High-grade astrocytoma with piloid features: A clinical and genomic analysis of prognostic factors using a large cohort

Abstract

Background

High-grade astrocytoma with piloid features (HGAP) is a recently defined tumor type that is not well understood. Prognostic factors of clinical outcomes are not well established.

Methods

Methylation profiling was performed on tumor samples, many at the National Cancer Institute (NCI) Laboratory of Pathology, and others from publicly available sources. Methylation classifier scores of ≥0.90 to the HGAP class on the NCI-Bethesda classifier version 3 were included. Clinical features were collected from the medical record. Survival analyses were performed using the Kaplan-Meier and Cox-proportional hazards methods.

Results

The cohort comprised 421 patients. There were high rates of ATRX alteration (62%), CDKN2A/B homozygous loss (78%) and MGMT promoter methylation (53%). MAPK alterations were identified in 74% of evaluable samples. The median age was 46 years, and posterior fossa location was predominant (52%). The median overall survival was 88 months. Older age (P = .01) and the presence of an ATRX alteration (P = .04) were found to be negative prognostic factors. The presence of cystic features on magnetic resonance imaging (MRI) was found to be favorably prognostic (P = .01). Factors that were not significantly associated with prognosis included histologic high-grade features, CDKN2A/B homozygous deletion, MGMT promoter methylation, extent of resection, and presence of NF1 syndrome.

Conclusions

This large cohort establishes relative frequencies of several important markers. Additionally, older age, the presence of an ATRX alteration, and cystic features on MRI were found to be prognostic. Our work may aid in optimizing treatment regimens for patients with this tumor type.

Graphical Abstract

Graphical Abstract.

Key Points.

• A large cohort of high-grade astrocytoma with piloid features (HGAP) patients was analyzed by methylation profiling.

• The median overall survival was 88 months.

• Older age and the presence of an ATRX alteration are poor prognostic factors in HGAP, while cystic features on magnetic resonance imaging are positively prognostic.

Importance of the Study

High-grade astrocytoma with piloid features (HGAP) was recently described and codified into the World Health Organization classification of central nervous system tumors. This tumor type is not fully characterized, and clinically relevant biomarkers are not well understood. Here, we studied a multi-institutional cohort of 421 unique patients with HGAP confirmed by methylation profiling, to refine the histological, molecular, clinical, and treatment landscape, and to explore potential prognostic factors. The median overall survival of the cohort was found to be 88 months. Older age and the presence of an ATRX alteration were found to be poorly prognostic molecular factors. Histologic high-grade features were not significantly associated with clinical outcomes. Additional variables that were not associated with outcome included MGMT promoter methylation and CDKN2A/B homozygous loss. Cystic features on magnetic resonance imaging were found to be favorably prognostic. This work sheds light on the genomic features of HGAP and provides insights into the clinical course of this tumor type.

High-grade astrocytoma with piloid features (HGAP) is a recently characterized tumor type, first described in 2018¹ and is in the current 2021 World Health Organization (WHO) Classification of Tumors of the Central Nervous System (CNS). High-grade astrocytoma with piloid features is an IDH-wildtype glial neoplasm, and a tumor type that can only be identified based upon its unique epigenetic profile using DNA methylation. High-grade astrocytoma with piloid features has thus far been characterized as an entity that shares some histological characteristic features with pilocytic astrocytomas. A CNS WHO grade is currently not assigned to HGAP, but its clinical behavior most closely correlates to a CNS WHO grade 3 neoplasm.² Frequent molecular alterations include CDKN2A/B homozygous deletions, ATRX alterations and activating alterations of the MAPK pathway.^1,3 While HGAP is associated with these alterations, these alterations are not unique to this entity and for this reason methylation profiling is considered essential to make the diagnosis. Clinically, outcomes in HGAP have been found to be more adverse than that of pilocytic astrocytoma and less aggressive compared to glioblastoma, IDH wildtype, but remain incompletely characterized.^1,3 Possibly due to the diagnostic requirement for methylation profiling, most recent literature on HGAP is focused on case reports or small case series.^4-18 The natural history and clinical outcome of patients with these tumors are therefore incompletely elucidated. Thus, clinically relevant prognostic factors have therefore been difficult to identify and remain incompletely validated. Additionally, differences in HGAP arising as part of neurofibromatosis syndrome type 1 compared to those that arise sporadically are not fully known. To our knowledge, there is a lack of large cohort studies with available treatment data to survey the treatment landscape. Here, we fostered a collaborative effort and present a comprehensive survey of the molecular, prognostic, and clinical outcome, including treatment landscape of HGAP, and importantly, identify several prognostic factors by interrogating one of the largest international cohorts to date with 421 unique patients.

Methods

Cohort Selection

The NIH Institutional Review Board approved this study. Samples in the cohort included tumor samples processed for clinical methylation testing at the National Cancer Institute (NCI) Laboratory of Pathology, as well as samples uploaded to the Methylscape site (https://methylscape.ccr.cancer.gov/). IDAT files available from public data sources were also queried^19-25 and included GSE157397, GSE215240, GSE215438, GSE215240, GSE215741, GSE103659, GSE109381, GSE10938, GSE175543, GSE143843, GSE109381, and TCGA. Criteria for inclusion included availability of IDAT files, as well as a classifier match to HGAP with a confidence score ≥0.90 on the Bethesda version 3 (v3) classifier.

Duplicate samples and primary recurrent pairs from the same patient were identified using single nucleotide polymorphisms (SNPs) to ensure that each patient was represented once in the study.

Grading, Immunophenotype Evaluation

Tumor samples were centrally reviewed by K.A. and M.Q. for high-grade features. Histological grade was codified as consistent with a glioma grade 3 or higher (“high grade”) vs a tumor that did not meet this threshold (“low grade”). Histologic grade (high vs low) was determined based upon the presence of brisk mitotic activity, tumor necrosis, and microvascular proliferation. Consensus grading was annotated. Outside reports were referenced for grading as appropriate. Microvascular proliferation alone was not sufficient to upgrade the tumor. Samples were also annotated for GFAP, Olig2, BRAF, ATRX status on immunohistochemistry based upon submitted and available surgical pathology reports.

DNA Methylation Profiling, Copy Number Analyses, and MGMT Promoter Analysis

Formalin-fixed paraffin embedded (FFPE) tumor tissue was processed for DNA methylation profiling as previously described.²⁶ Briefly, tissue was collected from unstained slides and DNA extraction was performed using the QIAamp DNA FFPE Tissue Kit (Qiagen). Bisulfite conversion was performed using EZ DNA Methylation Kit, Zymo Research, followed by restoration processing with Infinium FFPE DNA Restore Kit (Illumina). Converted samples were then processed on Infinium MethylationEPIC beadchips (Illumina) using the Infinium HD FFPE Methylation Assay automated protocol (Illumina).

Classification of samples was performed using DKFZ CNS tumor classifier as well as the Bethesda version 3 of the NCI-Bethesda Classifier.²⁷ Samples classified with calibrated score 0.9 and higher as HGAP were included in this study. Somatic copy number plots were generated using the previously described Conumee 2.0 method.²⁸  CDKN2A/B homozygous deletion was queried with a reference log2 ratio threshold of −0.415, as seen prior in the literature.²⁹ A focused set of genes (MDM4, GLI2, PDGFRA, KIT, KDR, MET, CDK4, MDM2, and PPM1D) were found to be recurrently amplified and each of these genes was systematically queried for amplification, where a gene was considered amplified with a log2 ratio threshold of 1.0 or greater. Unsupervised clustering was performed using the top 10 thousand most variable methylated probes, followed by unsupervised UMAP clustering with the top 1000 methylated probes. MGMT promoter methylation status was determined using the STP-27 method.^30,31 Volcano plots were performed to compare good (defined as uncensored overall survival [OS] status better than the median OS) vs poor (defined as uncensored OS status worse than the median OS) survivors.

Genomic Alterations

Gene alterations were scored and identified from outside reports or medical records, as well as from next generation sequencing of NCI-processed samples where available. National Cancer Institute samples in FFPE were processed using the AllPrep DNA/RNA FFPE Kit (Qiagen) and then next generation sequencing was performed using TruSight Oncology 500 panel (Illumina). Libraries were subsequently prepared and hybridized, and paired-end reads were sequenced. Bioinformatic processing was performed as previously described.²⁶

Clinical Features and Outcomes

Clinical features of the patient samples were collected from medical records, including age, sex, neurofibromatosis type 1 (NF1) syndrome, location of tumor, initial magnetic resonance imaging (MRI) features by imaging report, presence and type of surgery, presence and type of systemic agent, as well as presence of radiotherapy and concurrent chemoradiation. Outcomes were collected from medical records where available, including date of diagnosis (surgical tissue confirmation of tumor), date of progression (radiographic or clinical), and date of death. In a few select cases where the remote specific date of initial tumor diagnosis or remote first progression was not available but month and year were available, the first of the month was chosen and if only year was available, then July 1 was chosen. Univariate and multivariate analyses were performed. Kaplan-Meier curves were generated (R version 4.5.1, survminer package). Biomarker-treatment interaction effect was calculated (R version 4.5.1, survival package). Chi-squared analyses were performed for correlates; and Wilcoxon rank tests and the former analyses were performed to compare variables among groups (methylome clusters, and sporadic vs NF1 syndrome-related HGAPs) (R version 4.5.1). A P-value of <.05 was deemed statistically significant.

Results

Identification of HGAP Cases

From an initial group of 434 tumor samples that matched to HGAP at confidence score of ≥0.90, sample/patient duplicates initially screened based upon Bv3 classifier and patient duplicates confirmed by SNPs were removed; 421 unique patients in this cohort were identified. No diagnoses of HGAP were found in recurrent samples from paired patient samples with an initial diagnosis of pilocytic astrocytoma identified via the Bethesda v3 classifier.

All cases matched to HGAP on the DKFZ classifier, with the exception of 9 cases (2%) (Supplementary Material S2). The specifics of the 9 discrepant cases, including ATRX and CDKN2A/B status are shown in Supplementary Material S3.

Grading and Immunophenotyping

Of the 421 tumor samples, histology or a report of histologic features was available in 282 samples. High-grade histology was most prevalent, although some samples showed low-grade histopathology (Table 1). Most (82.1%) had “any mitoses present” (total available, n = 196); 35.8% of samples were noted to have necrosis (total n = 190); and 60.1% had the presence of microvascular proliferation (total n = 198). GFAP was positive in 97.4% of samples (total n = 152) and Olig2 was positive in 97.5% (total n = 122).

Table 1.

Histological grade and key molecular markers

	Count, n (%)	Total, n
Histological grade		282
High grade	185 (65.6)
Low grade	97 (34.4)
ATRX alteration		199
Yes	123 (61.8)
No	76 (38.2)
CDKN2A/B loss		421
Yes	329 (78.1)
No	92 (21.9)
MGMT promoter methylation status		421
Methylated	225 (53.4)
Unmethylated	196 (46.6)

Molecular Characteristics

ATRX alteration status (either mutation or loss of tumor cell expression) was available for 199 samples (Table 1). There were 123 cases with ATRX alteration (61.8%). CDKN2A/B homozygous loss status, MGMT promoter methylation status, and gene amplification status were available for all 421 patients (Table 1). CDKN2A/B homozygous deletion was found in 78.1% of tumors and the MGMT promoter was methylated in 53.4%. There were 87 cases with at least 1 known genomic alteration based upon available outside NGS reports, and the most common alterations (outside of ATRX and CDKN2A/B loss) were NF1 (58.6%) and FGFR1 (14.9%). Sixty-four (73.5%) samples examined for genomic alterations had at least 1 MAPK pathway driver alteration (ie NF1, BRAF, FGFR alterations). TERT promoter mutation was found in 2.3% of samples (n = 2). The amplification frequencies of queried genes were: MDM2 amplification (9.7%, n = 41), CDK4 amplification (7.6%, n = 32), PDGFRA amplification (3.3%, n = 14), PPM1D amplification (3.1%, n = 13), KIT amplification (2.6%, n = 11), KDR amplification (2.6%), MDM4 amplification (2.1%, n = 9), MET amplification (1.9%, n = 8), and GLI2 amplification (0.23%, n = 1). RNA fusions based upon available reports were uncommon and included KIAA1549::BRAF (n = 5), FGFR1::TACC1 (n = 3), KANK2-NTRK2 (n = 1), and NCOA4::RET (n = 1). Of the cases on which we had primary-recurrent pairs, we were able to evaluate 2 patients who were referred through the NIH clinical methylation pipeline: 1 case had ATRX and NF1 alterations at initial surgery and upon recurrent surgery 2 years later retained the same alterations and had acquired CDKN2A/B homozygous loss; the second patient had CDKN2A/B homozygous loss on initial surgery but did not have CDK4 or MDM2 amplification on initial surgery, and then 4 months later at recurrence with repeat surgery was noted to have developed these 2 amplifications along with persistence of retained CDKN2A/B homozygous loss.

Clinical Characteristics

The median age of the cohort was 46 years (range 4-88). The age distribution is shown in Figure S1. Males were more frequent (n = 244, 58.0%) than females (Table 2). The median age at diagnosis was similar in males (median age 46 years) and females (median age 47 years). Data on primary vs recurrent status of tumor samples were available in 262 cases, and 179 (68.3%) were from primary tumors. NF1 syndrome status was known for 193 cases, and 46 (23.8%) cases arose in the setting of NF1 syndrome (Table 2). Tumor location data were available for 301 cases and 52.5% were located in the posterior fossa compared to those located in the supratentorial space or the spine (Table 2). There were 181 available initial MRI reports, and among them 93.4% (n = 169) were noted to show contrast enhancement. Obstructive hydrocephalus on imaging was noted in 23.2% (n = 42). Cystic features were found in 23.8% (n = 43), and 8.8% (n = 16) presented with hemorrhage. Leptomeningeal disease on MRI was noted in 5.5% (n = 10). Treatment data for each case as available are shown in Table S1, and 46.6% of patients received biopsy (102 out of 219 patients); 85.8% (127 out of 148) received radiation. Chemotherapy treatment (any, n = 111) mostly consisted of temozolomide (n = 103); however, some patients received lomustine, carboplatin, etoposide, bleomycin, irinotecan, carmustine, vincristine, and vinblastine at some point of the treatment course. Targeted therapy included selumetinib (n = 13), everolimus (n = 5), plixorafenib (n = 1), dabrafenib/trametinib (n = 4), ulixertinib (n = 1), mirdametinib (n = 1), binimetinib (n = 2), and encorafenib (n = 1) and patients may have received several targeted therapies. In addition, 5 patients were noted to have received immunotherapy (immune checkpoint inhibitors nivolumab or pembrolizumab were the only 2 reported treatments in these 5 patients).

Table 2.

Clinical characteristics

Characteristic		Total, n
Age, years (median, range)	46 (4-88)	288
Sex (n, %)		421
Male	244 (58.0)
Female	177 (42.0)
Neurofibromatosis type 1 syndrome status, known status (n, %)		193
Yes	46 (23.8)
No	147 (76.2)
Tumor location (n, %)		301
Posterior fossa	157 (52.2)
Spine	23 (7.6)
Supratentorial	121 (41.2)

Methylation Subgroups

Unsupervised methylation analysis identified 2 overall HGAP subgroups (Figure S2), although inspection shows subgroup overlap for some samples. Methylation subgroups differed in histologic high-grade features and copy-number alteration burden, including focal amplification events (Table S2).

Survival and Prognosis

The median progression-free survival (PFS) of our cohort was 21.0 months (n = 202) and median OS was 88.0 months (n = 204) (Figure S3). Volcano plots were constructed and stratified by good vs poor performing HGAPs (using the median survival time 88 months as the cutoff criteria as defined in the “Methods” section) and no markers were found to be significantly different between the 2 prognostic groups (Figure S4).

Histological and molecular factors

Histologic grade (high grade vs low grade), regardless of reviewer and via consensus grading, was not significantly associated with patient outcome (Table 3). The presence of an ATRX alteration (defined as a pathogenic mutation and/or demonstration of loss of expression in tumor cells) was associated with decreased PFS (P = .002, Figure 1A) and OS (P = .04, Figure 1B) on univariate log rank analysis. The median PFS of patients with an ATRX alteration (n = 86) was 13.0 months, while those without ATRX alteration (n = 49) had a median PFS of 34.0 months. The median OS of those with ATRX alteration (n = 83) was 53.8 months, while for those without ATRX alteration (n = 50) was not reached. On Cox-proportional hazards, the hazard ratio (HR) was 1.96 for the presence of ATRX alteration for PFS (P = .002), and the HR was 1.97 for the presence of ATRX alteration for OS (P = .03). Recognizing that ATRX is most commonly assessed by immunohistochemistry in routine practice, we conducted additional analyses using only the cases where an ATRX immunohistochemical result was available. For both OS (n = 116 cases) and PFS (n = 114 cases) loss of ATRX on immunohistochemistry was associated with poorer patient outcome (data not shown). MGMT promoter methylation was not significantly associated with outcome for either PFS or OS regardless of treatment status (Table 3, Figure S5), and there was no statistically significant interaction effect between MGMT promoter methylation and temozolomide (P = .24). The presence of any gene amplification among those queried trended toward worse outcome for PFS (P = .07, Table 3, Figure S6) but not OS, and no significant association was found for any of the queried genes individually on OS (Table 3). Methylome cluster assignment was not significantly associated with PFS or OS (Table 3, Figure S7). CDKN2A/B homozygous loss, MAPK alterations, BRAF alteration, NF1 alteration, and FGFR1 alteration were not prognostic (Table 3).

Table 3.

Prognostic factors

Characteristics	Median progression free survival (months), n	P-value	Median overall survival (months), n	P-value
Histological factors
Histological grade at initial diagnosis, neuropathologist reviewer 1		.6		.4
High grade	14.8, 69		25.2, 70
Low grade	19.0, 43		48.5, 41
Histological grade at initial diagnosis, neuropathologist reviewer 2		.3		.09
High grade	19.5, 60		48.5, 60
Low grade	12.0, 46		29.5, 46
Histological grade at initial diagnosis, consensus grading		.8		.8
High grade	21.0, 50		27.0, 50
Low grade	12.0, 33		40.3, 31
Mitoses presence, any		.9		.6
Yes	23.6, 110		93.7, 110
No	12.2, 25		79.2, 24
Necrosis presence		.8		.4
Yes	25, 38		93.7, 37
No	18, 92		79.2, 90
Microvascular proliferation		.4		.6
Yes	29.3, 80		Not reached, 80
No	13.9, 56		59.1, 53
Molecular factors
ATRX alteration		.002		.04
Yes	13.0, 86		53.8, 83
No	34.0, 49		Not reached, 50
ATRX loss on immunohistochemistry only		.004		.03
Yes	13.9, 73		48.5, 70
No	35.5, 43		Not reached, 44
BRAF alteration		.2		.2
Yes	63.5,   6		54, 6
No	18.0, 62		Not reached, 62
NF1 alteration		.9		.9
Yes	19, 48		54, 50
No	29, 20		Not reached, 19
FGFR1 alteration		.1		.2
Yes	67.2, 14		Not reached, 14
No	19.0, 49		53.8, 49
MAPK alteration (any)		0.6		.6
Yes	23.6, 68		67.2, 69
No	21.0, 134		112.5, 135
CDKN2A/B loss		.2		.8
Yes	19, 162		88.0, 163
No	34, 40		79.2, 41
MGMT promoter status		.6		.2
Methylated	21.0, 106		59.1, 109
Not methylated	21.0, 96		112.5, 95
Gene amplification, any		.07		.1
Yes	20.3, 38		26.0, 41
No	21.0, 164		93.7, 163
CDK4 amplification		.05		.15
Yes	84.6, 17		Not reached, 16
No	20.3, 185		71.1, 188
MDM2 amplification		.3		.8
Yes	21.0, 19		Not reached, 19
No	21.0, 183		79.2, 185
CDK4 or MDM2 amplification		.1		.5
Yes	21.0, 21		Not reached, 21
No	21.0, 181		79.2, 183
PDGFRA amplification		.4		.4
Yes	14.8, 7		25.2, 7
No	21.0, 195		88.0, 197
KIT amplification		.2		.9
Yes	14.8, 6		25.2, 6
No	21.0, 195		88.0, 198
KDR amplification		.2		.9
Yes	14.8, 6		25.2, 6
No	21.0, 196		88.0, 198
MET amplification		.9		.1
Yes	7.6, 3		24.5, 4
No	21.0, 199		88.0, 200
Methylome cluster		.7		.13
Cluster 1	19.0, 119		288.2, 118
Cluster 2	25.0, 83		59.1, 86
Clinical factors
Age		.4		.01
Above 46 years	18.9, 98		32.6, 100
Below 46 years	23.6, 103		240.1, 103
Sex		.2		.5
Females	14.8, 81		64, 81
Males	23.6, 121		130, 123
NF1 syndrome status		.2		1
Yes	12.6, 38		54.0, 37
No	23.6, 99		88.0, 100
Tumor location		.6		.7
Posterior fossa	20.3, 104		67.2, 106
Spine	19.0, 18		240.1, 18
Supratentorial	24.0, 71		79.2, 69
Tumor characteristics (MRI)—obstructive hydrocephalus		.2		.2
Yes	18.9, 29		67.2, 29
No	19.0, 80		53.8, 79
Tumor characteristics (MRI)—hemorrhage		.9		.9
Yes	11.1, 15		93.7, 15
No	18.9, 89		48.5, 88
Tumor characteristics (MRI)—Cystic		.6		.02
Yes	23.6, 33		Not reached, 32
No	19.0, 78		53.8, 78
Treatment factors
Extent of resection		.9		.1
Biopsy	13.9, 67		240.1, 67
Resection (total or subtotal)	23.6, 93		93.7, 90
Radiation		.3		.2
Received, any	19.5, 119		67.2, 122
Did not receive	12.0, 19		79.2, 20
Chemotherapy		.5		1
Received, any	19.0, 106		59.1, 106
Did not receive	22.0, 35		88.0, 35
Temozolomide		.8		.7
Received, any	19.5, 98		64.0, 99
Did not receive	22.0, 33		79.2, 33
Upfront chemoradiation		.8		.3
Yes	19.0, 38		40.3, 38
No	11.0, 28		Not reached, 27
Adjuvant temozolomide		.06		.9
Yes	21.0, 30		48.5, 31
No	8.0, 38		59.1, 37
Targeted therapya		.7		.07
Received, any	24.0, 31		93.7, 32
Did not receive	13.0, 93		59.1, 92
Bevacizumab		.5		.1
Received, any	23.6, 31		Not reached, 31
Did not receive	13.0, 94		59.1, 93
Immunotherapy (immune checkpoint inhibitor)		.1		.02
Received, any	4.8, 4		12.0, 4
Did not receive	19.0, 119		79.2, 118

^aKnown targeted therapy: selumetinib (n = 13), everolimus (n = 5), plixorafenib (n = 1), dabrafenib/trametinib (n = 4), ulixertinib (n = 1), mirdametinib (n = 1), binimetinib (n = 2), and encorafenib (n = 1) and patients may have received several targeted therapies.

Figure 1.

Prognostic factors. (A) ATRX alteration Kaplan-Meier curve for progression-free survival. (B) ATRX alteration Kaplan-Meier curve for overall survival. (C) Age grouping Kaplan-Meier curve for progression-free survival. (D) Age grouping Kaplan-Meier curve for overall Survival. (E) MRI cystic features Kaplan-Meier curve for progression-free survival. (F) MRI cystic features Kaplan-Meier curve for overall survival. “+” is censored.

Clinical factors

Increased age was found to be a poor prognostic factor for OS (P = .007; Figure 1D) but was not associated with PFS (Figure 1C). Sex and NF1 syndrome status were not statistically significant factors on survival (Table 3). Tumor location (cerebral vs posterior fossa vs spinal location) was not significantly associated with survival (Table 3).

The presence of cystic features on initial MRI was found to be a favorable prognostic factor for OS (P = .02; Figure 1E), although it was not associated with PFS (Figure 1F). Overall survival was not reached for those with cystic features on MRI, while those without cystic features had a median OS of 53.8 months. The hazard ratio on univariate Cox-proportional hazards on OS was HR 0.37 (P = .01). Additional imaging characteristics, including the presence of obstructive hydrocephalus or hemorrhage, did not correlate with PFS or OS on univariate analysis (Table 1).

Extent of resection (biopsy vs resection, available for 219 patients) did not correlate with PFS (P = .9) or OS (P = .1) (Table 3). Receipt of targeted therapy did not correlate with PFS (P = .7), however, trended toward improved OS (P = .07; Table 3, Figure S8). Adjuvant temozolomide treatment trended toward improved PFS (P = .06; Table 3); however, it did not correlate with OS (P = .9). Immunotherapy (immune checkpoint inhibitors) administration did not correlate with PFS but was significantly associated with OS (P = .02; Table 3).

Multivariate analysis, including the variables found to be significant on univariate analyses using a Cox-proportional hazards model demonstrated that patient age (P = .006), the presence of ATRX alteration (P = .003), and the presence of cystic features on MRI (P = .035) remained significant and were independent predictors of OS (Table 4). ATRX alteration remained a poor predictor on multivariable analysis for PFS (Table S3).

Table 4.

Multivariate Cox-proportional hazards for overall survival

Variable	Hazard ratio (HR)	95% CI for HR	P-value
Age (continuous)	1.03	1.01-1.05	.006
ATRX alteration (present)	4.05	1.63-10.04	.003
Cystic features (present)	0.32	0.11-0.92	.035
Concordance = 0.72 (SE = 0.044) Likelihood ratio test = 20.04 (df = 3, P = 2e-04) Wald test = 16.16 (df = 3, P = .001) Score (log-rank) test = 16.56 (df = 3, P = 9e-04)

Correlates of ATRX Alteration Status

Given the association of ATRX alteration with prognosis, we examined its correlation with clinical and genomic factors in our cohort. Thirty-nine patients were known to have NF1 syndrome, and 34/39 (87%) of these cases showed an ATRX alteration. By comparison, of the 135 nonsyndromic cases, slightly more than one-half (74/135, 55%) had an ATRX alteration, revealing ATRX alteration positively correlated with patients with NF1 syndrome (chi-square test P = .0005; Table S4). ATRX alteration status also inversely correlated with the presence of amplification of the genes interrogated—those without gene amplification more often had ATRX alterations (P = .0436; Table S5). Further, patients younger than 46 years (median age of the total cohort) had a higher rate of ATRX alteration (P = .0042; Table S6), and tumors located in the posterior fossa had a higher rate of ATRX alterations (P = .0313; Table S7). The presence of an ATRX alteration additionally correlated with CDKN2A/B loss (P = .0028; Table S8). ATRX status did not significantly correlate with MGMT promoter methylation status, histological grade, or the presence of cystic features on imaging.

HGAP Arising in Neurofibromatosis Type 1 Syndrome

In patients with HGAP arising in NF1 syndrome, age was younger at diagnosis, with a median age of 34.5 years compared to those with sporadic HGAP, where median age was 50.0 years (P = .001). Those with reported NF1 syndrome were less often treated with temozolomide (P = .009) and more often treated with targeted therapy (P < .001; Table S9). Other molecular and clinical characteristics are detailed in Table S9.

Histological Interpretations of HGAP

Currently the WHO guidelines require DNA methylation profiling to reach an HGAP diagnosis. In that context, it is of interest to summarize the initial histologic interpretations of known HGAP cases. Histological interpretations were culled from outside pathology reports of cases sent to the NCI for methylation profiling (n = 230; Table 5). Many of the terms were “descriptive” of broad tumor categories for example “Glioma” or “Neuroepithelial tumor”, while others were specific (“glioblastoma” and “pilocytic astrocytoma”). High-grade astrocytoma with piloid features was suspected in a subset of the cases. Overall, the range of histologic interpretations demonstrate the wide variety of histologic appearance in tumors found to represent HGAPs based on methylation profiling.

Table 5.

Initial outside diagnosis on formal pathology report before DNA methylation diagnosis for 230 cases in the cohort

Outside initial diagnosis	n (%)
High-grade glioma	75 (32.6)
Glioma	64 (27.8)
Astrocytic tumor	29 (12.6)
HGAP, suspected	25 (10.9)
Pilocytic astrocytoma	12 (5.2)
Glioblastoma	11 (4.8)
Neuroepithelial tumor	7 (3.0)
Pleomorphic xanthoastrocytoma	4 (1.7)
Glioneuronal tumor	3 (1.3)

Discussion

Here, we interrogated a cohort of 421 unique HGAP cases to characterize the histologic, molecular, clinical, and treatment landscape and identify prognostic factors. To our knowledge, this is the largest HGAP cohort assembled to date for this relatively newly defined tumor type.

Histologically, there has been undefined characterization of the exact grade of this tumor. Using “high-grade” (histologic grade 3 or 4) vs “low-grade” (histologic grade 2) determinations, we found that most HGAPs appeared as histologically high-grade gliomas. However, in this cohort, we could not identify prognostic significance of histologic grade. Notably there was a wide range of histologic interpretations for HGAP (Table 5). However, we recognize DNA methylation is not widely available at this moment in time. Based upon our study results, those with IDH-wildtype tumor with ATRX alteration, CDKN2A/B deletion and posterior fossa location are features that might be suggestive of an HGAP. We did not find any patients with initial methylation-profiling proven pilocytic astrocytomas confirmed by the NIH/Bethesda v3 classifier who later recurred with tissue again analyzed by the NIH/Bethesda v3 that resulted in HGAP at recurrence. However, we cannot rule out the possibility that some HGAPs can evolve from pilocytic astrocytomas, since many pilocytic astrocytomas are not submitted routinely for methylation profiling analyses and therefore a possible selection bias may have contributed to our findings.

Molecularly, we found a high rate of ATRX alterations, CDKN2A/B loss, MAPK pathway alterations, NF1 alterations, and presence of MGMT promoter methylation, as confirmed prior in the literature.^1,3 The frequency of ATRX alterations in our large cohort in comparison to prior large cohort studies was slightly higher than that of prior large cohort studies (45%-59%)^1,3 though MAPK alterations and CDKN2A/B frequency was similar, around 75% and 80%^1,3 and MGMT promoter methylation frequency only slightly higher at 53.4% compared to the only other large cohort study noting this to be 45%.¹ Importantly, the presence of an ATRX alteration was found to be a poor prognostic factor of survival (P = .04), while CDKN2A/B loss and MGMT promoter methylation were not significantly associated with outcome in our cohort. While MGMT promoter methylation is a well-known prognostic/predictive marker in glioblastoma, IDH-wildtype, it is noted that most glioblastomas show loss of chromosome 10 (where the MGMT gene is located), which is not observed in the majority of HGAP. The specific reasons why MGMT promoter methylation is prognostic in glioblastoma but not in HGAP remain elusive. Although methylome subgroup assignment was not prognostic for survival in this cohort, these findings support the presence of epigenetic heterogeneity within HGAP that tracks with NF1 status, anatomic site, and copy-number alteration/amplification burden (Table S2).

Clinically, we found the median OS of this cohort to be 88 months, which corroborates other reports suggesting this behaves similarly to a grade 3 glioma.^1,2 Prior large cohort studies found the survival of HGAP to have a median survival comparable to IDH-mutant astrocytoma, grade 3 or 4, and have compared survival with glioblastoma and pilocytic astrocytomas, but have not been able to provide median OS estimates.^1,3 Importantly, we found that older age is a significant prognostic factor, which has not previously been identified in the literature. We confirmed there is a slight prevalence for this tumor in male patients, similar to previous studies.^1,3 While there continues to be a predilection for posterior fossa location, supratentorial HGAPs (41% in our cohort) were higher in frequency than previously reported.^1,3 We also found that about a quarter of HGAPs in our series arose in the setting of NF1 syndrome, and there was no statistically significant difference in survival of those compared to sporadic HGAPs.

To our knowledge, there has also been sparse literature on MRI features of HGAP with limited case reports and single/dual institutional studies.^4,11,16 The frequency and rate of imaging features and their significance have been difficult to characterize. Here, we found a higher-than-expected reported rate of hemorrhage (9%) and cystic features (23%), and the latter also being positively prognostic (P = .01). Importantly, the rate of leptomeningeal disease at presentation (5.5%) has also been described for the first time in a large cohort analysis, and this reflects relatively similar rates compared to IDH-wildtype glioblastoma (10%-11%).³² Although leptomeningeal disease is a well-known poor prognostic factor in gliomas,³² this was not found to be a statistically significant prognostic factor in our cohort, likely owing to small numbers of patients reported with leptomeningeal disease (n = 10). Leptomeningeal disease likely cannot be discounted as a poor prognostic factor in this entity based on what we know of other glial types.

Lastly, this is the first study to our knowledge to report large cohort treatment data in HGAP, though numbers still remain limited. Extent of resection (biopsy or resection) was not associated with survival in HGAPs and interestingly is inconsistent with prior studies that have identified gross total resection as a favorable prognostic factor in glial neoplasm—which has largely guided decisions around upfront surgical management of tumors. While this may be an exception in HGAPs, we need prospective studies to validate this signal, and to better parse out gross total resection as a specific factor. Outside of surgery, most patients received radiation (85%), chemotherapy (75%), and about half received adjuvant therapy (45%). About a quarter of patients received targeted therapy (25%) or bevacizumab (24%). Targeted therapy trended toward a positive effect on OS (P = .07) and immunotherapy was significantly associated with a poor prognosis (P = .02) in a very small subset (n = 5). However, in retrospective series, it is difficult to separate confounding by severity, the propensity to provide treatment given to the tumor based on the suspected natural history/clinical status of the patient; immortal time bias, the ability of people who live longer to have greater chance to be exposed to any given treatment; or true response. Other treatments did not affect PFS or OS, which may reflect the limitations of a retrospective study with decentralized radiographic/clinical response metrics and limited numbers. Therefore, prospective studies to better characterize treatment response and validate preliminary signals in HGAPs are imperative and urgently needed.

A clinical question specifically for HGAP treatment often faced by our neuro-oncologists is the utility of temozolomide in context of optimizing disease-specific therapy for HGAP and utility of further temozolomide for those with known MGMT statuses. Currently, clinicians often use MGMT statuses as in glioblastoma as a surrogate and extrapolate to determine how much temozolomide patients with other gliomas, including HGAP, should receive. However, there is a lack of definitive, interrogated evidence that MGMT outside of glioblastoma is clinically relevant in other IDH-wildtype high-grade gliomas as a marker of response to therapy. While retrospective series are certainly difficult to guide clinical treatment and prospective clinical trials are urgently needed in this entity—in this context, we did not find evidence that MGMT methylation is an HGAP-specific predictive marker for prolonged temozolomide treatment in this large cohort study. Clinically, this finding may be useful at junctures where discussions relate to whether further therapy is needed, in particular for patients who are highly vulnerable (poor performance status, older age) to side effects secondary to aggressive treatment and may offer some evidence to safely avoid overtreatment.

Limitations of this study include its retrospective nature and data availability. Clinical variables were unavailable for a subset of cases (particularly those derived from public datasets), which may introduce selection bias. This is important to keep in mind when interpreting prognostic factors identified and also factors that were not found to be statistically prognostic, such as sex and NF1 syndrome status. Prognostic factors predictive of OS did not always track with PFS, likely due to the difficulty with standard assessment of response across the different institutions and the biases inherent in retrospective nature. Nevertheless, one of the goals in this study was to retrospectively identify signals that might be interrogated and validated in future prospective studies in an entity that has largely relied on case reports and a few single institution studies, with few large cohort studies currently available.

In summary, we identified older age as well as presence of ATRX alteration as poor prognostic factors in HGAP, while cystic features on MRI were found to be favorably prognostic. We hope these prognostic factors will aid in treatment stratification while future prospective studies may confirm or refine our findings. We also were able to refine and characterize the molecular and clinical features of HGAP in a relatively large cohort for this uncommon but important tumor entity. More prospective studies are urgently needed, and while colleagues have begun to actively open trials that include HGAPs (NCT05503797; NCT06413706), these tumors are primarily included as part of basket trials or NF1 studies, and, to the best of our knowledge, there are no dedicated natural history studies or umbrella trials at this point in time. Further, investigation into pathways and drivers in this rare entity (including in sporadic HGAPs and of those arising in NF1 syndrome) are necessary, so future treatments may achieve the best chance of success in the context of strong biological understanding of this entity with sparse patient data. Finally, clear characterization, diagnosis, and treatment response of emerging tumors such as HGAPs have implications in current ongoing glioma trials and studies—this tumor may share histological and molecular features from pilocytic astrocytomas and glioblastomas to pleomorphic xanthoastrocytomas and other neuroepithelial tumors (Table 5), but the clinical behavior is distinct from these tumor types. It is imperative that samples with such initial diagnoses undergo DNA methylation for not only clinical diagnosis but also for purity of clinical trial eligibility and appropriate tissue sample identification in translational science. We hope this work will help lay the foundation for future prospective and translational studies in an effort to define and standardize treatment regimens and work toward better outcomes for these patients.

Supplementary Material

Supplementary material is available online at Neuro-Oncology (https://academic.oup.com/neuro-oncology).

Author Contributions

All authors listed in the manuscript contributed significantly to this manuscript. K.A. and M.Q. performed central histopathology review of available NIH-referred cases. All authors were also involved in the writing of the manuscript and have read and approved the manuscript.

Conflict of Interest Statement

K.A. reports serving on the Advisory Board for Servier. R.V.L. reports research support from BMS (drug only), serving on the Data Safety Monitoring Board of GT Medical Technologies, speaker’s bureau for Merck, Novocure, Servier, and consultant for Cardinal Health, Catalyst Pharmaceuticals, GT Medical Technologies, Merck, Novocure, Servier, Telix, and honoraria for editing for EBSCO, Elsevier, Medlink Neurology.

Funding

Open Access funding is provided by the National Institutes of Health (NIH). This research was supported (in part) by the Intramural Research Program of the NIH. The contributions of the NIH authors were made as part of their official duties as NIH federal employees, are in compliance with agency policy requirements, and are considered Works of the United States Government. However, the findings and conclusions presented in this paper are those of the authors and do not necessarily reflect the views of the NIH or the U.S. Department of Health and Human Services.

Ethics Statement

The NIH Institutional Review Board approved this study.

Data Availability

Data will be made available upon reasonable request.