Prognostic stratification of patients with anaplastic gliomas according to genetic profile

Abstract

BACKGROUND.

There is a need to improve the current, controversial, and poorly reproducible classification of anaplastic gliomas, which represent a highly heterogeneous entity in terms of survival.

METHODS.

The impact of the most common genetic alterations on survival was investigated based on 156 anaplastic gliomas: Among the patients who were included, the gender ratio was 1.32, the median age was 45.5 years (range, 20–83 years), and the median Karnofsky performance status was 70 (range, 40–100). Genetic analysis included a search for loss of heterozygosity (LOH) on chromosomes 1p and 19q; amplification of chromosomes 9p and 10q and of the epidermal growth factor receptor (EGFR), cyclin-dependent kinase 4 (CDK4) and mouse double-minute (MDM2) genes; and p53 expression.

RESULTS.

The median survival was 33.5 months, and the median progression-free survival was 15.8 months. In a univariate analysis, LOH on 1p and 19q was correlated with longer survival, whereas p53 expression, LOH on 9p, LOH on 10q, amplified EGFR, and deleted CDKN2A were correlated with shorter survival. LOH on 1p and 19q were associated with oligodendrogliomas, LOH on 10q was related to EGFR amplification, and LOH on 1p and 19q was mutually exclusive with EGFR amplification and LOH on 10q. In a multivariate analysis, the significant prognostic factors were age, histology, LOH on 1p and 19q, and P16/CDKN2A deletion. Recursive partitioning analysis (RPA) divided the whole group hierarchically into 3 distinct prognostic subgroups: Group A with 1p19q codeletion (median survival, 98 months), Group B with EGFR amplification (median survival, 17 months), and Group CC (median survival, 31 months), providing a basis for a genetically based prognostic subclassification for patients with Grade III gliomas.

CONCLUSIONS.

The search for 1p19q codeletion and EGFR receptor amplification provides a simple, clinically relevant prognostic subclassification of grade III gliomas. Cancer 2006. © 2006 American Cancer Society.

The morphologic classification of gliomas remains controversial and insufficient.1 The discovery over the last 15 years of the main genetic alterations has made a major contribution to understanding the molecular pathways involved in oncogenesis. One of the consequences of this progress is that the genetic profile may be helpful for diagnosing and partially for predicting prognosis and response to treatment in patients with glioma, as illustrated by the loss of chromosome 1p in patients with anaplastic oligodendrogliomas and low-grade gliomas.2–5 Anaplastic gliomas, including Grade III oligodendrogliomas (OIII), Grade III oligoastrocytomas (OAIII), and Grade III astrocytomas (AIII), comprise a highly heterogeneous subgroup in terms of prognosis.6 There is a need for future clinical trials and for clinical practice to identify homogeneous prognostic subgroups. In the current study, commonly reported molecular alterations were analyzed in a large series of anaplastic gliomas in the hope of detecting molecular predictors of survival that may be useful in conjunction with well identified clinical prognostic factors.

MATERIALS AND METHODS

Selection of Patients

From January 1997 to March 2005, clinical information on patients who were treated in our department for a primary brain tumor was collected in a data base. Patients who fulfilled the following inclusion criteria were selected for this study: 1) a histologic diagnosis of OIII, OAIII or AIII according to the World Health Organization (WHO) classification system, 2) detailed clinical data at diagnosis and during follow-up, 3) blood and tumor DNA available, and 4) written informed consent obtained.

Pathologic Review

All pathologic slides of tumor samples were reviewed by M.K. Tumor cell type identification and grading conformed to the WHO classification system.

Molecular Analysis

Tumor and paired blood samples were obtained after patients gave their informed consent for molecular analysis. Loss of heterozygosity (LOH) on chromosomes 1p, 19q, 10q, and 9p was detected by microsatellite analysis on blood and tumor DNA, as reported previously.7

The expression of p53 was detected on 5-μm section of formalin fixed and paraffin embedded tissues, as reported previously.8 Expression of p53 was defined as a moderate-to-strong (++ to +++) staining of >50% of nuclei.

Epidermal growth factor receptor (EGFR) amplification of tumor DNA was detected by using real-time polymerase chain reaction (PCR) analysis with an internal probe. The EGFR primers (forward, GTGCAGATCGCAAAGGTAATCAG; reverse, GCAGACCGCATG-TGAGGAT; probe, CCCCTCCCCGTATCTC; FAM Dye labeled) amplified a 79-base pair (bp) genomic fragment. The reference primers amplified a genomic fragment from RNase P (TaqMan RNase P Detection Reagents [FAM Dye], no. 4316831; Applied Biosystems, Foster City, CA). Real-time PCR cycles were performed as follows: 50°C for 2 minutes; 95°C for 15 minutes; Cycles 1 through 40, 95°C for 15 seconds; and 60°C for 1 minute. The same technique was used for the mouse double-minute gene (MDM2) with primers that amplified a 91-bp genomic fragment (forward, TTGGTTTCTAGACCATCTACCTCATCT; reverse, AAAAGCTGTGTGAATGCGTCAAAT; probe, ACCTGTCTC-ACTAATTGC [FAM Dye labeled]) and for the cyclin dependent kinase 4 gene (CDK4) with primers that amplified a 102-bp genomic fragment (forward, GCCACT-AAGCAGTAACCATTCAACT; reverse, CGGCTTCAGAGTTTCCACAGAA; probe, CCTTCTCACCTTAGGCC [FAM Dye labeled]). Homozygous deletion of P16/CDKN2A was detected by using real-time PCR analysis, as described previously.9

Statistical Methods

Frequency distributions and summary statistics were calculated for all clinical and molecular variables for the entire population. The chi-square test and the Fisher exact test for small samples were used to determine the association between molecular alterations. The distribution of molecular markers according to age was investigated with the Wilcoxon test. Probability estimates for the overall survival (OS) curves were calculated according to the Kaplan–Meier method. OS was defined as the time from the date of first surgery until death or last follow-up. Patients who were lost to follow-up were censored on the last known day of life, and patients who remained alive were censored on the day of their last visit. The log-rank test was used to test for equality of survival distributions. The following variables were investigated: age; gender; LOH on chromosomes 1p, 19q, 10q, and 9p; p53 expression; EGFR amplification; P16/CDKN2A deletion; MDM2 amplification; and CDK4 amplification. A Cox proportional regression model was used to explore the relation between OS and clinical and molecular explanatory variables. Factors that were significant in univariate analysis (with a P value ≤0.05) were entered as candidate variables in the multivariate Cox proportional-hazards regression model analysis. Median discontinuation time of follow-up and relative risks were presented with their 95% confidence interval (95% CI). To define homogeneous prognostic subgroups, we performed a recursive partitioning analysis (RPA) on the whole group, entering variables the variables that we identified as associated with survival in the univariate analysis, as described previously.10

RESULTS

Clinical and Histologic Data

One hundred fifty-six patients from the data base met the inclusion criteria, including 89 males and 67 females (gender ratio, 1.32) who ranged in age from 20 years to 83 years (median, 45.5 years). The pretreatment Karnofsky performance status (KPS) ranged from 40 to 100 (median, 70). A previous history of low-grade glioma was noted in 23 patients, was absent in 124 patients, and was unknown in 9 patients. The clinical, radiologic, histologic, and therapeutic characteristics of the patients studied are reported in Table 1 along with survival. The median follow-up was 57.3 months (95% CI, 52.9–71.2 months). The overall median survival was 33.5 months (95% CI, 28.1–42.8 months), and the median progression-free survival (PFS) was 15.8 months (95% CI, 14.2–21.2 months). Ninety-five patients (61%) were dead at the time of the current analysis. Age younger than 45.5 years, a KPS >80, oligodendroglial phenotype, and tumor removal were associated with longer survival (Table 1).

Table 1. Clinical and Radiologic Characteristics of 156 Patients and Impact on Survival

Characteristic	No. of patients (%)	Median survival, months	P	95% CI
95% CI indicates 95% confidence interval; AIII, Grade III astrocytoma; OAIII, Grade III oligoastrocytoma; OIII, Grade III oligodendroglioma; KPS, Karnofsky performance status.
Age, y			.012
≤45.5	78 (50)	40.6		31–98.5
>45.5	78 (50)	25		19.5–39.2
Previous low-grade glioma history			.31
No	124 (79.5)	30.6		24.4–36.7
Yes	23 (14.7)	42.4		25 to ∞
Radiology: contrast enhancement			.32
No	13 (9.6)	50.4		23.1 to ∞
Yes	122 (90.4)	31		25–39.2
Histology			.0002
AIII	14 (9)	24.5		18.4 to ∞
OAIII	55 (35)	28.5		30.8 to ∞
OIII	87 (56)	67.1		23.1–34.4
Surgery			.026
Biopsy	36 (23)	18.7		14.2–52.8
Complete or partial removal	104 (77)	33.5		29.1–46.8
Pretreatment KPS			.031
>80	59 (61)	52.1		28.1 to ∞
≤80	38 (39)	19.6		14.3–69.9

Correlations with Molecular Markers

The frequencies of the main molecular alterations were as follow: p53 expression, 56.4%; LOH on 9p, 39.4%; LOH on 10q, 38.1%; LOH on 19q, 33.3%; LOH on 1p, 29.3%; P16/CDKN2A deletion, 22.2%; EGFR amplification, 17.3%; CDK4 amplification, 3.6%; and MDM2 amplification, 1.4%. Associations and exclusions between the molecular alterations are indicated in Table 2. Loss of 1p and 19q (both highly associated; P < .0001) was associated with oligodendrogliomas (P = .005), whereas loss of 10q was related to EGFR amplification (P = 3.5 10⁻⁹). Loss of 1p and 19q was mutually exclusive with EGFR amplification (P = .004) and LOH on 10q (P = .03).

Table 2. Associations and Exclusions between Genetic Alterations*

	Percentage of tumors with alterations
	LOH 1p	LOH 19q	LOH 9p	LOH 10q	LOH 13q	IHC p53	EGFR	MDM2	CDKN2A
LOH indicates loss of heterozygosity; IHC, immunohistochemistry; EGFR, epidermal growth factor receptor gene; MDM2, mouse double-minute gene; CDKN2A, cyclin-dependent kinase N2A gene; CDK4, cyclin-dependent kinase 4 gene. * Association (P < .0001). † Exclusion (P = .04). ‡ Exclusion (P = .002). § Association (P < .0039).
LOH 19q	19*
LOH 9p	16	14
LOH 10q	9	14	16*
LOH 13q	7	7	8	9
p53	11	17	31	27	15
EGFR	2†	5	11‡	16*	5	11
MDM2	0	0	0	0	0	2	0
CDKN2A	6	6	15*	18*	7	18	8§	0
CDK4	0	<1	2	0	0	0	<1	<1	0

The correlations and exclusions between genetic alterations and locations are indicated in Table 3, which underlines the mutual exclusion between 1p19q loss (frontal location) and EGFR amplification (occipital location). Genetic profile was also tightly correlated with histology: LOH1p (81% in OIII vs. 46% in AIII and OAIII; P < .0001) and LOH19q (71% in OIII vs. 46% in AIII and OAIII; P < .001) were associated with oligodendrogliomas whereas LOH10q was associated with astrocytic and mixed tumors (64% in AIII and OAIII vs. 43% in OIII; P = .02).

Table 3. Correlation between Genetic Profile and Tumor Location

Variable	Tumor location
	Frontal	Temporal	Occipital
EGFR indicates epidermal growth factor receptor gene; NS, not significant.
Loss of 1p/19q	Association	Exclusion	Exclusion
P	.01	.008	.03
EGFR amplification	Exclusion	NS	Association
P	<.001	.07	.003

In univariate analysis (Table 4), LOH on 1p and 19q correlated with longer survival; whereas p53 expression, LOH on 9q, LOH on 10q, EGFR amplification, and P16/CDKN2A deletion were predictive of a poor prognosis. The only genetic alterations that were correlated with age (i.e., more frequent in older patients) were LOH on 9p (age 44.3 years vs. 49.4 years; P = .03), 10q (age 44 years vs. 51.2 years; P = .002), and EGFR amplification (age 44.4 years vs. 56 years; P = 3.10⁻⁵).

Table 4. The Impact of Genetic Alterations on Survival: Univariate Analysis

Variable	No. of deaths/No. observed	Median survival, months	95% CI, months	P
95% CI indicates 95% confidence interval; LOH, loss of heterozygosity; EGFR, epidermal growth factor receptor gene; CDKN2A, cyclin-dependent kinase N2A gene.
LOH 1p
No	66/101	30.1	24.4–36.7	3.2 10^−3
Yes	18/42	81.1	67.1 to ∞
LOH 19q
No	64/96	28.1	20.8–36.5	1.8. 10^−3
Yes	21/48	81.1	46.8 to ∞
p53 Expression
Yes	29/44	26.5	19.6–46.8	.045
No	15/34	98.5	30.3 to ∞
LOH 9p
No	46/86	36.7	30.3 to ∞	.045
Yes	38/56	24.5	19.5–46.8
LOH 10q
No	48/86	24.5	30.8–69.9	1.9 10^−3
Yes	34/53	42.4	18.4–39.2
EGFR
Normal	70/124	39.2	30.3–67.1	1.5 10^−4
Amplified	19/26	17.4	12.6–33.5
CDKN2A
Normal	63/112	37.1	29.7–69.9	.020
Deleted	23/32	20.8	18.4–39.2
LOH 1p/LOH 19q
No	75/116	29.1	23.1–34.8	2.3 10^−4
Yes	9/27	98.5	81.1 to ∞

Then, we considered separately the effects of genetic changes on prognosis within each histologic subtype. For the 87 patients with OIII, the discriminating prognostic factors were 1p19q codeletion (LOH on 1p and 19q: median survival, 98.5 months vs. 30.8 months; P = .02), LOH on 9p (median survival, 24.1 months vs. 119.7 months; P = .02), p53 expression (median survival, 21.4 months vs. not reached; P = .005); and EGFR amplification (median survival, 18.9 months vs. 81.1 months; P = .005). For the 55 patients with OAIII, only EGFR amplification was statistically significant (median survival, 12.4 months versus 29.7 months; P = .005). In the small group of 14 patients with astrocytomas, no molecular factor reached significance.

In the multivariate analysis (Cox proportional hazards regression model analysis), prognostic impact remained significant for age, histology, 1p and 19q loss, and P16/CDKN2A deletion. Table 5 shows that the stronger predictor of survival was LOH on 1p and 19q (relative risk of death, 0.30).

Table 5. Impact of Genetic Alterations on Survival: Cox Multivariate Analysis Model

Characteristic	P	RR
RR indicates relative risk; OIII, Grade III oligodendroglioma; OAIII, Grade III oligoastrocytoma; AIII, Grade III astrocytoma; LOH, loss of heterozygosity; CDKN2A, cyclin-dependent kinase N2A gene.
Age >45.5 y	.018	1.73
Histology: OIII vs. OAIII and AIII	.01	0.54
LOH 1p and LOH 19q	.0025	0.30
CDKN2A deletion	.0120	1.98

To stratify patients into homogeneous prognostic groups, a recursive partitioning analysis (RPA) was applied to the whole group. Figure 1 shows that the most discriminating prognostic criterion was the combined loss of 1p and 19q: The 27 patients (Group A) with 1p19q loss had a median survival of 98 months, whereas the 129 patients without 1p19q loss had a median survival of 29 months. In the latter group, the stronger predictor of survival was EGFR amplification: The 26 patients with EGFR amplification (Group B) had a median survival of 17 months, whereas the other 102 patients had a median survival of 31 months. In this last group (Group C), the next most discriminating factor was histology: Patients with oligodendroglioma (Group C1) had a better survival than patients with astrocytomas and mixed gliomas (Group C2; median survival, 40.6 months vs. 28.5 months; P = .01).

Figure 1. This is an illustration of the recursive partitioning analysis for the entire group: For each subgroup, the drawing shows the relative risk of death (black boxes), the number of deaths/number of patients evaluable, and the median survival (MS). EGFR indicates epidermal growth factor receptor; Nal, normal; ampl, amplified; OIII, Grade III oligodendroglioma; AIII, Grade III astrocytoma; OAIII, Grade III oligoastrocytoma.

EGFR amplification and 1p19q loss were mutually exclusive, and no patient had both alterations. Survival curves for Groups A, B, and C are shown in Figure 2 (P = 7.3 10⁻⁶).

Figure 2. This chart shows survival curves for Groups A, B and C. EGFR + indicates epidermal growth factor receptor positive.

In Group A (median age, 47.0 years), the majority of patients (23 of 27 patients) had OIII (85%), 3 patients had OAIII (11%), and 1 patient had AIII (4%). The 26 patients in Group B were older (median age, 56 years) and included 14 patients with OIII (54%), 11 patients with OAIII (42%), and 1 patient with AIII (4%). Among 90 patients in Group C (median age, 42.5 years), 42 patients had OIII (47%), 37 patients had OAIII (41%), and 11 patients had AIII (12%).

DISCUSSION

The objective of this study was to stratify patients with glioma into homogeneous prognostic subgroups according to clinical characteristics, histologic profiles, and genetic profiles. The histologic diagnosis of glioma is controversial, and there is a particular need to refine the current diagnosis for patients with anaplastic gliomas because of their heterogeneous prognosis. In this setting, our data outlined the usefulness of accurate, centralized histologic analysis, indicating that patients with OIII were associated with a median survival twice as long as the median survival of patients with OAIII and AIII; this difference was not detected before centralized reading because of the heterogeneity and subjectivity of the diagnosis (data not shown). Anaplastic gliomas can be divided into 2 histoprognostic subgroups: 1) patients with OIII, who have a median survival of 67 months, and 2) patients with OAIII and AIII, who have a median survival of 28.5 months. This prognosis-based subdivision is in accordance with previous studies that were based on the expression of Olig2 and glial fibrillary acidic protein (GFAP) that separated pure oligodendrogliomas (which have a homogeneous tumor cell population; Olig2-positive/GFAP-negative) from both oligoastrocytomas and astrocytomas (which display 2 distinct populations, Olig2-positive/GFAP-negative and Olig2-negative/GFAP-positive).11

In the univariate analysis, our data confirmed that age, performance status, and extent of surgical treatment (removal vs. biopsy) were the main clinical prognostic factors (Table 1).10 Genetic analysis showed that molecular data correlated with each other and with clinical and histologic data: Whereas 1p loss was associated strongly with 19q loss, it was mutually exclusive with EGFR amplification, which was associated with 9p loss, 10q loss, and CDKN2A deletion (Table 2). Molecular and histologic data also were correlated with location12–14: A positive association was observed between frontal location, 1p19q deletion, and pure oligodendroglial phenotype; whereas a posterior tumor location was correlated with 10q loss with or without EGFR amplification and astrocytic or mixed phenotype (Table 3).

In the univariate analysis, most genetic alterations were correlated with survival (Table 4): LOH on 1p and 19q was associated with longer survival, and p53 expression, LOH on 9p, LOH on 10q, amplified EGFR, and deleted CDKN2A were associated with shorter survival. In contrast to the other markers, which are related to anaplasia, P53 alteration and 1p19q codeletion are early events that frequently affect low-grade gliomas and are mutually exclusive, referring to different subtypes of gliomas, both in terms of location and histology. The relatively poorer prognosis for patients with p53-expressing tumors may be related to the finding that they represent a greater proportion of astrocytic and mixed gliomas compared with oligodendrogliomas. This difference in survival in patients with 1p19q-deleted tumors versus p53-expressing tumors also is present at the stage of low-grade gliomas.8

In the multivariate analysis, the most significant prognostic factors were age, histology, LOH on 1p and 19q, and CDKN2A homozygous deletion. With a relative risk of death of 0.30, 1p19q codeletion appeared to be the strongest prognostic factor for survival. EGFR amplification did not appear in Table 5, because it was related tightly to age: Both variables appeared redundant and neutralized each other on multivariate analysis, but each variable was highly significant when it was considered alone.

We used RPA to identify the most discriminating factors and to define homogeneous prognostic subgroups. The RPA orders hierarchically all possible cut-off points, which divide the data set into 2 samples that differ significantly with respect to survival. The most discriminating markers were the molecular alterations 1p19q loss and EGFR amplification, which defined 3 subgroups that clearly were distinct in respect to survival. With a median survival of 98 months, patients who had anaplastic gliomas with 1p19q loss represented a distinct prognostic subgroup independent of the type of treatment, as demonstrated recently by a large, randomized European Organization for Research and Treatment of Cancer trial in patients with anaplastic oligodendrogliomas and oligoastrocytomas.15 However, it remains unknown whether 1p19q codeletion is an intrinsic prognostic factor in the absence of any treatment, because most patients in that study (141 of 156 patients) received radiotherapy and/or chemotherapy. It is important to consider 1p19q codeletion (which usually involves the whole arm) and not the isolated, often partial deletion of regions 1p or 19q. In contrast to complete 1p19q codeletion, partial 1p deletion is more frequent in astrocytic or mixed tumors and bears a pejorative prognostic value, as demonstrated recently in a comparative genomic hybridization (CGH) array.16 Indeed, in the current study, the 1p19q loss was correlated with pure oligodendrogliomas, confirming previous data, whereas mixed gliomas were more frequent in Groups B and C, and pure astrocytoma was more frequent in Group C.

The older age of patients in Group B may explain in part why they had the poorest prognosis: Most patients in Group B, in addition to EGFR amplification, also had a 10q deletion. This association of both alterations is very frequent in patients with glioblastoma.17 Indeed, the median survival in Group B (17 months) was very close to that of patients with glioblastoma.18 This subgroup may correspond to small cell astrocytoma (Grade III and IV), which displays the same morphologic features as anaplastic oligodendroglioma (i.e., clear halo, chicken-wire vasculature, perineuronal satellitose, and microcalcifications) but is associated with poor survival and is characterized by EGFR amplification and 10q deletion.19

Group C is the largest and most likely the most heterogeneous subgroup. It can be subdivided further, according to prognosis, into patients with oligodendroglioma and patients with astrocytomas and mixed gliomas. However, it probably will require a more extensive genetic analysis to subdivide Group C into subgroups of patients with prognostically relevant entities.

If the current data are confirmed, then they suggest that patients with anaplastic glioma should be divided not only according to histologic diagnosis (which is highly subjective) but also according to molecular profile into at least 3 groups. It is noteworthy that unsupervised CGH-array analysis of a large series of gliomas led to a similar partition into 3 groups: patients with 1p19q codeletion, patients with EGFR amplification, and a third group of patients with 10q deletion, 9p deletion, and gain of chromosome 7 as the most frequent alteration (unpublished observations). There is little doubt that a heterogeneous entity like anaplastic gliomas needs to be subclassified according to these and other molecular criteria for future clinical trials and clinical practice.

Acknowledgements

We thank Dr. Karima Mokhtari for providing histologic samples and Anne-Marie Lekieffre, Murielle Brandel, Marc Ouzounian, and Tristan Salmon-Legagneur from the Association pour la Recherche sur les Tumeurs Cerebrales (ARTC) for their valuable assistance