P16 immunohistochemistry is a sensitive and specific surrogate marker for CDKN2A homozygous deletion in gliomas

While CDKN2A homozygous deletion (HD) has been recognized as both a diagnostic and a prognostic marker in gliomas and meningiomas, its detection is not widely accessible and cost effective. In this current study, we examined whether simple quantification of p16 immunoreactivity can serve as a surrogate marker for CDKN2A loss in gliomas. Our results demonstrate strong correlation between the degree of p16 immunostaining and the presence of CDKN2A HD across IDH-wildtype and IDH-mutant tumors of all grades. In tumors with pathologist-scored p16 greater than 20%, we found 100% specificity for excluding CDKN2A HD, and in tumors with p16 equal to or less than 5%, we found 100% specificity for predicting CDKN2A HD. Thereby, our study provides a cost effective and convenient method for evaluating CDKN2A homozygous loss status in glioma, as an alternative to expensive genomic sequencing.

Our results build on several prior studies, which use FISH or PCR to detect CDKN2A gene copy loss and immunohistochemistry to correlate with p16 expression, many of them using the same antibody clone [23,24,25,26,27,28]. Earliest studies by Rao et al. and Burns et al. used multiplex PCR to detect CDKN2A deletion in brain tumors and correlated it with p16 expression in astrocytomas, where a strong correlation was found between p16 negative tumors and homozygous loss of CDKN2A [23]; as well as in glioblastomas, where diffuse p16 immunostaining was found to confidently exclude CDKN2A deletion but p16 immunonegativity did not always correlate with CDKN2A deletion [24]. A following study by Parkait et al. did find significant association between p16 immunonegativity and CDKN2A deletion detected by FISH in glioblastoma [25]. Subsequently, Park et al. found only moderate correlation between p16 expression (performed on tissue microarrays) and CDKN2A loss as determined by FISH, but demonstrated the strong prognostic value of p16 expression in IDH-mutant astrocytomas [26]. Most recently, Suman et al. and Geyer et al. showed evidence for the strong negative predictive value of p16 in detecting CDKN2A deletion, also using FISH for determining CDKN2A status [27, 28]. Some of the reported limitations in the above studies include false positive FISH results due to partial hybridization failure, artifacts, and sub-optimal p16 cutoff values, hampering the standardized use of p16 as a surrogate marker for CDKN2A homozygous deletion in gliomas.

By leveraging the superior sensitivity of next-generation DNA sequencing [29] with semi-quantitative scoring methodologies and digital pathology, our study puts forward specific threshold values for p16 expression as a surrogate marker of CDKN2A HD status, enabling greater standardization of this cost-effective tool in glioma diagnostics. Given the diagnostic and prognostic implications when CDKN2A HD is detected in a lower grade glioma, we favored a conservative threshold p16 expression value of 5%, which optimizes both test specificity and positive predictive value for CDKN2A HD detection, over a threshold of 10% or higher, which leads to occasional overcalling of CDKN2A HD (i.e. false positives). By introducing a second cutoff of 20% for the exclusion of homozygous loss and continuing to sequence cases within the 6–20% gray zone, we find virtually perfect concordance between pathologist-scored p16 expression and CDKN2A HD status, without any false positives or false negatives.

Recently, an analogous analysis in meningiomas by Tang et al. showed that loss of p16 expression is a sensitive marker of CDKN2A loss determined by next-generation sequencing [32]. Similarly to meningiomas, CDKN2A HD is a molecular signature for highest grade in IDH-mutant astrocytomas (grade 4) and in IDH-mutant and 1p/19q-codeleted oligodendroglioma (grade 3), regardless of histology [1]. In our cohort, 3 out of 20 IDH-mutant astrocytomas and none out of 8 IDH-mutant oligodendrogliomas contained CDKN2A HD, overall consistent with prior reported frequencies [9] (Additional file 1: Data 2). Importantly, the presence of CDKN2A HD (with pathologists’ p16 score of 1%) upgraded one IDH-mutant astrocytoma without microvascular proliferation or palisading necrosis to grade 4 (Additional file 1: Data 1). Moreover, CDKN2A HD was detected in 1 out of 5 pilocytic astrocytomas (with pathologists’ p16 score of 1–2%). This pilocytic astrocytoma displayed atypical features, including elevated mitotic activity and increased MIB1 proliferation index, as well as an aggressive clinical behavior with recurrence only 10 months after initial resection. Of note, the tumor classified as a posterior fossa pilocytic astrocytoma rather than a high-grade astrocytoma with piloid features by orthogonal DNA methylation analysis. This confirms the diagnostic and prognostic value of CDKN2A HD as previously established [1, 13]. As p16 in both cases was less than 5%, it further demonstrates the utility of p16 as a surrogate marker of CDKN2A HD in clinical neuropathology, enabling quicker final diagnosis and circumventing expensive molecular testing.

Our study is not without limitations. While we found perfect correlation between CDKN2A HD status and pathologist-scored p16 expression in the 0–5% and 21–100% p16 score ranges, sensitivity and specificity were lower in the 6–20% range (so-called gray zone) with several false positive and false negative cases present in this range. A few of the cases in this gray zone were moved to the 0–5% and 21–100% ranges after unblinded consensus re-scoring. For example, two CDKN2A HD cases in the blinded study were over scored, but consensus discussion deemed the positive p16 staining to be mostly limited to neurons and/or glia (Fig. 4a) or endothelial cells (Fig. 4b). These examples highlight the potential confounding factor of background non-neoplastic brain tissue, which has been previously reported to show nuclear and cytoplasmic reactivity for p16 in scattered astrocytes, OPCs, and/or neurons, related to cellular senescence [33,34,35]. In our own experience with p16, we have observed occasional and inconsistent immunoreactivity in only scattered glia, neurons, and endothelium. To minimize non-neoplastic background in our scores, we evaluated the most densely cellular tumor area, correlated it to its H&E, and subtracted p16 reactivity when confidently recognized as endothelial or neuronal. We cannot exclude the possibility of rare p16 reactivity contributed by entrapped non-neoplastic glia within the tumor bulk, as reactive and neoplastic glia are extremely challenging to discriminate. A pattern of p16 staining in which positive cells are scarce and equally distributed from one another, rather than overlapping and clustering, was suggestive of non-neoplastic background (Fig. 4a). Importantly, QuPath analysis was unable to perform a similar background subtraction. Conversely, few cases without CDKN2A HD were found to be under scored after unblinding our analyses. This was most often due to the tumor representing a small biopsy composed of mostly normal brain with only few tumor cells at the infiltrative edge in an otherwise low-grade glioma (Fig. 4c). Even after unblinding ourselves to CDKN2A status, such cases remained in the gray zone, as we could not confidently distinguish normal from neoplastic cells. QuPath analysis also underscored p16 expression in such tumors (Fig. 4c). Thus, areas of high tumor cellularity may be necessary for interpretation of p16 immunoreactivity, as it is hard to discriminate scattered infiltrating tumor cells amidst mostly non-neoplastic glia, especially in small biopsy specimens and when using digital software for scoring.

Another caveat in correlating p16 expression to CDKN2A inactivation are the occasional tumors in which p16 expression is lost due to epigenetic silencing of the CDKN2A locus, rather than homozygous deletion [2]. We cannot exclude that some of the false positive cases in the 6–20% gray zone may indeed have had inactivated CDKN2A transcription through an epigenetic mechanism, leading to the loss of p16 expression in the absence of genomic loss at the 9p21 locus. This caveat is especially important to consider in tumors with global epigenetic alterations. Thus, our study concludes a strong correlation between p16 expression and CDKN2A homozygous deletion, rather than between p16 expression and CDKN2A inactivation. Finally, while the utilized next-generation sequencing technology has high sensitivity for capturing homozygous CDKN2A loss with lower false positives compared to FISH, it did not include calls for tumors with a single allele (hemizygous) CDKN2A loss. Indeed, we cannot exclude that some of the cases without CDKN2A homozygous deletion may have had loss of one of the CDKN2A alleles. Given that CDKN2A encodes tumor suppressors and the current literature correlates only homozygous CDKN2A loss with prognosis and grade in gliomas and meningiomas, determining hemizygous loss in our cohort was deemed irrelevant. In all, this study supports other recent findings [23,24,25,26,27,28, 32] for the role of p16 as a surrogate marker of CDKN2A loss, and establishes a cutoff p16 value of 5% for detecting homozygous CDKN2A deletion with robust sensitivity and specificity, and a cutoff p16 value of 20% for excluding homozygous CDKN2A deletion, in both low and high-grade gliomas.