An integrative model for recurrence in ovarian cancer

Ovarian cancer is one of the commonest cancers in women and the leading cause of death from gynaecological malignancy in the western world. About 205,000 cases of ovarian cancer are diagnosed worldwide each year [ 1 ]. It accounts for 3% of female cancers in Ireland with over 350 new cases each year [ 2 ]. Marked heterogeneity is a hallmark of the disease, not only in tumor histotype and grade but also in response to chemotherapy and overall prognosis [ 3 ]. Over 90% of cases arise from the surface epithelium. Serous adenocarcinomas are the commonest and account for 40%–50% of malignant neoplasms [ 4 ]. The majority of ovarian cancers present in advanced stages (III or IV) and are treated by surgery and systemic chemotherapy, most frequently carboplatin and paclitaxel. Conventional chemotherapy is still unsatisfactory as it ignores aspects of tumor biology. Despite an initial 70–80% response rate, current therapy is frequently followed by recurrence which is often resistant to chemotherapy, as demonstrated by the 5–20% long-term survivors [ 5 ]. Understanding the biological mechanisms underlying recurrence of ovarian cancer and addressing chemoresistance is of the utmost importance for improving treatment and outcome of the disease. Previous studies using single gene biomarkers to predict tumor response have been inconclusive. Patterns of gene expression for recurrence are likely to involve multiple gene pathways and integration of these pathways. High throughput discovery tools such as DNA microarrays have enabled the study of gene expression profiles of large numbers of cancer samples. Several groups have successfully applied expression array technology to the molecular classification of ovarian cancers [ 6 , 7 ], confirming a process of differentiation in the progression of ovarian cancer [ 8 ] and others have attempted to predict outcome and chemotherapeutic response [ 9 , 10 ]. Previous studies in our laboratory focused on assays and markers to predict response to chemotherapy [ 11 , 12 ]. While numerous studies have characterized primary ovarian cancers, less information is available regarding expression patterns of recurrent ovarian cancers. The aim of this study was to determine whether primary and recurrent ovarian tumors could be distinguished based on their gene expression profiles, by using gene expression arrays and to identify potential biomarkers of recurrence. A flow chart of our study design is shown in Figure 1 . To identify potentially important mediators of recurrence in the most frequent histological pattern of ovarian cancer, we performed cDNA microarray experiments on a homogenous set of primary and recurrent serous papillary ovarian tumors (cohort 1). Gene expression profiling revealed a total of 907 genes as differentially expressed between primary and recurrent samples at p < 0.01. Using the more stringent false discovery rate (FDR 0.1), this list was narrowed down to 182 genes. Included in this FDR list (with the exception of CLDN16 , which was the top differentially expressed gene in the p value list), were BTC , S100B , IL27RA , CSRP2 , ARFRP1 , PVRL2 , WASF1 , STARD10 , LASS4 , LGALS3BP , CASK , IFNGR1 , PGM2L1 , USF2 , PERP , ESM , CHORDC1 , RNPC1 , MGAT4B and CACNA1D which were selected for validation. Table 1 displays the fold changes observed for these genes, the corresponding probe identification and the p values. Flow chart of our study design. 2 cohorts were used in this study: In the first one, we selected a homogeneous series of primary and recurrent serous papillary adenocarcinomas from different patients(Between patient cohort). The second cohort consisted of 3 paired ovarian cancers (primary and recurrent samples coming from the same patient) but of different histology (Within patient cohort). Selected genes identified from microarray experiments were validated for both cohorts and a subset of these genes (n = 12) were validated in an independent set (test set) of 13 serous papillary adenocarcinomas using TaqMan PCR. To address whether recurrence follows similar patterns and to avoid individual genetic variation, we profiled paired samples from the same patient (cohort 2). A total of 586 genes were differentially expressed between primary and recurrent at p < 0.05 and this was reduced to 75 genes at p < 0.01. Genes with a fold change >4, upregulated in recurrent compared to primary tumors included: Septin 6 , ZNF218 , S100A8 , MMP9 , FOXF1 and ILIR2 (Table 2 ). Hierarchical heat maps, presented in Figures 2a and 2b , for both cohorts, demonstrated distinct gene expression patterns between primary and recurrent ovarian cancers. Hierarchical cluster heatmaps demonstrating distinct patterns of gene expression between primary and recurrent ovarian tumors. (a) Heatmap of the ovarian tumors in cohort 1 based on the FDR0.1 list with the primary clustering on the left and the recurrent samples on the right. Vertical bars represent the samples and the horizontal bars represent the genes. Green bars reflect downregulated genes and red bars upregulated genes. (b) Heat map discriminating recurrent (left) and primary (right) ovarian tumours in cohort 2 based on the p0.01 list. P, primary tumours; R, recurrent tumors. Notably, upregulated genes in the recurrent compared to primary tumors in cohort 1 and 2 segregated in the same gene families. Included among these genes are S100B and S100A8 belonging to the S100 family of calcium binding cytoplasmic proteins, TJP3 and CLDN16 belonging to the family of tight junction proteins, BTC and NRG2 belonging to the family of EGFR ligands, and interleukin receptors IL1R2 and IL27RA (Figure 3 ). Gene families involved in the molecular regulation of recurrence in ovarian cancer. Some of the upregulated genes in recurrent compared to primary ovarian carcinomas that we validated in cohort 2 belong in the same gene families with some of the upregulated genes validated in cohort 1. Upregulation of tight junction proteins and EGFR ligands, development of a cytokine response via interleukin receptors and intracellular signaling via calcium binding S100 proteins seem to contribute to the "recurrent" signature and possibly have a role in drug resistance. Using TaqMan PCR we validated dysregulated genes against different interrogation sets in order to select those likely to represent markers of recurrence (Figure 4 ). A list of the genes chosen for validation together with their molecular function and biological processes is shown in additional file 1 . Correlation was carried out using Spearman correlation co-efficient. The fold changes in the arrays were plotted against relative quantitation from the TaqMan analyses of recurrent versus primary tumors. High concordance was revealed between TaqMan and microarray experiments in cohort 1 (r = 0.874, p < 0.01) (Figure 5 ) and cohort 2 (r = 0.845, p < 0.05) (Figure 6 ). TaqMan PCR validation of target genes identified in both training and test sets. Gene selection for TaqMan validation was based on the most differentially expressed genes from the p and FDR value list with a fold change > 4 but also included genes that had a 2–4 fold change and also some genes involved in the most differentially expressed pathways. Priority was given to selection of genes upregulated in recurrent compared to primary samples, which might provide "recurrence" signatures in ovarian cancer. Upregulated genes validated in both cohorts were alternatively interrogated (external validation) and further advanced for validation in the test set. Independent validation on a test set refers to completely distinct samples of serous histology that were not previously employed in marker development (n = number of gene targets selected for validation). TaqMan PCR validation of microarray experiments in cohort 1. The fold changes in the arrays were plotted against the relative quantitation from TaqMan in recurrent vs primary tumours. The TaqMan values are displayed in blue and the array results in red. Spearman correlation r showed high concordance between the 2 experiments. TaqMan PCR validation of microarray experiments in cohort 2. A similar concordance was observed as in cohort 1. IL1R2 and ZNF218 identified in cohort 2 as upregulated in recurrent, when validated in samples from cohort 1, gave the best distinction with fold changes of 2.81 and 2.94 respectively (Figure 7 ). No significant difference was observed between recurrent and primary samples for the remaining 13 which is in accordance with the array results. External validation of a subset of upregulated genes in cohort 2 that validated in cohort 1. Bars indicate the relative overexpression of target genes in recurrent vs primary tumors. IL1R2 and ZNF218 gave the best distinction between recurrent and primary tumors with greater than twofold changes. Consecutively independent validation of a subset of the above genes (n = 12) from both cohorts was carried out in our test set of primary and recurrent serous papillary adenocarcinomas (n = 13) using TaqMan PCR to identify if these targets were possible markers of recurrence for serous papillary adenocarcinomas (Figure 8 ). BTC and FGF2 provided the best distinction between recurrent and primary tumors with fold changes of 2.8 and 2.71 respectively. Adding 2 recurrent samples of different histologies to the previously homogeneous histological sample cohort conferred no statistical significance for any of the validated genes (fold changes < 2). These were subsequently excluded from the analysis to preserve homogeneity in the test set. Independent TaqMan PCR validation of a set of selected genes from both cohorts in a test set of serous papillary adenocarcinomas of varying grade and stage. BTC and FGF2 provided the best distinction between recurrent and primary tumours with fold changes of 2.8 and 2.71 respectively. The results confirm the utility of the derived set of markers as potential markers of recurrence. Recurrence is rather multifactorial as indicated by non identification of a single biochemical pathway to relate the above targets using the ingenuity program [ 13 ].