Value of Molecular Classification for Prognostic Assessment of Adrenocortical Carcinoma

Guillaume Assié; Anne Jouinot; Martin Fassnacht; Rossella Libé; Simon Garinet; Louis Jacob; Nadim Hamzaoui; Mario Neou; Julien Sakat; Bruno de La Villéon; Karine Perlemoine; Bruno Ragazzon; Mathilde Sibony; Frédérique Tissier; Sébastien Gaujoux; Bertrand Dousset; Silviu Sbiera; Cristina L Ronchi; Matthias Kroiss; Esther Korpershoek; Ronald De Krijger; Jens Waldmann; Marcus Quinkler; Magalie Haissaguerre; Antoine Tabarin; Olivier Chabre; Michaela Luconi; Massimo Mannelli; Lionel Groussin; Xavier Bertagna; Eric Baudin; Laurence Amar; Joel Coste; Felix Beuschlein; Jérôme Bertherat

doi:10.1001/jamaoncol.2019.1558

. 2019 Jul 11;5(10). doi: 10.1001/jamaoncol.2019.1558

Value of Molecular Classification for Prognostic Assessment of Adrenocortical Carcinoma

Guillaume Assié ^1,², Anne Jouinot ^1,^2,³, Martin Fassnacht ^4,⁵, Rossella Libé ^1,², Simon Garinet ¹, Louis Jacob ¹, Nadim Hamzaoui ⁶, Mario Neou ¹, Julien Sakat ¹, Bruno de La Villéon ¹, Karine Perlemoine ¹, Bruno Ragazzon ¹, Mathilde Sibony ^1,⁷, Frédérique Tissier ^1,⁸, Sébastien Gaujoux ⁹, Bertrand Dousset ⁹, Silviu Sbiera ⁴, Cristina L Ronchi ^4,¹⁰, Matthias Kroiss ⁵, Esther Korpershoek ¹¹, Ronald De Krijger ^11,¹², Jens Waldmann ¹³, Marcus Quinkler ¹⁴, Magalie Haissaguerre ¹⁵, Antoine Tabarin ¹⁵, Olivier Chabre ¹⁶, Michaela Luconi ¹⁷, Massimo Mannelli ¹⁷, Lionel Groussin ^1,², Xavier Bertagna ^1,², Eric Baudin ¹⁸, Laurence Amar ¹⁹, Joel Coste ²⁰, Felix Beuschlein ^21,²², Jérôme Bertherat ^1,^2,^✉

¹Institut Cochin, INSERM U1016, CNRS UMR8104, Paris Descartes University, Paris, France

²Endocrinology, Assistance Publique Hôpitaux de Paris, Hôpital Cochin, Paris, France

³Medical Oncology, Assistance Publique Hôpitaux de Paris, Hôpital Cochin, Paris, France

⁴Division of Endocrinology and Diabetes, Department of Internal Medicine I, University Hospital, University of Würzburg, Würzburg, Germany

⁵Comprehensive Cancer Center Mainfranken, University of Würzburg, Würzburg, Germany

⁶Department of Oncogenetics, Assistance Publique Hôpitaux de Paris, Hôpital Cochin, Paris, France

⁷Department of Pathology, Assistance Publique Hôpitaux de Paris, Hôpital Cochin, Paris, France

⁸Department of Pathology, Assistance Publique Hôpitaux de Paris, Hôpital Pitié Salpétrière, Paris, France

⁹Department of Digestive and Endocrine Surgery, Assistance Publique Hôpitaux de Paris, Hôpital Cochin, Paris, France

¹⁰Institute of Metabolism and System Research, University of Birmingham, Birmingham, United Kingdom

¹¹Department of Pathology, Erasmus MC University Medical Center, Rotterdam, the Netherlands

¹²Department of Pathology, University Medical Center Utrecht, Utrecht, the Netherlands

¹³Department of Surgery, University Hospital Giessen and Marburg, Campus Marburg, Marburg, Germany

¹⁴Endocrinology in Charlottenburg, Berlin, Germany

¹⁵Department of Endocrinology, Diabetes and Metabolic Diseases, University Hospital of Bordeaux, Bordeaux, France

¹⁶Department of Endocrinology, University Hospital of Grenoble, Grenoble, France

¹⁷Department of Experimental and Clinical Biomedical Sciences, University of Florence, Florence, Italy

¹⁸Department of Nuclear Medicine and Endocrine Oncology, Institut Gustave Roussy, Villejuif, France

¹⁹Hypertension Unit, Hôpital Européen Georges Pompidou, Assistance Publique-Hôpitaux de Paris, Paris, France

²⁰Biostatistics and Epidemiology Unit, Hôtel Dieu, Assistance Publique-Hôpitaux de Paris, Paris, France

²¹Medizinische Klinik und Poliklinik IV, Ludwig-Maximilians-Universität München, Munich, Germany

²²Klinik für Endokrinologie, Diabetologie und Klinische Ernährung, Universitätsspital Zürich, Zurich, Switzerland

Accepted for Publication: March 29, 2019.

^✉

Corresponding Author: Jérôme Bertherat, MD, PhD, INSERM U1016, CNRS UMR8104, Paris Descartes University, 24 Rue du Faubourg St Jacques, Paris 75014, France (jerome.bertherat@aphp.fr).

Published Online: July 11, 2019. doi:10.1001/jamaoncol.2019.1558

Author Contributions: Drs Assié and Jouinot contributed equally to this work. Drs Assié and Jouinot had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Assié, Jouinot, Fassnacht, de La Villéon, Dousset, Mannelli, Coste, Bertherat.

Acquisition, analysis, or interpretation of data: Assié, Jouinot, Fassnacht, Libé, Garinet, Jacob, Hamzaoui, Neou, Sakat, de La Villéon, Perlemoine, Ragazzon, Sibony, Tissier, Gaujoux, Sbiera, Ronchi, Kroiss, Korpershoek, De Krijger, Waldmann, Quinkler, Haissaguerre, Tabarin, Chabre, Luconi, Groussin, Bertagna, Baudin, Amar, Coste, Beuschlein, Bertherat.

Drafting of the manuscript: Assié, Jouinot, Sakat, Tissier, Korpershoek, Groussin, Bertherat.

Critical revision of the manuscript for important intellectual content: Jouinot, Fassnacht, Libé, Garinet, Jacob, Hamzaoui, Neou, de La Villéon, Perlemoine, Ragazzon, Sibony, Gaujoux, Dousset, Sbiera, Ronchi, Kroiss, De Krijger, Waldmann, Quinkler, Haissaguerre, Tabarin, Chabre, Luconi, Mannelli, Bertagna, Baudin, Amar, Coste, Beuschlein, Bertherat.

Statistical analysis: Assié, Jouinot, Jacob, Neou, Coste.

Obtained funding: Fassnacht, Sakat, Sbiera, Bertherat.

Administrative, technical, or material support: Jouinot, Fassnacht, Perlemoine, Sibony, Tissier, Gaujoux, Sbiera, Ronchi, Korpershoek, De Krijger, Haissaguerre, Tabarin, Luconi, Mannelli, Groussin, Bertagna, Amar, Beuschlein, Bertherat.

Supervision: Assié, Jouinot, Libé, Gaujoux, Dousset, Waldmann, Baudin, Bertherat.

Conflict of Interest Disclosures: Dr Assié reported receiving personal fees from Novartis Pharma and HRA Pharma outside the submitted work and having a patent pending. Dr Jouinot reported receiving grants from the French National Cancer Institute (L’Institut Thématique Multi-Organisme Cancer d'Aviesan) during the conduct of the study. Dr Fassnacht reported receiving grants from the German Research Foundation during the conduct of the study and grants from HRA Pharma, Millendo, German Cancer Aid, and the German Research Foundation outside the submitted work. Dr Sbiera reported receiving grants from Else Kröner-Fresenius Stiftung during the conduct of the study. Dr Chabre reported receiving personal fees and registration fee to congress from Novartis, HRA Pharma, and IPSEN outside the submitted work. Dr Beuschlein reported receiving grants from FP7 during the conduct of the study. Dr Bertherat reported grants from the French Ministry of Health, European Community, and French National Cancer Institute during the conduct of the study and grants, personal fees, and nonfinancial support from Novartis, personal fees from HRA Pharma and Atterocor, and nonfinancial support from Ipsen outside the submitted work. No other disclosures were reported.

Funding/Support: This work was supported by the European Network for the Study of Adrenal Tumor (ENSAT) Cancer Health F2-2010-259735 program (FP7 program) to ENSAT, the Programme Hospitalier de Recherche Clinique to the Cortico Medullo-surrénale Tumeur Endocrines (COMETE) network (Towards an Easy-to-use Adrenal Cancer/Tumor Identity Card), the Brou de Lauriere Foundation (to Dr Bertherat’s laboratory), the Promex Foundation (to Dr Bertherat’s laboratory), the Institut National de la Santé et de la Recherche Médicale (Dr Assié received a contrat d’interface), and L’Institut Thématique Multi-Organisme Cancer d'Aviesan as part of the Plan Cancer 2014-2019 (Dr Jouinot received a PhD grant). This work was also supported in part by the Deutsche Forschungsgemeinschaft within the CRC/Transregio 205/1 (The Adrenal: Central Relay in Health and Disease) (Drs Fassnacht, Kroiss, and Beuschlein).

Role of the Funder/Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: The tumor bank of Cochin Hospital (Benoit Terris), the Centre de Ressources Biologiques of Bordeaux University Hospital, the Endocrinology Departments of Ambroise Paré Hospital (Marie Laure Raffin-Sanson), Le Mans Hospital (Noha Saad), St Nazaire Hospital (Sylvie Regnier-Le Coz), Lyon University Hospital (Francoise Borson-Chazot, MD, PhD, and Jean Louis Peix, MD, PhD), Toulouse University Hospital (Delphine Vezzosi, MD), La Rochelle Hospital (Frederique Duengler, MD), and the Oncogenetic Unit of Cochin Hospital (Michel Vidaud, PharmD, PhD) helped with sample collection. The members of our laboratories and COMETE and ENSAT provided support and discussions, and the staff of the clinical and pathology departments were involved in patient care. No compensation was given for this work.

^✉

Corresponding author.

PMCID: PMC6624825 PMID: 31294750

This biomarker analyss of patients with adrenocortical carcinoma compares the molecular classification for prognostic assessment of adrenocortical carcinoma with other known prognostic factors.

Key Points

Question

Is the prognostic molecular classification of adrenocortical carcinoma (ACC) more accurate than other known prognostic factors?

Findings

In this prognostic study of 368 patients who had undergone surgical removal of ACC localized tumors, molecular classification was an independent prognostic factor in patients stage I to III disease after complete surgery, and the combination of tumor stage, tumor grade, and molecular class provided the best prognostic model of disease-free survival. The prognostic value of molecular classification was confirmed using targeted molecular measurements in a large independent cohort.

Meaning

The findings suggest that in patients with low-grade and stage I to III ACC, molecular classification can better discriminate poor outcome ACC, which may spare patients from unnecessary adjuvant treatment.

Abstract

Importance

The risk stratification of adrenocortical carcinoma (ACC) based on tumor proliferation index and stage is limited. Adjuvant therapy after surgery is recommended for most patients. Pan-genomic studies have identified distinct molecular groups closely associated with outcome.

Objective

To compare the molecular classification for prognostic assessment of ACC with other known prognostic factors.

Design, Setting, and Participants

In this retrospective biomarker analysis, ACC tumor samples from 368 patients who had undergone surgical tumor removal were collected from March 1, 2005, to September 30, 2015 (144 in the training cohort and 224 in the validation cohort) at 21 referral centers with a median follow-up of 35 months (interquartile range, 18-74 months). Data were analyzed from March 2016 to March 2018.

Exposures

Meta-analysis of pan-genomic studies (transcriptome, methylome, chromosome alteration, and mutational profiles) was performed on the training cohort. Targeted biomarker analysis, including targeted gene expression (BUB1B and PINK1), targeted methylation (PAX5, GSTP1, PYCARD, and PAX6), and targeted next-generation sequencing, was performed on the training and validation cohorts.

Main Outcomes and Measures

Disease-free survival. Cox proportional hazards regression and C indexes were used to assess the prognostic value of each model.

Results

Of the 368 patients (mean [SD] age, 49 [16] years), 144 were in the training cohort (100 [69.4%] female) and 224 were in the validation cohort (142 [63.4%] female). In the training cohort, pan-genomic measures classified ACC into 3 molecular groups (A1, A2, and A3-B), with 5-year survival of 9% for group A1, 45% for group A2, and 82% for group A3-B (log-rank P < .001). Molecular class was an independent prognostic factor of recurrence in stage I to III ACC after complete surgery (hazard ratio, 55.91; 95% CI, 8.55-365.40; P < .001). The combination of European Network for the Study of Adrenal Tumors (ENSAT) stage, tumor proliferation index, and molecular class provided the most discriminant prognostic model (C index, 0.88). In the validation cohort, the molecular classification, determined by targeted biomarker measures, was confirmed as an independent prognostic factor of recurrence (hazard ratio, 5.96 [95% CI, 1.81-19.58], P = .003 for the targeted classifier combining expression, methylation, and chromosome alterations; and 2.61 [95% CI, 1.31-5.19], P = .006 for the targeted classifier combining methylation, chromosome alterations, and mutational profile). The prognostic value of the molecular markers was limited for patients with stage IV ACC.

Conclusions and Relevance

The findings suggest that in localized ACC, targeted classifiers may be used as independent markers of recurrence. The determination of molecular class may improve individual prognostic assessment and thus may spare unnecessary adjuvant treatment.

Introduction

The prognosis of adrenocortical carcinoma (ACC) varies widely, ranging from tumors curable by complete surgery to those that are unresectable and fast growing with a high potential for metastatic spread. Current prognostic factors include mainly tumor stage, best assessed by the European Network for the Study of Adrenal Tumors (ENSAT) staging system,¹ and tumor proliferation index, best measured by mitotic count^2,3 or Ki67 immunostaining.⁴ Other prognostic factors commonly reported include cortisol secretion and age.^5,6,7 Despite such prognostic factors, it is currently not possible to predict whether a patient is cured after complete surgery. Therefore, adjuvant treatment with mitotane is commonly prescribed,^8,9 although most patients may not benefit from it.

Integrated pan-genomic studies^10,11 have been performed in ACC, showing the convergence of genomic alterations into distinct molecular subtypes. These molecular subgroups are associated with clinically relevant differences in outcome. In particular, 1 patient subgroup is associated with a poor outcome, which has been characterized by a specific transcriptomic signature^12,13,14 called C1A, a methylome signature of hypermethylation on CpG islands¹⁵ called CIMP, a high number of chromosome alterations called noisy, and an accumulation of mutations among a limited number of recurrent driver genes. Conversely, another patient subgroup has been associated with a better outcome, characterized by a C1B transcriptomic signature.

In the current study, our aim was to compare molecular classification with other known factors used for prognostic assessment of ACC. In addition, given the limited use of pan-genomic measures in clinical oncogenetics departments, we proposed targeted molecular classifiers that reflect the ACC genomic subtypes.

Methods

Patients

In this prognostic study, 2 cohorts of patients with ACC were included: a training and a validation cohort. The training cohort consisted of 144 patients (eTable 1 in the Supplement) previously included in the ENSAT or The Cancer Genome Atlas (TCGA) study.^10,11 The validation cohort consisted of 224 patients (eTable 2 in the Supplement) from 21 centers in the ENSAT network in France, Germany, the Netherlands, and Italy. The study was performed from March 1, 2005, to September 30, 2015, with a median follow-up of 35 months (interquartile range, 18-74 months). (eMethods in the Supplement). Data were analyzed from March 2016 to March 2018. Written informed consent for the molecular analysis and the collection and use of the clinical data was obtained from all patients within ENSAT. The present analysis was approved by the local institutional review board of each clinical center. All data were deidentified.

Pan-genomic Molecular Measures

A molecular classification was determined on the training cohort, previously characterized by pan-genomic approaches. Tumors that accumulated genomic alterations associated with poor outcome (C1A transcriptome, CIMP methylome, and noisy chromosome alteration profiles) were assigned to a poor outcome class, in agreement with previous reports.^10,11 Conversely, tumors with the C1B transcriptome profile were assigned to a better outcome class.^10,11 For other tumors, prognostic classes were defined as the best combinations of pan-genomic measures that fit disease-free survival (DFS) and overall survival (OS) (eMethods and eFigure 1 in the Supplement).

Targeted Molecular Profiling

Targeted markers, aiming to recapitulate pan-genomic profiles, were assessed for 72 patients with ACC from the training cohort and for the entire validation cohort.

Gene Expression

Targeted gene expression consisted of a 2-gene molecular predictor measuring the differential expression of BUB1B (OMIM 602860) and PINK1 (OMIM 608309) by quantitative reverse-transcriptase polymerase chain reaction (eMethods in the Supplement), classifying tumors into poor and better outcome subgroups, as previously published.^12,16

DNA Methylation

Targeted methylation was measured for 4 genes (PAX5 [OMIM 167414], GSTP1 [OMIM 134660], PYCARD [OMIM 606838], and PAX6 [OMIM 607108]) by methylation-specific multiplex ligation-dependent probe amplification (eMethods in the Supplement), classifying tumors into poor and better outcome subgroups, as previously described.¹⁷

Gene Mutations, Homozygous Deletions, and Amplifications

Eighteen driver genes previously reported to be recurrently altered in ACC^10,11,18 were sequenced by targeted next-generation sequencing (NGS; Life Technologies) with a dedicated panel (eMethods in the Supplement). In addition to sequence variations, homozygous deletions and gene amplifications were assessed by analyzing both single-nucleotide polymorphism array data and the DNA copy number from NGS using TARGOMICs.¹⁹ Tumors were classified into poor or better outcome subgroups, depending on the mutational status (at least 1 vs no mutation) in pathways associated with prognostic value, that is, cell-cycle (TP53 [OMIM 191170], RB1 [OMIM 614041], CDK4 [OMIM 123829], CCNE1 [OMIM 123837], and CDKN2A [OMIM 615914]) and Wnt/β-catenin genes (CTNNB1 [OMIM 116806] and ZNRF3 [OMIM 612062]) (eTable 3 in the Supplement).

Chromosome Alterations

Chromosome alteration profiles were determined by single-nucleotide polymorphism array, based on the targeted status of 9 chromosome arms (eMethods in the Supplement). Tumors were classified into poor or better outcome subgroups, depending on the chromosome profile (noisy vs chromosomal or quiet).

Targeted Molecular Classifiers

Targeted molecular measures were combined into 2 distinct molecular classifiers that recapitulate the pan-genomic classification: a 3-dimensional (3-D) targeted classifier or a DNA-based targeted classifier. The 3-D targeted classifier used tumor RNA and DNA and combined targeted gene expression, targeted methylation, and targeted measures of chromosome alterations (eMethods, eFigure 1, and eFigure 2A in the Supplement). The DNA-based targeted classifier used tumor DNA only and combined targeted methylation, targeted chromosome alteration profile, and mutational status (eMethods, eFigure 1, and eFigure 2B in the Supplement). For both the 3-D and DNA-based targeted classifiers, poor, intermediate, and better outcome prognostic classes were defined by the combination of targeted statuses that showed the best concordance with the pan-genomic classification using the κ coefficient.

Statistical Analysis

Calculations were performed using R statistical software (stats, irr, pheatmap, survival, and survcomp packages; R Foundation) (source codes are given in the eMethods in the Supplement). Comparisons between groups were assessed using the 2-tailed, unpaired t test or parametric analysis of variance for quantitative variables and the Fisher or χ² test for qualitative variables. Correlations were assessed using the Spearman correlation coefficient. The Cohen κ concordance coefficient was used to determine the best combinations of targeted molecular measurements as surrogates for pan-genomic classification.

Survival curves were obtained with Kaplan-Meier estimates and compared using the log-rank test. Cox proportional hazards regression was used for the DFS and OS analyses. Overall survival was defined as the time from diagnosis until death or last follow-up. Disease-free survival was analyzed only in patients with stage I to III disease after complete surgery and was defined as the time from primary tumor resection until recurrence or last follow-up. Univariate analyses tested the association of demographic and known prognostic factors with DFS and OS, along with molecular classification. Significant variables were combined into multivariable models. Performance of the prognostic models was further quantified by the Harrell C index. The best prognostic model was then tested in the validation cohort. The proportional hazards assumption was confirmed for each model constructed. All P values were 2-sided, and the level of significance was set at P < .05.

Results

Patient Characteristics

Of the 368 patients (mean [SD] age, 49 [16] years), 144 were in the training cohort (100 [69.4%] female) and 224 were in the validation cohort (142 [63.4%] female). The training cohort patients were from the ENSAT and TCGA genomic studies,^10,11 and the validation cohort patients were from ENSAT only. The baseline characteristics of the patients are presented in eTable 4 in the Supplement. The median follow-up was 39 months for the training cohort and 35 months for the validation cohort, with 16 patients (11.1%) in the training cohort and 23 patients (10.3%) in the validation cohort unavailable for follow-up. Patients in the validation cohort were older (mean age, 51 vs 47 years), had a higher tumor proliferation index (77 [43%] of 179 vs 30 [26%] of 116 high-proliferation tumors), and received less adjuvant mitotane (73 [57%] of 129 vs 61 [80%] of 76).

Pan-genomic Classification of ACC (Training Cohort)

Meta-analysis of ENSAT and TCGA Pan-genomic Studies

The ENSAT and TCGA consortia had reported concordant molecular classes of ACC. These 2 cohorts were thus merged into 1 because they did not include any patients in common. Each tumor was classified according to its transcriptomic (C1A or C1B), chromosome alteration (noisy or chromosomal/quiet), and methylome profile (CIMP or non-CIMP). Combining these 3 omics-based profiles resulted in discrete molecular classes (A1, A2, and A3-B) (eFigure 1 in the Supplement) with major differences in outcomes (Figure, A and B): 5-year OS was 9% for the A1 group, 45% for the A2 group, and 82% for the A3 to B group (log-rank P < .001). This 3-D pan-genomic classification correlated with cortisol secretion (r = 0.47), ENSAT stage (r = 0.40), and tumor proliferation (r = 0.40) (eTable 5 and eTable 6 in the Supplement).

Figure. — A, Disease-free survival (DFS) among patients with R0 stage I to III disease, according to molecular group, using pan-genomic classifier (training cohort). B, Overall survival (OS) among patients with stage I to IV disease, according to molecular group, using pan-genomic classifier (training cohort). C, DFS among patients with R0 stage I to III disease, according to molecular group, using the 3-dimensional (3-D) targeted classifier (validation cohort). D, DFS among patients with R0 stage I to III disease, according to molecular group, using the DNA-based targeted classifier (validation cohort). Survival curves were obtained with Kaplan-Meier estimates.

Prognostic Value of the Pan-genomic Classification

For patients with localized ACC (R0 and stage I-III), ENSAT stage III, a high tumor proliferation index, cortisol secretion, and pan-genomic groups A1 and A2 were identified as adverse prognostic factors of DFS by univariate analysis (eTable 7 in the Supplement). In the multivariable model (eFigure 3 in the Supplement), the 3-D pan-genomic classification was the only independent prognostic factor of DFS (group A1 vs A3-B), with a hazard ratio (HR) of 55.91 (95% CI, 8.55-365.40; P < .001) (eTable 8 in the Supplement). The 3-D pan-genomic classification remained associated with DFS after adjustment of adjuvant mitotane treatment (group A1 vs A3-B: HR, 55.06; 95% CI, 5.80-467.10; P < .001). The combination of ENSAT stage, tumor proliferation index, and 3-D pan-genomic classification provided the most discriminant prognostic model (C index, 0.88) (eFigure 4 in the Supplement).

In contrast to localized ACC, the prognostic value of the 3-D pan-genomic classification was weaker for patients with stage IV disease (univariate analysis of OS for group A1 vs A3-B: HR, 5.40; 95% CI, 1.16-25.11; P = .03) (eTable 8 and eFigure 7 in the Supplement).

Targeted Molecular Classification Using Tumor DNA and RNA

Targeted measures were adopted, reflecting each dimension of the 3-D pan-genomic classifier. These measures included the expression of 2 genes, BUB1B and PINK1, quantified by quantitative reverse-transcriptase polymerase chain reaction, and DNA methylation in CpG islands of 4 genes, GSTP1, PYCARD, PAX6, and PAX5, assayed by methylation-specific multiplex ligation-dependent probe amplification, along with the chromosome alteration profile.

From 3-D Pan-genomic to 3-D Targeted Classification (Training Cohort)

In the training cohort, targeted molecular measures of gene expression and DNA methylation were associated with the pan-genomic classes (eFigure 1 in the Supplement). The 3-D targeted classifier, which combined targeted gene expression, targeted DNA methylation, and the chromosome alteration profile, was concordant with the pan-genomic classes (κ = 0.76, P < .001) (eFigure 1 in the Supplement).

Prognostic Value of the 3-D Targeted Classifier (Validation Cohort)

For patients with localized ACC (R0 and stage I-III), the molecular class determined by the 3-D targeted classifier (eFigure 2A in the Supplement) was confirmed to be a prognostic factor in univariate analysis (Figure, C, and eTable 9 and eFigure 6A in the Supplement). The median DFS was 14 months for the poor prognostic group, 29 months for the intermediate prognostic group, and not reached (exceeding the follow-up period) for the better prognostic group (log-rank P < .001) (Figure, C). After multivariable modeling, including tumor stage and tumor proliferation index (eFigure 3 in the Supplement), the 3-D targeted classifier was identified as an independent prognostic factor (Table). The HRs for recurrence were 3.43 (95% CI, 1.22-9.67; P = .02) for the intermediate and 5.96 (95% CI, 1.81-19.58; P = .003) for the poor prognostic groups. Combination of the ENSAT stage, tumor proliferation index, and molecular class, determined by the 3-D targeted classifier, was confirmed to be the best discriminant model (C index, 0.77). In contrast to patients with localized ACC, the prognostic value of the 3-D targeted classifier was not significant for patients with stage IV ACC (eTable 10 and eFigure 6B in the Supplement).

Table. Multivariable Prognostic Models of Disease-Free Survival in the Validation Cohort^a.

Variable	3-D Targeted Classifier		DNA-Based Targeted Classifier		Expression Marker		Methylation Marker
Variable	HR (95% CI)	P Value	HR (95% CI)	P Value	HR (95% CI)	P Value	HR (95% CI)	P Value
ENSAT stage III vs I-II	1.48 (0.97-2.25)	.07	1.50 (1.13-1.99)	.005	1.56 (1.03-2.36)	.03	1.52 (1.14-2.02)	.004
Tumor proliferation index high vs low	3.30 (1.42-7.66)	.005	2.04 (1.13-3.69)	.02	2.88 (1.28-6.48)	.01	2.13 (1.18-3.83)	.01
Molecular class
Intermediate vs better	3.43 (1.22-9.67)	.02	1.85 (0.94-3.64)	.08	NA	NA	NA	NA
Poor vs better	5.96 (1.81-19.58)	.003	2.61 (1.31-5.19)	.006	3.71 (1.37-10.00)	.01	2.41 (1.35-4.32)	.003

Open in a new tab

Abbreviations: 3-D, 3-dimensional; ENSAT, European Network for the Study of Adrenal Tumors; HR, hazard ratio; NA, not applicable.

^{^a}

Multivariable models were generated for patients with R0 stage I to III disease. The HR for recurrence and the 95% CI are based on a Cox proportional hazards regression model with ENSAT stage, tumor proliferation index, and molecular classifiers as explanatory variables. High tumor proliferation index was defined by mitotic count of 20 or more per 50 high-power fields and/or Ki67 staining in 20% of cells or more. Molecular class was determined by targeted classifiers using tumor RNA and DNA (3-D targeted classifier) or tumor DNA only (DNA-based targeted classifier) (details in eMethods in the Supplement).

Targeted Molecular Classification Using Tumor DNA Only

Compared with the 3-D targeted classifier, the DNA-based targeted classifier did not consider gene expression but included instead the mutational status of the selected genes.

From 3-D Pan-genomic to DNA-Based Targeted Classification (Training Cohort)

Somatic mutations within pathways associated with the prognostic value (eTable 3 in the Supplement), that is, cell cycle and Wnt/β-catenin, were associated with the pan-genomic classes (eFigure 1 in the Supplement). The DNA-based targeted classifier, which combined targeted measures of methylation, chromosome alteration profiles, and mutations, was concordant with the pan-genomic classes (κ = 0.73, P < .001) (eFigure 1 in the Supplement).

Prognostic Value of the DNA-Based Targeted Classifier (Validation Cohort)

In patients with localized ACC (R0 and stage I-III), the molecular class determined by the DNA-based targeted classifier (eFigure 2B in the Supplement) was confirmed to be a prognostic factor in univariate (Figure, D, and eTable 7 and eFigure 6C in the Supplement) and multivariable models (Table). In contrast to patients with localized ACC, the prognostic value of the DNA-based targeted classifier was not as strong for patients with stage IV ACC (eTable 8 and eFigure 6D in the Supplement).

Discussion

The study findings support the prognostic value of targeted molecular classifiers, reflecting the pan-genomic classification of ACC. To date, clinical factors, including tumor stage and tumor proliferation index, are considered to be the strongest and most consensual prognostic factors for patients with ACC.²⁰ However, risk stratification within subgroups is still variable.^21,22,23 In particular, the individual risk of recurrence for patients with stage I to III disease after complete surgery cannot be properly estimated. In addition, these prognostic factors are not specific to ACC. Tumor stage is a globally obvious prognostic factor, especially the presence of metastases, and the tumor proliferation index is associated with survival for most cancers.²⁴ We proposed complementing these standard prognostic factors with an approach based primarily on the molecular biology of ACC. Molecular subgroups are supported by distinct pathophysiologic mechanisms and translate into clinically relevant differences in outcome. The results of 2 independent integrated genomic studies,^10,11 one from ENSAT and the other from the TCGA consortium, converged on a molecular classification of ACC mainly based on transcriptomic profiles. Concordance between the 2 studies^10,11 made it possible to merge the genomic data into a single meta-analysis to define a unique molecular classification for these 2 cohorts.

This study supports the prognostic value of molecular classification in localized ACC. Conversely, the prognostic value of molecular classification was weaker in patients with stage IV disease. The presence of metastases appeared to be the strongest pejorative prognostic factor in the cohort, in line with previous clinical studies.^1,7 Of note, almost all metastatic tumors were classified into molecular classes of poor and intermediate outcome. Thus, the association between molecular class and survival was hard to assess, especially considering the limited size of the cohort with stage IV ACC .

For the transfer to clinical routine, we developed 2 combinations of targeted measures, the 3-D targeted classifier and the DNA-based targeted classifier. The former provided information closest to the pan-genomic classification but could be evaluated for only half of the patients because of limited access to tumor RNA in the retrospective cohort. This is a potentially important limitation for routine transfer. However, many centers propose systematic freezing of tumor samples. In addition, techniques measuring RNA levels on formalin-fixed paraffin-embedded samples have been developed, adapted to the degraded nature of RNA in these samples,²⁵ with promising results in cancer oncogenetics.²⁶ In ACC, adaption of the targeted RNA signature to formalin-fixed paraffin-embedded samples remains to be done and tested. Considering this potential limitation, the DNA-based classifier was developed. This classifier can be assessed using frozen or formalin-fixed paraffin-embedded samples²⁷ and measured by targeted NGS.¹⁹ Finally, beyond the intrinsic performance of any molecular classifier, the final choice of a technique was also determined by its local availability and feasibility, with major differences among centers.

Combining molecular classification with standard prognostic factors, such as tumor stage and proliferation index, provided the best prognostic discrimination in patients with localized ACC in the training (C index, 0.88) and validation (C index, 0.77) cohorts. Therefore, the findings suggest that molecular classification should be integrated into patient care. This method raises several questions concerning the consequences for patients. For example, should patients with a better outcome receive a simplified follow-up after surgery for R0 disease and receive locoregional treatments in case of recurrence? In more common cancers, such as breast carcinoma, molecular classification has radically changed practice, resulting in less use of aggressive and costly treatments for less aggressive tumors.^28,29 Thus, patients with a better outcome would probably benefit from a less aggressive management after complete tumor removal and avoid toxic adjuvant treatment, such as mitotane and/or cytotoxic chemotherapy. Conversely, perhaps it would be possible for patients with an intermediate or poor outcome to be followed up more closely and receive more aggressive systemic adjuvant therapies. In addition to mitotane, adjuvant cytotoxic chemotherapy in patients with poor outcomes could be considered. Clinical trials are needed to address these questions, and future randomized trials in ACC, especially in the adjuvant setting, should add molecular classification to the set of standard prognostic factors.

Limitations

A limitation of this study includes the number of patients unavailable for follow-up, approximately 10% for both the training and validation cohorts, followed up for less than 2 years. However, unavailability of patient follow-up does not affect the interaction of the molecular markers with survival and other prognostic factors. The differences in baseline characteristics between the training and validation cohorts may have also affected the results of this study. The tumor proliferation index was significantly higher in the validation cohort. However, the overall prognosis and prognostic value of tumor proliferation were similar in the training and validation cohorts. Thus, the differences between the 2 cohorts may reflect the evolution of the techniques for determining tumor proliferation. First, mitotic count was used for the oldest tumors (training cohort) and Ki67 for the most recent ones (validation cohort). Second, Ki67 is globally increasing over time (Pearson r = 0.37, P < .001). An additional limitation is related to the Ki67 cutoff chosen to discriminate high- and low-proliferation indexes. Both Ki67 cutoffs at 10% and 20% have been investigated previously.^4,7,9 In our study, the Ki67 cutoff at 20% was more concordant with a cutoff of more than 20 mitoses per 50 high-power fields to define high proliferation index tumors and was retained for this reason. Another limitation is the lack of standardization of adjuvant and palliative treatments. However, therapeutic decisions were not based on molecular markers. In addition, the limited efficacy of current treatments suggests a limited influence of this potential bias on this study. Third, these limitations are intrinsic to a retrospective and multicentric design, which is shared by all large prognostic studies on ACC.^5,7,30 Given the rarity of the disease, these limitations are challenging to overcome. Beyond these limitations, to our knowledge, this cohort was the largest collection of adrenal cancers investigated at the molecular level.

Conclusions

The findings suggest that molecular classifiers can be considered as independent prognostic markers in localized ACC. Combining these markers with tumor stage and proliferation index may provide the most accurate individual prognostic determination and may thus orient adjuvant therapy decisions after complete surgery. Molecular markers appear to be suitable for clinical routine use and may be proposed in standard care for patients with ACC.

Supplement.

eMethods. Patients, Pan-genomic Molecular Measures, Targeted Molecular Profiling, and Source Codes

eTable 1. Clinical and molecular data of the training cohort

eTable 2. Clinical and molecular data of the validation cohort

eTable 3. Prognostic value of recurrently altered genes in ACC (training cohort)

eTable 4. Patient characteristics

eTable 5. Correlation between pan-genomic classification and standard prognostic variables

eTable 6. Association between pan-genomic classes and standard prognostic variables

eTable 7. Uni- and multivariable analyses of disease-free survival in the training cohort

eTable 8. Univariate analysis of overall survival in stage IV patients of the training cohort

eTable 9. Univariate analysis of disease-free survival in the validation cohort

eTable 10. Univariate analysis of overall survival in stage IV patients of the validation cohort

eFigure 1. Pan-genomic classification of ACC (training cohort)

eFigure 2. Targeted molecular classification of ACC (validation cohort)

eFigure 3. Patients’ inclusion for survival models in training (A) and validation (B) cohorts

eFigure 4. Harrell’s C-Indexes for disease-free survival in R0 stage I-III patients of the training cohort

eFigure 5. Pan-genomic classification and overall survival in stage IV patients of the training cohort

eFigure 6. Molecular classifiers and overall survival in the validation cohort

Click here for additional data file.^{(2.1MB, pdf)}

References

1.Fassnacht M, Johanssen S, Quinkler M, et al. ; German Adrenocortical Carcinoma Registry Group; European Network for the Study of Adrenal Tumors . Limited prognostic value of the 2004 International Union Against Cancer staging classification for adrenocortical carcinoma: proposal for a revised TNM classification. Cancer. 2009;115(2):1440-1447. doi: 10.1002/cncr.24030 [DOI] [PubMed] [Google Scholar]
2.Weiss LM, Medeiros LJ, Vickery AL Jr. Pathologic features of prognostic significance in adrenocortical carcinoma. Am J Surg Pathol. 1989;13(3):202-206. doi: 10.1097/00000478-198903000-00004 [DOI] [PubMed] [Google Scholar]
3.Miller BS, Gauger PG, Hammer GD, Giordano TJ, Doherty GM. Proposal for modification of the ENSAT staging system for adrenocortical carcinoma using tumor grade. Langenbecks Arch Surg. 2010;395(7):955-961. doi: 10.1007/s00423-010-0698-y [DOI] [PubMed] [Google Scholar]
4.Beuschlein F, Weigel J, Saeger W, et al. . Major prognostic role of Ki67 in localized adrenocortical carcinoma after complete resection. J Clin Endocrinol Metab. 2015;100(3):841-849. doi: 10.1210/jc.2014-3182 [DOI] [PubMed] [Google Scholar]
5.Abiven G, Coste J, Groussin L, et al. . Clinical and biological features in the prognosis of adrenocortical cancer: poor outcome of cortisol-secreting tumors in a series of 202 consecutive patients. J Clin Endocrinol Metab. 2006;91(7):2650-2655. doi: 10.1210/jc.2005-2730 [DOI] [PubMed] [Google Scholar]
6.Berruti A, Fassnacht M, Haak H, et al. . Prognostic role of overt hypercortisolism in completely operated patients with adrenocortical cancer. Eur Urol. 2014;65(4):832-838. doi: 10.1016/j.eururo.2013.11.006 [DOI] [PubMed] [Google Scholar]
7.Libé R, Borget I, Ronchi CL, et al. ; ENSAT network . Prognostic factors in stage III-IV adrenocortical carcinomas (ACC): a European Network for the Study of Adrenal Tumor (ENSAT) study. Ann Oncol. 2015;26(10):2119-2125. doi: 10.1093/annonc/mdv329 [DOI] [PubMed] [Google Scholar]
8.Terzolo M, Angeli A, Fassnacht M, et al. . Adjuvant mitotane treatment for adrenocortical carcinoma. N Engl J Med. 2007;356(23):2372-2380. doi: 10.1056/NEJMoa063360 [DOI] [PubMed] [Google Scholar]
9.Fassnacht M, Dekkers O, Else T, et al. . European Society of Endocrinology clinical practice guidelines on the management of adrenocortical carcinoma in adults, in collaboration with the European Network for the Study of Adrenal Tumors. Eur J Endocrinol. 2018;(July). doi: 10.1530/EJE-18-0608 [DOI] [PubMed] [Google Scholar]
10.Assié G, Letouzé E, Fassnacht M, et al. . Integrated genomic characterization of adrenocortical carcinoma. Nat Genet. 2014;46(6):607-612. doi: 10.1038/ng.2953 [DOI] [PubMed] [Google Scholar]
11.Zheng S, Cherniack AD, Dewal N, et al. ; Cancer Genome Atlas Research Network . Comprehensive pan-genomic characterization of adrenocortical carcinoma. Cancer Cell. 2016;29(5):723-736. doi: 10.1016/j.ccell.2016.04.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
12.de Reyniès A, Assié G, Rickman DS, et al. . Gene expression profiling reveals a new classification of adrenocortical tumors and identifies molecular predictors of malignancy and survival. J Clin Oncol. 2009;27(7):1108-1115. doi: 10.1200/JCO.2008.18.5678 [DOI] [PubMed] [Google Scholar]
13.Giordano TJ, Kuick R, Else T, et al. . Molecular classification and prognostication of adrenocortical tumors by transcriptome profiling. Clin Cancer Res. 2009;15(2):668-676. doi: 10.1158/1078-0432.CCR-08-1067 [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Tömböl Z, Szabó PM, Molnár V, et al. . Integrative molecular bioinformatics study of human adrenocortical tumors: microRNA, tissue-specific target prediction, and pathway analysis. Endocr Relat Cancer. 2009;16(3):895-906. doi: 10.1677/ERC-09-0096 [DOI] [PubMed] [Google Scholar]
15.Barreau O, Assié G, Wilmot-Roussel H, et al. . Identification of a CpG island methylator phenotype in adrenocortical carcinomas. J Clin Endocrinol Metab. 2013;98(1):E174-E184. doi: 10.1210/jc.2012-2993 [DOI] [PubMed] [Google Scholar]
16.Fragoso MCBV, Almeida MQ, Mazzuco TL, et al. . Combined expression of BUB1B, DLGAP5, and PINK1 as predictors of poor outcome in adrenocortical tumors: validation in a Brazilian cohort of adult and pediatric patients. Eur J Endocrinol. 2012;166(1):61-67. doi: 10.1530/EJE-11-0806 [DOI] [PubMed] [Google Scholar]
17.Jouinot A, Assie G, Libe R, et al. . DNA methylation is an independent prognostic marker of survival in adrenocortical cancer. J Clin Endocrinol Metab. 2017;102(3):923-932. doi: 10.1210/jc.2016-3205 [DOI] [PubMed] [Google Scholar]
18.Juhlin CC, Goh G, Healy JM, et al. . Whole-exome sequencing characterizes the landscape of somatic mutations and copy number alterations in adrenocortical carcinoma. J Clin Endocrinol Metab. 2015;100(3):E493-E502. doi: 10.1210/jc.2014-3282 [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Garinet S, Néou M, de La Villéon B, et al. . Calling chromosome alterations, DNA methylation statuses, and mutations in tumors by simple targeted next-generation sequencing: a solution for transferring integrated pangenomic studies into routine practice? J Mol Diagn. 2017;19(5):776-787. doi: 10.1016/j.jmoldx.2017.06.005 [DOI] [PubMed] [Google Scholar]
20.Jouinot A, Bertherat J. Management of endocrine disease: adrenocortical carcinoma: differentiating the good from the poor prognosis tumors. Eur J Endocrinol. 2018;178(5):R215-R230. doi: 10.1530/EJE-18-0027 [DOI] [PubMed] [Google Scholar]
21.Ayala-Ramirez M, Jasim S, Feng L, et al. . Adrenocortical carcinoma: clinical outcomes and prognosis of 330 patients at a tertiary care center. Eur J Endocrinol. 2013;169(6):891-899. doi: 10.1530/EJE-13-0519 [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Else T, Williams AR, Sabolch A, Jolly S, Miller BS, Hammer GD. Adjuvant therapies and patient and tumor characteristics associated with survival of adult patients with adrenocortical carcinoma. J Clin Endocrinol Metab. 2014;99(2):455-461. doi: 10.1210/jc.2013-2856 [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Kerkhofs TMA, Verhoeven RHA, Van der Zwan JM, et al. Adrenocortical carcinoma: a population-based study on incidence and survival in the Netherlands since 1993. Eur J Cancer. 19902013;49(11):2579-2586. doi: 10.1016/j.ejca.2013.02.034 [DOI] [PubMed]
24.Yerushalmi R, Woods R, Ravdin PM, Hayes MM, Gelmon KA. Ki67 in breast cancer: prognostic and predictive potential. Lancet Oncol. 2010;11(2):174-183. doi: 10.1016/S1470-2045(09)70262-1 [DOI] [PubMed] [Google Scholar]
25.Narrandes S, Xu W. Gene expression detection assay for cancer clinical use. J Cancer. 2018;9(13):2249-2265. doi: 10.7150/jca.24744 [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Kwa M, Makris A, Esteva FJ. Clinical utility of gene-expression signatures in early stage breast cancer. Nat Rev Clin Oncol. 2017;14(10):595-610. doi: 10.1038/nrclinonc.2017.74 [DOI] [PubMed] [Google Scholar]
27.de Leng WWJ, Gadellaa-van Hooijdonk CG, Barendregt-Smouter FAS, et al. . Targeted next generation sequencing as a reliable diagnostic assay for the detection of somatic mutations in tumours using minimal DNA amounts from formalin fixed paraffin embedded material. PLoS One. 2016;11(2):e0149405. doi: 10.1371/journal.pone.0149405 [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Paik S, Shak S, Tang G, et al. . A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. N Engl J Med. 2004;351(27):2817-2826. doi: 10.1056/NEJMoa041588 [DOI] [PubMed] [Google Scholar]
29.Sparano JA, Gray RJ, Makower DF, et al. . Adjuvant chemotherapy guided by a 21-gene expression assay in breast cancer. N Engl J Med. 2018;379(2):111-121. doi: 10.1056/NEJMoa1804710 [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Assié G, Antoni G, Tissier F, et al. . Prognostic parameters of metastatic adrenocortical carcinoma. J Clin Endocrinol Metab. 2007;92(1):148-154. doi: 10.1210/jc.2006-0706 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials