Adrenocortical carcinoma survival rates correlated to genomic copy number variants

Adrenocortical carcinoma (ACC) is one of the least commonly occurring but deadliest cancers. With an incidence of less than 0.77 in 1 million people, fewer than 250 new cases of ACC are seen within the United States each year (1). Carcinoma of the adrenal cortex accounts for between 0.02% and 0.2% of all cancer deaths (2, 3). The prognosis is poor with a 5-year survival rate of 20% to 45% due to typically late presentation and the limited effectiveness of broad-spectrum chemotherapy (4, 5). The standard chemotherapeutic treatment for this disease, since 1960, remains mitotane (Bristol-Myers Squibb), a derivative of the pesticide DDT (6). Response rates are poor (∼22%), although survival for those who do respond is improved from 14 to 50 months (7). Due to the lack of effective treatments for ACC, increased knowledge of the cellular processes involved in oncogenesis is necessary for the development and implementation of targeted therapies.

Some have argued that the development and growth of ACC tumors is a multistep progression with accumulation of genetic alterations (8). Although ACC does occur in the context of inherited syndromes, such as the Li-Fraumeni and Beckwith-Wiedemann syndromes, most cases are of a sporadic nature. The Li-Fraumeni syndrome is traced to a germ-line mutation in the p53 tumor suppressor gene on chromosome 17p13 (9, 10). The genetic basis for the Beckwith-Wiedemann syndrome is an allelic loss at chromosome 11p15 (11). Characteristic mutations of both syndromes have been observed in some sporadic adrenocortical cancers (12, 13).

Knowledge of genomic gains, losses, and translocations associated with sporadic adrenocortical cancer has not yet revealed an ACC oncogenic pathway. Studies of the chromosomal and molecular abnormalities in ACC have relied on technologies, such as conventional comparative genomic hybridization (CGH) or fluorescence in situ hybridization, to identify genomic gains and losses.

Recent advances in modern array-based molecular genetic technologies offer the opportunity to study tumors with greater resolution and in a high-throughput fashion with the end goals of understanding oncogenesis, enabling the development of diagnostics and targeted therapies. Zhao et al. suggested that gains in DNA copy number in chromosome 17 or 17q represent the earliest genomic changes and are seen in benign adrenocortical adenomas only (14). Additional studies have identified loss of heterozygosity at 17p13 and 11p15 along with increased expression of insulin-like growth factor 2 as molecular events associated with progression toward a malignant phenotype (15). As reported by Dohna et al., gains and high-level amplifications in chromosomes 7, 14, and 19 were seen only in ACC and not in benign adrenocortical neoplasms, suggesting that these events may be late genetic perturbations in tumorigenesis (16). Deletions have also been found within chromosomes 1p and 17p implicating potential losses of tumor suppressor genes in these regions, marking the progression of benign adrenal adenoma to ACC (17). Similar to other cancers, there is evidence that the number of genetic aberrations increases as a factor of tumor growth and this size progression has been correlated with malignancy (14, 18, 19).

Herein, we use high-density CGH array analysis to identify genomic aberrations associated with poor survival as potential prognostic markers and therapeutic targets in aggressive ACC.

The total set of 25 ACC tumors includes 21 flash-frozen tumor samples collected between 1987 and 2003 at the Mayo Clinic. The subsequent 4 ACC tumors were acquired through collaborations with the University Hospital Essen and the University of Calgary. Research consent was obtained at the respective institutes through accepted institutional review board protocols and accepted into our research program and tumor bank under the Western Institutional Review Board–approved protocol no. 20051769. Extensive clinical and demographic information was collected at time of diagnosis and surgery (Supplementary Data 1).⁷

DNA extraction from flash-frozen ACC tumor samples was done following the standard protocol for the Qiagen DNeasy Blood and Tissue Kit (Qiagen). Oligo-based CGH was done according to manufacturer’s instructions for the Agilent Human Genome Microarray Kit 44B (Agilent Technologies) starting with 800 ng DNA from clinical samples and a separate reference sample of 800 ng control male genomic DNA (Promega). Experimental samples labeled with Cy5 and reference samples labeled with Cy3 (GE Healthcare) were combined and hybridized to a single CGH array. Deviating from the Agilent Oligo aCGH Hybridization Kit protocol, samples were incubated with Cot1 (1 mg/mL), 10× blocking solution, and 2× hybridization buffer at 99°C for 3 min, 37°C for 30 min, and centrifuged at high speed for 1 min. The samples were hybridized according to Agilent protocol. Arrays were washed, scanned, and data extracted using Agilent Feature Extraction 8.1. Data for arrays fulfilling standard Agilent quality requirements were included in further statistical analyses.

The log₂ of Cy5 and Cy3 ratio (log-ratio) was used as the measure of copy number ratio between tumor and normal samples throughout the analysis. To detect the regions of copy number aberration, log-ratio values of a sliding window of 25 probes in length were compared using the Wilcoxon's rank-sum test with the log-ratio values of an artificial reference chromosome constructed for each tumor (20). The reference chromosome was created by recursively accepting 25 randomly sampled log-ratio values from the autosomal regions with P value of Wilcoxon's signed rank test (against 0) > 0.05 until the size reached 1,000. The P values and corresponding false discovery rates (or q values) for the copy number gain or loss (indicated as P_gain and Q_gain or P_loss and Q_loss) were calculated independently for each sliding window and assigned to the center position of the window.

Commonly shared genomic aberrations in ACC tumors were identified by gathering overlapping sliding windows that were gained or lost by ≥50% of all ACC patients. A sliding window was called gained if P_gain < 10⁻⁵ and Q_gain < 10⁻⁴ and lost if P_loss < 10⁻⁵ and Q_loss < 10⁻⁴. The statistical significance of an identified region was evaluated by testing the distribution of copy number changed tumor fractions of sliding windows contained in the region against the ratio cutoff 0.5 using Wilcoxon's signed rank test.

The clinical implication of the observed genomic aberrations was examined in detail by identifying chromosomal regions whose gains or losses were associated with the overall survival difference using the log-rank test for all sliding windows across the genome. Sliding windows whose gains or losses were linked to poor prognosis (P ≤ 0.05) and occurred in ≥20% of all patients were called survival relevant and retained for further analysis.

Unsupervised clustering of a summary matrix was done to detect the coappearance of survival relevant genomic aberrations from different chromosomal regions. For each chromosome that contained survival relevant sliding windows, the survival relevant region was identified first by collecting all survival relevant sliding windows across all patients within it. Then a summary indicator was scored for the chromosome of a patient as +1 (gain), 0 (no change), or −1 (loss) if >50% of sliding windows that fell within the survival relevant region of the patient harbored corresponding genomic aberrations. In the clustering, the distance between two vectors was measured by counting the number of elements that were not the same and the complete linkage was used to determine the distance between two intermediate clusters.

Genomic aberrations are visualized as the histogram of copy number change frequencies over all ACC tumors (Fig. 1A) and the heat map of chromosomal gain and loss for individual patients (Fig. 1B). The copy number aberration frequency plot (Fig. 1A) represents the fraction of patients with P ≤ 10⁻⁵ and q ≤ 10⁻⁴ for each sliding window. The results clearly indicate that ACC tumors have an extensive level of genomic aberrations as has been reported previously (Table 1). In most of the chromosomes, common aberrations (gains or losses) encompassing large fractions of the samples are observed, indicating high occurrence of genomic change. The most distinctive aberrant regions include gains of chromosomes 5, 6q, 7, 8q, 12, 16q, and 20 and losses of chromosomes 1, 2q, 3, 6p, 7p, 8p, 9, 10, 11, 13q, 14q, 15q, 16, 17, 19q, and 22q (visualized in Fig. 1 and detailed in Supplementary Data 2).

Chromosomal regions were further analyzed to determine if the occurrence of genomic aberrations was associated with survival difference. Figure 2 shows the distribution of log-rank test P values across the genome. The P values above and below the X axis represent the association between gains and losses (respectively) and survival difference. Specific genomic aberrations associated with worse survival rates are detailed in Table 2 with gains in 6q, 7q, 12q, and 19p and losses in 3, 8, 10p, 16q, 17q, and 19q. Survival curves created for each aberration linked to poor survival can be seen in Supplementary Data 3.

To clarify possible coappearances of survival relevant genomic aberrations, unsupervised hierarchical clustering was done with sliding windows in regions relevant to survival difference (see Materials and Methods for details). The results shown in Fig. 3A indicate that genomic aberrations relevant to survival difference in unique chromosomes do not cosegregate. Furthermore, the clustering of samples results in three sample groups, each of which is associated with varying presence of selected aberrations. The groups are differentially associated with survival trends given by a Kaplan-Meier analysis P value of 7.4 × 10⁻⁶ (Fig. 3B). Additional tumor characteristics (Fig. 3A) are not associated with the described survival difference, indicating that the survival clusters are dependent on accumulation of genomic aberrations alone.

With few therapeutic options and a poor understanding of ACC oncogenesis, emerging high-density genomic scanning technologies enable identification of genomic events linked to disease. Table 1 summarizes previous genomic analyses of ACC tumors presented in five publications. A new, more precise high-resolution technology is used in the present study to identify genomic aberrations in ACC with >44,000 60-mer oligonucleotide probes spaced across the entire genome. Compared to the conventional CGH or interphase fluorescence in situ hybridization, which offer a resolution of ≤10 Mb, new Agilent chip-based array CGH offers a much improved resolution with probes on average every 35 kb (21). The previous studies of chromosomal amplifications and deletions used sample sets between 8 and 14 adrenocortical tumors. In the present study, we used high-resolution CGH to analyze 25 ACC samples for genetic gains and losses within the tumor population.

Array-based CGH analysis identifying chromosomal aberrations shared among a high percentage of ACC tumors gives a superior evaluation of areas characteristically linked to ACC oncogenesis. Genomic aberrations within the ACC genome (Fig. 1A) illustrate the overall gains within chromosomes 5, 6q, 7, 8q, 12, 16q, and 20 and losses within chromosome 1, 2q, 3, 6p, 7p, 8p, 9, 10, 11, 13q, 14q, 15q, 16, 17, 19q, and 22q (Supplementary Data 2). These high-resolution results recapitulate genomic abnormalities presented in past research (see Table 1). Of note, the analysis identified unique losses in 10q, 14q, and 15q, giving a more comprehensive characterization of the ACC genomic landscape. Genes within these regions can give insight into the pathways of oncogenesis active within ACC. Among other deleted regions, literature suggests that the loss of chromosomes 1p and 17p are linked to progression from benign to malignant ACC (17). Recent studies have identified a chromodomain helicase DNA-binding protein 5 within 1p36.31 as a tumor suppressor gene (22, 23). The absence of this protein allows uncontrolled cell growth and immortalization through the p19^Arf/p53 pathway (22). With mutations of p53 observed in Li-Fraumeni syndrome (with associated development of ACC), the loss of chromodomain helicase DNA-binding protein 5 may act as an additional hit within this pathway, increasing malignant growth, proliferation, and a lack of apoptosis. It is the elucidation and validation of these proposed pathways that will lead to an understanding of ACC oncogenesis.

Linking genomic events to clinical phenotype further increases the characterization of cancerous processes. Analysis of Kaplan-Meier survival curves for common aberrations identified a set of genomic gains and losses occurring specifically within populations with shortened survival rates. As illustrated in Fig. 2, genomic aberrations in red show selective presence in ≥20% of patients with short postsurgical survival. The regions include gain of 6q, 7q, 12q, and 19p and loss of 3, 8, 10p, 16q, 17q, and 19q (Table 2). These regions further cluster by accumulation of aberrations (Fig. 3A) and survival analysis identified an association between level of accumulation and survival rate (Fig. 3B). It is hypothesized that an increase in total aberrations, and not a combination of specific aberrations, links to shortened survival. This situation would indicate increasing genomic aberration with progressing disease. Multistep cancer progression cannot be illustrated in this data set due to the absence of analysis on benign adrenal tumors. Addition of benign and early-stage ACCs will allow development of a cancer progression model.

Disease variables, such as tumor size, tumor weight, tumor grade, functionality, and localization, were not shown to be significantly linked to the described survival differences (Fig. 3A). It is not clear whether these clinical variables truly have no effect on survival or whether there is insufficient statistical power (due to small sample size) to identify the underlying clinical stratifications.

Herein, high-resolution CGH has been used to characterize genomic differences between a group of 25 ACC tumors and normal DNA. These areas of gain and loss give insight into the mechanisms of disease and identify genomic regions containing potential targets for therapeutic intervention. Accumulation of gains and losses in specific regions can also be linked with survival difference, creating a genetic footprint for ACC survival and progression of disease.