Global DNA methylation profiling reveals chromosomal instability in IDH-mutant astrocytomas

Diffusely infiltrating gliomas are among the most common central nervous system tumors in adults. Over the past decade, the subcategorization of these tumors has changed to include both traditional histologic features and more recently identified molecular factors. However, one molecular feature that has yet to be integrated is the presence/absence of chromosomal instability (CIN). Herein, we use global methylation profiling to evaluate a reference cohort of IDH-mutant astrocytomas with and without prior evidence of CIN (n = 42), and apply the resulting methylation-based characteristics to a larger test cohort of publicly-available IDH-mutant astrocytomas (n = 245). We demonstrate that IDH-mutant astrocytomas with evidence of CIN cluster separately from their chromosomally-stable counterparts. CIN cases were associated with higher initial histologic grade, altered expression patterns of genes related to CIN in other cancers, elevated initial total copy number burden, and significantly worse progression-free and overall survival. In addition, in a grade-for-grade analysis, patients with CIN-positive WHO grade 2 and 3 tumors had significantly worse survival. These results suggest that global methylation profiling can be used to discriminate between chromosomally stable and unstable IDH-mutant astrocytomas, and may therefore provide a reliable and cost-effective method for identifying gliomas with chromosomal instability and resultant poor clinical outcome. Supplementary Information The online version contains supplementary material available at 10.1186/s40478-022-01339-2.


Introduction
With the recent introduction of the 2021 5th Edition of the WHO Classification of Tumours of the Central Nervous System, diffuse gliomas in adults are now defined by both histologic and molecular features, and in many cases molecular features outrank traditional histologic classification in both diagnosis and grade [18]. Oligodendrogliomas (WHO grades 2-3) are now defined by the presence of simultaneous deletion of chromosomal arms 1p and 19q, as well as mutation in either IDH1 or IDH2. Moreover, what was previously classified by histology as "astrocytoma" and "glioblastoma" has now been divided into IDH-mutant astrocytoma (WHO grades 2-4) and IDH-wildtype glioblastoma (WHO grade 4). Notably, IDH-mutant astrocytomas with homozygous CDKN2A deletion are now considered WHO grade 4 regardless Open Access *Correspondence: richardsont2@uthscsa.edu Page 2 of 11 Liu et al. Acta Neuropathologica Communications (2022) 10:32 of histologic features [2,18], although numerous other mutations, specific copy number variants, and other molecular features have been suggested as relevant to prognosis and potentially tumor grade [21]. One potentially useful molecular feature that may help explain the underlying heterogeneity in clinical outcome between IDH-mutant astrocytomas of the same WHO grade is chromosomal instability (CIN). Previously, we have demonstrated that overall copy number variation (CNV), distributed across the entire genome, tends to increase with histologic grade in IDH-mutant astrocytomas [28], and IDH-mutant grade 2-3 astrocytomas with poor clinical outcomes have incongruously elevated overall CNV at the time of initial diagnosis [27,29]. This high CNV occurs frequently in cases with other aggressive molecular features including CDK4 amplification and homozygous CDKN2A deletion, although in many cases elevated CNV is the only molecular factor present to suggest poor prognosis [22,27,29]. These IDH-mutant astrocytomas can be stratified solely by CNV level with thresholds of approximately 310-465 megabase pairs (Mbp) (~ 10-15% of the total genome) separating patient outcomes [23,28,32]. Additionally, CNS and other solid tumors with previously identified CIN [35] can reliably be identified based on mRNA profiles of specific gene sets (CIN25, CIN70) in cases with available whole exome sequencing [6,28], although this method can add significant processing time and costs. There is a wealth of evidence suggesting that CIN is a contributing factor to the aggressiveness of a subset of otherwise lower-grade IDHmutant astrocytomas, however because CIN is a dynamic and ongoing process and a single biopsy or resection specimen represents only a snapshot of the temporal and spatial molecular evolution of the neoplasm, detection may be difficult and impractical in many clinical settings [35,36].
Global DNA methylation profiling is a technique that has become relatively common in the past 5 years to diagnose and categorize CNS neoplasms (as well as identify new entities) based on epigenetic features [5,24]. In this report, we leverage global methylation profiling to distinguish CIN and chromosomal stable (CS) phenotypes in IDH-mutant astrocytomas. We utilized a small cohort of IDH-mutant astrocytomas (n = 42) with known CIN or CS status to identify epigenetic differences between the two groups, and then applied this method using an unbiased approach to a larger cohort of publicly-available IDH-mutant astrocytomas (n = 245) to verify these epigenetic signatures and investigate the clinical and molecular differences between cases clustering as CIN or CS by methylation profiling.

Case selection
Reference cohort: We analyzed 42 IDH-mutant astrocytomas (2021 WHO grades 2-4) from previous studies with multiple previously established lines of evidence suggesting either chromosomal instability (CIN) or chromosomal stability (CS) status, including mRNA profiling of gene panels linked to CIN (CIN70) [6,28], evidence of cell-to-cell chromosomal heterogeneity [29], mutations in genes with known roles in the maintenance of chromosomal stability [22,27,28], copy number profile evidence of significant spatial chromosomal heterogeneity [19], and significantly elevated CNV with multiple biopsies/resections reflective of rapid chromosomal alteration (Additional File 1: Fig. 1 ). All cases represent the first resection specimen, before treatment was initiated, and were classified according to WHO 2021 integrated histologic/molecular criteria [18].
The fraction of CNV was calculated from data in cBio-Portal as the fraction of the genome (expressed as percent of nucleotide base pairs) with log2 of copy number > 0.3 (gain or loss) following the procedure described by Gao et al. and visualized with the cBioPortal interface [14]. CNV was quantified by percentage of the total genome, as previously described [13,22,27]. Mutation analysis was performed on whole exome sequencing and included all recurrent mutations as well as genes with previously identified roles in cell proliferation, cancer and malignant progression in CNS neoplasms and other cancers, and maintenance of chromosomal stability [22,27]. Gene variant annotation was performed using The 1000 Genomes Project, COSMIC, dbSNP, ClinVar, CanProVar 2.0, and FATHMM-MKL [12,15,16,30,31,37].

Statistical analysis
Differences in patient age at diagnosis and CNV were evaluated using student's t-test. The significance of Kaplan-Meier survival curves was calculated using the Mantel-Cox test (Log-rank test). Proportion of cases with mutations specifically associated with genome instability, as well as gender, CIN70, MGMT promoter methylation, and 2016/2021 WHO grade were calculated using Fisher's Exact test. All statistical calculations were performed with GraphPad Prism version 9 (GraphPad, La Jolla, CA).

Methylation profile, molecular features, and clinical characteristics of reference cohort
Methylation profiling was performed on 14 IDH-mutant astrocytomas with previously established evidence of chromosomal instability (CIN) [19,22,28,29] and 28 "chromosomally stable" (CS) cases ( Fig. 1), demonstrating two distinct methylation clusters with UMAP   Table 1). In addition, cluster 1 cases also had significantly higher overall copy number variation (CNV) compared to cluster 2 (20.2 ± 5.8% vs 7.0 ± 1.0%; p = 0.0018), significantly more frequent cases with high CIN70 mRNA expression levels (p < 0.0001), a set of 70 genes with tight correlation between high expression and the presence of CIN [6,28], higher 2016 WHO grade (p = 0.0037), and higher 2021 WHO grade (p = 0.0224). No significant differences were identified in terms of patient gender, age at diagnosis, MGMT promoter methylation status, or other mutations or copy number changes.
Of particular interest are the three cases with known CIN that clustered with CS cases (Fig. 1a). All three cases demonstrated MGMT promoter methylation, and were 2021 WHO grade 2, 3, and 4. No difference in age was identified between these three cases and cluster 1 or the other cluster 2 cases. These cases have significantly higher overall CNV (15.3% ± 8.1%) compared to the other cluster 2 cases (p = 0.0285), but statistically equivalent CNV to cluster 1 (p = 0.6845). No significant differences were noted in PFS or OS between these three cases and cluster 1 or cluster 2, although there was a non-significant trend toward shorter PFS and OS in these cases compared to cluster 2 and longer PFS and OS compared to cluster 1.

Methylation profile, molecular features, and clinical characteristics of test cohort
Methylation profiling was performed with UMAP analysis on 245 IDH-mutant astrocytomas, and identified 57 cases (23.3%) with a similar pattern of methylation probe levels to cluster 1 of the reference cohort (CIN) and 188 cases with similar methylome characteristics to cluster 2 of the reference cohort (CS) (Fig. 2a-c). UMAP plotting demonstrated that all reference cohort cases aligned with their respective test cohort clusters. UMAP and t-SNE plotting methods demonstrated statistically equivalent results.

Discussion
Diffuse gliomas in adults remain a largely incurable disease [21,25]. However, more recently identified molecular features and epigenetic classification systems have helped to refine the diagnostic criteria beyond traditionally used histologic features, as well as identify novel, potentially targetable genetic alterations, some of which have now been codified into official diagnostic algorithms as the new standard of care [18,26]. Still, there is significant variation in terms of individual patient morbidity and mortality that is partially unaccounted for by recent changes to the WHO diagnostic and grading system.
Chromosomal instability (CIN) is a known mechanism for malignant progression in many systemic cancers. This process results in large scale and rapid alterations to whole chromosomes or chromosomal regions that may affect numerous genes and cell processes and may cause significant intratumoral and cellto-cell molecular heterogeneity, producing clones that may vary in malignancy, tendency to invade surrounding tissue, ability to evade immune regulation, and/ or resistance to traditional therapies [36]. While this process is relatively well known and the mechanisms underlying it have long been studied, detecting the presence of CIN from a single biopsy or resection specimen may be challenging and impractical to perform from a technical standpoint, and a number of direct and indirect methods for CIN detection have been proposed [35]. Additionally, while CIN frequently results in more aggressive behavior in cancer, the process itself may be targeted by a number of therapeutic agents that are currently FDA-approved or in clinical trials for other cancer types [35], raising the possibility that these drugs may be useful as adjuvant therapies in diffuse gliomas with CIN. While there are few studies that investigate the effect of CIN and overall CNV in diffuse gliomas, we and others have used large publicly-available and institutional cohorts of IDH-mutant astrocytoma and IDHwildtype glioblastoma to demonstrate that CIN can be readily identified through whole genome copy number profiling and whole exome sequencing in certain subsets. Importantly, stratification of cases using these methods has as good or better predictive power as previously codified histologic or molecular features [22,23,[27][28][29]32].
In this report, we analyzed a small cohort of IDHmutant astrocytomas with known CIN or CS status, determined by multiple distinct methods, to determine if this one characteristic would yield separate methylation profile clusters (Fig. 1). We then analyzed a larger cohort to verify that the identified methylome characteristics and resulting clustering pattern would be corroborated in a blindly evaluated cohort (Fig. 2). Our findings demonstrate that methylation profiling can distinguish IDH-mutant astrocytomas with and without CIN: cluster 1 had a distinct methylation profile, significantly greater proportion of cases with at least one measure of CIN, significantly higher CNV levels at the time of initial surgery, and significantly worse PFS and OS compared to cluster 2 (Table 1). It is worth noting that cluster 1 cases tended to be higher grade at first diagnosis, which perhaps is to be expected since higher grades tended to be associated with more genomic instability [9,28]. On the other hand, cases with CIN may present at higher histologic grades simply because they progress more rapidly and so on average may be identified later in their evolution [29]. However, methylation profiling identified 15 grade 2 cases and 26 grade 3 cases in the CIN cluster, suggesting that methylation profiling is a potentially viable method for identifying cases with CIN and resulting poor prognosis even in these lower-grade astrocytomas ( Table 2).
This report helps to further solidify the concept that CIN and CNV are features that can be measured and used as prognostic factors in subsets of IDH-mutant astrocytomas in certain situations. The fact that the CIN cluster cuts across 2021 WHO grades and the fact that the CIN cluster has significantly elevated overall CNV, even in grade 2 tumors, supports the idea that CIN is present in a subset of these tumors and may be detected before histologic indicators of tumor progression or aggressive behavior are present, and suggests that CIN is present early in tumor development, representing a fundamentally altered biology compared to IDHmutant astrocytomas without this feature. Incongruously elevated CNV identified in newly diagnosed lower-grade IDH-mutant astrocytomas likely results from underlying CIN and may be a driver of poor clinical behavior and outcome, rather than simply being a reflection of other molecular processes.
These data represent only a small sample of IDHmutant astrocytomas with methylation profiling paired with other molecular and clinical data and as such does not necessarily define the full molecular parameters of these tumors. Nonetheless, our findings do serve as a proof-of-concept that methylation profiling, an already widely accepted modality for diagnosing CNS neoplasms, may be employed to identify this underlying characteristic and separate out cases with CIN which may have more aggressive clinical courses. Given the role that global methylation profiling is beginning to play in clinical practice, this may be an ideal way to routinely screen for characteristics such as CIN, rather than relying on more intensive and less practical methods, such as single cell sequencing to identify the minority of cases in which CIN may be present [35]. Still, more work is needed to further define the epigenetic signature associated with CIN in much larger and more diverse cohorts of IDH-mutant astrocytoma, as well as to place these CIN-positive diffuse gliomas in the greater context of all CNS neoplasms by methylome profiling [24]. It is important to rapidly identify genetic and epigenetic characteristics that affect diagnosis and prognosis to ensure appropriate treatment, and multiple studies have recently shown that this type of in-depth molecular data can be available in time to guide neurooncological treatment [11,17,26].
In conclusion, these data further support the idea that the presence of CIN and elevated CNV levels are crucial data points in predicting clinical outcome, and may one day be useful targets for personalized therapies. Methylation profiling has proven to be a powerful tool for CNS neoplasm diagnosis and may potentially identify IDHmutant astrocytomas with underlying CIN, even if histologically consistent with 2021 WHO grade 2 or 3, to help further refine prognosis and improve therapeutic design in the future.