Grading of diffuse gliomas is a critical component of the diagnostic process that guides therapeutic decision-making and should, ideally, be based on prognostically significant measures of tumor behavior. Based on recently published recommendations by the cIMPACT-NOW working group [3, 12], CDKN2A status is likely to be incorporated into future grading systems for IDHm astrocytomas, but it remains unclear how to interpret the results of a commonly used testing modality, FISH, for this purpose. In order to define an appropriate cutoff value for interpreting CDKN2A FISH results as positive for homozygous deletion in routine clinical service, and to assess the anticipated impact of applying these results in our patient population using the proposed cIMPACT-NOW grading schemes, we performed a retrospective analysis of IDHm astrocytomas that had prospectively undergone CDKN2A FISH analysis. Multivariate Cox-PH analysis revealed that the prognostic significance of CDKN2A in IDHm astrocytomas is optimized by using a cutoff value of ≥ 30% of tumor cells with homozygous deletion in order to define tumor-wide “homozygous deletion” status. Applying this cutoff value, very few primary treatment-naïve tumors in our institutional cohort showed this level of deletion, and of the tumors that did exceed this threshold, the majority were already considered grade 4 by histology. Compared to primary tumors, CDKN2A homozygous loss was more frequent in recurrent grade 3 and grade 4 tumors. While both histologic grade and CDKN2A loss were independent predictors of survival in our cohort, the inclusion of CDKN2A as a grading criterion failed to improve survival stratification, at least in the setting of tumors not meeting grade 4 criteria histologically. With the ≥ 30% cutoff, only two lower grade tumors would be upgraded to grade 4, and both patients were still living at the time of analysis, having already exceeded the median survival of histologic grade 4 tumors.
At first blush, it may seem counterintuitive that the Cox-PH model developed using the full cohort of tumors spanning all grades confirms CDKN2A deletion as an independently significant prognostic factor, yet the optimal cutoff of 30% was exceeded by only a very small fraction of grade 2 and 3 tumors. These findings can be reconciled, however, if the overall model is being largely driven by prognostic differences among morphologically grade 4 tumors. Indeed, unlike lower grade tumors, grade 4 tumors with ≥ 30% CDKN2A deletion showed significantly shorter overall survival than grade 4 tumors with intact CDKN2A, in both primary and first recurrence tumors (Fig. 5a and b). In light of the published studies claiming dismal prognosis for lower grade tumors with CDKN2A loss [7, 10, 11], we wanted to make sure that the model driven by grade 4 effects was not inappropriately excluding certain grade 2–3 tumors from being classified as a higher molecular grade; perhaps a different, lower threshold might be more appropriate for stratifying the lower grade tumors. We addressed this question in two ways: by applying a lower threshold and examining whether hypothetically-upgraded tumors behave as the adjusted grade would suggest (Fig. 5c and d) and by creating a Cox-PH model using only grade 2 and 3 tumors (excluding grade 4) (Fig. 6). Contrary to the hypothesis that a lower threshold would perform better for lower grade tumors, both analyses show the opposite—in our cohort of patients, there is no evidence that a threshold at or below 30% homozygous deletion improves prognostication of lower-grade lesions. We note, however, that it remains possible that lower grade IDHm astrocytomas with high level (≥ 30%) CDKN2A deletion by FISH do in fact have a poor prognosis, but given the rarity of grade 2 or 3 tumors exceeding this threshold, there is insufficient evidence to determine the prognostic impact of this finding in our cohort.
Our findings differ from those studies showing a significant prognostic impact of CDKN2A deletion in both grade 3 and grade 4 IDHm astrocytomas, as detected by array-based techniques [7, 10]. While our data show a similar poor prognosis of CDKN2A deletion in grade 4 IDHm astrocytomas, these findings do not extend to grade 3 tumors. Using the proposed cIMPACT-NOW 5/6 criteria and our statistically-defined FISH cutoff of ≥ 30%, the hypothetical upgrade rate of grade 3 tumors in our patient cohort is markedly lower (1/31, 3%) than that in Shirahat et al. (15/90, 17%) or Appay et al. (35/211, 17%). While it is possible that technical differences between various testing modalities could have an unexpectedly large effect, a likelier cause of this discrepancy could be differences in the population characteristics of the tumors being analyzed. The histologic grades at presentation for tumors in the cohorts of Shirahat et al. (26% grade 2, 43% grade 3, 32% grade 4) and Appay et al. (grade 2 not included in primary analysis, 50% grade 3, 50% grade 4) are skewed towards higher grades, while our single institution cohort had a larger proportion of lower grade tumors (51% grade 2, 29% grade 3, and 20% grade 4). The distribution of cases in our study, with the highest proportion presenting as grade 2 and the lowest as grade 4, is in broad agreement with previously published large all-comer cohorts of IDHm astrocytomas [4]. Similarly, the frequency at which homozygous CDKN2A loss was detected in our cohort at each grade (grade 2: 1.8%, grade 3: 3.2%, grade 4: 27%) is similar to the proportion of IHDm astrocytomas in The Cancer Genome Atlas (TCGA) combined low grade glioma and glioblastoma cohorts (grade 2: 3.5%, grade 3: 6.7%, grade 4: 18.8%; primary tumors, multiple methods of CDKN2A assessment) [8]. The enrichment in grade 3 and grade 4 tumors in the Shirahat et al. and Appay et al. studies, along with the elevated frequency of CDKN2A loss in grade 3, suggests that these cohorts may represent a different population of tumors, possibly including increased numbers of diagnostically difficult, borderline, or clinically aggressive cases sent for expert consultation. It is also not clear whether these studies limited their analyses to only primary treatment-naïve specimens. Inclusion of recurrent/treated tumors could be a significant confounder, especially given the overall worse prognosis and increased frequency of CDKN2A loss seen in recurrent tumors in our cohort.
The existing literature on use of FISH specifically to detect CDKN2A homozygous deletion in gliomas is sparse. Perhaps the most relevant is a recent study by Yang et al. that used FISH to examine CDKN2A, CDK4, and PDGFRA copy number alterations in grade 2 and 3 astrocytomas [11]. The authors used a cutoff of ≥ 20% of tumor cells showing homozygous deletion, but a rationale for this threshold was not provided. The study found CDKN2A homozygous deletion of ≥ 20% at similar rate in grade 2/3 tumors as our cohort (15% Yang et al. versus 10% for our cohort) with both grade 2 and grade 3 tumors showing deletion. The survival effect of CDKN2A in the Yang et al. cohort cannot be directly evaluated, as the survival analysis in that study grouped tumors with CDKN2A deletion with those showing CDK4 amplification (i.e. alteration in the RB1 pathway), rather than assessing each gene independently. While these RB1-altered tumors did show shorter overall survival than the RB1-intact group, they nonetheless appear to have a significantly better prognosis than the morphologic grade 4 tumors in our study. This comparison with the published literature further supports our findings that grade 2/3 astrocytomas with CDKN2A homozygous deletion as assessed by FISH have longer survival than histologic grade 4 tumors.
The ability to accurately and reliably detect CDKN2A homozygous deletion in IDHm astrocytomas is necessary if this is to be included as a grade defining criteria in the next revision of the WHO Classification. The cIMPACT-NOW recommendations do not offer definitive guidance as to which testing modalities should be used, and there are many possible methods to detect loss. FISH is perhaps the oldest and most widely accepted technique for detecting copy number alterations. FISH testing has many benefits, including the ability to definitively identify homozygous deletion in infiltrating tumors, and to separate true homozygous deletion from hemizygous loss and complex copy number alterations. Like all testing modalities, however, FISH has certain intrinsic technical limitations, including insensitivity to deletions smaller than the region covered by the probe, and artifactual loss of signal due to partial sectioning of nuclei when performed on FFPE tissue sections. As an example of the latter, consider a nucleus with a single CEP9 signal and no 9p21 signals. This result could accurately reflect monosomy 9 with additional 9p21 deletion, but could also arise in other ways: (1) homozygous 9p21 deletion with artifactual loss of one CEP9 signal; (2) monosomy 9 with artifactual loss of one 9p21 signal; (3) hemizygous 9p21 loss paired with artifactual loss of one CEP9 and one 9p21 signal; (4) wild type chromosome 9 with artifactual loss of three probes. The possibility of artifactual signal loss highlights the necessity of interpretative guidelines. In our laboratory, a nucleus with no 9p21 signals and at least one CEP9 signal is interpreted as homozygous deletion, which favors sensitivity for detecting true absence of CDKN2A over specificity for excluding monosomy 9 with artifactual 9p21 loss. A more stringent criteria requiring two CEP9 signals would lead to even fewer tumors being identified as having homozygous CDKN2A deletion.
In addition to FISH, there are numerous variations of genomic microarrays in use for analyzing brain tumor samples. Comparative genomic hybridization arrays (aCGH) excel at detecting copy number alterations in aggregate tissue samples, but can struggle at detecting loss in the context of a sparsely infiltrating tumor. Single nucleotide polymorphism (SNP) arrays are often used in combination with either true or virtual aCGH to help separate hemizygous loss and complex alteration events from true homozygous loss (although not at a single cell level). Finally, the data obtained from methylation or other NGS arrays can often be analyzed to provide output similar to a combination aCGH/SNP array. As with FISH, each of these techniques have their own interpretive subtleties and technical limitations, which complicates cross-modality comparisons of the prognostic implications of test results. Clear and specific descriptions of the algorithms used to judge homozygous deletion status would be helpful for this purpose, but are frequently lacking. As a result, direct comparison between these various methods is often not possible, and it is not clear which technique should represent the “gold standard.”
An important point of concordance between our study and the existing literature is the lack of prognostically meaningful CDKN2A homozygous deletion in histologically grade 2 tumors. The lone patient with a grade 2 tumor with over 30% homozygous deletion in our study is still alive at 61 months from diagnosis, and neither of the above array-based studies identified any histologic grade 2 tumors with CDKN2A homozygous deletion (Shirahat et al. n = 54, Appay et al. n = 20). The wording of the cIMPACT-NOW 6 proposal for updated grading criteria is not clear regarding whether or not morphologic grade 2 IDHm tumors will require CDKN2A testing for formal grading; clarification on this point will be essential in any upcoming WHO update. Results from our cohort agree with the previous studies and suggest that testing grade 2 tumors for deletion would be very low yield at best, might lead to inappropriate upgrading of indolent tumors, and could impose an undue financial burden on the global healthcare system.
In conclusion, CDKN2A homozygous deletion is a marker of poor prognosis in histologic grade 4 IDHm astrocytomas, but the impact of this finding in histologic grades 2 and 3 tumors is less clear. Different techniques for determining CDKN2A status may provide markedly different results between and even within individual institutions. Specific criteria for determining the presence of homozygous deletion across different testing modalities will be essential if CDKN2A homozygous deletion is included as a grading criterion in the next revision of the WHO Classification.