Peritoneal carcinomatosis from ovarian cancer: chemosensitivity test and tissue markers as predictors of response to chemotherapy

Thirty-two patients with peritoneal carcinomatosis from primary advanced (7 cases) or recurrent (25 cases) ovarian cancer were recruited for the in vitro chemosensitivity assay and for analysis of biomarkers potentially predictive of resistance to platinum compounds. Patients underwent surgical resection at Pierantoni Hospital in Forlì and or at Bentivoglio Hospital in Bologna. Inclusion criteria were histological confirmation of advanced or recurrent ovarian cancer and pre- or a postsurgery chemotherapy based on a platinum compound (carboplatin/taxol or cisplatin/adriamycin or carboplatin/gemcitabine or carboplatin as monochemotherapy). It was not possible to perform the in vitro chemosensitivity test in 2 patients due to insufficient material. The remaining 30 patients all had serous tumor subtypes. Median age of patients was 60 ± 13.3 years (range 32-81).

Clinical response was evaluated by measuring circulating CA125 levels before each treatment cycle. Tumor imaging was performed every three cycles using ultrasonography or CT/MRI scans. The same clinical and instrumental evaluation was carried out every 3 months after the end of treatment.

Immediately after surgical resection, tumor specimens were sampled and analyzed (under sterile conditions) by a pathologist to confirm the tumor representativity of the samples. A part of the tissue was then stored in RNAlater ^® Tissue Collection (Invitrogen, Carlsbad, CA) at a temperature of +4°C to preserve mRNA integrity, while another part was used immediately for the chemosensitivity test.

Total RNA was extracted from fresh surgical biopsies using TRIzol^® Reagent within 2 or 3 hours of surgery, in accordance with the manufacturer's instructions (Invitrogen). Reverse transcription (RT) reactions were performed in a 20-μl volume containing 800 ng of total RNA using iScript TM cDNA Synthesis kit (Bio-Rad Laboratories, Hercules, CA) and analyzed by Real Time RT-PCR (MyiQ System, Bio-Rad) to detect the expression of the genes MGMT, BRCA1, ERCC1, GSTP1, and XPD. Primers for mRNA amplification were designed using Beacon Designer Software (version 4, BioRad) and sequences are listed in Table 1. The standard reaction volume was 25 μl containing 2 μl of cDNA template, 1 × SYBR Green Mix and 5 μM of forward and reverse primers. The mixture was subjected to the following cycling conditions: 95°C for 1 min and 30 s, followed by 40 cycles of amplification for 15 s at 95°C and 30 s at 59°C (for XPD) or 60°C (for MGMT, BRCA1, ERCC1, GSTP1, β ₂ -microglobulin, and hypoxanthine phosphoribosyltransferase (HPRT)). The amount of mRNA of each marker was normalized to the endogenous references β ₂ -microglobulin and HPRT using Gene Expression Macro Software (Version 1.1) (BioRad). Commercial RNA control derived from a pool of normal ovarian tissue mRNA was used as calibrator.

The efficiency of amplification, which never exceeded 5% variability in the different experiments, was used to determine the relative expression of mRNA and was calculated using Gene Expression Macro Software (Version 1.1) (BioRad). The reproducibility of Real-Time PCR results was verified in triplicate, and the coefficient of variation (CV), calculated from the three C_t values, was always < 1.5%.

A cell suspension was obtained after 4-16 hours of enzymatic digestion of fresh tumor tissue. Cells were counted and plated at a density of 1,000,000 cells/well in 96-well flat-bottomed microtiter plates (100 μl of cell suspension/well). Experiments were run in octuplicate. The optical density of treated and untreated cells was determined at a wavelength of 540 nm using a fluorescence plate reader.

Cells were exposed for 72 hours to 1, 10 and 100 μM of cisplatin or adriamycin; 8, 80 and 800 μM of carboplatin; 4, 40 and 400 μM of gemcitabine; and 0.6, 6 and 60 μM of taxol. Drugs were used at concentrations corresponding to peak plasma levels and were also tested at doses equivalent to one-tenth of and tenfold the peak plasma value. Drug activity was assessed by sulforhodamine B assay according to the method of Skehan et al [4]. PC3 tumor cell line, for which the dose-response curve to the anticancer agents used is known, was used as an internal control in all single experiments performed.

The relationship between continuous (gene expression) and dichotomous variables was analyzed using a non-parametric ranking statistic (median test) [26]. Spearman's correlation coefficient (r_s) was used to investigate the correlation between the mRNA expression of different genes, such as MGMT, BRCA1, ERCC1, GSTP1 and XPD, considered as continuous variables. Receiver operating characteristic (ROC) analysis was performed for both individual markers and their combinations. We considered an algorithm that renders a single composite score using the linear predictor fitted from a binary regression model. This algorithm has been justified to be optimal under the linearity assumption [27, 28] that the ROC curve is maximized (i.e., best sensitivity) at every threshold value. The chi-square test was used to compare dichotomous variables.

All statistical analyses were performed with SAS Statistical Software (version 9.1, SAS Institute Inc., Cary, NC). Two-sided p values < 0.05 were considered significant.

Of the 5 genes analyzed, MGMT and XPD expression was directly and significantly related to resistance to cisplatin-including regimens (p = 0.03 and p = 0.04, respectively) (Table 3). In particular, median expression values of MGMT and XPD in tumors were about fourfold higher in non-responders than in responders.

All 5 genes were generally poorly correlated with each other; with correlation coefficients (r_s) ranging from 0.577 to 0.074. In particular, of the two genes whose expression was maximally predictive of sensitivity or resistance to clinical treatment, XPD was not significantly related to ERCC1 or GSTP1, and showed borderline clinical significance with MGMT. The second, MGMT, was significantly related, albeit with a very poor correlation coefficient, to the other four genes (Table 4). The accuracy in predicting sensitivity or resistance to clinical treatment was analyzed for each single gene and for combinations of genes not significantly correlated with each other. Results were expressed as the area under the curve (AUC) and in terms of sensitivity, specificity and overall accuracy (Table 5). AUC values were maximum for MGMT (0.73; 95% CI 0.53-0.94) and XPD (0.70; 95% CI 0.48-0.91), and different gene combinations did not provide more accurate information. Only the 5 markers considered together slightly improved the AUC value (0.79; CI 0.62-0.97).

These results were paralleled by those expressed as overall accuracy: 78.5% and 75% for MGMT and XPD, respectively and 75% for the 5 markers considered together. XPD expression was characterized by the highest sensitivity (89.4%) but very low specificity (44.4%), while MGMT showed both high sensitivity (78.9%) and specificity (77.8%).

In parallel, a molecular profile of chemosensitivity to all the drugs used in the clinical treatment was generated for each tumor. Patients were subdivided into responders (complete or partial clinical response and stable disease), or non-responders (progressive disease), to evaluate the correlation between in vitro chemosensitivity assay and clinical response to platinum-containing treatments (Table 6). Seventeen patients (56.6%) were treated with carboplatin and taxol, of whom 6 had primary advanced and 11 recurrent ovarian cancer. We did not observe any significant differences in either in vitro or clinical sensitivity or resistance between primary and recurrent cancers. Considering the 2 subgroups together, concordance between in vitro results and clinical response was observed in 14 cases (3 in terms of sensitivity, 11 in terms of resistance). The 3 cases in whom there was no correspondence between in vitro and in vivo results were all in vitro sensitive to one drug (carboplatin or taxol); two showed clinical progression and one stable disease (Table 6). Similarly, in the subgroup of 6 patients treated with cisplatin and adriamycin, 3 were in vitro- sensitive to both drugs and showed a clinical response, while 3 were in vitro resistant to both drugs and showed disease progression. Patients treated with carboplatin (3 cases: 1 primary and 2 recurrent), carboplatin and gemcitabine (2 cases), or carboplatin, taxol and gemcitabine (2 cases) were in vitro resistant to all the drugs and all had disease progression.

Results of the clinical response predictivity of the most relevant markers, considered singly or in combination, and of the in vitro chemosensitivity test are shown in Table 7. Significant predictivity in terms of sensitivity and resistance to the different cisplatin-based regimens was observed for MGMT expression (75.0% and 72.5%, respectively; p = 0.03), while high predictivity with regard to resistance (90.9%), but very low predictivity in terms of sensitivity (37.5%) (p = 0.06) were observed for XPD. The combined analysis of the five markers gave the highest predictivity with regard to resistance but very low predictivity in relation to sensitivity (100% and 33.3%, respectively; p = 0.07).