Introduction

Diffuse gliomas of adults comprise ~60ā€“70% of primary malignant intracranial tumors in Japan [1]. These tumors are primarily graded according to histological findings, such as mitotic activity, microvascular proliferation, and/or necrosis [2,3,4]. In addition, the current World Health Organization (WHO) classification (WHO 2016) [5] incorporates genetics for defining histological types of gliomas, which enables a more accurate prediction of clinical behavior. Accordingly, adult-type diffuse glioma is subdivided into distinct categories based on isocitrate dehydrogenase (IDH) 1/2 mutation [6, 7] and 1p/19q codeletion [8]. Specifically, glioblastomas are categorized as IDH-wildtype (~90%) or IDH-mutant (~10%). Lower-grade diffuse gliomas are classified into three groups, IDH-mutant diffuse glioma without 1p/19q codeletion (the vast majority of diffuse astrocytic tumors), IDH-mutant diffuse glioma with 1p/19q codeletion (the vast majority of diffuse oligodendroglial tumors), and IDH-wildtype diffuse gliomas [9,10,11].

Homozygous deletion (HD) of the gene encoding cyclin-dependent kinase inhibitor 2ā€‰A (CDKN2A) is a recently reported biomarker for predicting poor prognoses of patients with IDH-mutant diffuse gliomas [12,13,14,15,16,17]. The Consortium to Inform Molecular and Practical Approaches to CNS Tumor Taxonomy (cIMPACT-NOW) update 5 recently recommended that IDH-mutant astrocytoma with CDKN2A/B HD be considered as a WHO grade IV tumor, even in the absence of microvascular proliferation or necrosis [18]. However, the prognostic value of CDKN2A HD in oligodendroglioma and IDH-wildtype glioma has not been established. Currently, CDKN2A HD is usually assessed using fluorescence in situ hybridization (FISH), array comparative genomic hybridization, or multiplex ligation-dependent probe amplification (MLPA), but newer techniques such as next generation sequencing and DNA methylation profiling microarray have also been used [19]. However, use of these molecular assays is limited due to their high costs, long turnaround time, and non-availability at all centers. This highlights the need to develop surrogate assays, such as immunohistochemistry-based approaches.

A historical candidate as a potential surrogate biomarker was the p16 protein, which is encoded by CDKN2A (9p21) and plays an important role in blocking G1 to S phase transition via CDK4 and CDK6 inhibition. However, whether the loss of p16 immunoreactivity correlates with CDKN2A HD is controversial [20,21,22,23]. More recently, the expression of methylthioadenosine phosphorylase (MTAP) has been extensively studied in various types of epithelial neoplasms [24,25,26,27,28] and malignant mesothelioma [29,30,31,32].

MTAP is essential for adenosine monophosphate and methionine salvage. As MTAP catalyzes the conversion of methylthioadenosine to adenine and methylthioribose-1-phosphate with the latter compound being further metabolized to methionine, it is expressed in virtually all cells [33, 34]. MTAP is located on 9p21 only 165ā€‰kb telomeric to CDKN2A, and in tumors, it is often simultaneously deleted with CDKN2A HD. MTAP HD results in the loss of MTAP immunohistochemical staining, which is accepted as a reliable surrogate marker of CDKN2A HD for differentiating malignant mesothelioma from reactive mesothelial cell proliferation (sensitivity: 74ā€“93%, specificity: 96ā€“100%) [35,36,37]. However, the predictive role of MTAP immunohistochemistry for CDKN2A HD has not been fully investigated in glioma. Therefore, we undertook the current study to determine (1) whether the loss of MTAP immunoreactivity could be used as a surrogate for CDKN2A HD in adult-type diffuse glioma and (2) whether MTAP staining had prognostic value for IDH-mutant astrocytoma. For comparison, we also evaluated in parallel p16 immunostaining.

Materials and methods

Patients and tissue specimens

The study was approved by the institutional review board at the National Cancer Center, Tokyo, Japan (No. 2013-042). A total of 189 adult-type diffuse glioma specimens were obtained between 1997 and 2019 based on tissue availability. The status of IDH, 1p/19q, and CDKN2A was previously available for a subset of the tumors with the remaining tumors being newly tested as part of the current study. IDH-mutant gliomas that were previously determined to harbor CDKN2A HD at original diagnosis were preferentially selected for improved immunohistochemistry evaluation. All cases were diagnosed at the National Cancer Center Hospital (Tokyo, Japan), and their diagnoses were reviewed and confirmed according to the WHO 2016 classification [5], except for some IDH-mutant astrocytomas for which the loss of ATRX expression and strong diffuse p53 immunopositivity were considered equivalent to the absence of 1p/19q codeletion in accordance with the cIMPACT-NOW update 2 [38]. Among 189 tumors, 30 tumors were subjected to BRAF V600E testing by pyrosequencing, but none harbored the mutation. None of the cases in this study was screened for FGFR rearrangement. Of the 189 diffuse glioma specimens tested, 178 tumors were successfully scored using MTAP immunohistochemistry and CDKN2A MLPA or FISH, as detailed in the results section. Hence, these 178 tumors we used for subsequent analyses to determine the molecular/immunohistochemical correlation and prognostic implications.

The entire 178 tumor cohort was divided into three groups, the IDH-mutant astrocytoma group, the IDH-mutant oligodendroglioma group, and the IDH-wildtype group. The IDH-mutant astrocytoma group consisted of 77 tumors from 74 unique patients, including IDH-mutant gliomas that lacked 1p/19q codeletion as determined by MLPA or FISH and IDH-mutant gliomas that exhibited a loss of ATRX nuclear expression and strong diffuse p53 immunopositivity. According to the WHO 2016 classification, the IDH-mutant astrocytoma group had 15 IDH-mutant diffuse astrocytomas (DA), 35 IDH-mutant anaplastic astrocytomas (AA), and 27 IDH-mutant glioblastomas. Two of the IDH-mutant glioblastomas and one of the AA were recurrences of DAs. The IDH-mutant oligodendroglioma group consisted of 13 tumors from 12 unique patients. Based on the WHO 2016 classification, five were oligodendrogliomas, IDH-mutant and 1p/19q-codeleted, and eight were anaplastic oligodendrogliomas, IDH-mutant and 1p/19q-codeleted. There was one pair of primary and recurrent anaplastic oligodendrogliomas from the same patient. Finally, the IDH-wildtype group consisted of 88 tumors from 88 unique patients. These included infiltrating gliomas without the IDH mutation as determined by either IDH1 (R132H) immunohistochemistry or sequencing. All these tumors corresponded to glioblastomas, IDH-wildtype, according to the WHO 2016 classification.

Molecular testing

Molecular testing was performed using either frozen or formalin-fixed paraffin-embedded (FFPE) specimens. DNA was extracted from the specimens using a DNeasy Blood & Tissue Kit (Qiagen, Hilden, Germany) according to the manufacturerā€™s instructions. DNA concentrations were determined using a Qubit DNA quantification system (Invitrogen, Carlsbad, CA, USA).

The IDH1/2 mutation was analyzed using pyrosequencing as previously described [39]. Briefly, an 86-bp fragment of IDH1 containing the targeted codon 132 and an 85-bp fragment of IDH2 containing the targeted codon 172 were amplified. The amplified fragments then underwent allele quantification analysis, which is an analysis mode available in the PyroMark Q24 software (Qiagen).

CDKN2A and 1p/19q testing was performed using MLPA and/or FISH. SALSA MLPA P088 Oligodendroglioma 1p-19q probemix (MRC-Holland, Amsterdam, The Netherlands) was used according to the manufacturerā€™s instructions. The kit contained 19 probes for loci on 1p and three probes for loci on 1q, two probes for loci on 19p and 11 probes for loci on 19q, and three probes for loci on CDKN2A and two probes for loci on CDKN2B. The data were analyzed using Coffalyser.Net (MRC-Holland) to determine the occurrence of 1p/19q codeletion and CDKN2A HD.

Dual-color CDKN2A or 1p/19q FISH studies were carried out using 4ā€‰Ī¼m-thick tissue specimens, either on whole tissue sections or tissue microarray (TMA). The probes used were Vysis LSI CDKN2A Spectrum Orange/CEP9 Spectrum Green probes (Abbott Molecular, Abbott park, IL, USA) and the 1p/19q probe set (Leica Biosystems, Tokyo, Japan). Images were captured using the Metafer Slide Scanning Platform (MetaSystems, Altlussheim, Germany) and a minimum of 100 non-overlapping tumor cells were examined. Tumors in which >10% of the cells showed loss of signal for both CDKN2A genes and at least one CEP9 signal (0 Spectrum Orange/1ā€“2 Spectrum Green) were considered HD [40]. Lymphocytes or endothelial cells with two CDKN2A and two CEP9 signals served as internal controls. For 1p/19q codeletion, a minimum of 100 non-overlapping tumor cell nuclei with at least two green (1q or 19q) signals were evaluated. A 1p/19q codeletion was considered to be present when (1) >50% of tumor nuclei had only one orange (1p or 19q) signal or two orange and four green signals; and (2) the ratio of 1p/1q to 19q/19p was ā‰¤0.8.

Although we used FISH and/or MLPA to obtain molecular data, these two methods have different capabilities to determine CDKN2A and MTAP copy number. The commercial FISH probe we used covered ~222ā€‰kb in the 9p21 region and hybridized to MTAP, CDKN2A, and CDKN2B, whereas the MLPA assay employed probes of smaller size (153, 190, and 253ā€‰bp) designed to hybridize to a specific region (~20ā€‰kb) within CDKN2A and did not hybridize to MTAP. Regardless of this difference, for the sake of brevity, we used the term CDKN2A HD for all cases in this paper unless otherwise specified. This choice was made based on the previous finding that in 564 diffuse gliomas, MTAP HD without CDKN2A/B HD is extremely rare (0.5%) [41], according to cBioPortal analysis [42, 43]. In addition, previous studies used a similar FISH assay to demonstrate the prognostic significance of CDKN2A HD in glioma [13, 16].

Immunohistochemistry

Immunohistochemistry staining was performed using either whole tissue sections or previously prepared TMAs with 2.0ā€‰mm cores. A 4ā€‰Āµm-thick section cut from the FFPE block was deparaffinized and the sections exposed to 3% hydrogen peroxide to block endogenous peroxidase activity. After heat-induced epitope retrieval in citrate buffer or Targeted Retrieval Solution, pH 9.0 (Dako), slides were incubated for 1ā€‰h at room temperature with primary antibodies, and subsequently labeled using the EnVision System (Dako, Glostrup, Denmark). Diaminobenzidine was used as the chromogen and hematoxylin as the counterstain.

For MTAP staining, we primarily used a rabbit monoclonal antibody EPR6893 (1:1000; Targeted Retrieval Solution, pH 9.0, Abcam, Cambridge, UK) that was one of the few reagents published on mesothelioma at the time this project was initiated [35]. However, because more recent publications on mesothelioma have favored different clones [36], we additionally used one of such newer clones 2G4 (1:200; Targeted Retrieval Solution, pH 9.0, Abnova, Taiwan) and compared the performance between these two antibodies using the same 178 samples. For each antibody, the staining protocol was optimized using several different types of malignant tumors (e.g., mesothelioma) whose CDKN2A statuses were known. Normal cerebral tissue was diffusely positive using either protocol as expected. For p16 staining, we used mouse monoclonal antibody G175-405 (1:10; citrate buffer, BD Biosciences, San Jose, CA).

MTAP immunohistochemistry revealed cytoplasmic staining with or without nuclear labeling in normal cells, such as endothelial cells or lymphocytes. The loss of MTAP immunoreactivity in tumor cells was defined by the absence of cytoplasmic staining in the presence of reactivity of the internal positive controls. Tumors that exhibited cytoplasmic MTAP immunoreactivity but no nuclear staining were considered to retain MTAP staining [36]. We scored the staining only when the internal positive control was properly labeled, and any samples that failed to stain positive for endothelial cells/lymphocytes were considered uninterpretable. For all cases, the staining was independently interpreted by two pathologists (KS and AY) blinded to the molecular results. All scoring results were subsequently compared and any discrepancy or difficulty was discussed to reach consensus. In addition, for the cases evaluated using whole tissue sections, the level of difficulty in MTAP interpretation was graded as ā€œlowā€ or ā€œhigh.ā€ The difficulty was considered low when the interpretation could accurately be performed even at low magnification (Ɨ10 objective). The rest of the cases definitely required higher magnification for scoring and were judged as high.

For p16 immunohistochemistry, nuclear and cytoplasmic stainings were evaluated separately. The p16 staining was often heterogeneous with a mosaic-like appearance, even if the staining was positive. Therefore, only complete loss of staining was considered p16-deficient with all other patterns of reactivity being considered as retained p16 expression. Interpretation of the p16 staining was performed by one pathologist (KS) blinded to the molecular results.

Statistical analysis

For statistical analysis, variables were compared using the Mannā€“Whitney U-test or Chi square test. To analyze performance, we evaluated the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for MTAP and p16 staining. Overall survival (OS) and progression free survival (PFS) after the first surgery were analyzed using the Kaplanā€“Meier method and compared using a log-rank test. Pā€‰<ā€‰0.05 was considered statistically significant. All statistical analyses were performed using the Easy R (EZR) plug-in (Saitama Medical Center, Jichi Medical University, Saitama, Japan) [44], which is a graphical user interface for R (The R Foundation for Statistical Computing, Vienna, Austria).

Results

CDKN2A HD

CDKN2A HD was present in 44% (78/178) of all diffuse gliomas evaluated in our study. Specifically, it was present in 22% (17/77) of IDH-mutant astrocytomas, 46% (6/13) of IDH-mutant oligodendrogliomas, and 63% (55/88) of IDH-wildtype glioblastomas (TableĀ 1). The frequency was significantly correlated with higher age at diagnosis (Pā€‰<ā€‰0.001) and IDH-wildtype status (Pā€‰<ā€‰0.001). Among IDH-mutant astrocytomas, 0% (0/15) of DA, 11% (4/35) of AA, and 48% (13/27) of IDH-mutant glioblastomas harbored CDKN2A HD. CDKN2A HD was significantly associated with WHO grade (Pā€‰<ā€‰0.001; TableĀ 1), but not related to patient age or sex or tumor location (Supplementary Table). Among the selected histological variables analyzed, microcystic change was significantly less common in tumors with CDKN2A HD than in those without CDKN2A HD (Pā€‰=ā€‰0.018). Interestingly, two of the IDH-mutant DAs with intact CDKN2A recurred 7 and 13 years later as IDH-mutant glioblastoma with CDKN2A HD, suggesting a role for this deletion in tumor progression. The frequency of CDKN2A HD in IDH-mutant oligodendroglioma detected in our study was higher than that in previous studies [13, 15], likely because tumors with CDKN2A HD were preferentially selected by the study design.

Table 1 Clinicopathological characteristics of all diffuse glioma cases in relation to CDKN2A homozygous deletion status.
Full size table

MTAP immunohistochemistry (clone EPR6893)

Of 189 diffuse glioma samples initially tested, 107 tumors were included on TMAs. Excluding five tumors whose cores were unavailable, 80 of the 102 tumors (78%) were successfully scored. The remaining 22 tumors were difficult to score using TMAs due to high mixtures of reactive glial cells, relatively high background staining, or weak to no internal normal control positivity. The latter 22 tumors and 4 of the 5 tumors whose cores were lacking on the TMA, along with the remaining 82 tumors not included in the TMA, were evaluated using whole tissue sections. Of these 108 tumors evaluated, 100 tumors (100/108, 93%) were successfully scored, of which 2 cases were further excluded because CDKN2A data were unavailable. Overall, 178 tumors were successfully scored and formed the basis of the analyses.

Loss of MTAP immunoreactivity was observed in 40% (72/178) of all diffuse gliomas. The staining loss was uniformly seen in the evaluable tumor areas. Two representative gliomas with MTAP loss and CDKN2A deletion are illustrated in Fig.Ā 1Aā€“D. The loss was seen in 21% (16/77) of IDH-mutant astrocytomas, 54% (7/13) of IDH-mutant oligodendrogliomas, and 56% (49/88) of IDH-wildtype glioblastomas (TableĀ 2). MTAP immunoreactivity was retained in the remaining cases, and one representative case with retained MTAP and intact CDKN2A copy number is illustrated in Fig.Ā 1E, F. Among the IDH-mutant astrocytomas, 15/17 of the tumors with CDKN2A HD showed MTAP deficiency, while 59/60 of the tumors lacking CDKN2A HD retained MTAP expression (sensitivityā€‰=ā€‰88%, specificityā€‰=ā€‰98%, PPVā€‰=ā€‰94%, NPVā€‰=ā€‰97%; TableĀ 2). MTAP expression was retained in two IDH-mutant glioblastomas with CDKN2A HD (Fig.Ā 2). Although one MTAP-negative IDH-mutant AA showed no evidence of CDKN2A HD based on MLPA (Supplementary Fig.Ā 1A, B), that tumor did show 9p21 HD based on FISH analysis (Supplementary Fig.Ā 1C). For the IDH-wildtype group, 49/55 of the tumors with CDKN2A HD were deficient for MTAP, while all 33 tumors with no evidence of CDKN2A HD retained MTAP expression (sensitivityā€‰=ā€‰89%, specificityā€‰=ā€‰100%, PPVā€‰=ā€‰100%, NPVā€‰=ā€‰85%). The concordance between MTAP immunohistochemistry and CDKN2A HD for the IDH-mutant oligodendroglioma group was poorer than that for the other groups (sensitivityā€‰=ā€‰67%; specificityā€‰=ā€‰57%, PPVā€‰=ā€‰57%, NPVā€‰=ā€‰67%, TableĀ 2).

Fig. 1: MTAP immunohistochemistry and molecular tests for CDKN2A.
figure 1

a An IDH-mutant astrocytoma demonstrating the loss of cytoplasmic MTAP staining. Positively labeled endothelial cells serve as an internal control. b The same tumor from (a) demonstrating CDKN2A homozygous deletion (HD) according to MLPA. The arrow head indicates the region of CDKN2A probes. c An IDH-mutant astrocytoma showing loss of cytoplasmic MTAP staining. d The same tumor in (c) demonstrates CDKN2A HD by FISH. No orange CDKN2A signal was observed in the presence of two green CEP9 signals. e An IDH-mutant astrocytoma retaining cytoplasmic MTAP staining. f Based on MLPA, this tumor harbored an intact copy number of CDKN2A. The arrow head indicates the region of CDKN2A probes.

Full size image
Table 2 Correlation between CDKN2A homozygous deletion and MTAP or p16 immunohistochemistry.
Full size table
Fig. 2: Discrepancy between MTAP immunohistochemistry and CDKN2A copy number status.
figure 2

a An IDH-mutant glioblastoma exhibiting retained MTAP cytoplasmic immunoreactivity. b This tumor showed CDKN2A HD by MLPA. The arrow head indicates the region of CDKN2A probes.

Full size image

The level of difficulty in MTAP interpretation of specimens evaluated using whole tissue sections was low in 61% (60/98) of the cases and high in 39% (38/98) of the cases. Among 98 cases, 10 cases required discussion to reach consensus (10%, 10/98). As a result of internal positive control cells staining weakly, the MTAP scoring required close examination of more than one high-power field of tissue in 28% of the tumors (Fig.Ā 3A, B). In 14% of tumors, the staining intensity within a slide was heterogeneous with the periphery of the tissue sections showing weak or no positivity in the internal normal controls. This was particularly notable for large resection specimens and was likely an artifact. Even in well-prepared samples, scoring required particular care in less-cellular, low-grade tumors to distinguish intermixed non-neoplastic glial cells from infiltrating tumor cells (Fig.Ā 3C). Recognizing positive reactivity in smaller tumor cells with less cytoplasm was more challenging than in large pleomorphic cells with ample cytoplasm. Perinuclear halos were another challenge for interpretation and were present in 6% of tumors (Fig.Ā 3D).

Fig. 3: Challenges in interpretation of MTAP immunohistochemistry.
figure 3

a An IDH-mutant astrocytoma showing positive MTAP staining in fine cytoplasmic processes with weakly positive internal normal controls (arrow head). b Loss of MTAP staining in an IDH-mutant astrocytoma showing weak positivity for the internal normal controls (arrow head). c An IDH-mutant astrocytoma showing loss of MTAP staining with intermixed non-neoplastic reactive astrocytes retaining MTAP positivity. d An IDH-mutant oligodendroglioma, with perinuclear halos that made MTAP interpretation challenging. The case was interpreted as MTAP deficient, although the tumor lacked CDKN2A HD.

Full size image

p16 immunohistochemistry

Loss of nuclear expression of p16 was observed in 57% (102/178) of all gliomas evaluated, and loss of cytoplasmic expression was seen in 51% (90/178) of glioma specimens. Loss of nuclear and cytoplasmic expression of p16 was observed in 42% (32/77) and 40% (31/77) of IDH-mutant astrocytomas, respectively, 77% (10/13) and 77% (10/13) of IDH-mutant oligodendroglioma, respectively, and 68% (60/88) and 56% (49/88) of IDH-wildtype glioblastomas, respectively (TableĀ 2, Supplementary Fig.Ā 2). Among the IDH-mutant astrocytomas, p16 nuclear and cytoplasmic immunohistochemistry showed sensitivities of 94% and 94%, respectively, and specificities of 73% and 75%, respectively (TableĀ 2). Similarly, for IDH-wildtype glioblastomas, p16 nuclear and cytoplasmic immunohistochemistry showed sensitivities of 91% and 78%, respectively, and specificities of 70% and 82%, respectively. The parameters of statistical relevance for IDH-mutant oligodendrogliomas were even poorer (TableĀ 2). The combination of MTAP and p16 did not provide superior accuracy to MTAP staining alone for predicting CDKN2A HD in all gliomas (nā€‰=ā€‰178), IDH-mutant astrocytomas (nā€‰=ā€‰77), IDH-mutant oligodendrogliomas (nā€‰=ā€‰13), or IDH-wildtype glioblastomas (nā€‰=ā€‰88).

Survival analysis

Among the 178 tumors evaluated in the study, three pairs of IDH-mutant astrocytomas and one pair of IDH-mutant oligodendrogliomas were resected from the same patients at different times. Hence, the cohort was composed of 174 unique patients. The initially resected tumors from those four pairs were used for survival analysis with the subsequently resected specimens being excluded. All patients were treated with a combination of chemotherapy and radiotherapy in addition to surgical resection or biopsy. Survival data were available for all patients, including 74 patients with IDH-mutant astrocytoma, 12 patients with IDH-mutant oligodendroglioma, and 88 patients with IDH-wildtype glioblastomas. The median age of all 174 patients at diagnosis was 51.5 years (range, 17ā€“87ā€‰y). The cohort included 97 men and 77 women. One teenage patient was included in the analysis as the tumor was of adult-type with IDH mutation. During the median follow-up of 17.2 months (range, 0.8ā€“193.8 months), 118 patients died.

In the IDH-mutant astrocytoma group, both CDKN2A HD and MTAP immunohistochemical deficiency were significantly poor prognostic factors for OS (Pā€‰<ā€‰0.001 each; Fig.Ā 4A, B). In contrast, neither CDKN2A HD nor MTAP immunohistochemical deficiency stratified survival of the patients with IDH-mutant oligodendroglioma (Pā€‰=ā€‰0.799 and Pā€‰=ā€‰0.671, respectively; Fig.Ā 4C, D). This was similar for patients with IDH-wildtype glioblastomas (Pā€‰=ā€‰0.158 and Pā€‰=ā€‰0.484, respectively; Fig.Ā 4E, F).

Fig. 4: Kaplanā€“Meier analysis and log-rank tests regarding overall survival (OS) of patients with IDH-mutant astrocytoma, IDH-mutant oligodendroglioma, and IDH-wildtype glioblastoma stratified by CDKN2A HD and MTAP staining status.
figure 4

a Among patients with IDH-mutant astrocytomas, shorter OS correlated with CDKN2A HD. b Shorter OS also correlated with MTAP immunohistochemical deficiency in patients with IDH-mutant astrocytomas. In contrast, OS was not significantly stratified by CDKN2A HD (c) or MTAP immunohistochemical deficiency (d) in IDH-mutant oligodendrogliomas. OS was also not significantly stratified by CDKN2A HD (e) or MTAP immunohistochemical deficiency (f) in IDH-wildtype glioblastomas.Ā Survival time is in months.

Full size image

For IDH-mutant astrocytomas, WHO grade was a significant prognostic factor for both OS (Pā€‰=ā€‰0.0237; Fig.Ā 5A) and PFS (Pā€‰=ā€‰0.0331; Fig.Ā 5D). However, when the analysis was limited to WHO grades IIā€“III only, the survival differences did not reach statistical significance for either OS (Pā€‰=ā€‰0.307) or PFS (Pā€‰=ā€‰0.0551), although grade II tended to do better than grade III (Fig.Ā 5A, D). In this WHO grade IIā€“III subgroup, both CDKN2A HD and MTAP loss remained significant prognostic factors for OS (CDKN2A HD: Pā€‰=ā€‰0.0124; Fig.Ā 5B; MTAP: Pā€‰<ā€‰0.01; Fig.Ā 5C) and PFS (CDKN2A HD: Pā€‰<ā€‰0.01; Fig.Ā 5E, MTAP: Pā€‰<ā€‰0.01; Fig.Ā 5F). Among IDH-mutant glioblastomas, MTAP deficiency was associated with a marginally significant shorter OS (Pā€‰=ā€‰0.0495, Fig.Ā 5C) and a trend toward shorter PFS (Pā€‰=ā€‰0.165, Fig.Ā 5F), while CDKN2A HD was associated with a trend toward both shorter OS (Pā€‰=ā€‰0.128, Fig.Ā 5B) and PFS (Pā€‰=ā€‰0.428,Ā Fig. 5E). Notably, IDH-mutant lower-grade astrocytoma with CDKN2A HD behaved similarly to IDH-mutant glioblastoma, with their differences not being significant for either OS (Pā€‰=ā€‰0.654; Fig.Ā 5B) or PFS (Pā€‰=ā€‰0.713; Fig.Ā 5E). Likewise, IDH-mutant lower-grade astrocytoma with deficient MTAP immunoreactivity showed overlapping unfavorable behavior with IDH-mutant glioblastoma, with their differences also not being significant for either OS (Pā€‰=ā€‰0.572; Fig.Ā 5C) or PFS (Pā€‰=ā€‰0.646; Fig.Ā 5F).

Fig. 5: In-depth survival analysis of the IDH-mutant astrocytoma cohort.
figure 5

Kaplanā€“Meier analysis and log-rank tests of overall survival (OS) (a) and progression free survival (PFS) (d) among patients with IDH-mutant astrocytoma in relation to WHO grade. OS (b) and PFS (e) were significantly stratified by CDKN2A HD in lower-grade (grades IIā€“III) IDH-mutant astrocytoma. OS (c) and PFS (f) were significantly stratified by loss of MTAP immunoreactivity in lower-grade IDH-mutant astrocytoma (grades IIā€“III). Note that neither OS nor PFS was significantly different between IDH-mutant glioblastoma and IDH-mutant lower-grade astrocytoma with CDKN2A HD (b, e) or loss of MTAP immunoreactivity (c, f).Ā Survival time is in months.

Full size image

The results of p16 immunohistochemistry failed to provide significant prognostic stratification of any groups of patients with diffuse gliomas. The p16 immunohistochemistry results are shown in Supplementary Fig.Ā 3.

Comparison with an alternative MTAP antibody (clone 2G4)

We performed MTAP staining using clone 2G4 on the same 178 samples and compared the performance with that of EPR6893. The scoring results were concordant in all but one sample (177/178, 99%). The discrepant case was an IDH-mutant glioblastoma with CDKN2A HD, in which MTAP immunohistochemistry was scored as retained using clone 2G4 but as lost using clone EPR6893. Hence, among the IDH-mutant astrocytomas, the performance parameters of clone 2G4 were slightly lower (sensitivityā€‰=ā€‰82%, specificityā€‰=ā€‰98%, PPVā€‰=ā€‰93%, NPVā€‰=ā€‰95%) than those of EPR6893 (sensitivityā€‰=ā€‰88%, specificityā€‰=ā€‰98%, PPVā€‰=ā€‰94%, NPVā€‰=ā€‰97%).

Regarding the level of interpretational difficulty based on whole tissue sections only, it was considered as low in 66% (65/98) of cases using 2G4 and in 61% of cases (60/98) using EPR6893. Specifically, 2G4 enabled easier interpretation in 34% (13/38) of specimens that were hard to interpret using EPR6893, but the interpretation became harder with 2G4 in 13% (8/60) of samples that were easy to interpret using EPR6893. The specific challenges associated with 2G4 were similar to those associated with EPR6893, but overall, 2G4 provided a stronger staining and fewer cases associated with weak internal positive control cells and heterogeneous staining intensity than did EPR6893. Distinguishing staining of tumor cells from that of reactive glia and interpretation of cells with perinuclear halo were equally challenging for both 2G4 and EPR6893.

Discussion

Discovery of the IDH mutation has revolutionized the diagnosis and management of diffuse glioma in adults as the mutation portends significantly better prognosis and response to therapy [45, 46]. However, risk stratification of patients with IDH-mutant infiltrating astrocytomas remains an issue, with the assessment of lower-grade tumors being particular problematic [10, 11, 14, 45, 47,48,49,50,51]. CDKN2A HD has been reported as one of the most promising prognostic biomarkers in IDH-mutant lower-grade astrocytomas [13], IDH-mutant astrocytic tumors of all grades [14], and IDH-mutant astrocytic and oligodendroglial tumors [15]. Our current data are in line with this emerging view. We revealed a significant correlation between CDKN2A HD and OS in the IDH-mutant astrocytoma group. Moreover, when we focused the analysis on WHO grades IIā€“III IDH-mutant astrocytoma, CDKN2A HD remained a significant adverse prognostic factor, while histological grade failed to significantly stratify prognosis. Furthermore, IDH-mutant lower-grade astrocytoma with CDKN2A HD exhibited overlapping survival with IDH-mutant glioblastoma in our cohort, which, albeit based on a small number, supports the cIMPACT-NOW update 5 recommendation that CDKN2A HD in IDH-mutant astrocytoma should be given a grade IV designation, even in the absence of necrosis or microvascular proliferation [18]. In contrast to a previous report [17], our study failed to demonstrate a significant survival difference among patients with IDH-mutant glioblastoma in relation to CDKN2A HD. This may have been due to the smaller size of the cohortĀ evaluated.

In our current study, we evaluated the utility of MTAP immunohistochemistry as a proxy for CDKN2A HD in the assessment of diffuse glioma. We showed that the loss of MTAP staining provided a high concordance with CDKN2A HD among IDH-mutant astrocytomas (sensitivityā€‰=ā€‰88%, specificityā€‰=ā€‰98%) and IDH-wildtype glioblastomas (sensitivityā€‰=ā€‰89%, specificityā€‰=ā€‰100%). In addition, MTAP immunohistochemistry enabled a comparable degree of prognostic stratification as that of CDKN2A HD for the entire IDH-mutant astrocytoma and IDH-mutant lower-grade astrocytoma subgroup. Similar to CDKN2A HD, loss of MTAP staining in IDH-mutant lower-grade astrocytoma indicated an unfavorable survival, overlapping with that of IDH-mutant glioblastoma. These results suggest that MTAP immunohistochemistry may serve as a useful surrogate for CDKN2A molecular testing of infiltrating astrocytoma.

Our study also revealed several caveats in the use of MTAP immunohistochemistry in the assessment of diffuse glioma. First, there were a few cases of discordance between MTAP staining and CDKN2A HD. MTAP staining loss had overall lower sensitivity than specificity, which is consistent with published data that CDKN2A HD occurs more commonly than MTAP HD, with a small subset harboring only CDKN2A HD. It has been reported that in 230 diffuse gliomas with CDKN2A HD, MTAP copy number is retained in 10% of the specimens, while MTAP deletion without CDKN2A HD is rare (0.5%) [41], according to analysis using the cBioPortal database [42, 43]. In our current study, two IDH-mutant glioblastomas and six IDH-wildtype glioblastomas retained MTAP staining, despite CDKN2A HD. These cases may have harbored only CDKN2A HD. An alternative explanation for two of these cases is subclonal CDKN2A deletion, as the CDKN2A status was evaluated via MLPA on fresh frozen tissues whose origin may not match the exact site of samples used for IHC. However, in the remaining six cases, CDKN2A status was obtained via FISH on the very tumor sites showing retained MTAP expression, which makes it unlikely that discordance was caused by subclonal genetic changes. Conversely, the only IDH-mutant AA that was MTAP deficient, despite intact CDKN2A (determined by MLPA), may have harbored only MTAP HD as that case showed a loss of 9p21 signals by FISH. Overall, our data suggest that CDKN2A molecular testing of astrocytoma may be omitted when MTAP staining is lost, but there may be a small chance of discordance when MTAP expression is retained. Whether such a minor discordance should always be resolved using confirmatory CDKN2A molecular testing may be a subject of debate as MTAP staining itself provided a comparable degree of prognostic stratification to CDKN2A HD in IDH-mutant astrocytoma. Although MTAP HD is often perceived as a passenger genetic variant to CDKN2A HD, recent data suggest that MTAP dysfunction itself may actively contribute to gliomagenesis via epigenetic alteration [52]. The clinical impact of discordant CDKN2A/MTAP status in glioma is currently unclear, in part due to the relatively small number of events. Future validation using a larger cohort of IDH-mutant astrocytoma would be necessary to firmly establish the role of MTAP staining in diagnostic algorithms.

Second, the interpretation of MTAP immunohistochemistry was not always straightforward. The results were equivocal in a fifth of the samples initially tested using TMA, requiring re-testing using whole tissue sections. Even then, the staining remained uninterpretable in 7% of the specimens. Overall, MTAP immunohistochemistry was considered easy to interpret (i.e., low difficulty) in only 61% of the specimens successfully scored, with the remaining specimens presenting variable degrees of difficulty. Specific challenges included weak staining of internal positive controls, artifactual heterogeneous labeling, perinuclear halos, tenuous tumor cytoplasm in low-grade tumors, and differentiating tumor cells from admixed non-neoplastic elements. This could be partly alleviated using an alternative clone such as 2G4, which is becoming more favored than EPR6893 in the field of malignant mesothelioma [36], because it provided a stronger staining with a larger proportion (66%) of cases becoming easy to interpret, while maintaining a virtually comparable power to predict CDKN2A HD. Nonetheless, it had some similar challenges to EPR6893, and, in a small number of samples, the scoring was considered more difficult than when using EPR6893. Although sample size, tumor cell fraction, and tissue quality dictate success rates for any biomarker testing, we believe that these variables affected MTAP immunohistochemistry more profoundly than did many other immunohistochemical antibodies that are routinely used in practice, including IDH1 (R132H) immunohistochemistry. Future studies are required to determine whether different combinations of staining platforms and/or antibody clones may resolve these issues, and interobserver variability studies would also be necessary before translating MTAP staining into practice.

The third caveat concerns the poor performance of MTAP immunohistochemistry in IDH-mutant oligodendrogliomas. This likely was primarily related to the fact that the staining was less reliable for scoring oligodendroglioma than for scoring astrocytoma. Oligodendroglioma nuclei are surrounded by an artifactual perinuclear halo with only a small amount of cytoplasm [53]. Therefore, cytoplasmic MTAP staining may be scarcely visible and dismissed, even if MTAP is expressed. The practical impact of this limitation is currently uncertain as the prognostic implication of CDKN2A HD in oligodendroglioma is controversial [13, 15, 54] and CDKN2A molecular assays have not been recommended in this setting.

We also found that p16 staining was of limited use in predicting CDKN2A HD. Consistent with previous reports [20, 21, 23], loss of p16 staining had low specificity. Consequently, neither nuclear nor cytoplasmic staining was found to be prognostic for diffuse glioma. Low specificity may be caused by CDKN2A being inactivated by other mechanisms or due to the antibody having low epitope affinity [55].

A minor, yet novel histological observation was that microcystic change inversely correlated with the presence of CDKN2A HD in IDH-mutant astrocytoma. Microcystic change has been reported as a favorable prognostic factor in glioblastoma [56] and is often observed in slowly growing gliomas [5]. The absence of microcystic change in tumors with CDKN2A HD might be a histological correlate of more aggressiveness.

In conclusion, we showed that the loss of MTAP immunoreactivity could serve as a reasonable surrogate for predicting CDKN2A HD in IDH-mutant astrocytomas and IDH-wildtype glioblastomas. MTAP immunohistochemical deficiency may provide a significant prognostic value for IDH-mutant astrocytoma, comparable to CDKN2A HD. Despite several caveats, MTAP immunohistochemistry may be applicable to diagnostic practices of brain tumors, as a rapid, economical, and accessible tool.