Total copy number variation as a prognostic factor in adult astrocytoma subtypes

Diffuse gliomas are among the most common primary CNS tumors, representing approximately 27% of all primary brain tumors [29, 30]. Due to their infiltrative nature, these tumors are surgically incurable, although the exact prognosis depends on numerous histologic and molecular factors. The standard of care now dictates molecular classification of gliomas based on IDH1/2 mutation status as IDH-mutant gliomas have a significantly better prognosis than their IDH-wildtype grade-matched counterparts [25]. While histologic grade shows correlation with overall survival within these molecular groups, there remains significant heterogeneity in clinical outcome.

Since the widespread adoption of the 2016 WHO classification system, much work has been done to find further molecular markers to sub-stratify both IDH-mutant and IDH-wildtype astrocytomas in hopes of better predicting tumor behavior and outcome, including identification of secondary mutations, focal genetic alterations, methylation patterns, and multivariate prognostic models [3, 24, 42, 44]. Within the IDH-wildtype groups, these studies have suggested that lower-grade gliomas (LGG) with EGFR amplification, gain of chromosome 7 and loss of 10, or TERT promoter mutations will have aggressive clinical courses and outcomes similar to IDH-wildtype glioblastoma, regardless of histologic features. In IDH-mutant groups, lower-grade tumors with alterations in genes in the retinoblastoma pathway, including amplification of CDK4 and deletion of CDKN2A/B, demonstrate significantly worse clinical behavior and shorter patient survival [1, 5, 8, 33].

Previous work has demonstrated that IDH-mutant glioblastomas have higher levels of total copy number variation (CNV) across the entire genome and evidence of more frequent chromothripsis than lower-gradeIDH-mutant astrocytomas [9]. We subsequently showed that in IDH-mutant grade II and III astrocytomas, this increased level of CNV was present before progression to glioblastoma in cases with exceptionally poor outcomes, defined by rapid progression to glioblastoma and short survival times after initial diagnosis [36, 37]. The poor outcome appeared to be directly correlated with overall CNV, but not other factors, including mutation burden or differences in methylation profiles, suggesting that this large scale CNV pattern could potentially override the beneficial effect of IDH-mutant status.

To better understand the effect of CNV, we analyzed 135 astrocytic tumors from The Cancer Genome Atlas (TCGA) (67 IDH-wildtype and 68 IDH-mutant cases) with respect to clinical outcome, CNV levels, chromosomal and specific gene amplification and deletion events, chromothripsis, total mutation load, specific mutations in known glioma/GBM genes, and mutations in genes associated with overall genomic instability. Building on our previous results, we performed wide scale genomic analysis, on a framework of pre-established prognostic factors including grade, IDH1/2-status, and the presence of CDK4 amplifications or CDKN2A/B deletions. With the exception of 2 IDH1/2-wildtype cases, CDK4 amplification and CDKN2A/B deletion were found to be mutually exclusive. We divided the cases into 5 groups: IDH1/2-mutant LGG without CDK4 amplification or CDKN2A/B deletion (Group 1), IDH1/2-mutant LGG with either CDK4 amplification or CDKN2A/B deletion LGG (Group 2), IDH1/2-mutant GBM (Group 3), IDH1/2-wildtype LGG (Group 4), and IDH1/2-wildtype GBM (Group 5).

We demonstrate that higher levels of CNV and chromothripsis are correlated with clinical outcome in the IDH-mutant groups, while the IDH-wildtype groups had uniformly high CNV levels and poor outcomes. Other prognostic factors appear to be inconsistent. We also identified a significantly higher number of mutations in genes involved with overall genomic stability, paralleling levels of overall CNV and chromothripsis, in the cases with worse prognosis. While defining the exact role of genes involved in progression may still be needed for development of individualized targeted therapies, use of CNV could potentially serve as a clinically impactful model for prognostication of different astrocytoma subtypes, and may aid in our understanding of the underlying biology of these tumor types.

Diffuse gliomas represent approximately 27% of all primary brain tumors and approximately 81% of all malignant brain tumors [29, 30], making them an intense subject of study and public health expenditure. The recent changes to glioma classification in the 2016 WHO classification system are based around the beneficial role of IDH-mutation in gliomas [25]; however, significant molecular heterogeneity exists within the lower-gradeIDH-mutant and wildtype gliomas. More work is necessary to further stratify IDH-mutant astrocytomas [44], and there is evidence that many IDH1/2-wildtype LGGs may be biologically identical to IDH1/2-wildtype glioblastomas [17, 34]. In addition, new methods to analyze whole genome genetic and epigenetic signatures are leading to new definitions for many of these tumor groups with significant prognostic implications [4, 38, 43].

We previously reported that increased CNV is associated with a more aggressive biological behavior and poor overall survival in IDH-mutant LGGs [36, 37]. With whole genome analysis in the current study, we show that CNV correlates with clinical outcome, and was significantly lower in the IDH-mutant LGGs compared to the IDH-mutant LGGs with CDK4 or CDKN2A/B alterations or IDH-mutant GBMs. (Figs. 3a and 4). These results confirm our previous findings, in which IDH-mutant LGG cases selected solely on the basis of poor clinical outcome displayed significantly higher levels of CNV before progression to GBM than a cohort with more conventional progression-free and overall survival [36]. The elevated CNV levels in IDH-mutant LGGs with CDK4 or CDKN2A/B alterations and IDH-mutant GBM represent a heterogenous assortment of genomic alterations within the IDH-mutant group with only a few consistent areas of gains and losses (Fig. 5b-c) whereas a large fraction of the CNV in IDH-wildtype tumors arose from consistent amplifications in chromosome 7p (containing EGFR), and deletions in chromosomes 9p and 10 (Fig. 6).

Although the overall CNV changes seem to occur before histologic progression to GBM in cases with other negative prognostic factors and/or clinically demonstrated poor outcomes, there is still uncertainty in the exact connection to elevated levels of CNV and the driving force behind this poor progression. Our data also agrees with the previously demonstrated data that CDK4 and CDKN2A/B alterations are prognostic factors within the IDH-mutant LGGs [44]. While worse prognosis seems to correlate with CDK4 or CDKN2A/B status, our earlier study [36] showed only a fraction of the rapidly progressing tumors had these specific alterations, yet all of them had high overall CNV, indicating that it may be an earlier event or a separate phenomenon altogether. Further analysis of CNV data may help determine if the IDH-mutant LGGs with CDK4 and/or CDKN2A/Balterations are actually early GBMs or simply under-sampled tumors, similar to current thinking on many IDH-wildtype LGGs [3, 42]. While it is reasonable to argue that our cohort of IDH-mutant LGGs without CDK4 or CDKN2A/B alterations show low CNV because they selectively exclude tumors with specific known amplifications/deletions to enrich the other cohorts, if this were to hold true, the clinical outcome would likely also follow the same pattern and would show worse outcome within the other groups containing CDK4 amplification or CDKN2A/B deletion. CDK4 and CDKN2A/B did not show a prognostic difference in IDH-mutant GBMs or IDH-wildtype LGGs or GBMs, and the overall CNV was not different between these two groups (Fig. 2a-c), so the effect of both of these alterations seems limited to IDH-mutant LGG cases. CDK4 amplification and CDKN2A/B deletion also appear to be mutually exclusive, with only two total cases (2.3%) having both molecular alterations (Additional file 4: Figure S4 and Additional file 5: Figure S5).

An additional finding in these tumor groups is the trend toward more frequent mutations in genes associated with overall chromosomal stability in groups with worse clinical outcomes (groups 2–5) compared to the group with relatively favorable outcomes (group 1) (Fig. 8b, Table 2). This correlates positively with the trends toward increased CNV levels and number of cases with chromothripsis and inversely with the progression-free and overall survival in these groups (Table 1). The number of mutations in genes with chromosomal stability functions and cases with chromothripsis are somewhat lower in the IDH-wildtype cohorts compared to groups 2 and 3 in the IDH-mutant cohorts, despite having statistically identical CNV levels (Fig. 8). This difference may be explained by the fact that a large portion of the CNV in these IDH-wildtype groups is more homogeneously associated with specific chromosomal regions (7, 9p, 10) instead of more diffusely distributed as seen in the IDH-mutant groups with high CNV and poor outcome (Figs. 5 and 6).

This process also provides a potential mechanistic explanation for the widespread genomic alterations and the worse prognosis associated with this increase in CNV in at least a subset of cases. Inactivating mutations in genes associated with maintenance of genetic and chromosomal integrity, and the resulting increase in CNV, allows for rapid and widespread changes to the genome, including chromothripsis, and has the potential to cause more frequent gains of oncogenes and loss of tumor suppressor genes and drive tumor formation and progression towards malignancy [11, 19, 20, 41, 46]. This may also suggest a different molecular mechanism underlying total CNV levels in IDH-mutant and IDH-wildtype groups. At this point, however, we can only state that these factors are all correlated with poor clinical outcome, but no causative links can definitively be made.

The present study reinforces our previous findings [36, 37] demonstrating that elevated CNV is associated with poor outcome in grade II and III IDH-mutant astrocytomas, and presents this as a potential prognostic factor. We demonstrate for the first time that higher CNV is associated with previously established prognostic factors within the IDH-mutant LGG subgroup, such as CDK4 amplification and CDKN2A/B deletion. This study is also the first to demonstrate a significant quantitative difference in mutations of genes related to chromosomal stability in groups with higher CNV and worse clinical outcomes (Fig. 8b).

It is important to note that while many of the genetic and epigenetic methods used to generate these data are currently only used for research purposes, recent proof-of-concept studies have demonstrated that specific and large-scale genetic and epigenetic alterations can be identified rapidly and relatively inexpensively [12, 18], including overall methylation patterns indicative of IDH1/2 status, methylation of key gene promotors, CNV, mutations, and gains and losses of key genes and chromosomal regions. These studies have demonstrated that with newer techniques these molecular factors can be identified in approximately the time that it takes to make a histologic diagnosis. It is therefore conceivable that CNV and other molecular factors identified in this report could soon be used clinically at the time of initial diagnosis to help guide prognosis and treatment strategies.

Our results support our previous findings that IDH-mutant lower-grade astrocytomas with higher total CNV are associated with poor clinical outcome and behave more consistently with IDH-mutant GBM than other IDH-mutant LGGs with low CNV, and suggest that CNV could be a viable prognostic factor in these tumors alongside IDH1/2 mutations, CDK4 amplifications, and CDKN2A/B deletions. We demonstrated that high CNV occurs in IDH1/2-wildtype astrocytomas and glioblastomas which also have poor prognoses, although the reason underlying elevated CNV may be different in IDH-mutant and IDH-wildtype tumors. We also provide a possible mechanism for the overall CNV differences in these astrocytoma subgroups, as the CNV levels seem to correlate with numbers of mutations in genes with roles in maintaining genomic stability. These results suggest that high overall CNV negate the beneficial effects of IDH1/2 mutation, and could potentially be used as a prognostic marker in IDH-mutant astrocytomas in the future.