Prognostic value of DNA methylation subclassification, aneuploidy, and CDKN2A/B homozygous deletion in predicting clinical outcome of IDH mutant astrocytomas

Abstract

Diffuse gliomas account for 18.8% of all central nervous system (CNS) tumors and of these, IDH1/2 mutant astrocytomas account for 76.8% of CNS World Health Organization (WHO) grade 2, 87.5% of grade 3, and 2.4% of grade 4 astrocytomas.¹ As recommended by the National Comprehensive Cancer Network (NCCN) guidelines, treatment recommendations and the decision between observation versus radiation and adjuvant chemotherapy, are guided based on the histologic grade as defined by the WHO criteria.^2–4 According to the 2021 WHO classification of CNS tumors, 5th edition for isocitrate dehydrogenase (IDH) mutant astrocytoma, WHO grade 2 tumors are defined by infiltration without mitotic activity, necrosis, or microvascular proliferation; WHO grade 3 tumors are defined by infiltration with significant mitotic activity, with no specified threshold, and without necrosis or microvascular proliferation; and WHO grade 4 tumors are defined by the presence of either necrosis or microvascular proliferation.⁵ Multiple studies have shown that histological grading criteria alone show limited prognostic value in IDH mutant astrocytoma.^6–8 Further studies failed to determine a definitive threshold for mitotic activity or Ki-67 proliferation index that would correlate with survival in patients with IDH mutant astrocytoma.^6,7

Previous studies have focused on predictive markers that would correlate well with prognosis of IDH1/2 mutant astrocytomas, including the extent of surgical resection, ploidy, and CDKN2A/B homozygous deletion status.^8–10 The extent of surgical resection has been shown to be an important independent predictor of prognosis in IDH1/2 mutant tumors.^11,12 The presence of the homozygous deletion of CDKN2A/B was shown to be highly associated with poor prognosis and; therefore, included in the 2021 CNS WHO as a molecular biomarker of grade 4.^8,9 There are additional molecular alterations associated with a worse prognosis, including RB1 mutation and CDK4 amplification, suggesting that other molecular alterations may play a role in aggressive behavior.⁵ DNA methylation profiling has emerged as a powerful diagnostic tool in CNS tumors, with some entities benefiting from additional prognostic information, that is, ependymoma and medulloblastoma.^13,14 DNA methylation separates IDH mutant astrocytomas into 2 epigenetically distinct groups, designated astrocytoma IDH lower grade (A_IDH_LG) and astrocytoma IDH high grade (A_IDH_HG).¹⁵ A_IDH_LG class is composed of astrocytic neoplasms with IDH1 or IDH2 mutations that are histologically WHO grade 2 or grade 3 and do not have codeletion of 1p and 19q. A_IDH_HG class contains primarily histologic WHO grade 4 infiltrating astrocytic tumors with IDH1 or IDH2 mutations, and lack codeletion of 1p and 19q, but with numerous and complex chromosomal alterations and copy number changes.

Given the poor prognostic value of histology, we sought to compare the prognostic significance of DNA methylation, CDKN2A/B homozygous deletion, and ploidy on survival in IDH mutant astrocytoma.

Methods

Study Design and Sample Selection

We performed a retrospective analysis of all tumors with a clinical and molecular diagnosis of IDH1/2 mutant astrocytoma diagnosed between the years of 2014 and 2022 at NYU Langone Health. All 98 tumors with complete clinical, and pathological data were included, and were profiled by whole genome DNA methylation profiling and the DNA methylation class retrieved from electronic medical records. Histologic diagnoses were updated according to the 5th edition of the CNS WHO. Institutional review board approval was obtained (IRB# S14-00948). Independent validation cohorts were obtained from The Cancer Genome Atlas (TCGA, N = 231) and Heidelberg (N = 77) and analyzed using the same criteria.

DNA Methylation and Mutational Analysis

DNA was extracted from archival formalin-fixed paraffin-embedded tissue using the Maxwell Promega. Clinically validated DNA methylation profiling was performed at the NYU Department of Molecular Pathology CLIA certified laboratory using the Illumina 450k or EPIC array as described previously¹⁶ and analyzed using the clinically validated DNA methylation classifier.¹⁵ Copy numbers were analyzed using the “conumee” R package in bioconductor as described previously.^15,17 Mutational analyses to confirm mutational status were performed by the clinically validated NYU Oncomine Focus panel or NYU Langone Genome PACT, a 510(k) FDA cleared (K202304) matched tumor-normal 607 gene panel. CDKN2A/B homozygous deletion and ploidy analysis was performed by visual assessment of the copy number plots from each case derived from DNA methylation profiles using previously described criteria (Figure 1)^8,17,18 to account for tumor cell fraction, which may dilute the copy number signal. In brief, copy number plots derived from the DNA methylation were considered complex if 3 or more chromosomal arms showed copy number gains or losses, regardless of the size. To avoid interobserver variability, the primary and both validation data sets were quantified by the same observer (K.G.).

Figure 1.

Copy number plots. We analyzed ploidy using the copy number plots from the DNA methylation analysis using previously published methods.^8,17,18 Cases were stratified into complex copy number plot, notably without CDKN2A/B homozygous deletion (A) and simple copy number plot (B). CDKN2A/B homozygous deletion (arrow) status on chromosome 9p was also analyzed via the copy number plot (C).

Open in new tabDownload slide

Radiology Review of Extent of Resection

Extent of tumor resection was performed by analysis of postoperative and preoperative imaging by a board-certified neuroradiologist who was blinded to tumor classification and survival data. All cases were reviewed utilizing standard clinical MRI examinations at 1.5 or 3.0 Tesla utilizing various scanner parameters. Exams were reviewed utilizing FLAIR (3D when available). The earliest available postoperative MRI was utilized for assessment. One examination was only available for review with postoperative CT scans, but there was clear residual tumor on that examination. Two patients did not have preoperative imaging available for review. Equivocal cases were treated conservatively and considered subtotal resections, as were biopsy-only cases. Cases were bifurcated in a binary fashion as gross total resection (GTR, 100% resection) or subtotal resection (STR, <100% resection) utilizing the standard of clear residual infiltrating signal beyond the resection margins where there was preoperative signal abnormality.¹⁹ As many cases did not demonstrate contrast enhancement but all cases demonstrated preoperative FLAIR signal abnormality at or beyond the margins of enhancement, FLAIR was utilized to uniformly categorize the extent of resection as GTR or STR.

Statistics

Kaplan-Meier analysis for progression free survival (PFS) and overall survival (OS) were performed using GraphPad Prism software (Version 9.4.1). Univariate Cox regressions were conducted to assess the impact of each covariate on OS and PFS. The covariates included 5 major factors: DNA methylation subgroup (A_IDH_LG, A_IDH_HG), WHO grade (grades 2, 3, and 4), CDKN2A/B status (deleted, retained), copy number plot (complex, simple), and extent of resection (subtotal resection, gross total resection). WHO grade is one-hot encoded in the regression. The likelihood ratio test was performed to test the overall significance of WHO grade. For other factors, we used the Wald Chi-squared test to test the statistical significance.^20,21

Results

Cohort Description

We performed a retrospective analysis of 98 clinically diagnosed IDH mutant astrocytomas that were profiled by DNA methylation. The patient’s age ranged from 11 to 80 years with a median age of 38 years. The cohort had a male-to-female ratio of 1.2:1 (Table 1). Our cohort included 39 WHO grade 2 (40%), 31 WHO grade 3 (32%), and 28 WHO grade 4 (28%) IDH mutant astrocytomas (Table 1, Figure 2A). A GTR was achieved in 41 cases (42.8%) and a STR in 56 cases (57.2%). Extent of resection could not be determined in one case. The majority of patients received chemotherapy and/or radiation (N = 80, 81%; Table 1). The extent of resection did not have a statistically significant impact on overall or PFS.

Figure 2.

Cohort stratified by biomarkers. (A) Histology and (B) DNA methylation classification into clinically validated low- (A_IDH_LG) and high-grade (A_IDH_HG) subgroups. (C) Association of WHO grade and DNA methylation class. (D) WHO grade 2 tumors and the A_IDH_LG methylation class were more likely to have simple copy number plots (85% and 74%, respectively). The WHO grade 4 tumors and the A_IDH_HG methylation class were more likely to have complex copy number structure (86% and 85%, respectively). The WHO grade 3 tumors were evenly split between simple (45%) and complex copy number plots (55%). (E) The majority of WHO grade 2 tumors and the A_IDH_LG methylation class showed maintenance of CDKN2A/B (85% and 79%, respectively). Similar to the DNA methylation classification, WHO grade 3 tumors were almost evenly separated into a CDKN2A/B retained subgroup (42%) and a CDKN2A/B homozygous deletion subgroup (58%). WHO grade 4 tumors did not show enrichment for the CDKN2A/B retained (46%) or deleted categories (54%). The majority of the A_IDH_HG methylation class had CDKN2A/B homozygous deletion (68%), while 32% maintained CDKN2A/B. (F) In our cohort, only 16% of cases with simple copy number plots had isolated homozygous deletion of CDKN2A/B, while 63% of cases with complex copy number had CDKN2A/B deletion (63%).

Open in new tabDownload slide

All IDH mutant astrocytomas were subclassified by DNA methylation into 2 previously established subgroups, A_IDH_LG (N = 58, 59%) and A_IDH_HG (N = 40, 41%) using the DNA methylation classifier as described previously (Figure 2B).¹⁵ Homozygous deletion of CDKN2A/B was observed in 39 cases (40%) and retention in 59 cases (60%). Complex copy number plots were observed in 49 cases (50%) and simple copy number plots were observed in 49 cases (50%; Table 1).

Association of Histological Grade and Molecular Features

When comparing WHO histological grade and DNA methylation categories, most WHO grade 2 tumors classified with the lower grade methylation class, A_IDH_LG (N = 36, 92%) and most grade 4 tumors classified with the higher grade methylation class, A_IDH_HG (N = 22, 79%). Strikingly, the WHO grade 3 tumors evenly divided into the 2 DNA methylation subgroups, A_IDH_LG (N = 16, 52%) and A_IDH_HG (N = 15, 48%; Figure 2C).

The majority of WHO grade 2 tumors showed maintenance of CDKN2A/B (N = 33, 85%) and as was seen by DNA methylation, WHO grade 3 tumors were separated into a CDKN2A/B retained subgroup (N = 13, 42%) and a CDKN2A/B homozygous deletion subgroup (N = 18, 58%). Interestingly, the WHO grade 4 tumors did not show enrichment for the CDKN2A/B retained (N = 13, 46%) or deleted categories (N = 15, 54%), suggesting that multiple molecular mechanisms beyond CDKN2A/B loss contribute to WHO grade 4 features.

Comparison Between DNA Methylation Classes and CDKN2A/B Status

When comparing DNA methylation data and copy number data, most A_IDH_LG methylation class tumors showed retained CDKN2A/B (N = 46, 81%). However, A_IDH_HG methylation class tumors showed enrichment for homozygous deletion of CDKN2A/B (N = 27, 70%; Figure 2E). CDKN2A/B homozygous deletion has been shown previously to portend a worse prognosis, and our data shows that CDKN2A/B homozygous deletion was enriched in the A_IDH_HG methylation class. Nevertheless, several cases within the A_IDH_HG cohort retained the CDKN2A/B locus (N = 13, 33%), and several tumors in the A_IDH_LG cohort carried a CDKN2A/B homozygous deletion (N = 12, 21%).

Comparison Between DNA Methylation and Ploidy

Previous studies have indicated that chromosomal instability is associated with aggressive disease.²² In our cohort, the WHO grade 2 tumors and the A_IDH_LG methylation class were more likely to have simple copy number structure (N = 31, 79% and N = 43, 74%), a finding consistent with the description of the A_IDH_LG class in the DKFZ classifier.^15,17 The WHO grade 4 tumors and the A_IDH_HG methylation class were more likely to have complex copy number structure (N = 24, 86% and N = 34, 85%). However, the WHO grade 3 tumors were evenly split between simple (N = 14, 45%) and complex copy number structure (N = 17, 55%; Figure 2D).

While both CDKN2A/B homozygous deletion and complex copy number plot have been shown to portend a worse prognosis, our data show they do not necessarily occur together. In our cohort, 8 (16%) cases with simple copy number plots had isolated homozygous deletion of CDKN2A/B and 18 cases (37%) with complex copy number plots retained CDKN2A/B (Figure 2F)

Biomarkers and Survival Analysis

Survival analysis showed that there was no significant difference in OS in IDH1/2 mutant astrocytoma when stratified by WHO grade alone (P value = .0699; Figure 3A) or by copy number complexity (P value = .0588; Figure 3D). OS was significantly different when patients were stratified either by CDKN2A/B homozygous deletion or by DNA methylation subclass (P value = .0286 and .0016, respectively). None of the molecular biomarkers were associated with significantly better PFS, although DNA methylation classification showed a trend (P value = .0534; Figure 3B and C).

Figure 3.

Survival data. Survival analysis showed that there was no significant difference in overall survival (OS) in IDH1/2 mutant astrocytoma when stratified by WHO grade alone (P value = .0699) (A) or by copy number complexity (P value = .0588) (D). OS was significantly different when patients were stratified either by CDKN2A/B homozygous deletion or by DNA methylation subclass (P value = .0286 and .0016, respectively) (B and C). None of the molecular biomarkers were associated with significantly better progression free survival (PFS), although DNA methylation classification showed a trend to significance (P value = .0534).

Open in new tabDownload slide

Since WHO grade 3 astrocytomas were almost evenly separated into 2 groups when using DNA methylation subclass, CDKN2A/B homozygous deletion, or copy number, we hypothesized that more accurate assessment of grade 3 tumors by molecular techniques could lead to improved diagnosis and prognosis. To evaluate this, we stratified WHO grade 3 astrocytomas using DNA methylation subclass, CDKN2A/B status, and copy number complexity and performed survival analysis. DNA methylation was a significant predictor of OS (P value = .0204) in grade 3 tumors but not PFS (P value = .1746; Figure 4A). There was no statistically significant difference between OS or PFS in WHO grade 3 tumors that were stratified by CDKN2A/B homozygous deletion (P value = .3515 and .0954, respectively) or copy number complexity alone (P value = .2039 and .2465, respectively; Figure 4B and C).

Figure 4.

IDH mutant WHO grade 3 astrocytoma overall and progression free survival (PFS) stratified by methylation subclass (A), CDKN2A homozygous deletion (B), and copy number (C). Overall survival (OS) was significantly different only using DNA methylation as a marker (P value = .0204), while PFS was not. There was no statistically significant difference between OS and PFS in WHO grade 3 tumors that were stratified by CDKN2A/B homozygous deletion (P value = .3515 and .0954, respectively) or copy number complexity alone (P value = .2039 and .2465, respectively).

Open in new tabDownload slide

Multivariate Analysis

Cox proportional hazards model showed that DNA methylation was the only significant predictor of OS (P value = .01) and near-significant predictor of PFS (P value = .05). WHO grade 3 was significant for OS (P value = .04) but not PFS (.69) and WHO grade 4 was borderline significant for PFS (P value = .05) but not OS (P value = .08). CDKN2A/B status (P value: OS = .13, PFS = .17) and copy number complexity (P value: OS = .11, PFS = .06) were not statistically significant for OS or PFS (Table 2).

Validation Cohorts

The TCGA data set included 231 IDH mutant astrocytomas, of which 117 were WHO grade 2 (51%), 105 were WHO grade 3 (46%), and 8 were WHO grade 4 (3%; Supplementary Figure 1A). DNA methylation subclassified these tumors into A_IDH_LG (N = 198, 86%) and A_IDH_HG (N = 33, 14%; Supplementary Figure 1B). Homozygous deletion of CDKN2A/B was detected in 22% of cases (N = 50) and 80% of tumors showed complex copy number profile (N = 184), while 20% of tumors were simple (N = 47). Similar to the NYU cohort, survival analysis showed that there was no significant difference in OS in IDH mutant astrocytoma when stratified by WHO grade alone (P value = .0572) or by copy number complexity (P value = .0510). However, OS was significantly different when IDH mutant astrocytomas were stratified either by DNA methylation (P value = .0001) and CDKN2A/B homozygous deletion (P value = .0097; Supplementary Figure 2).

We then performed survival analysis on histologically diagnosed WHO grade 3 IDH mutant astrocytomas. We found that OS was statistically significant when WHO grade 3 tumors using DNA methylation subclassification (P value = .0102), and borderline significant using copy number complexity (P value = .0490) but was not statistically significant for CDKN2A/B homozygous deletion (P value = .0755; Supplementary Figure 3).

The Heidelberg validation cohort data set included 77 IDH mutant astrocytomas, of which 29 were WHO grade (38%), 30 were WHO grade 3 (N = 39%), and 18 were WHO grade 4 (N = 23%; Supplementary Figure 4A). DNA methylation subclassified these tumors into A_IDH_LG (N = 54, 70%) and A_IDH_HG (N = 23, 30%; Supplementary Figure 4B). Homozygous deletion of CDKN2A/B was present in 26% of cases (N = 20), 77% of tumors showed complex (N = 59), and 23% showed simple copy number profiles (N = 18). Survival analysis showed that OS was significantly associated with WHO grade (P value = .0033), DNA methylation (P value = .0025), and homozygous deletion of CDKN2A/B (P value = .0476). OS was not significantly associated with copy number complexity (P value = .8397; Supplementary Figure 5).

Survival analysis on histologically diagnosed WHO grade 3 IDH mutant astrocytomas stratified by DNA methylation, copy number complexity, and CDKN2A/B homozygous deletion showed no statistically significant difference in OS using DNA methylation (P value .6002), CDKN2A/B homozygous deletion (P value = .3113), or copy number complexity (P value = .1845; Supplementary Figure 6).

Discussion

Grading of astrocytic gliomas has been a topic of research and discussion since the late 1940s.²³ It evolved in the pre-IDH era to its current scheme largely based on histologic criteria.⁵ The IDH1/2 gene mutations were found to be present in a distinct subgroup of glioma without 1p/19q codeletion and with ATRX gene and TP53 gene mutations that had a better prognosis than their IDH1/2 wild-type counterparts.^24,25IDH1/2 mutation-based classification was incorporated into the WHO CNS tumors classification in 2016, and since then, multiple studies have shown that the WHO grading system for these tumors could be improved using molecular biomarkers.^6–8,25

Many studies have tried to elucidate the biomarkers that could improve the current grading scheme, proposing CDKN2A/B homozygous deletion and ploidy as potential biomarkers.^9,10 As a result, the new 5th edition of the WHO added homozygous deletion of CDKN2A/B to the WHO grading of IDH1/2 mutant astrocytomas as a default grade 4 biomarker. DNA methylation has become an important tool in CNS tumor diagnostics, with the 5th edition of the WHO requiring its use in many tumor entities, and suggesting its use in prognostication in other tumors such as ependymoma and medulloblastoma.^{5,13,14,26,27}

Our data suggest that the current WHO grading system does not correlate well with OS or PFS. In addition, ploidy does not correlate significantly with OS or PFS, suggesting that chromosomal complexity is not a robust molecular biomarker. However, we found that DNA methylation stratification by subclasses, A_IDH_LG and A_IDH_HG correlates significantly with OS and has borderline significance for PFS, which may be resolved with a larger cohort. In addition, we found that homozygous deletion of CDKN2A/B correlates with OS, although less significantly than DNA methylation, and not PFS; however, again that might be improved with a larger cohort. However, clinical and radiological criteria for progression vary between institutions and may confound the prognostic value of molecular biomarkers for PFS across various studies.

Our cohort identified multiple tumors that were high grade both by histology and DNA methylation but did not have homozygous deletion of CDKN2A/B. This suggests that there are other mechanisms for high-grade behavior in addition to CDKN2A/B loss that may contribute to high-grade DNA methylation signature and should be identified in future studies. This also indicates a limitation of CDKN2A/B homozygous deletion as the only molecular biomarker of WHO grade 4.

The most striking finding in our cohort is that grade 3 IDH mutant astrocytomas, which is histologically an intermediate group, can be subclassified using molecular techniques into a low-grade subgroup with a better prognosis and a high-grade subgroup with a worse prognosis, which was confirmed in the TCGA cohort, albeit not in the Heidelberg cohort. In our cohort, WHO grade 3 tumors evenly split into low-grade and high-grade subgroups by every metric studied, including CDKN2A/B homozygous deletion, copy number, and DNA methylation subclass. Most importantly, separation into molecular high- and lower-grade subgroups correlated with survival, that is, A_IDH_LG classifying WHO grade 3 tumors had a better OS than A_IDH_HG classifying WHO grade 3 tumors. This suggests that, biologically and clinically, there is no intermediate, grade 3, IDH1/2 mutant astrocytoma, and these tumors can be reliably separated into distinct lower- and high-grade subgroups using molecular analyses.

We and others show that the current WHO grading system could be improved to optimize the clinical management of patients with IDH mutant astrocytomas.^6,8 We demonstrate that DNA methylation subclass followed by CDKN2A/B copy number status show a significant correlation with OS in patients with IDH mutant astrocytoma in 3 independent cohorts. We also show that WHO grade 3 astrocytoma divide into 2 groups with very different prognoses. Elimination of the WHO grade 3 category using molecular analysis may decrease the need for adjuvant therapy for molecularly low-grade tumors and identify molecularly high-grade tumors in the absence of high-grade histology or CDKN2A/B deletion.

Funding

The study was supported by the Friedberg Charitable Foundation, Gray Family Foundation, Making Headway Foundation, and NIH (grant R01-CA226527).

Conflict of interest statement

D.G.P. and NYU Grossman School of Medicine own an EU and Hong Kong patent titled “Method for treating high-grade gliomas,” which is unrelated to the context of this manuscript. D.G.P. has received consultant fees from Tocagen, Synaptive Medical, Monteris, Robeaute, and Advantis. M.S. is a scientific advisor and shareholder of C2i Genomics, Heidelberg Epignostix and Halo Dx, and a scientific advisor of Arima Genomics, and InnoSIGN, and received research funding from Lilly USA. A.v.D. is a scientific advisor and shareholder of Heidelberg Epignostix. Other authors declare no conflict of interest.

Authorship statement

Study concept and design: K.G., M.S.; acquisition of cases and data collection: K.G., M.G., M.S., D.P., D.G.P., C.M.W., A.F., D.Z., J.G.G., D.A.O., M.B., M.E., Mi.Sh., D.R., A.v.D., Q.T.T., Z.A., B.A.O.; experiments: C.S., K.G.; analysis of data: S.W., Y.F., J.S., E.L., A.C., R.J.; manuscript review: K.G., M.S., S.W., Y.F., E.L., A.C.; wrote manuscript: K.G., S.W., Y.F.,. E.L., M.S.; all authors read and approved the final paper.

Data Availability

Histological and corresponding molecular data for all cases are available as a Supplementary Table 1. No new data were generated for this research.

References