological and immunohistochemical methods. Nondiffuse low-grade astrocytic tumors such as PA and glioneuronal tumors were not included in our study.  After normalization of the assay against normal cerebellar tissue (FFPEmaterial), a difference of minus threefold standard deviation from the mean was considered as significant loss. MYB-FISH analysis was performed with commercial probes (Cytocell) as published previously. 23

| RNA-analysis for KIAA1549-BRAF and other recurrent fusions by Nanostring assay
The information of probes for detection of fusion genes is given in Table S1. A total of 100 ng RNA was added to the nCounter Elements TagSet in hybridization buffer and incubated at 67 C for 20 hours. Samples were then processed on the nCounter Preparation Station and cartridges were scanned on the nCounter Digital Analyzer. Raw NanoString counts were subjected to normalization using counts obtained for positive control probe sets. The normalized data was then subjected to background noise subtraction. A statistical outlier detection method was used to detect the presence of fusion/duplication events.

| Molecular inversion probe assay
To identify copy-number gains and losses, we used a molecular inversion probe (MIP) array including approximately 330 000 inversion probes (Thermo Fisher, Santa Clara, CA). The MIP assay was performed according to the protocols of the manufacturer. The raw MIP data file was analyzed using the Nexus Copy Number 8.0 Discovery Edition software (Bio-Discovery, El Segundo, CA). SNPFASST2 segmentation algorithm was used to make copy-number and loss of heterozygosity (LOH) calls. KIAA1549-BRAF-fused, BRAF-V600E-mutant, wild-type for these genes). Patients aged <1 year at diagnosis (n = 2) or with Histone3-K27M-mutation (n = 4) were excluded from multivariable analysis.

Extent of resection was included as time-dependent variable for OS
and EFS coming into effect at the date of surgery. Results are reported as hazard ratios (HR) with 95% confidence interval (CI) and P-value of likelihood-ratio test for selected variables.
Analyses were exploratory, and P values were considered as descriptive measures to detect and study meaningful effects. In particular, no significance level was fixed. Data for the remaining 100 patients are summarized in Table 1.

| RESULTS
Median age at diagnosis was 9.5 years; NF1 was diagnosed clinically in four patients. Thirty-two tumors were negative for these investigated genetic alterations. Extended analyses could be performed in 20/32 cases wild-type for these genes and where sufficient tumor material was available, but revealed no additional genetic alteration. In particular, methylationbased classification did not result in a confident score for any of the defined methylation tumor subgroups according to Capper et al., 27 MYB-FISH, molecular inversion probe (MIP) and Nanostring assays for rare fusions were negative. The association between genetic parameters and clinical features are depicted in Table 2.
Median follow-up for the 65 patients with genetically classified tumors was 8.5 years. Median age at diagnosis was 9.6 years and highest for patients with IDH1-mutated tumors (13.2 years).
The four histological DG2 with Histone3-K27M-mutation (6%) were localized in the midline (thalamus, 2 cases; caudal brainstem, 2 cases), with a median tumor diameter of 3.6 cm. After initial biopsy, adjuvant F I G U R E 1 Diffuse glioma Grade 2-cohort. Patient numbers and distribution among strategic subgroups. *Sixteen patients had radiological progression after initial incomplete resection/biopsy or no intervention, but did not receive nonsurgical treatment: 10 had salvage surgery and six were observed without surgical intervention Epidemiologic data, DG2-cohort    Survival data including the results of univariable analyses are detailed in Table 3 and illustrated in Figure 3A-D.   were among the most frequently reported presenting symptoms comparable to the observation in young adults. 33 Seizures related to better survival in adults. 34 Although this was not investigated in our cohort directly, patients with DG2 of the cerebral hemispheres, but also of the cerebellum and spinal cord, had a better OS than those with tumors in the supratentorial midline and brainstem. A higher rate of complete/subtotal resection is accepted as the major contributing factor to better PFS and/or OS for all pediatric LGG 4,5,12,13,33 and had been achieved for 48% to 70% for hemispheric, cerebellar and spinal tumors in this cohort. In addition to site, tumor size and the infiltrative nature of DG2 contribute to resectability. The distribution of "small" and "large" tumors (below/above 3 cm, as defined for ODG 35 ) did not differ between the sites in our cohort ( These results implied an unfavorable response to salvage treatment for progressive tumors beyond first-line therapy, "linking progression to survival", 4 and suggested the existence of prognostically relevant molecular genetic subtypes.

| Treatment results differentiated for the molecular genetic subgroups
We identified five genetically and clinically distinct major subtypes among 65 patients with sufficient tissue for molecular testing. This cohort did not show demographical differences to the cohort without residual material and therefore can be considered representative for DG2 (Table S3).

| Histone3-K27M-mutation
The most striking finding was the presence of Histone3-K27M-mutation in four midline tumors (6%). Histone3-mutations are a hallmark of malignant diffuse midline glioma WHO-grade IV according to the WHO-classification, 1 but have also been described in low-grade circumscribed 41 and diffuse, mostly thalamic and brainstem tumors 23,42 preceding malignant evolution 43 and death. 6,23,41 In the retrospective analysis of 289 pediatric low-grade glioma by Yang et al, 23 they constituted 6% and were allocated to a "high-risk group". Co-occurrence with other genetic alterations was described for various low grade histologies, 10,41 but was not seen in our tumors that were uniformly progressive within few months and all patients died. We advocate to test all diffuse midline gliomas for this mutation to identify and exclude Histone3-K27M-mutated midline diffuse gliomas in future pediatric LGG protocols, even if they lack signs of anaplasia.

| Impact of molecular-genetic classification
Despite an extensive genetic testing for rare fusions on the RNAlevel, methylation profiling and DNA-based molecular inversion probe assays we did not find distinct genetic events in half of all tumors, while smaller series detected at least one alteration in most DG2. 10,15 The increasing number of genetic alterations identified for each histologic variant calls for an increasingly differentiated diagnostic program. 9,15 Yet, even the extensive characterization of LGG in the series of Yang could not detect molecular changes in 27% of tumors (78/ 289). 23 In that paper recurrent TERT-mutations were associated with worse prognosis in older children. In our series, TERT-promotor mutations were absent indicating different tumor cohorts. Corroborating their findings of an "intermediate" biologic behavior for this subgroup, 23 the majority of our wild-type tumors were managed by surgery and "wait and see" strategy, although almost one-third needed adjuvant treatment and 6/9 patients received multimodal therapy including multiple surgeries. Although 28/32 patients were alive at last follow-up, four tumors had evolved to high-grade, including one with radiographic signs of gliomatosis cerebri and three patients had died.
While the unfavorable course for Histone3-K27M-mutated tumors is indisputable, no differences were found with respect to EFS after diagnosis or to OS for the other genetic DG2-entities.
Reflecting treatment stratification, primary chemotherapy was rather applied in younger patients and associated with multiple lines of treatment for 77% of patients. Nevertheless, the favorable outcome of our treatment groups with a 5-years OS of 72% and 82% after primary chemotherapy and RT, respectively, underscores the appropriateness of the study strategy for pediatric DG2.
Malignant transformation was seen, however, in all groups, except for the KIAA1549-BRAF-fusion positive tumors. Thus, rebiopsy of recurrent or progressing lesions is warranted even in early progression. Malignant transformation of pediatric LGG has been shown for BRAF-V600E-mutated tumors and was linked to the concurrent presence of CDKN2A-deletion. 6,23,48,49 Although the underlying molecular alteration was already traced in the initial biopsy in the report of Mistry et al, 49 and identical genetic alterations were found in primary and recurrent tumors in the series of Bergthold et al, 47 its later evolution cannot be ruled out. An unfavorable outcome of tumors with CDKN2A-deletion was linked to specific tumor sites. 23 However, none of our tumors had concomitant CDKN2A-loss. IDH-mutations are the hallmark of adult diffuse glioma, but are present in secondary glioblastoma, as well. 6,8 The propensity for malignant transformation has to be accepted for IDH-mutated tumors often in association with ATRX-mutation or p53-aberration. 8 Therefore, DG2 should be genetically classified upfront employing appropriate methods for diagnosis and forecasting outcome. In addition, genetic classification will be a prerequisite for targeted therapy within future protocols.
We found no significant differences with respect to outcome as compared to WHO-grade I tumors. Thus, the current LGG treatment strategy seems appropriate for all DG2-entities, with the exception of Histone3-K27M-mutated tumors that require a HGG-related treatment. More than half of all patients were safely followed by observation, while multimodal adjuvant treatment controlled progressive tumors in most cases.