Cell PRESS
The pathophysiology, diagnosis and prognosis of adrenocortical tumors revisited by transcriptome analyses
Guillaume Assie, Marine Guillaud-Bataille, Bruno Ragazzon, Xavier Bertagna, Jérôme Bertherat and Eric Clauser
Department of Endocrinology, Metabolism and Cancer, Institut Cochin, INSERM U567, University Paris Descartes, CNRS UMR8104, Paris, France
Analyzing gene expression (transcriptome) in tissue is now reliable using industrial pangenomic microarrays. Accumulating data on adrenal cortex and adrenocortical tumor transcriptomes have already identified striking transcriptome differences not only between adenoma and carcinoma but also between two sets of carcinoma, which have very different prognoses. These findings result in the development of diagnostic and prognostic molecular predictors, which improve the outcome deter- mination compared with standard clinical and patho- logical tools. These transcriptome data observing adrenocortical tumor phenotype in great but complex detail, combined with genomic and proteomic infor- mation, will function for future research investigating the pathophysiology of their tumorigenesis and hormo- nal secretion.
Pathophysiological and clinical applications of tumor transcriptomes
For several decades, the molecular analysis of pathophy- siology of a tissue or organ has been limited to a candidate gene approach. In the mid-1990s, technical improvements of molecular biology tools made it possible to measure the expression of several thousand genes at the same time, in the same tissue, in a single experiment. The transcriptomic era was born. After several years of improvement, tech- nology has now reached maturity, with industrial DNA chips and standardized procedures that greatly reduce experimental variability.
Applied to tumor tissues, transcriptome studies have two major goals - allowing a better understanding of tumor pathophysiology and identifying clinically relevant mar- kers for diagnosis, prognosis or treatment. Information for each tumor is analyzed by either unsupervised or super- vised bioinformatics methods. Unsupervised clustering classifies the tumors according to similarity of expression profiles. The tumors are ordered in a tree, called the dendrogram (Figure 1), where each branch corresponds to a tumor and the height of the branches (vertical bars) reflects the similarities of the tumor transcriptomes. Such a dendrogram identifies clusters of tumors, which, when compared with clinical annotations, usually show good agreement between pathology and transcriptomes. For
example, for several tumor types, benign and malignant tumors usually segregate in two different clusters [1]. In rare cases, unsupervised clustering identifies a cluster, which a posteriori corresponds to a specific form of the disease in terms of biology, prognosis or potential targeted therapy (class discovery). This is, for example, the case of
Glossary
Adrenal primary aldosteronism: excessive secretion of aldosterone by the adrenal cortex. This secretion is independent of external factors (ACTH and angiotensin II) and is the consequence of a benign tumor or hyperplasia of zona glomerulosa.
Aldosterone producing adenoma (Conn adenoma): benign tumor of the adrenal cortex, which produces an excess of aldosterone.
Dendrogram: tree, resulting from the classification of the adrenocortical tumors according to their transcriptome. Each branch represents a tumor type and the height of the branches (vertical bars) represents the similarity between the tumors.
Macronodular hyperplasia: bilateral hyperplasia of the adrenal cortex, with macronodules producing an excess of cortisol. This disease is potentially the consequence of an excessive expression of one or several G-protein-coupled receptors activating the cAMP pathway.
McFarlane stage: extension stage of an adrenal tumor, classified in four stages, depending on the local extension (stages I and II), regional extension (stage III) or general extension (metastasis, stage IV).
Primary pigmented nodular adrenal disease: bilateral hyperplasia of the adrenal cortex, with micronodules producing an excess of cortisol. This disease is often associated with other tumors (cardiac myxomas, lentiginosis), which is called Carney complex and is the consequence of germinal inactivating mutations of PRKAR1A, a gene coding a regulating subunit of the cAMP dependent protein kinase (PKA).
Supervised comparison: bioinformatics analysis of tumor transcriptomes, used to identify the most differentially expressed genes between two groups, which have been chosen a priori (examples: benign and malignant tumors or metastatic and non-metastatic tumors).
Training cohort: group of patients, representative of the disease, in which each is tested for the hypothesis of a statistical link between the level of expression of one or several genes and a clinical trait of the disease.
Unsupervised clustering: bioinformatic classification of the transcriptome of a series of samples (tumors) according to the similarity of their transcriptome and independently of clinical and pathological aspects. The resulting dendrogram classifies samples according to their transcriptome similarities and identifies closely related clusters of biological samples (tumors). Compar- ing this classification with the clinical and pathological annotations allows linking these phenotypic traits to transcriptome profiles.
Validation cohort: group of patients representative of the disease and independent of the training cohort, for which the statistical link is validated between the expression level of one or several genes and a clinical trait of the disease identified in the training cohort.
Weiss score: histopathological score of an adrenal tumor, corresponding to nine criteria of proliferation, nuclear abnormality and extension. Tumors with a Weiss score of 0-2 are considered benign, tumors with a Weiss score >3 are considered malignant, and tumors with a Weiss score of 2 or 3 have an undetermined status.
Corresponding author: Clauser, E. (eric.clauser@inserm.fr).
or death (yes = black, no = white) status of each tumor was added. Almost all tumors with no recurrence and Weiss scores between 0 and 2 form a first cluster (benign tumors), whereas a large majority of tumors with recurrence, 1 histopathological status using automated bioinformatics software. The resulting dendrogram (upper) identifies different clusters of tumors. Histopathological (Weiss score from 0 to 9) and clinical (recurrence; yes = black, no = white)
O 0 6000 1 O L 1 TRENDS in Endocrinology & Metabolism O / 00 00010 0000 100 1 I 1 U 0 1 1 0 1 11 0 0 P 2 0 + - 0 + 00 w ~ 1 0 0 + Good prognosis 00 UP 0 5 CO co 10 5 ” Adenomas Carcinomas 4 7 5 00 6 00 7 Bad prognosis CO 2 ₼ 4 4 0 Death Pathology: Weiss score Recurrence / metastases V 1 Figure 1. Unsupervised clustering of adrenocortical tumors. The transcriptomes of 92 adrenocortical tumors have been classified according to the similarity of their transcriptome and independently of their clinical or L 7 2 A A
death and Weiss score >3 are in the other cluster (malignant tumors). The group of malignant tumors is further divided in two subclusters with very different outcomes. Adapted from Figure 1 in Ref. [22] .
F
H
.
4
4
3.5
3
2.5
2
1.5
1
0.5
the “basal-like” subgroup of breast carcinoma, which over- expresses the EGF receptor and was identified by unsu- pervised clustering [2].
Supervised methods analyze the link between gene expression and clinical features among the tumors (Glos- sary). For example, lists of differentially expressed genes can be generated from comparing malignant and benign or metastatic and non-metastatic tumors [1]. These lists might decipher the signaling and molecular pathways involved in the pathophysiology of the disease. They might also help to characterize gene signatures that are specific to diagnostic, prognostic or therapeutic classes of tumors.
These approaches have been used to analyze the tran- scriptome of the adrenal cortex and its tumors. The adrenal cortex produces three types of steroids (cortisol, aldoster- one and androgens) in three different structural layers of the gland (Figure 2). The tumoral development of the gland presents several forms: micro- or macronodular hyperpla- sia, benign tumors, which are frequent (1-10% of autopsy series), and malignant tumors, which are rare but of very poor prognosis. In addition, these tumors and hyperplasia might produce in excess one or several of the steroids mentioned above. In the past decade, several studies attempted to answer several important questions concern- ing these adrenal tumors, such as: what are the molecular mechanisms of hormonal overproduction? What is the physiopathology of benign and malignant tumorigenesis and are they sequential? And, are there molecular markers that can help with the diagnosis and prognosis of malig- nant tumors? The goal of this review is to analyze the benefit of such approaches for understanding the patho- physiology and clinical care of patients with adrenocortical tumors.
Transcriptome of normal adrenocortical tissue
Quantification of gene expression (transcriptome) of a tissue provides information on the function(s) and differ- entiation potential of that tissue. Hu et al. first charac- terized the human adrenal cortex transcriptome by extensive sequencing of a complementary DNA (cDNA) library [3]. We and others generated cDNA libraries of adrenocortical tissue samples [4,5] using a closely related method called Serial Analysis of Gene Expression [6,7]. These sequencing-based methods use the sequence infor- mation of each transcript to identify and count it. Quanti- fication is then absolute (i.e. expressed as a copy number for a given transcript over the total number of sequenced transcripts).
The adrenal cortex transcriptome shows several specific features. First, steroidogenic enzymes are among the most abundant transcripts, indicating the high transcriptional cost of steroidogenic differentiation (Table 1). Second, transcripts for chaperones and proteins of the glutathione system are highly expressed, showing the importance of oxidative stress associated with active steroidogenesis and of cellular responses to this stress. Third, the number of mitochondrial transcripts reflects the mitochondrial activity associated with steroid production. Finally and perhaps surprisingly, neuroendocrine markers are also expressed. Although this might reflect medullar contami- nation of the cortex sample or the presence of medullar
| Tissue Author Year, Reference | Normal cortex | Zona glomerulosa Assie et al. 2005 [4] | PPNAD Horvath et al. 2006 [5] | APA Assie et al. 2005 [4] | ||
|---|---|---|---|---|---|---|
| Hu et al. 2000 [3] | Horvath et al. 2006 [5] | |||||
| Steroidogenic enzymes | ||||||
| Steroidogenic acute regulatory protein | STAR | 161 | 24 | 7 | 42 | 9 |
| 3B-Hydroxysteroid dehydrogenase type 2 | HSD3B2 | 108 | 0 | 32 | 5 | 36 |
| Aldosterone synthase | CYP11B2 | 13 | 0 | 22 | ||
| 11-beta-Hydroxylase | CYP11B1 | 4 | 6 | 10 | 2 | |
| 21-Hydroxylase | CYP21A2 | 1 | 8 | 7 | 180 | |
| 17-Hydroxylase/17,20 lyase | CYP17 | 79 | 55 | |||
| Steroidogenic substrates providing | ||||||
| Adrenodoxin | ADX | 11 | 6 | 16 | ||
| Cytochrome P450 oxydoreductase | POR | 4 | 4 | 0 | 8 | |
| Glutathione and antioxidant | ||||||
| Aldose reductase | 47 | |||||
| Glutathione peroxidase | 11 | |||||
| Heat shock 90-kDa protein 1 | HSPCB | 18 | 29 | |||
| Heat shock 70-kDa protein 6 | HSPA6 | 1 | 0 | |||
| Heat shock 27-kDa protein 7 | HSPB7 | 0 | 2 | |||
| Heat shock 70-kDa protein 8 | HSPA8 | 5 | 15 | |||
| Glutathione S transferase theta 1 | GSTT1 | 1 | 3 | |||
| Peroxiredoxin 6 | PRDX6 | 0 | 3 | |||
| Neuroendocrine markers | ||||||
| Chromogranin B | CGB | 9 | 6 | 73 | ||
Gene expression of the different players of steroidogenesis is analyzed in total adrenal cortex, in the zona glomerulosa, which produces aldosterone, cortisol producing tumors (primary pigmented nodular adrenocortical disease or PPNAD) or aldosterone producing adenoma (APA). Different transcriptome publications have quantified these adrenocortical or endocrine transcripts and expressed them as number for 10,000 transcripts.
islets within the cortex, immunohistochemical analysis shows colocalization of steroidogenic enzymes and neuro- endocrine markers, such as chromogranin B, in authentic adrenocortical hyperplasia [5].
Such a tissue-specific transcriptome signature might help discriminate between a primary adrenocortical carci- noma (ACC) and a secondary cancer, as reported for other tumor types [8]. Indeed, primary ACCs can be difficult to distinguish from clear cell carcinomas, hepatocellular car-
cinomas, pheochromocytomas or melanomas. For this differential diagnosis, pathologists currently use immuno- histochemistry, targeting Melan-A, synaptophysin and chromogranin A [9-11]. As an example, we recently experi- enced the case of a malignant adrenocortical tumor, diag- nosed as an ACC, the transcriptome of which was not specific of the gland but was largely enriched in hepatic transcripts. The follow-up of the patient revealed a primary hepatocarcinoma (personal data).
(a) Cholesterol
(b)
CYP11A1
3ßHSD2
CYP21A1
Pregnenolone
Progesterone
Deoxycorticosterone
CYP11B2
Zona:
Aldosterone
CYP17A1
Glomerulosa
CYP21A1
CYP11B1
17OH-Pregnenolone
17OH-Progesterone → Deoxycortisol
Cortisol
3ßHSD2
Fasciculata
CYP17A1
CYP17A1
DHEA
3ßHSD2
44Androstenedione = adrenal androgens
SULT2A1
Reticularis
SDHA
TRENDS in Endocrinology & Metabolism
Transcriptome and hormone secretory profile of adrenocortical tumors
The adrenal cortex produces mineralocorticoids, glucocor- ticoids and androgens. Specific hormonal production is finely tuned in time and space, in relation with the struc- tural and functional organization of the adrenal cortex (Figure 2). This production is altered in adrenocortical tumors and becomes excessive and not regulated by their physiological stimuli, namely adrenocorticotrophic hor- mone (ACTH) and angiotensin II.
Several studies on candidate genes showed the link between the expression level of specific genes and the hormone secretory pattern. For example, the expression level of steroidogenic acute regulatory protein (StAR) is correlated to the rate of steroidogenic enzymes and of steroid hormone secretion (cortisol and aldosterone in the adrenal cortex and estradiol or testosterone in the gonads) after acute stimulation (minutes) by their respect- ive stimuli (ACTH, angiotensin II or luteinizing hormone, LH) [12]; aldosterone synthase (CYP11B12) expression level is correlated to aldosterone secretion after chronic (hours) stimulation by angiotensin II or K+ [13]. Several other steroidogenic enzymes and substrate supplying enzymes are necessary for steroidogenesis, and some of these genes facilitate steroidogenesis when overexpressed [14]. Thus, there is a clear correlation between gene expres- sion profile of steroidogenic enzymes and related proteins and steroidogenesis functional activity of the tissue, indi- cating that the transcriptome is useful for characterizing the hormonal status of an adrenocortical tumor.
Primary aldosteronism (PAL)
The transcriptome of primary aldosteronism [15] has been characterized through studies of aldosterone producing adenoma (APA), also called Conn adenoma [4,16,17], but no information is available for the rare form of PAL called bilateral adrenal hyperplasia. Aldosterone synthase (CYP11B2) is highly, significantly and consistently over- expressed in these adenomas compared with normal tis- sues. Other enzymes of the aldosterone synthesis pathway (CYP21A2, HSD3B2) are moderately or slightly overex- pressed, orienting steroidogenesis towards aldosterone production. Surprisingly, other steroidogenic enzymes involved in cortisol or androgen production, such as 17- hydroxylase/17,20-lyase (CYP17A1) and 11ß-hydroxylase (CYP11B1) are often equally or overexpressed in APA compared with adrenal cortex. This observation contrasts with the exclusive aldosterone production in APA [18].
Expression of these genes varies within and between studies, reflecting the variability among tumors. A recent transcriptome study clearly distinguishes two types of APAs, with or without high aldosterone synthase expres- sion [17]. These findings raise the question of the definition of an APA: does it correspond to a unique aldosterone- producing tumor surrounded by “normal” adrenal cortex, or does it refer to a global aldosterone-producing adrenal disease with adenoma? In several series of APA, one cannot exclude that some cases correspond to non-secreting ade- noma surrounded by hyperplasia of the glomerulosa. This is illustrated by an in situ hybridization study [19], which shows different aspects of aldosterone synthase expression
in PAL, including aldosterone synthase expression restricted to a unique adenoma surrounded by non-labeled glomerulosa or non-labeled adenomas surrounded by labeled glomerulosa. Finally, the variations in steroido- genic enzyme expression levels can also reflect differences in the chosen control tissue (adjacent zona glomerulosa or normal and entire adrenal cortex).
Other genes identified as overexpressed in APA com- pared with normal adrenal include: genes involved in providing substrates to steroidogenesis (cholesterol with high-density lipoprotein receptor), electrons with the adre- nodoxin (ADX) and the cytochrome P450 oxydoreductase (POR) [4]; genes involved in calcium signaling (calcium signaling proteins and cofactors) are highly expressed in APA [4]; and G-protein-coupled receptor (GPCR) genes (LH receptor and several other GPCRs) are highly expressed in APA [20,21]. This raises the question of the presence of illegitimate receptors, as described for macronodular hyperplasia. The functional relevance of these expressions remains to be determined.
ACTH-independent cortisol production
The transcriptome of several tumors, producing cortisol independently from ACTH regulation and therefore corre- sponding to primary defects of the adrenal cortex, has been reported. Some of these include adenomas (CPA) [16], a unique benign tumor overproducing cortisol; macronodu- lar hyperplasia [22,23], with multiple, bilateral supracen- timetric benign nodules, potentially the consequence of GPCR illegitimate expression or overexpression; primary pigmented nodular adrenocortical disease (PPNAD) [5], a hyperplasia of the adrenal cortex with pigmented micro- nodules as a result of constitutive activation of cAMP- dependent protein kinase; and ACC [24-26], a rare but dramatic cancer, which might also produce androgens and aldosterone. All these tissues express the steroidogenic enzymes required for cortisol production, including 3B- hydroxysteroid dehydrogenase type 2 (HSD3B2), which is highly expressed [5]. In contrast, aldosterone synthase (CYP11B2) expression is not reported in the transcriptome of these tissues. This indicates that the register of expressed steroidogenic enzymes correlates with the speci- ficity of cortisol production.
However, important variations in the levels of expres- sion of the steroidogenic enzymes are observed between the different types of adrenal tumors: steroidogenic enzymes are highly expressed in small and well differentiated tumors (CPA and PPNAD), whereas low expression is reported in larger and/or less differentiated tumors (macronodular hyperplasia and cortisol producing ACC). These findings, garnered from the transcriptome profiles of the tumors, correlate well with the level of cortisol pro- duction per gram of tissue, which is much higher in small tumors compared with larger tumors.
Macronodular hyperplasia (AIMAH) is a peculiar con- dition clinically characterized by an unusual increase of cortisol secretion in response to various hormonal stimu- lations, including gastric inhibitory polypeptide (GIP) [27], catecholamines and vasopressin (illegitimate response). AIMAH is characterized biologically by the abnormal expression of different GPCRs for these hormones. Receptor
expression can be either increased compared with the nor- mal adrenal cortex or illegitimate, when the normal tissue does not express the receptor.
Interestingly, transcriptome analysis of AIMAH adds new information for the pathology of the disease. Unsu- pervised clustering classifies the AIMAH into several groups. In each group, AIMAH presents closely related transcriptomes and similar types of clinical illegitimate responses [22,23]. In addition, some familial cases of AIMAH have been reported, and the illegitimate response profile is conserved between affected family members [28,29]. Together, these data suggest that AIMAH can be classified into different pathophysiological groups, each group characterized by a specific pattern of clinical illegi- timate responses (intermediate phenotype) and a specific transcriptome, possibly related to GPCR expression.
The cyclic AMP pathway is an important player in regulating cortisol secretion via ACTH, and some of these tumors (PPNAD) are the consequence of genetic mutations of proteins in this pathway. Therefore, it was interesting to make a supervised analysis of the transcriptome oriented toward the expression of the different proteins involved or regulated by this signaling pathway in such tumors [30]. However, no signature of cyclic AMP pathway activation has yet been identified in the transcriptome of these tumors.
The transcriptome of adrenal androgen production
Steroidogenic genes involved in androgen production has been reported in the zona reticularis [31] and fetal adrenal transcriptomes (Box 1) [32]. The androgen production signature in the adrenal cortex includes low expression of HSD3B2 and high expression of sulfotransferase 2A1 (SULT2A1) and cytochrome P450 side chain cleavage (CYP11A1). No transcriptome has been described yet for the rare adrenocortical androgen-producing tumors.
How does the transcriptome reveal the malignant status of an adrenocortical tumor?
Discriminating between a frequent benign adrenocortical tumor and a rare but dramatic carcinoma is sometimes not an easy task. It is evoked today on clinical and imaging features (tumor size, hormonal secretory profile, Magnetic resonance imaging (MRI) density). The diagnosis is con- firmed by pathology; the pathological Weiss score [9] is the most commonly used. The Weiss score ranges from 0 to 9 according to several criteria of cell morphology, prolifer- ation and invasiveness. Tumors with a Weiss score of 0 or 1 are benign, a score of 4 or more reflects malignancy, and scores of 2 or 3 suggests undetermined malignancy. Immu- nohistochemical (Ki67) or molecular (17p13 LOH and IGF- II overexpression) markers might also aid this diagnosis.
Benign and malignant ACTs have a different transcriptome
Comparing transcriptomes of benign and malignant adre- nocortical tumors was expected to clarify this issue. Sev- eral recent publications compared transcriptomes in series of benign and malignant adrenocortical tumors [25,26,33,34]. The data were first classified by unsuper- vised clustering, according to the similarity of their tran-
scriptomes and independent of their clinical or pathological status. In all studies, this process identified two major clusters of tumors. By analyzing the clinical status of tumors in each of these two groups, it was evident that one group contains benign tumors and the other malignant tumors. As an example, the hierarchical unsu- pervised clustering of the largest series published by our group (Figure 1) classifies 98% of the clinically identified benign tumors in one cluster and 96% of the clinically identified malignant tumors in the other cluster. Interest- ingly, 86% of the tumors with an intermediate phenotype (Weiss score 2 or 3) cluster into the malignant group. These data clearly indicate that the transcriptomes of benign and malignant adrenocortical tumors differ.
The malignancy signature
The clear discrimination of benign and malignant ACTs using unsupervised clustering demonstrates that a large number of genes are differentially expressed in benign versus malignant tumors (Tables 2 and 3). These genes constitute a malignancy signature, which can be split in two categories: a general proliferation signature that is common to all cancer types, and a signature specific to ACTs. The general proliferation signature is mainly com- posed of cell cycle regulators and cell cycle effectors [1]. Table 2 reports the most important corresponding genes found in ACTs.
The adrenal-specific malignancy signature includes growth factor pathways (Table 2), among which is the insulin-like growth factor II (IGF-II) pathway. Different transcriptome analyses confirmed IGF-II overexpression in a majority of ACCs and also demonstrated that among growth factors, IGF-II is the most overexpressed. Tran- scriptome analyses also provided a global view of the IGF- II pathway, including receptors and binding proteins. It appears that IGF-I receptor, which mediates the trophic effects of IGF-II, is expressed at the same level in benign and malignant tumors. More recently, using H295R cells, a human adrenocortical cancer cell line, grafts of these cells in nude mice and IGF-I receptor inhibitors, two studies provided further convincing evidence of a major role of IGF-II in adrenal malignancy [15,35]. Several other growth factors are overexpressed in ACC (Table 2), but their functional relevance remains to be determined.
Clinical application: diagnosis of malignancy
Because major differences exist between the transcriptome of benign and malignant tumors, it should be possible to set up diagnostic tools that can be transferred to clinical prac- tice. With that aim, several authors proposed to reduce the number of eligible genes and use quantitative PCR-based approaches rather than microarray approaches to reduce costs and facilitate the development of hospital tests.
Initial attempts classified the tumors in two groups (benign and malignant) according to their Weiss scores and made a supervised analysis of the transcriptome to identify the most differentially expressed genes in the two groups. Using this strategy, Slater et al. perfectly discriminated the tumors in their cohort with the top 74 most differentially expressed genes [36]. However and by definition, with such an approach the gene signature of
| Pathway | Protein class | Gene symbol | Protein name | Fold change | ||||
|---|---|---|---|---|---|---|---|---|
| Reference | [25] | [26] | [33] | [24] | [38] | |||
| Steroidogenesis | HSD3B2 | Hydroxy-delta-5-steroid dehydrogenase, | 14 | 5 | ||||
| 3 beta- and steroid delta-isomerase 2 | ||||||||
| CYP11B1 | cytochrome P450, family 11, subfamily B, polypeptide 1 | 14 | 6 | + ☒ | ||||
| Electron transport | FMO2 | Flavin containing monooxygenase 2 | 7 | |||||
| (non-functional) | ||||||||
| RSPO3 | R-spondin 3 homolog (Xenopus laevis) | 6 | ||||||
| CYBRD1 | Cytochrome b reductase 1 | 5 | 4 | |||||
| AOX1 | aldehyde oxidase 1 | 4 | 7 | |||||
| STEAP4 | STEAP family member 4 | 4 | ||||||
| Metabolism | Lipid | APOC1 | apolipoprotein C-I | 5 | ||||
| PLTP | phospholipid transfer protein | 5 | ||||||
| Carbohydrate | B4GALT6 | UDP-Gal:betaGlcNAc beta 1,4-galactosyltransferase, polypeptide 6 | 5 | |||||
| MAN1A1 | Mannosidase, alpha, class 1A, member 1 | 4 | ||||||
| Others | AADAC | Arylacetamide deacetylase (esterase) | 14 | 7 | ||||
| ADH1B/ADH1C | Alcohol dehydrogenase IB/alcohol | 13 | 25 | |||||
| dehydrogenase 1C | ||||||||
| ALDH1A1 | Aldehyde dehydrogenase 1 family, member A1 | 7 | 12 | 5 | ||||
| AMDHD1 | Amidohydrolase domain containing 1 | 5 | ||||||
| INMT | Indolethylamine N-methyltransferase | 5 | ||||||
| C7orf10 | Chromosome 7 open reading frame 10 | 4 | ||||||
| EPHX2 | Epoxide hydrolase 2, cytoplasmic | 3 | 4 | |||||
| Cell cycle | Apoptosis | SEMA6A | Sema domain, transmembrane domain (TM) | 4 | ||||
| regulation/apoptosis | and cytoplasmic domain 6A | |||||||
| Others | NOV | Nephroblastoma overexpressed gene | 9 | |||||
| DCN | Decorin | 7 | ||||||
| HOXA5 | Homeobox A5 | 5 | ||||||
| SERPINF1 | Serpin peptidase inhibitor, clade F | 4 | ||||||
| (alpha-2 antiplasmin,), member 1 | ||||||||
| Transcription | Nuclear | NR4A2 | Nuclear receptor subfamily 4, group A, | 7 | ||||
| receptors | member 2 | |||||||
| RARRES2 | Retinoic acid receptor responder | 7 | 6 | 3 | ||||
| (tazarotene induced) 2 | ||||||||
| Transporters and channels | ABCG1 | ATP-binding cassette, sub-family G (WHITE), | 6 | |||||
| member 1 | ||||||||
| KCNK2 | Potassium channel, subfamily K, member 2 | 6 | ||||||
| SLC16A9 | Solute carrier family 16, member 9 | 6 | ||||||
| (monocarboxylic acid transporter 9) | ||||||||
| MRC1/MRC1L1 | Mannose receptor, C type 1/mannose receptor, | 6 | ||||||
| C type 1-like 1 | ||||||||
| ABCA1 | ATP-binding cassette, subfamily A (ABC1), member 1 | 6 | ||||||
| SCNN1A | Sodium channel, nonvoltage-gated 1 alpha | 5 | ||||||
| KCNQ1 | Potassium voltage-gated channel, KQT-like subfamily, member 1 | 5 | 4 | |||||
| HEPH | Hephaestin | 4 | ||||||
| ABCC3 | ATP-binding cassette, subfamily C (CFTR/MRP), member 3 | 3 | 4 | 6 | ||||
| SLC9A3R1 | Solute carrier family 9 (sodium/hydrogen exchanger), member 3 regulator 1 | 3 | 4 | |||||
| KCNJ8 | Potassium inwardly rectifying channel, subfamily J, member 8 | 3 | 8 | |||||
Benign tumors differ from malignant tumors by the increased expression of registers of genes involved in steroidogenesis, apoptosis, metabolism and transport. The most differentially expressed genes of these pathways are listed. These genes are overexpressed in adenoma compared with carcinoma by fold change, indicated in the right columns and depending on the publication. The symbol ”+” is provided when the ratio is not quantified but the gene is clearly overexpressed in adenoma.
malignancy cannot be a better test than the Weiss score, which is the reference. To overcome this limitation, some authors proposed to use another definition of malignancy - the probability of recurrence. de Fraipont et al. found two clusters of genes that predicted recurrence - the IGF-II cluster and the steroidogenesis cluster [24]. These molecu- lar predictors were as efficient as the prediction based on pathology, using a Weiss score cut-off value of 4. More
recently, we identified a new molecular predictor by clas- sifying a cohort of 47 adrenocortical tumors (training cohort) in two groups (with or without recurrence) and analyzing their differentially expressed genes [25]. Exten- sive bioinformatics and statistical analyses of all the genes and combinations of genes indicates that the subtraction between the expressions of DGL7 and PINK1 genes is the best molecular predictor of recurrence and malignancy.
| Pathway | Protein class | Gene symbol | Protein name | Fold change | ||||
|---|---|---|---|---|---|---|---|---|
| Reference | [25] | [26] | [33] | [24] | [38] | |||
| Proliferation | Cyclins | CCNB1 | Cyclin B1 | 7 | ||||
| signature | CCNB2 | Cyclin B2 | 6 | 9 | ||||
| CDKN3 | Cyclin-dependent kinase inhibitor 3 | 5 | 7 | |||||
| CCNE1 | Cyclin E1 | 4 | 10 | |||||
| CCNA2 | Cyclin A2 | 3 | ||||||
| Other regulators | CDC2 | Cell division cycle 2, G1 to S and G2 to M | 6 | 8 | ||||
| GAS2L3 | Growth arrest-specific 2 like 3 | 5 | + | |||||
| MAD2L1 | MAD2 mitotic arrest deficient-like 1 (yeast) | 4 | 4 | |||||
| Chromosome | PTTG1 | Pituitary tumor-transforming 1 | 6 | |||||
| segregation and | ||||||||
| duplication | ||||||||
| UBE2C | Ubiquitin-conjugating enzyme E2C | 6 | ||||||
| MLF1IP | MLF1 interacting protein | 6 | ||||||
| RRM2 | Ribonucleotide reductase M2 polypeptide | 6 | ||||||
| PRC1 | protein regulator of cytokinesis 1 | 5 | ||||||
| TPX2 | TPX2, microtubule-associated, homolog (Xenopus laevis) | 5 | ||||||
| TOP2A | Topoisomerase (DNA) II alpha 170-kDa | 4 | 9 | |||||
| Growth factors and receptors | IGF2 | Insulin-like growth factor 2 (somatomedin A) | 27 | 106 | 432 | + | 6 | |
| FGFR1 | Fibroblast growth factor receptor 1 | 2 | 4 | + | 4 | |||
| FGFR4 | Fibroblast growth factor receptor 4 | 2 | + | + | ||||
| Signal transduction | PLA2G1B | Phospholipase A2, group IB (pancreas) | 7 | |||||
| PBK | PDZ binding kinase | 6 | ||||||
| AK3L2 | Adenylate kinase 3-like 2 | 5 | ||||||
| RACGAP1 | Rac GTPase activating protein 1 | 5 | ||||||
| AK3L1/AK3L2 | Adenylate kinase 3-like 1/adenylate kinase 3-like 2 | 5 | ||||||
Malignant tumors differ from benign tumors by their increased expression of genes involved in proliferation and growth. The most differentially expressed genes of these pathways are listed. The genes are overexpressed in carcinoma compared with adenoma by fold change, indicated in the right columns and depending on the publication. The symbol ”+” is provided when the ratio is not quantified but the gene is clearly overexpressed in carcinoma.
Discs large homolog 7 (drosophila) gene (DLG7) is a gene of unknown function, which is involved in stem cell survival and carcinogenesis (colon, liver); it is overex- pressed in ACC. PTEN induced putative kinase (PINK1) is activated by the tumor suppressor gene PTEN and is involved in mitochondrial integrity; its expression is reduced in ACC. The accuracy of this predictor DLG7- PINK1 was tested next on an independent cohort (vali- dation cohort) of 94 non-metastatic tumors (adenomas and carcinomas). This molecular predictor appears to be at least as good as the Weiss score-based prediction for the diagnosis of carcinoma versus adenoma and can be used routinely at hospitals.
Several clinical benefits are expected from this molecu- lar predictor (or others). First, such a predictor can confirm pathological results. In addition, it will be crucial in many hospitals where pathological expertise for this disease is not available, because ACC is so rare. Finally, a multi- variate statistical analysis seems to indicate that the molecular predictor contains some additional and indepen- dent information for the diagnosis that pathology (Weiss score) cannot provide. This suggests that this molecular predictor can improve the currently available diagnostic tools. To confirm the importance of this molecular predictor for the diagnosis of adrenocortical tumors, the next steps are to set up multicentric and prospective studies on larger cohorts in a routine base at the hospital. These studies will also determine the interest of this molecular predictor for the diagnosis of malignancy of tumors with intermediate Weiss scores (2 and 3).
A new classification of adrenocortical carcinomas
The prognosis of ACC is very poor (15-45% survival at 5 years). The main prognostic factor is the tumor extension. The McFarlane stage can be used to reflect this extension (stage I and II: localized tumor < or > to 5 cm, stage III: regional extension, stage IV: metastasis or invasion of adjacent organs). Up to recently, this factor was the only factor able to predict the prognosis of an ACC.
The two types of adrenocortical carcinomas
Using unsupervised hierarchical clustering, we recently identified two different groups of ACCs with different tran- scriptomes (Figure 1) [25]. By analyzing the clinical differ- ences between these two groups, we were surprised to observe a major difference in terms of overall survival; 5- year survival was 91% in the good prognosis group and 20% in the bad prognosis group (Log-rank p-value = 1.2 x 10-4). Other clinical parameters were unable to distinguish these two groups (Figure 3). The Weiss score was often higher in tumors of the “bad prognosis group” compared with the ‘good prognosis group’, but this score does not predict the prognosis. Among the 1850 genes that are significantly differently expressed between the two groups, ~500 are overexpressed more than 2-fold in the “bad prognosis” group and 250 in the “good prognosis” group. Bioinformatics analysis revealed that the “bad prognosis” group overex- presses many transcription factors. For example, zinc finger protein 711 and 521, nuclear receptor subfamily 4-group A-member 3 and RAR-related orphan receptors A and B are in the top 100 overexpressed genes, and mitotic cell cycle
(a)
Overall survival molecular prediction (BUB1-PINK1)
100
1
80
Good prognosis
60
40
20
Bad prognosis
0
Log-rank p. : 1.68e-06
0
1
2
3
4
5
6
7
8
9
10
11
Years
(b)
9
..
8
..
…
Tumor pathology (Weiss score)
7
…
6
5
..
..
4
3
·
2
1
Good prognosis
Bad prognosis
(c)
4
.
Tumor stage (McFarlane)
3
…
2
…
1
…
Good prognosis
Bad prognosis
TRENDS in Endocrinology & Metabolism
genes (CDK6, cell division cycle 2 G1 to S and G2 to M and cyclin B2) are in the top 100 genes. In the “good prognosis” group, an enrichment in cell metabolism was observed; 4- aminobutyrate aminotransferase, acyl-CoA synthase mem- ber 4 and Glucose 6 phosphate dehydrogenase are in the top 100 overexpressed genes, whereas genes involved in intra- cellular transport (fatty acid transporter member 2, aro- matic amino acid transporter and potassium voltage-gated channel subfamily H member 2) are in the top 100, as well as apoptosis and cell differentiation genes. Interestingly, in this “good prognosis” group several signaling proteins are also more expressed, including protein kinase Ca, phospho- lipase C X domain containing 3 and the prolactin receptor, the latter of which is 25 times more expressed in this group than in the “bad prognosis” group.
Other recent studies also identified two groups of car- cinomas using unsupervised clustering of transcriptome
Box 1. Pediatric adrenocortical carcinoma
Beyond the age of onset, childhood ACCs differ from adult ACC in different ways [10]:
. These tumors are more frequently associated with cancer predisposition syndromes (Li Fraumeni and Beckwith-Wiede- mann).
· The hormone secretion profile is different, with androgen excess occurring more often in children.
· The pathology is different, and discriminating benign from malignant tumors is even more challenging in children ACTs [38].
West et al. reported the only available microarray study of pediatric adrenocortical tumors [39]. By comparing ACTs to normal tissue, the authors could confirm or extend previous knowledge of the molecular biology of these pediatric tumors:
· IGF2 is one of the most overexpressed genes, similar to adult ACTs.
· FGFR4 is one of the most overexpressed genes in children ACTs.
· SF1, the locus of which has been demonstrated to be amplified in these ACTs, is also overexpressed.
· Among the most repressed genes are 3ß-hydroxysteroid dehy- drogenase type 2 and the related transcription factors, NURR1 and NGF-B.
The pediatric ACT transcriptome is closer to fetal or child normal adrenal transcriptome than adult adrenal tissue transcriptome, suggesting fetal origin of the pediatric tumors [32]. In contrast to that observed in adult ACTs, the authors did not report any clear discrimination between ACA and ACC using unsupervised cluster- ing methods. These results question the nature of the so-called ACAs in children: are these truly benign tumors?
data [34,37]. These two groups also showed a difference in terms of survival. Giordano et al. further characterized these two groups of carcinomas, showing that carcinomas with high degree of mitoses were more abundant in the ‘bad prognosis group’ than in the ‘good prognosis group’. In addition, using multivariate analysis, they confirmed that the transcriptome of an ACC contains some prognostic information that clinic, tumor extension (McFarlane stage) and pathology cannot provide [37]. Further studies are needed to understand the biological bases of this transcrip- tome-based prognostic information.
Clinical application
This discovery that two groups of carcinoma have different prognoses and different transcriptomes prompted us to develop a molecular predictor of prognosis, with a similar methodology than that developed for the molecular pre- dictor of diagnosis classifying adenoma and carcinoma. Using a cohort (training cohort) of 23 malignant or poten- tially malignant tumors, a clinical indicator (overall sur- vival), and an extensive statistical analysis to determine which genes or combination of genes are differentially expressed, we identified the best prognostic predictor as a subtraction between BUB1B (budding uninhibited by benzimidazoles 1 homolog beta) and PINK1 expression levels. BUB1B-PINK1 measured by RT-qPCR was vali- dated in an independent cohort of 35 malignant or poten- tially malignant tumors, showing an important discrimination between tumors predictive of good prog- nosis and those predictive of bad prognosis [25]. The pro- spective and multicentric validation of this molecular prognostic predictor is underway.
Concluding remarks and perspectives
In the past few years, the scientific community has accu- mulated an enormous amount of information with the pangenomic transcriptomes of a large number of adreno- cortical tumors. This information is purely observational and provides an extremely precise phenotype of these tumors, which is difficult to interpret owing to the vast amount of information. The major accomplishment is the identification of molecular predictors for the diagnosis of these tumors and the prognosis of ACC. However, these predictors have to be universally validated by increasing the number of patients in the cohorts (particularly for the prognostic predictor) and by multiplying the centers where such molecular predictor analyses can be done.
Understanding the pathophysiology of these tumors by reading lists of hundreds of differentially expressed genes and cellular pathways is extremely challenging. Transcrip- tome analysis can only provide directions and orientations for future functional studies of the mechanisms of tumor- igenesis and hormonal secretion in these adrenal tumors. By overexpressing or knocking out a gene or a pathway in adrenocortical cell lines or animals (transgenic mice) and further analyzing the transcriptome data to choose the good targets, we should be able in the next 5-10 years to better understand the pathophysiology of these tumors.
Therefore, further information should come from a deeper analysis of the different subgroups of ACT transcrip- tomes but also from the correlations of these transcriptome modifications with genomic alterations. In addition, we should keep in mind that these genomic and expression modifications do not always correlate with the levels and activities of the real players of the cell, which are the proteins. Therefore, additional and comparative proteomic data are required.
Finally, we should not forget that in addition to the characterization of well-identified groups of tumors, obser- vation of individual tumors is also sometimes very infor- mative in terms of pathophysiology. For these tumors, the future will be personalized medicine, with an individua- lized diagnostic and prognostic determination for clinical practice.
Acknowledgements
The authors would like to acknowledge the “Ligue Contre le Cancer” for supporting our transcriptome studies through the “Carte d’Identité des tumeurs” project, particularly Dr Aurélien de Reynies and Dr Jacqueline Godet. We are indebted to the Conny Maeva Fundation, which supported our lab activity. We would also like to thank Dr Rosella Libe and Dr Lionel Groussin for their thorough comments.
References
1 Whitfield, M.L. et al. (2006) Common markers of proliferation. Nat. Rev. Cancer 6, 99-106
2 Cheang, M.C. et al. (2008) Gene expression profiling of breast cancer. Annu. Rev. Pathol. 3, 67-97
3 Hu, R.M. et al. (2000) Gene expression profiling in the human hypothalamus-pituitary-adrenal axis and full-length cDNA cloning. Proc. Natl. Acad. Sci. U. S. A. 97, 9543-9548
4 Assie, G. et al. (2005) Steroidogenesis in aldosterone-producing adenoma revisited by transcriptome analysis. J. Clin. Endocrinol. Metab. 90, 6638-6649
5 Horvath, A. et al. (2006) Serial analysis of gene expression in adrenocortical hyperplasia caused by a germline PRKAR1A mutation. J. Clin. Endocrinol. Metab. 91, 584-596
6 Rebouissou, S. et al. (2009) Frequent in-frame somatic deletions activate gp130 in inflammatory hepatocellular tumours. Nature 457, 200-204
7 Velculescu, V.E. et al. (1995) Serial analysis of gene expression. Science 270, 484-487
8 Bender, R.A. and Erlander, M.G. (2009) Molecular classification of unknown primary cancer. Semin. Oncol. 36, 38-43
9 Lau, S.K. and Weiss, L.M. (2009) The Weiss system for evaluating adrenocortical neoplasms: 25 years later. Hum. Pathol. 40, 757-768
10 Ribeiro, R.C. and Figueiredo, B. (2004) Childhood adrenocortical tumours. Eur. J. Cancer 40, 1117-1126
11 Zhang, H. et al. (2008) Comparison of immunohistochemical markers in the differential diagnosis of adrenocortical tumors: immunohistochemical analysis of adrenocortical tumors. Appl. Immunohistochem. Mol. Morphol. 16, 32-39
12 Clark, B.J. and Stocco, D.M. (1997) Steroidogenic acute regulatory protein: the StAR still shines brightly. Mol. Cell. Endocrinol. 134, 1-8
13 Bassett, M.H. et al. (2004) The regulation of aldosterone synthase expression. Mol. Cell. Endocrinol. 217, 67-74
14 Miller, W.L. (2008) Steroidogenic enzymes. Endocr. Dev. 13, 1-18
15 Barlaskar, F.M. et al. (2009) Preclinical targeting of the type I insulin- like growth factor receptor in adrenocortical carcinoma. J. Clin. Endocrinol. Metab. 94, 204-212
16 Bassett, M.H. et al. (2005) Expression profiles for steroidogenic enzymes in adrenocortical disease. J. Clin. Endocrinol. Metab. 90, 5446-5455
17 Lenzini, L. et al. (2007) Heterogeneity of aldosterone-producing adenomas revealed by a whole transcriptome analysis. Hypertension 50, 1106-1113
18 Honda, T. et al. (2001) Combined primary aldosteronism and preclinical Cushing’s syndrome: an unusual case presentation of adrenal adenoma. Hypertens. Res. 24, 723-726
19 Enberg, U. et al. (2004) Postoperative differentiation between unilateral adrenal adenoma and bilateral adrenal hyperplasia in primary aldosteronism by mRNA expression of the gene CYP11B2. Eur. J. Endocrinol. 151, 73-85
20 Saner-Amigh, K. et al. (2006) Elevated expression of luteinizing hormone receptor in aldosterone-producing adenomas. J. Clin. Endocrinol. Metab. 91, 1136-1142
21 Ye, P. et al. (2007) G-protein-coupled receptors in aldosterone- producing adenomas: a potential cause of hyperaldosteronism. J. Endocrinol. 195, 39-48
22 Bourdeau, I. et al. (2004) Gene array analysis of macronodular adrenal hyperplasia confirms clinical heterogeneity and identifies several candidate genes as molecular mediators. Oncogene 23, 1575-1585
23 Lampron, A. et al. (2006) Whole genome expression profiling of glucose- dependent insulinotropic peptide (GIP)- and adrenocorticotropin- dependent adrenal hyperplasias reveals novel targets for the study of GIP-dependent Cushing’s syndrome. J. Clin. Endocrinol. Metab. 91, 3611-3618
24 de Fraipont, F. et al. (2005) Gene expression profiling of human adrenocortical tumors using complementary deoxyribonucleic acid microarrays identifies several candidate genes as markers of malignancy. J. Clin. Endocrinol. Metab. 90, 1819-1829
25 de Reynies, A. et al. (2009) Gene expression profiling reveals a new classification of adrenocortical tumors and identifies molecular predictors of malignancy and survival. J. Clin. Oncol. 27, 1108-1115
26 Giordano, T.J. et al. (2003) Distinct transcriptional profiles of adrenocortical tumors uncovered by DNA microarray analysis. Am. J. Pathol. 162, 521-531
27 Costa, M.H. and Lacroix, A. (2007) Cushing’s syndrome secondary to ACTH-independent macronodular adrenal hyperplasia. Arq. Bras. Endocrinol. Metabol. 51, 1226-1237
28 Lee, S. et al. (2005) Ectopic expression of vasopressin V1b and V2 receptors in the adrenal glands of familial ACTH-independent macronodular adrenal hyperplasia. Clin. Endocrinol. (Oxf.) 63, 625-630
29 Vezzosi, D. et al. (2007) Familial adrenocorticotropin-independent macronodular adrenal hyperplasia with aberrant serotonin and vasopressin adrenal receptors. Eur. J. Endocrinol. 156, 21-31
30 Libe, R. and Bertherat, J. (2005) Molecular genetics of adrenocortical tumours, from familial to sporadic diseases. Eur. J. Endocrinol. 153, 477-487
31 Wang, W. et al. (2001) Differentially expressed genes in zona reticularis cells of the human adrenal cortex. Mol. Cell. Endocrinol. 173, 127-134
Review
32 Rainey, W.E. et al. (2001) Gene profiling of human fetal and adult adrenals. J. Endocrinol. 171, 209-215
33 Velazquez-Fernandez, D. et al. (2005) Expression profiling of adrenocortical neoplasms suggests a molecular signature of malignancy. Surgery 138, 1087-1094
34 Laurell, C. et al. (2009) Transcriptional profiling enables molecular classification of adrenocortical tumours. Eur. J. Endocrinol. 161, 141-152
35 Almeida, M.Q. et al. (2008) Expression of insulin-like growth factor-II and its receptor in pediatric and adult adrenocortical tumors. J. Clin. Endocrinol. Metab. 93, 3524-3531
36 Slater, E.P. et al. (2006) Analysis by cDNA microarrays of gene expression patterns of human adrenocortical tumors. Eur. J. Endocrinol. 154, 587-598
37 Giordano, T.J. et al. (2009) Molecular classification and prognostication of adrenocortical tumors by transcriptome profiling. Clin. Cancer Res. 15, 668-676
38 Michalkiewicz, E. et al. (2004) Clinical and outcome characteristics of children with adrenocortical tumors: a report from the International Pediatric Adrenocortical Tumor Registry. J. Clin. Oncol. 22, 838-845
39 West, A.N. et al. (2007) Gene expression profiling of childhood adrenocortical tumors. Cancer Res. 67, 600-608