Abstract
Adrenocortical carcinoma (ACC) is a rare disease with an overall poor but heterogeneous prognosis. This heterogeneity could reflect different mechanisms of tumor development. Gene expression profiling by transcriptome analysis led to ACC being divided into two groups of tumors with very different outcomes. Somatic inactivating mutations of the tumor suppressor gene TP53 and activating mutations of the proto-oncogene β-catenin (CTNNB1) are the most frequent mutations identified in ACC. This study investigates the correlation between p53 and β-catenin alterations and the molecular classification of ACC by transcriptome analysis of 51 adult sporadic ACCs. All TP53 and CTNNB1 mutations seemed to be mutually exclusive and were observed only in the poor-outcome ACC group. Most of the abnormal p53 and β-catenin immunostaining was also found in this group. Fifty-two percent of the poor-outcome ACC group had TP53 or CTNNB1 mutations and 60% had abnormal p53 or β-catenin immunostaining. Unsupervised clustering transcriptome analysis of this poor-outcome group revealed three different subgroups, two of them being associated with p53 or β-catenin alterations, respectively. Analysis of p53 and β-catenin target gene expressions in each cluster confirmed a profound and anticipated effect on tumor biology, with distinct profiles logically associated with the respective pathway alterations. The third group had no p53 or β-catenin alteration, suggesting other unidentified molecular defects. This study shows the important respective roles of p53 and β-catenin in ACC development, delineating subgroups of ACC with different tumorigenesis and outcomes. Cancer Res; 70(21); 8276–81. ©2010 AACR.
Introduction
Adrenocortical carcinoma (ACC) is a rare and aggressive tumor with an overall poor prognosis: the 5-year survival rate is generally below 35% (1, 2). Despite this, overall poor prognosis outcomes and survival vary greatly. This heterogeneity is partially explained by the tumor stage at diagnosis, but also probably reflects intrinsic tumor biology and different mechanisms of tumor development.
Most sporadic adrenocortical tumors (ACT) are monoclonal, suggesting that a somatic genetic defect occurs early in tumorigenesis. Studies of rare genetic syndromes associated with ACC [i.e., Beckwith-Wiedemann due to insulin-like growth factor II (IGF2) overexpression, or Li-Fraumeni syndrome due to inactivating mutations of the tumor suppressor gene TP53] have greatly facilitated progress and increased our understanding of sporadic ACTs (3). IGF2 overexpression occurs in >85% of ACC (4, 5). TP53 somatic mutations are present in about a third of sporadic adult ACCs (6–8). The occurrence of a somatic TP53 mutation is associated with a worse survival (1), and allelic losses at or around the TP53 locus (17p13) are observed in >80% of ACC (4). The Wnt/β-catenin signaling pathway plays an important role in adrenal cortex development (9). Immunohistochemical studies have shown that the distribution of the β-catenin protein is abnormal in ACC, suggesting activation of this pathway (10). There are somatic activating mutations of CTNNB1 (β-catenin) in several types of benign and malignant ACTs, including ACC (10–12), and this pathway has recently been shown to play a role in the development of ACTs in transgenic mice (13).
Studies of the genomics of ACTs have helped improve our understanding of the pathophysiology and classification of these tumors (14–16). Transcriptome analysis clearly shows that the gene profile of benign ACTs differs from that of malignant ACTs (ACCs; reviewed in ref. 17). We and others have recently reported that transcriptome analysis also identified two groups of ACCs (14, 16), with contrasted outcome and survival: one having a poor overall survival (OS) rate (cluster C1A in ref. 16) and the other with a better survival rate (cluster C1B in ref. 16). The 5-year survival rates were 20% and 91% in the C1A and C1B clusters, respectively (18). This suggests that the molecular alterations occurring in these two groups of ACC are different and that they modulate the tumor phenotype.
We have therefore investigated a large cohort of adult sporadic ACC to determine the relationship between the molecular classification determined by transcriptome analysis and the two most frequent somatic genetic alterations described in ACC: inactivating mutations of the tumor suppressor gene TP53 and activating mutations of the proto-oncogene CTNNB1. As we found a clear correlation between these genetic alterations and the molecular classification, we analyzed their effect on gene expression profiles, and specifically the expression level of the respective target genes of the two pathways.
Materials and Methods
Tumor samples
Fifty-one tumors (whole cohort) were prospectively collected, and DNA and RNA were extracted as previously described (16) with signed informed consent and approval by the institutional review board of Cochin Hospital.
DNA preparation, sequencing, and immunohistochemical studies
The TP53 and CTNNB1 coding sequences of tumor DNA were sequenced as described previously (10, 19). Immunohistochemical staining for p53 and β-catenin was done using available paraffin-embedded tissue sections (48 of 51 tumors) as described previously (10, 19) and considered abnormal when positive for nuclear staining.
Expression analysis
The expression profiles of 34 ACCs derived from a previous study (16) were performed with the HG-U133 Plus 2.0 Affymetrix GeneChip arrays. All data are available on ArrayExpress Web site (http://www.ebi.ac.uk/arrayexpress, experiment E-TABM-311). Unsupervised clustering analysis of these 34 microarrays was performed as described (16). The association between clusters and biological variables was measured with Fisher's exact test. Prediction analysis was performed on a series of 51 ACCs analyzed by quantitative reverse transcription-PCR (RT-PCR; ΔCt data) using the PAM algorithm (pamr R package; ref. 20). The 34 ACCs (21 C1A and 13 C1B) from the expression profiling series were used to train several predictors and select a final four-gene predictor (MCM5, VEPH1, PINK1, and SLC2A1); this was then used to classify the remaining 17 samples. Differential expression was measured with Bayes moderated t test (limma R package). Survival curves were obtained by the Kaplan-Meier method. Survival differences were assessed with the log-rank test.
Results
Tumor characteristics and molecular classification
Among 51 ACCs (whole cohort), 34 were previously used for microarray analysis (ACC microarray cohort). We could identify two groups of ACCs with different outcomes: C1A group had poor outcomes and C1B group had good outcomes (16). We analyzed 41 genes by quantitative RT-PCR in the whole cohort to identify predictors of malignancy, disease-free survival (DFS), and OS (16).
We tested the suitability of these 41 genes for classifying the ACC microarray cohort in C1A and C1B groups. A predictor based on the combination of MCM5, VEPH1, PINK1, and SLC2A1 genes provided the best classification of these tumors; it was therefore used to classify the remaining 17 tumors. Finally, 31 tumors were classified as C1A and 20 were classified as C1B. As expected, the patients with C1A tumors had very poor outcomes. DFS and OS, represented by the Kaplan-Meier curves in Fig. 1A, were lower for the C1A group (n = 31) than for the C1B group (n = 20; log-rank test: P = 4.91 × 10−4 for DFS and P = 1.99 × 10−5 for OS), reinforcing our previous results (16).
Distribution of samples between C1A and C1B and characteristics of the whole cohort. A, DFS (left) and specific OS (right) in the C1A and C1B clusters. The P value of the log-rank test for differences between survival curves is shown. B, overall molecular characteristics of the whole ACC cohort: TP53 and CTNNB1 mutation (black rectangles), and abnormal p53 and β-catenin immunohistochemistry (IHC; hatched gray rectangles). Clinical annotations: specific death and metastasis or relapse (black = yes, white = no). Each molecular and clinical variable is associated with a P value calculated using Fisher's exact test, measuring its association with the distribution of the sample.
Distribution of samples between C1A and C1B and characteristics of the whole cohort. A, DFS (left) and specific OS (right) in the C1A and C1B clusters. The P value of the log-rank test for differences between survival curves is shown. B, overall molecular characteristics of the whole ACC cohort: TP53 and CTNNB1 mutation (black rectangles), and abnormal p53 and β-catenin immunohistochemistry (IHC; hatched gray rectangles). Clinical annotations: specific death and metastasis or relapse (black = yes, white = no). Each molecular and clinical variable is associated with a P value calculated using Fisher's exact test, measuring its association with the distribution of the sample.
Somatic TP53 and CTNNB1 mutations and immunohistochemistry
About the whole cohort, TP53 mutations and nuclear p53 immunostaining were present in 9 (17.6%) and 13 (27%) of the ACCs, respectively (Figs. 1B and 2A; see Supplementary Table S1 for details). Seven of the nine ACCs with TP53 mutations had nuclear p53 immunostaining. CTNNB1 activating mutations and abnormal β-catenin immunostaining were present in 8 (15.7%) and 12 (25%) of the ACCs, respectively (Figs. 1B and 2B; see Supplementary Table S1 for details). Seven of the eight ACCs with CTNNB1 mutations showed nuclear staining.
Abnormal accumulation of p53 and β-catenin in ACC. A, p53 immunohistochemistry in a tumor showing p53 overexpression (right, dark brown nuclear staining) and in a tumor without overexpression (left, blue negative nuclear staining). B, β-catenin immunohistochemistry in a tumor showing an abnormal/strong cytoplasmic and nuclear accumulation of β-catenin (right) and in a tumor with only a normal membrane location (left). Magnification, ×200.
Abnormal accumulation of p53 and β-catenin in ACC. A, p53 immunohistochemistry in a tumor showing p53 overexpression (right, dark brown nuclear staining) and in a tumor without overexpression (left, blue negative nuclear staining). B, β-catenin immunohistochemistry in a tumor showing an abnormal/strong cytoplasmic and nuclear accumulation of β-catenin (right) and in a tumor with only a normal membrane location (left). Magnification, ×200.
Interestingly, TP53 and CTNNB1 mutations, and β-catenin nuclear staining were exclusively found in the poor-outcome (C1A) group (29%, 25.8%, and 40% of C1A, respectively) in comparison with the good-outcome (C1B) group (Fisher's exact test: P < 0.008, P < 0.02, and P < 0.02, respectively). p53 nuclear staining was more frequent in C1A group (36.7%) than in C1B group (11.1%), but the difference was not statistically significant. The TP53 and CTNNB1 mutations seemed to be mutually exclusive, only one ACC (no. 9) showed both TP53 and CTNNB1 mutations.
There was at least one mutation in the TP53 or CTNNB1 genes in 52% of the poor-outcome (C1A) group, and 60% of tumors had abnormal p53 and/or β-catenin immunostaining. But 32.2% of these C1A tumors had neither mutations nor abnormal immunohistochemistry.
Subclassification of the poor-outcome ACC group (C1A cluster) by gene profiling
We carried out a specific unsupervised clustering analysis of the C1A group of the ACC microarray cohort to understand more clearly the molecular genetics of the poor-outcome ACC group. Hierarchical clustering identified three subgroups (Fig. 3).
The three subgroups of the poor-outcome (C1A) group identified by unsupervised analysis. Dendrogram of 34 ACCs based on the top 1% (n = 547) most varying (robust coefficient of variation) probe sets using Ward linkage, and (1 − Pearson coefficient) as distance. Molecular annotations for TP53 and CTNNB1 are the same as in Fig. 1 and are associated with a P value calculated using Fisher's exact test, measuring its association with the assignment of the sample to C1A(p53), C1A(x), or C1A(β-catenin) for the ACC microarray cohort.
The three subgroups of the poor-outcome (C1A) group identified by unsupervised analysis. Dendrogram of 34 ACCs based on the top 1% (n = 547) most varying (robust coefficient of variation) probe sets using Ward linkage, and (1 − Pearson coefficient) as distance. Molecular annotations for TP53 and CTNNB1 are the same as in Fig. 1 and are associated with a P value calculated using Fisher's exact test, measuring its association with the assignment of the sample to C1A(p53), C1A(x), or C1A(β-catenin) for the ACC microarray cohort.
TP53 mutations and β-catenin nuclear staining were clearly associated with this molecular subclassification (Fisher's exact test: P < 0.02 and P < 0.001, respectively): C1A(p53) group contained all tumors with a TP53 mutation and all tumors of C1A(β-catenin) group had β-catenin nuclear staining. We found no association with the third group [C1A(x)].
We investigated the associations of TP53 mutations and β-catenin nuclear staining with clustering effect target gene expression by gene set enrichment analysis for p53 and Wnt/β-catenin pathways. We compared each subgroup with C1B tumors. The p53 signaling pathway was enriched in C1A(p53) (P < 0.006) and the Wnt/β-catenin signaling pathway was enriched in C1A(β-catenin) (P < 0.005; Supplementary Table S3). Moreover, global expression of p53-positive target genes was altered in the C1A(p53) subgroup (Supplementary Table S3; P < 0.05). Thirty-four percent of p53 target genes were altered; in particular, RRM2B, TP53INP1, and MDM2 were strongly diminished (Fig. 4A). Similarly, global expression of β-catenin–positive target genes was altered in C1A(β-catenin) (Supplementary Table S3; P < 0.005). Fifty-eight percent of β-catenin target genes were altered; in particular, CLDN1, AXIN2, and LGR5 were strongly increased (Fig. 4B).
Expression of p53- and β-catenin–positive target genes in each C1A subgroup. A and B, three examples of p53 targets (RRM2B, TP53INP1, and MDM2) and three examples of β-catenin targets (CLDN1, AXIN2, and LGR5). Each panel contains three box plots representing the distributions of the log intensity values (microarray data) for the following groups of ACC: C1A(p53) (red box), C1A(β-catenin) (green box), and C1B (gray box). *, P < 0.05; **, P < 0.01; ***, P < 0.001 (limma test).
Expression of p53- and β-catenin–positive target genes in each C1A subgroup. A and B, three examples of p53 targets (RRM2B, TP53INP1, and MDM2) and three examples of β-catenin targets (CLDN1, AXIN2, and LGR5). Each panel contains three box plots representing the distributions of the log intensity values (microarray data) for the following groups of ACC: C1A(p53) (red box), C1A(β-catenin) (green box), and C1B (gray box). *, P < 0.05; **, P < 0.01; ***, P < 0.001 (limma test).
Discussion
Gene profiling allowed new classifications of cancers, unrevealing new genetic alterations. We and others have used transcriptome analysis to develop a new classification of these rare and clinically heterogeneous ACC tumors, with clinical relevance, for diagnosis and prognosis (14, 16).
We now identify a subgroup of poor-outcome ACCs (C1A) that are associated with frequent somatic genetic alterations observed in these tumors: TP53 inactivating mutations and CTNNB1 activating mutations, which would be late event promoting the development of aggressive tumors. Moreover, these mutated ACCs represent the majority of the C1A group. p53 and β-catenin immunohistochemistry suggest that other genetic alterations influence the protein profile, as previously suggested (10, 19). More importantly, unsupervised cluster analysis showed that the gene profiles of the tumors with p53 and β-catenin alterations were different, which supports this molecular classification and its effect on tumorigenesis. The C1A(p53) cluster contained all the tumors with a TP53 mutations, and all the tumors in the C1A(β-catenin) cluster had altered β-catenin pathway. The remaining tumors of this aggressive C1A cluster were all assigned to a specific cluster of ACC, in which there were neither p53 nor β-catenin alteration associations. This group was enriched in cell cycle and metabolism genes (data not shown). This suggests that there are other molecular defects that remain to be identified. By the same token, the tumors in the C1A(p53) and C1A(β-catenin) clusters in which no genetic alterations were identified might harbor genetic or epigenetic alterations of the genes controlling the same pathways.
Analysis of the expression of the p53 and β-catenin target genes in their respective clusters clearly showed that these alterations have a major influence on tumor biology. Some of these target genes (Supplementary Table S2) have been implicated in oncogenesis. The fact that the changes in the expressions of these genes are specific to each C1A subcluster indicates that the tumor biology and cancer development mechanisms of these subclusters are different. Clinically, all these tumors are associated with a poor outcome. This agrees well with previous observations about TP53 mutations in ACC (1). It also suggests that alterations to the Wnt/β-catenin pathway are a poor prognostic factor in ACC. The IGF2 overexpression that occurs in >85% of ACC (see review in ref. 3) is widespread among both the C1A and C1B ACCs (data not shown). Therefore, although IGF2 is important in adrenal cortex oncogenesis, additional molecular defects are needed for the development of the most aggressive tumors.
This study provides new insights into the molecular classification of ACC. It points to the major effect of TP53 and CNNTB1 mutations on tumor biology. This approach should help identify new genetic alterations that occur in ACC without mutations of these two genes. This new vision of the molecular classification of ACC should also guide the development of new therapeutic strategies by identifying those tumors most likely to respond to drugs targeting specific signaling pathways.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Acknowledgments
We thank Dr. Owen Parkes for editing the English text.
Grant Support: Programme Hospitalier de Recherche Clinique grant PHRC060251, Recherche Translationnelle DHOS/INCA 2009 grant RTD09024, and Contrat d'Initiation à la Recherche Clinique grant CIRC05045. B. Ragazzon was the recipient of fellowships from the Conny-Maeva foundation and S. Gaujoux was the recipient of a research fellowship from the Fond d'Etude et de Recherche du Corps Medical, AP-HP, Paris, France.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.