CDKN2A Loss Is Associated with Shortened Overall Survival in Lower Grade (World Health Organization II-III) Astrocytomas

Abstract

Lower grade (WHO II-III) gliomas vary widely in clinical behavior and are classified as astrocytic, oligodendroglial, or mixed forms. Anaplasia depends greatly on mitotic activity, with CDKN2A loss considered the most common mechanism for cell cycle dysregulation. We investigated whether loss of the CDKN2A gene is associated with overall survival across pathologically and genetically defined glioma subtypes. After adjusting for IDH mutation, sex, and age, CDKN2A deletion was strongly associated with poorer overall survival in astrocytomas but not in oligodendrogliomas or oligoastrocytomas. Molecular classification of astrocytomas by IDH mutation, TP53 mutation, and/or ATRX loss of expression revealed that CDKN2A loss in IDH/TP53 mutated tumors was strongly associated with worse overall survival. CDKN2A loss in IDH mutated tumors with ATRX loss was only weakly associated with worse overall survival. These findings suggest that CDKN2A testing may provide further clinical aid in lower-grade glioma sub-stratification beyond IDH mutation and 1p19q codeletion status, particularly in IDH/TP53 mutated astrocytomas.

INTRODUCTION

Infiltrating gliomas comprise approximately 60% of primary malignant intracranial tumors (1) and have a devastating course given their widespread invasiveness, tendency toward biological progression, and resistance to available adjuvant therapies. The World Health Organization (WHO) classifies infiltrating gliomas as diffuse (WHO grade II) astrocytoma (A-II), oligodendroglioma (O-II), or oligoastrocytoma (OA-II), anaplastic (WHO grade III) astrocytoma (A-III), oligodendroglioma (O-III), or oligoastrocytoma (OA-III), and glioblastoma ([GBM]; WHO grade IV) (2). While diagnostic reproducibility among neuropathologists is high for GBM, there is far lower concordance for cell type determination in other subtypes and in the distinction between WHO grades II and III. The latter difficulty stems in part from different grading criteria for astrocytic and oligodendroglial tumors, vagaries in WHO grading criteria, and inter-observer variability in the detection of mitotic figures and microvascular proliferation. Additionally, whereas molecular criteria have been published recently to distinguish diffuse astrocytomas from oligodendrogliomas in adult patients more objectively, it remains unclear whether previously devised grading criteria based solely on morphology can be appropriately applied to newly defined glioma subtypes, which are based mostly on molecular definitions (3). As such, the development of clinically useful “grading” biomarkers for molecularly defined glioma subsets is sorely needed to improve diagnostic reproducibility.

The large majority of grade II and III gliomas, as well as so-called secondary GBM that derive from lower grade gliomas, harbor mutations in an IDH gene, with IDH1 being most common (4–7). Mutations in TP53 and ATRX also occur frequently and are present in approximately 60% to 70% of tumors of astrocytic differentiation (2). In a recent study of lower-grade gliomas, ATRX mutation was closely associated with TP53 mutation and was restricted to IDH mutated tumors (8). A consequence of ATRX mutation is activation of the alternative lengthening of telomeres pathway and telomerase-independent immortalization of tumor cells (9, 10). Conversely, approximately 70% of oligodendrogliomas harbor 1p19q codeletion and lack TP53 and ATRX alterations (2, 8, 11–14). Approximately 80% of oligodendrogliomas have mutations in the TERT gene promoter, which reactivates telomerase activity (15). Mixed oligoastrocytomas remain poorly defined molecularly, with the vast majority showing classic molecular features of either astrocytoma or oligodendroglioma (16).

In addition to the alterations described above, homozygous and hemizygous losses involving 9p21 have been observed at high frequency in infiltrating gliomas, with homozygous loss being most common (17–20). One of the consequences of 9p21 deletion is loss of the cyclin-dependent kinase inhibitor CDKN2A gene, which results in cellular proliferation and dysregulation of pro-apoptotic pathways (21). Cell cycle progression from G1 to S phase relies on complex formation between cyclin dependent-kinases CDK4 or CDK6 and D-type cyclins, which subsequently leads to phosphorylation of the retinoblastoma (RB1) protein, release of the elongation factor (EF2) transcriptional factor, and activation of genes involved in G₁ to S transition (22, 23). Alterations in this pathway appear to be rare in grade II gliomas but are frequent in higher-grade (III and IV) tumors, suggesting a key role for the CDKN2A–CDK4-RB pathway in malignant progression (24–26). Since the main criteria for anaplastic designation in gliomas are either the presence of mitoses (astrocytomas) or high mitotic activity (oligodendroglial tumors), loss of the CDKN2A gene or p16 protein (the CDKN2A product) appears an ideal candidate for distinguishing the molecular phenotypes of WHO grade II and III gliomas.

Several studies have reported worse prognosis for CDKN2A loss in gliomas (27–33). However, these studies were conducted before the realization that IDH mutation is an independent favorable prognostic factor in adult type diffuse gliomas (6, 34, 35). Therefore, it is unclear whether CDKN2A loss remains a statistically significant prognostic factor after correcting for the effect of IDH mutation. Furthermore, prior studies have not explored the effect of CDKN2A loss on the newly defined ‘molecular’ grade II-III astrocytomas, i.e. IDH mutated, TP53 mutated, often with associated loss of ATRX protein expression, or ‘molecular’ grade II-III oligodendrogliomas, i.e. that are IDH mutated and 1p19q codeleted. Our study addresses these questions by testing the hypothesis that CDKN2A loss determined by fluorescence in situ hybridization (FISH) and/or loss of p16 protein expression determined by immunohistochemistry (IHC) can enhance prognostication of overall survival in molecularly-characterized lower-grade (WHO II-III) adult type diffuse gliomas. If so, these molecular biomarkers could potentially enhance future WHO grading criteria and improve inter-observer concordance rates among pathologists.

MATERIALS AND METHODS

Study Participants and Selection Criteria

Cases were selected from among participants of the San Francisco Bay Area Adult Glioma Study (AGS), which was conducted at the University of California, San Francisco, as previously described (36, 37). Briefly, the AGS enrolled patients who were newly diagnosed with a histologically confirmed glioma at age ≥18 years between 1991 and 2012. All participants gave informed consent, and the study was conducted under protocols approved by the UCSF Institutional Review Board.

The following criteria were used to select cases from the AGS for this analysis: Only cases classified as grade II or III astrocytoma, oligodendroglioma, or oligoastrocytoma during the study’s neuropathology review were included. In addition, only cases with sufficient available tissue for the planned assays were included. Because these subjects were also a subset of another study relating tumor markers to inherited single nucleotide polymorphism data, subjects were of European ancestry and had constitutive DNA available. Cases were prioritized with respect to available mutational status for isocitrate dehydrogenase (IDH1 or IDH2) obtained by DNA sequencing as previously described (37, 38). Tumor TP53 mutation was also available for many of the subjects and assayed as described previously (39).

FISH Analysis

We assessed CDKN2A copy number alterations by FISH using commercially available probes (LSI CDKN2A (9p21) orange and CEP9 green Spectrum; Abbott Laboratories, North Chicago, IL). In brief, green and orange fluorescent signals were enumerated under an Olympus BX60 fluorescence microscope with appropriate filters (Olympus, Melville, NY). One hundred non-overlapping nuclei were assessed for numbers of green and red signals, and cases were considered deleted regardless of whether the pattern suggested hemizygous or homozygous loss. An interpretation of deletion was made when the orange to green ratio was <0.8. Partial hybridization failure was ruled out by evaluating orange and green signals within vascular endothelial cells as an internal control. Failed or weak hybridizations were repeated; uninterpretable cases were considered “non-informative.” Interpretable results were available for 253 cases (93.7%).

FISH images were captured using a black and white, high-resolution COHU CCD camera, Z-stack motor and a CytoVision basic workstation (Applied Imaging, Santa Clara, CA) with sequential DAPI (1 level), FITC (5 levels), and rhodamine (5 levels) filter settings. The resulting images were reconstituted with blue, green, and orange pseudocolors using CytoVision software. 1p19q deletion status was assayed according to published methods (40, 41). Tumors classified histologically as astrocytoma were not assayed for 1p19q deletion.

Immunohistochemistry

Immunohistochemistry for ATRX and p16 was performed at the Brain Tumor Research Center and Clinical Immunohistochemistry Laboratories, University of California San Francisco, utilizing an automated staining processor (Ventana, Tucson, AZ). ATRX loss of expression was assessed using the rabbit polyclonal antibody (HPA001906; 1:100, Sigma, St. Louis, MO). Only cases with retained staining of endothelial cells (internal positive control) were considered interpretable. For p16 expression, the Mtm Lab kit (Ventana) and the JC8 mouse antibody (1:100, Santa Cruz Biotechnology, Dallas, TX) were used, with the majority of cases evaluated using the Mtm Lab kit. Quantitative analysis of the staining pattern for 10 cases using both antibodies demonstrated a high correlation (r² = 0.8; p = 0.0004). Nuclear staining was required for a positive p16 result.

Image acquisition for p16 quantification was done using an Olympus BX-41 microscope, 20X objective, and Olympus DP72 digital camera. Areas of maximal staining were photographed for assessing labeling index. JPEG images were imported into Image J (http://imagej.nih.gov/ij/), and the ImmunoRatio plug-in (http://153.1.200.58/sites/default/files/software/immunoratio-plugin/index.html) was used to determine labeling index. Approximately 1000 cells were required for an interpretable result.

Statistics and Survival Analysis

Comparisons among categorical variables were made using chi-square and Fisher exact tests. Cox proportional hazards regression was employed to estimate overall survival. Models built based on histologic subtype (astrocytoma, oligodendroglioma, oligoastrocytoma) or pathologic diagnosis (A-II, A-III, O-II, O-III, AO-II, AO-III) were adjusted for sex, age, and IDH status while models built based on molecular parameters (IDH, TP53, ATRX, 1p19q status) were adjusted only for sex and age. Molecular subtypes of astrocytoma were defined based on the presence of IDH/TP53 mutations (IDH/TP53 group), combined IDH/TP53 mutations with ATRX loss (IDH/TP53/ATRX group), or IDH mutation and ATRX loss (IDH/ATRX group). Molecular oligodendroglioma was defined as IDH mutated, 1p19q codeleted tumors. A logistic analysis was used to evaluate the association between CDKN2A loss by FISH and p16 labeling index by IHC, stratified by histologic subtype or pathologic diagnosis. The area under the curve was obtained from the receiver operating characteristic curve (ROC), which provides a graphical representation of the relationship between false-positive and true-positive rates (true-positive as defined by FISH). The optimal labeling index for p16 was extracted from the ROC table and used to dichotomize the continuous p16 labeling index data into "deleted" and "intact" groups. Results with a p value <0.05 were deemed significant.

RESULTS

Patient Characteristics

The clinical and pathological characteristics of the 270 study subjects are summarized in Table 1. The percentages of astrocytomas and oligodendrogliomas were similar (41.9% and 38.5%, respectively); tumors with mixed histology comprised 19.6% of cases. The median follow-up time was 10.6 years (mean: 10.5 years; range 0.2–22.3 years), with 105 deaths (38.9%) and 165 patients (61.1%) alive at last follow-up.

Table 1.

Distribution of histologic subtype, pathologic diagnosis, demographic, and clinical characteristics of glioma patients from the UCSF Adult Glioma Study.

	N	Percent	Mean age (y)	SE (y)	Median age (y)
All Cases	270	100.0%	42.2	0.7	41.0
Histology
Astrocytoma	113	41.9%	42.4	1.3	41.0
Oligodendroglioma	104	38.5%	44.5	1.0	44.5
Oligoastrocytoma	53	19.6%	36.9	1.3	36.0
Pathology
Diffuse astrocytoma (A-II)	49	18.2%	41.4	1.9	40.0
Anaplastic astrocytoma (A-III)	64	23.7%	43.2	1.8	41.5
Oligodendroglioma (O-II)	80	29.6%	43.9	1.2	44.0
Anaplastic oligodendroglioma (O-III)	24	8.9%	46.8	2.0	51.0
Oligoastrocytoma (OA-II)	42	15.6%	37.7	1.6	36.0
Anaplastic oligoastrocytoma (OA-III)	11	4.1%	34.1	2.4	32.0
Gender
Male	147	54.4%
Female	123	45.6%
Surgery
Biopsy	20	7.4%
Resection	250	92.6%
Adjunct therapy
Chemotherapy
No	107	40.7%
Yes	154	58.6%
Unknown	2	0.8%
Temozolomide	129	47.8%
Radiotherapy
No	127	47.0%
Yes	143	53.0%
Outcome
Alive	165	61.1%
Dead	105	38.9%

CDKN2A Loss Occurred More Frequently in Astrocytoma than Other Histologic Subtypes and Was Present in Relatively Similar Percentages in A-II and A-III

Examples of CDKN2A results by FISH are illustrated in Figure 1. The frequency of loss for each group is summarized in Table 2 and illustrated in Supplementary Figure 1. Astrocytomas demonstrated the highest frequency of CDKN2A loss (astrocytoma: 52/108 cases (55.3%); oligodendroglioma: 23/96 cases (24.5%); oligoastrocytoma: 19/49 cases (20.2%); p = 0.006, chisquare). No significant difference was seen in the relative percentage of deleted cases in A-II vs. A-III tumors (A-II: 21/47 cases (44.7%); A-III: 31/61 cases (50.8%); p = 0.6, Fisher exact test).

Figure 1. Fluorescence in situ hybridization for CDKN2A.

(A, B) Results of fluorescence in situ hybridization demonstrating intact (A: CEP9/CDKN2A ratio near one) and deleted (B) CDKN2A status. In the latter case, a non-neoplastic cell containing 2 green CEP9 signals and 2 red CDKN2A signals is seen (white arrow), whereas all the tumor cells have only CEP9 signals.

Table 2.

Distribution of CDKN2A, IDH, ATRX, TP53 and 1p19q status by pathologic diagnosis in tumors from glioma patients in the UCSF Adult Glioma Study

	All cases		Diffuse astrocytoma (A-II)		Anaplastic astrocytoma (A-III)		Oligodendroglioma (O-II)		Anaplastic oligodendroglioma (O-III)		Oligoastrocytoma (OA-II)		Anaplastic oligoastrocytoma (OA-III)
	N	Percent	N	Percent	N	Percent	N	Percent	N	Percent	N	Percent	N	Percent
CDKN2A
Intact	159	63%	26	55%	30	49%	57	77%	16	73%	25	63%	5	56%
Deleted	94	37%	21	45%	31	51%	17	23%	6	27%	15	38%	4	44%
Total	253		47		61		74		22		40		9
IDH
Wild type	65	24%	10	21%	26	41%	9	11%	8	33%	8	19%	4	36%
Mutated	202	76%	37	79%	37	59%	71	89%	16	67%	34	81%	7	64%
Total	267		47		63		80		24		42		11
ATRX
Intact	156	60%	22	47%	24	39%	68	86%	20	87%	18	44%	4	36%
Lost	106	40%	25	53%	37	61%	11	14%	3	13%	23	56%	7	64%
Total	262		47		61		79		23		41		11
TP53
Wild type	107	63%	9	31%	24	48%	42	89%	14	82%	13	65%	5	63%
Mutated	64	37%	20	69%	26	52%	5	11%	3	18%	7	35%	3	38%
Total	171		29		50		47		17		20		8
1p19q
Intact	65	42%	NA*		NA*		19	25%	5	23%	29	74%	8	89%
Codeleted	89	58%					58	75%	17	77%	10	26%	1	11%
Total	154						77		22		39		9

^*NA, 1p19q data not available for tumors classified histologically as astrocytoma.

The results for p16 expression by IHC are summarized in Supplementary Table 1. The p16 labeling index was highly variable across cases within a given histopathologic group.

CDKN2A Loss Was Associated with Poorer Survival in Astrocytomas But Not Oligodendrogliomas or Oligoastrocytomas

As hypothesized, CDKN2A loss determined by FISH was associated with significantly worse overall survival in grade II and III gliomas after adjusting for age, sex, and IDH mutation (Table 3; hazard ratio [HR] = 1.6, 95% confidence interval [CI] = 1.0–2.4, p = 0.03). This result was mostly related to a strong adverse effect of CDKN2A loss among astrocytoma patients; HR = 2.0, 95% CI = 1.1–3.5 (Table 3; Fig. 2A–D). Results were similar for grade II and III astrocytomas (Table 3). There were non-significant inverse associations of CDKN2A loss and overall survival among patients with oligodendroglioma or oligoastrocytoma (Table 3). Thus, CDKN2A loss was associated with poor survival among patients with astrocytoma, but this effect was not observed among patients with oligodendroglioma or oligoastrocytoma.

Table 3.

Association of CDKN2A loss by fluorescence in situ hybridization with overall survival stratified by histologic subtype, pathologic diagnosis, and molecular subtypes using Cox proportional hazards models in glioma patients from the UCSF Adult Glioma Study

	CDKN2A intact				CDKN2A deleted
	N	N deceased	Mean overall survival (years)	Median overall survival (years)	N	N deceased	Mean overall survival (years)	Median overall survival (years)	HR	95% CI	p-value	HR	95% CI	p-value
									Unadjusted			Adjusted for age and sex
All cases*	159	56	10.3	11.8	94	42	8.8	7.2	1.7	1.1–2.5	0.01a	1.6	1.0–2.4	0.03
Astrocytoma*	56	23	8.7	9.5	52	31	5.5	4.4	2.6	1.5–4.6	0.0008a	2.0	1.1–3.5	0.02
Oligodendroglioma*	73	20	11.1	15.4	19	4	7.0		0.7	0.2–1.7	0.4	0.7	0.2–2.00	0.5
Oligoastrocytoma*	30	13	9.8	10.6	23	7	11.5	17.2	0.9	0.3–2.2	0.7	0.8	0.3–2.4	0.7
A-II*	26	9	10.2	12.8	21	9	6.6	6.4	3.9	1.4–11.2	0.01a	2.7	0.8–8.8	0.1
A-III*	30	14	5.0	6.9	31	22	4.7	3.3	2.1	1.1–4.2	0.03a	1.8	0.8–3.6	0.1
O-II*	57	14	9.4	11.8	17	2	7.6	NA	0.4	0.07–1.6	0.3	0.4	0.07–1.5	0.2
O-III*	16	6	11.2	15.4	6	2	1.9	NA	1.4	0.2–6.6	0.7	1.6	0.2–9.9	0.6
OA-II*	25	12	9.6	10.6	15	5	12.3	17.2	0.7	0.2–2.0	0.5	1.0	0.3–3.1	0.9
OA-III*	5	1	1.0	NA	4	2	3.4	3.7	2.8	0.3–60.0	0.4	0.2	0.0004–21.2	0.6
IDH/TP53 mutated, ATRX lost	21	10	10.2	12.9	21	14	6.4	4.8	3.3	1.4–8.8	0.008a	3.4	1.4–8.4	0.008a
IDH/TP53 mutated	28	12	10.6	12.8	22	15	6.3	4.8	4.3	1.8–10.1	0.001a	4.4	1.8–10.3	0.0008a
IDH mutated, ATRX lost	49	19	9.6	9.2	39	18	8.7	7.8	1.6	0.8–3.0	0.2	1.6	0.8–3.0	0.2
IDH mutated, 1p19q codeleted**	56	16	9.2	10.6	16	0	NA	NA	NA	NA	NA	NA	NA	NA

HR, Hazard ratio; CI, Confidence interval; NA, Median survival is not available due to lack of events.

^*Hazard ratios for histologic groups and pathologic diagnosis were adjusted for IDH mutation status in addition to age and sex.

^**These tumors were also ATRX intact.

A-II, Diffuse astrocytoma; A-III, Anaplastic astrocytoma; O-II, Oligodendroglioma, O-III, Anaplastic oligodendroglioma, OA-II, Oligoastrocytoma; OA-III, Anaplastic oligoastrocytoma.

^astatistical significance at the 0.05 level.

Figure 2.

Effect of CDKN2A loss on overall survival of histologically classified lower grade (World Health Organization II-III) gliomas. (A–D) CDKN2A loss was associated with worse overall survival in lower grade gliomas not stratified by histology (A) and in astrocytomas (B), but not in oligodendrogliomas (C) or oligoastrocytomas (D).

Logistic analysis demonstrated a poor association between CDKN2A deletion by FISH and p16 expression by IHC (Supplementary Table 2). Furthermore, the loss of p16 expression by IHC was weakly associated with poor overall survival in lower-grade (II-III) gliomas (Supplementary Table 3).

CDKN2A Loss Was Strongly Associated With Poor Survival in Certain Molecular Subsets of Astrocytoma

As an alternative to using the histologic classification, we evaluated the effect of CDKN2A loss on the overall survival of certain molecular subsets of grade II-III astrocytoma, defined as IDH mutated tumors with both TP53 mutation and ATRX loss (termed IDH/TP53/ATRX group), or either combined with TP53 mutation (termed IDH/TP53 group) or ATRX loss (termed IDH/ATRX group) (Table 2; Supplementary Fig. 1). The co-occurrence of IDH, TP53, and ATRX alterations is provided in Table 4. Among 123 IDH mutated tumors with known TP53 and ATRX status, 42 cases (34.1%) harbored both mutation in TP53 and loss of ATRX expression, 21 harbored only ATRX loss (17.1%), and 9 harbored only TP53 mutation (7.3%) (Table 4). Fifty-one cases (41.5%) showed no alteration in either TP53 or ATRX, of which 38 (74.5%) were 1p19q codeleted “molecular oligodendroglioma.” Overall, a strong association was noted for the co-occurrence of TP53 mutation and ATRX loss (p < 0.0001, Fisher exact test). Similarly, a strong association was found for the co-occurrence of IDH mutation and ATRX loss both in IDH mutated cases with 1p19q intact status (p = 0.0004, Fisher exact test; Table 4) and IDH mutated cases not evaluated for 1p19q since they were classic astrocytomas on histopathology (p < 0.0001; Fisher exact test; Table 4); the latter was similar to observations made in IDH/TP53 mutated cases. These data support a strong association for the co-occurrence of IDH mutation with TP53 mutation and/or ATRX loss in grade II-III astrocytomas.

Table 4.

Distribution of IDH/TP53 mutation status and ATRX loss of expression in tumors of patients from the UCSF Adult Glioma Study

IDH mutated cases with both ATRX and TP53 data available (n = 123)
ATRX	TP53	N	Percent	Fisher Exact test p-value
Intact	Wild type	51	41.5%	<0.0001*
Intact	Mutated	9	7.3%
Lost	Wild type	21	17.1%
Lost	Mutated	42	34.1%
Total		123
Cases with both IDH and ATRX data available (N=259)
IDH	ATRX	N	Percent	Fisher’s Exact test p-value
Wild type	Intact	47	18.1%	0.0017*
Wild type	Lost	14	5.4%
Mutated	Intact	108	41.7%
Mutated	Lost	90	34.7%
Total		259
Cases with both IDH and ATRX data available (stratified by 1p19q status) 1p19q codeleted (N=87)
IDH	ATRX	N	Percent	Fisher’s Exact test p-value
Wild type	Intact	5	5.7%	0.3
Wild type	Lost	1	1.1%
Mutated	Intact	78	89.7%
Mutated	Lost	3	3.4%
Total		87
1p19q intact (N=63)
IDH	ATRX	N	Percent	Fisher’s Exact test p-value
Wild type	Intact	16	25.4%	0.0004*
Wild type	Lost	7	11.1%
Mutated	Intact	9	14.3%
Mutated	Lost	31	49.2%
Total		63
1p19q not assayed (tumors histologically classified as astrocytoma; N=101)
IDH	ATRX	N	Percent	Fisher’s Exact test p-value
Wild type	Intact	26	25.7%	<0.0001*
Wild type	Lost	6	5.9%
Mutated	Intact	17	16.8%
Mutated	Lost	52	51.5%
Total		101

^*Statistical significance at the 0.05 level.

In the IDH/TP53/ATRX group, CDKN2A loss was associated with worse overall survival after adjusting for age and sex (Fig. 3A; HR = 3.4, 95% CI = 1.4–8.4, p = 0.008; Table 3). CDKN2A loss was strongly associated with poor survival in IDH/TP53 mutated tumors (Fig. 3B; HR = 4.4, 95% CI = 1.8–10.3, p = 0.0008; Table 3), but this association did not reach statistical significance in the IDH/ATRX group (Fig. 3C; HR = 1.6, 95% CI = 0.8–3.0, p = 0.2; Table 3). The 2 groups overlapped partially, such that of the 115 IDH mutated cases with known status for CDKN2A, TP53, and ATRX, 42 cases (36.5%) showed ATRX loss and TP53 mutation, 7 cases (6.1%) were ATRX intact and TP53 mutated, and 20 cases (17.4%) showed ATRX loss without TP53 mutation (Fig. 4). The IDH/TP53 and IDH/ATRX groups contained a similar proportion of grade II and grade III tumors. Furthermore, the subjects in each group were of nearly identical age at the time of diagnosis (IDH/TP53: 36.5 ± 1.3 years; IDH/ATRX: 36.4 ± 1.2 years, mean ± SE). Thus, in both histologically and molecularly defined astrocytoma (in particular, in astrocytomas defined by the presence of IDH and TP53 mutations), CDKN2A loss was strongly associated with poor survival after adjusting for age and sex.

Figure 3.

Effect of CDKN2A loss on overall survival in subsets of molecularly defined astrocytomas and oligodendrogliomas. (A–C) CDKN2A loss by fluorescence in situ hybridization (FISH) was associated with worse overall survival in IDH/TP53 mutated tumors with ATRX loss (IDH/TP53/ATRX group) (A) and IDH/TP53 mutated tumors (IDH/TP53 group) (B) but not in IDH mutated tumors with ATRX loss (IDH/ATRX group) (C). (D) Patients with molecularly defined oligodendrogliomas in the CDKN2A deleted group appeared to have improved survival; however, the statistical significance of the association between CDKN2A loss and overall survival could not be accurately assessed because none of the subjects in the CDKN2A deleted group died during the follow-up period.

Figure 4.

Overlap between IDH/TP53 and IDH/ATRX groups. (A) The astrocytoma groups defined by IDH/TP53 mutation (IDH/TP53 group) or IDH mutated tumors with ATRX loss (IDH/ATRX group) showed partial overlap such that among 115 IDH mutated cases with known status for CDKN2A IDH TP53, and ATRX, 42 cases (36.5%) showed ATRX loss and TP53 mutation, 7 cases (6.1%) were TP53 mutated and ATRX intact; 20 cases (17.4%) were ATRX intact and TP53 wild type. (B) OncoPrint of molecular parameters grouped by histology and grade.

With respect to molecularly defined oligodendrogliomas, the effect of CDKN2A loss on IDH mutated, 1p19q codeleted tumors was examined (Fig. 3D). This group consisted of 16 patients with no deaths (Table 3); therefore, a comparison using Cox proportional hazards would not be reliable and an estimate of the association between CDKN2A loss and survival in molecular oligodendroglioma could not be determined.

DISCUSSION

This study demonstrates an independent adverse effect of CDKN2A loss on the overall survival of patients with lower grade (WHO II-III) astrocytomas after accounting for IDH mutation, age, and sex. Tumors for 270 patients were classified and graded according to WHO criteria by expert neuropathologists, and molecular data for CDKN2A, IDH, TP53, ATRX, and 1p19q were included. While most patients (61.1%) were alive at censoring, the survival analysis yielded unequivocal results with good concordance between the histologic and molecular analysis, in particular for the molecular subset of astrocytoma defined by IDH/TP53 mutation. Unfortunately, a conclusion could not be reached for molecular oligodendrogliomas because of the absence of deaths in the CDKN2A deleted group. This analysis will require a follow-up interval longer than the median 10.9 years obtained for the molecular oligodendroglioma subgroup included here.

In previous studies, loss of CDKN2A was associated with worse survival in both astrocytomas and oligodendrogliomas (27–33). This study shows that CDKN2A loss is associated with worse survival in astrocytomas, but this effect was not observed in oligodendrogliomas. The disagreement between the findings here and those of prior studies may stem from the powerful effect of IDH mutation on survival, which was not considered previously because the role of IDH mutation in gliomas was not yet known. However, we cannot exclude other factors such as limited statistical power in the oligodendroglioma subgroup since there were few deaths among these patients in our cohort.

Most cases of CDKN2A loss in infiltrating gliomas are homozygous deletions in the 9p21 region (17–20). Such genomic alterations would be predicted to result in loss of p16 protein expression. Surprisingly, the expression of p16 by IHC correlated poorly with CDKN2A deletion by FISH. Several factors could explain this discrepancy. For example, while hemizygous losses are relatively straightforward to detect by FISH, the loss of p16 expression might be minimal and result in no significant detectable change in p16 expression by IHC. Indeed, in relating our findings to those of the The Cancer Genome Atlas (TCGA) lower-grade glioma data set, (publically available at http://www.cbioportal.org/), a significant decrease in CDKN2A mRNA expression was observed only for homozygous deletion of the CDKN2A gene but not hemizygous deletion (Supplementary Fig. 2). Alternatively, CDKN2A gene expression may be lost through promoter hypermethylation (42, 43) or point mutations (26), which are seen in a small subset of cases but are undetectable by FISH. If both copies of CDKN2A were inactivated by either mechanism, p16 expression by IHC would be expected to decrease while the status of CDKN2A by FISH would be “intact.” Regardless of the reason(s) for the discrepancy, it should be emphasized that the strong association between CDKN2A loss by FISH and poor survival in astrocytoma, as opposed to the weak association between p16 loss of expression by IHC and poor survival in astrocytoma, argues strongly in favor of using FISH to evaluate the status of CDKN2A when prognosticating astrocytomas.

The analysis of molecular subsets of astrocytomas revealed a strong association between CDKN2A loss and poor survival in patients harboring IDH/TP53 mutations, but a weaker effect was observed in the subset of molecular astrocytomas defined by IDH mutation and ATRX loss of expression. The concordance between ATRX loss and TP53 mutation in our study (76%) was similar to that reported in previous studies on ATRX (range: 71%-78%) (8, 10, 44). Though a small fraction of cases with missense mutations (~10%) may be missed by IHC, the majority of mutations in ATRX result in loss of protein expression (9, 44). Prior data strongly suggest that biological differences may exist between the IDH/TP53 and IDH/ATRX groups (9, 44–46). For example, other pathways of gliomagenesis, including telomerase-dependent mechanisms (47, 48), could predominate in IDH/TP53 mutated tumors with intact ATRX. Alternatively, mutations in other genes involved in the alternative lengthening of telomeres pathway could be present. Future studies focusing on these questions should help clarify this issue.

Whether the combination of IDH mutation and TP53 mutation or the combination of IDH mutation and ATRX loss of expression should be considered the “gold standard” for defining molecular astrocytomas in the clinical setting remains unresolved. The strong association between CDKN2A loss and poor survival in IDH/TP53 astrocytomas reported here relied on determination of CDKN2A status by FISH and TP53 mutation by sequencing. The poor correlation between CDKN2A status by FISH and p16 loss by IHC shows that the assay of choice can play a crucial role. Since molecular information on these markers will be an integral component of the standard integrated diagnoses in clinical reports as recently proposed by the WHO (3), our findings suggest that comparative studies using both sequencing and IHC methods should be conducted to better define the standard of care and avoid confusion.

In conclusion, our study demonstrates that CDKN2A loss is highly prognostic in IDH/TP53 mutated astrocytomas. Testing for CDKN2A loss could be a potentially useful “grading” biomarker for this adult glioma subtype.