Purpose: The goal of this study was to determine whether distinct gene expression profiles are associated with intrinsic and/or acquired chemoresistance in epithelial ovarian carcinoma.

Experimental Design: Gene expression profiles were generated from 21 primary chemosensitive tumors and 24 primary chemoresistant tumors using cDNA-based microarrays. Gene expression profiles of both groups of primary tumors were then compared with those of 15 ovarian carcinomas obtained following platinum-based chemotherapy (“postchemotherapy” tumors). A theme discovery tool was used to identify functional categories of genes involved in drug resistance.

Results: Comparison of primary chemosensitive and chemoresistant tumors revealed differential expression of 85 genes (P < 0.001). Comparison of gene expression profiles of primary chemosensitive tumors and postchemotherapy tumors revealed more robust differences with 760 genes differentiating the two groups (P < 0.001). In contrast, only 230 genes were differentially expressed between primary chemoresistant and postchemotherapy groups (P < 0.001). Common to both gene lists were 178 genes representing transcripts differentially expressed between postchemotherapy tumors and all primary tumors irrespective of intrinsic chemosensitivity. The gene expression profile of postchemotherapy tumors compared with that of primary tumors revealed statistically significant overrepresentation of genes encoding extracellular matrix–related proteins.

Conclusions: These data show that gene expression profiling can discriminate primary chemoresistant from primary chemosensitive ovarian cancers. Gene expression profiles were also identified that correlate with states of intrinsic and acquired chemoresistance and that represent targets for future investigation and potential therapeutic interventions.

Platinum-based combination chemotherapy is the standard first-line treatment for advanced-stage epithelial ovarian carcinoma. For the ∼75% of patients diagnosed with advanced-stage disease, 20% to 30% progress on or rapidly become resistant to this treatment and subsequently show low response rates to other second-line agents (13). Early identification of this group of patients could lead to their enrollment in clinical trials or treatment with other experimental therapeutics because standard treatment affords them little benefit. Among initially chemosensitive patients, the vast majority will eventually relapse. Thus, chemoresistance may be present at the outset of treatment (intrinsic resistance) or may develop during treatment (acquired resistance). In practice, ovarian cancers are considered “platinum sensitive” if the clinical progression free interval is >6 months, and evidence suggests that the longer this interval, the higher the subsequent response rates to additional chemotherapy (46).

Understanding the biological mechanisms underlying chemoresistance is of utmost importance for improving the treatment and outcome of ovarian cancer. This topic has been the subject of intense research, and previous studies on chemoresistance in ovarian cancer have investigated potential involvement of molecules involved in drug transport, apoptosis, DNA repair, and detoxification pathways (711). Much of this research has been done using cell culture models and far fewer data are available on the relevance of these studies to, and biomarkers and potential mechanisms of drug resistance for, clinical samples.

The availability of new high-throughput screening techniques has allowed for more global investigations of molecular profiles associated with chemoresistance. In the present study, cDNA microarrays were used to investigate gene expression patterns associated with both intrinsic and acquired chemoresistance in ovarian cancer. The first aim of this investigation was to determine if intrinsically chemoresistant and chemosensitive tumors could be distinguished based on their gene expression profiles.

For this part of the investigation, a case was classified as intrinsic chemoresistant based on persistent or recurrent disease within 6 months of initiating first-line platinum-based combination chemotherapy. Chemosensitive tumors were classified as such based on a complete response to chemotherapy and a platinum-free interval of ≥13 months. These conservative clinical criteria for defining platinum sensitivity and resistance were employed to exclude tumors with intermediate levels of resistance. In the second part of the investigation, gene expression profiles of tumors obtained following chemotherapy (“postchemotherapy” samples) were compared with those of the chemosensitive and chemoresistant primary tumors. The postchemotherapy group consisted of tumors from nine patients treated with neoadjuvant chemotherapy who subsequently underwent an interval cytoreductive surgery and from six patients who had residual cancer present at the time of second-look surgery following chemotherapy.

Gene expression profiles of these postchemotherapy tumors were compared with those of our primary (i.e., chemonaive) chemosensitive and chemoresistant tumors. The rationale for this approach was 2-fold. First, after each cycle of cytotoxic chemotherapy, the “log kill” effect leads to a significant reduction in the number of tumor cells that are sensitive to the administered therapy (12, 13). Second, tumor cells that survive the treatment are likely to experience changes in gene expression that allow them to withstand the selective pressure of the drugs used. Hence, tumor samples obtained shortly following chemotherapy are enriched in resistant clones and are likely to display the molecular signature associated with chemoresistance.

Tissue specimens. This study was approved by the Memorial Sloan-Kettering Cancer Center Institutional Review Board. All tumor samples were obtained at the time of surgery, frozen in liquid nitrogen, and stored at −80°C until use. Information on treatment and response was obtained from patient chart review. Intrinsically chemoresistant tumors were defined as those associated with persistent or recurrent disease within 6 months of the initiation of first-line platinum-based combination chemotherapy. Chemosensitive tumors were defined as those with a complete response to chemotherapy and a platinum-free interval of ≥13 months. The postchemotherapy group is composed of patients who had either surgical debulking following chemotherapy (i.e., neoadjuvant chemotherapy) or residual tumor at the time of a second-look procedure. All of these tumor samples were obtained within 6 weeks of the last cycle of chemotherapy. Ovarian tissues from two postmenopausal women obtained at the time of salpingoophorectomy for benign indications were used for comparative purposes.

RNA preparation and cDNA microarray analysis. Isolation of RNA was done using the RNeasy column (Qiagen, Valencia, CA) according to the manufacturer's instructions. The integrity of RNA was verified by denaturing gel electrophoresis. Total RNA was linearly amplified using a modification of the Eberwine procedure (14). Briefly, total RNA was reverse transcribed by using a 63-nucleotide synthetic primer containing the T7 RNA polymerase binding site [5′-GGCCAGTGAATTGTAATACGACTCACTATAGGGA-GGCGG(T)24-3′]. Second-strand cDNA synthesis (producing double-stranded cDNA) was done with RNase H, Escherichia coli DNA polymerase I, and E. coli DNA ligase (Invitrogen, Carlsbad, CA). After cDNA was blunt ended with T4 DNA polymerase (Invitrogen), purification was accomplished by phenol/chloroform/isoamyl alcohol extraction and ammonium acetate/ethanol precipitation. The double-stranded cDNA was then transcribed using T7 polymerase (T7 Megascript kit, Ambion, Austin, TX), yielding amplified antisense RNA that was purified using RNeasy mini-columns. Commercially available pooled total RNA from 10 different human cell lines (Stratagene, La Jolla, CA) was amplified and used as the reference for cDNA microarray experiments.

Investigation of gene expression differences between primary chemosensitive and chemoresistant tumors was done using two separate cDNA microarrays to maximize the number of genes screened. The two cDNA microarrays contained 32,448 and 7,585 features each for a combined total of 40,033 transcripts. The comparison between the postchemotherapy samples and the primary tumors was done using the 7,585-feature cDNA microarray. All cDNA microarrays were manufactured at the National Cancer Institute. Amplified RNA (4 μg) was reverse transcribed and directly labeled using cyanine 5–conjugated dUTP (tumor RNA) or cyanine 3–conjugated dUTP (pooled reference RNA). Hybridization was done in the presence of 5× SSC and 25% formamide for 14 to 16 hours at 42°C. Slides were washed, dried, and scanned using an Axon Instruments 4000a laser scanner. A detailed protocol for RNA amplification as well as cDNA probe labeling and hybridization is available at http://nciarray.nci.nih.gov/reference/ (under “Alternative Methods and Protocols”). Genepix software (Molecular Devices, Sunnyvale, CA) was used to analyze the raw data that were then uploaded to a relational database maintained by the Center for Information Technology at the NIH.

Data analysis. The log expression ratios for the spots on each array were normalized by subtracting the median log ratio for the same array. Data were filtered to exclude spots with size <25 μm, intensity less than twice background or <500 units in both red and green channels, and any flagged or missing spots. In addition, any features found to be missing or flagged in >10% of the arrays were not included in the analysis. The Genepix software assigns intensity levels in arbitrary units with a range between 0 and 65,535. For reference, typical median background on arrays used in this investigation were 250, and the median probe signal was ∼4,500 (after background subtraction). Statistical comparison between tumors groups was done using the “BRB Array Tools” software (http://linus.nci.nih.gov/BRB-ArrayTools.html), consisting of a modified F test with P < 0.001 considered significant. This stringent P was selected in lieu of the Bonferroni correction for multiple comparisons, which was deemed excessively restrictive. Gene lists were interrogated using EASE software (15) to identify possible overrepresentation of genes belonging to the same biological or functional class.

Class prediction was done using a compound covariate predictor tool available as part of the BRB Array Tools software. This tool creates a multivariate predictor for one of two classes to each sample. The genes included in the multivariate predictor were those that were univariately significant at the selected significance cutoff of P < 0.0001. The multivariate predictor is a weighed linear combination of log ratios for genes that are univariately significant. The weight consists of the univariate t statistics for comparing the classes. A leave-one-out approach was then employed to test the ability of the compound covariate predictor to assign chemosensitive or chemoresistant class to each sample. A permutation test was used to assess the significance of our cross-validated error rate. The random permutations test the null hypothesis that there are no systematic differences in gene expression profiles of the chemoresistant and chemosensitive tumors. This assumption can be tested by randomly permuting labels among the gene expression profiles and determining what proportion of the permuted data sets have a misclassification error rate less than or equal to the observed error. This rate serves as the achieved significance level in a test against the null hypothesis. Detailed information about the compound covariate predictor and the permutation test for significance is provided by the Biometric Research Branch, National Cancer Institute and is available at http://www.healthsystem.virginia.edu/internet/obgyn/supplemental-figure.pdf.

Immunohistochemical analyses. Immunohistochemical staining was done on 5 μm sections of formalin-fixed, paraffin-embedded ovarian tumor specimens. After deparaffinization, sections were pretreated with steam for 20 minutes in citrate buffer (pH 6.0). Slides were stained with primary antibodies against Ki-67 (mouse monoclonal clone MIB-1; 1:75 dilution; DAKOCytomation, Carpinteria, CA), proliferating cell nuclear antigen (PCNA; mouse monoclonal clone PC10; ready-to-use; DAKOCytomation), and cathepsin D (CTSD; rabbit polyclonal; ready-to-use; DAKOCytomation). Staining was done on a DAKO Autostainer using the LSAB2 kit (DAKOCytomation) consisting of biotinylated anti-mouse and anti-rabbit ready-to-use secondary antibodies, streptavidin-horseradish peroxidase, and the chromogen diaminobenzidine. The slides were counterstained with methyl green, dehydrated, and mounted. Quantitative scoring was done as the product of two staining characteristics, the percentage of immunopositive cells and the intensity of the staining. Slides were examined microscopically at 20× power, and five separate areas of each tumor specimen were examined, with ∼100 cells per area analyzed. Scoring for the number of immunopositive cells was accomplished by assigning 0% to 25% as 1, 26% to 50% as 2, 51% to 75% as 3, and 76% to 100% as 4. Intensity was scored as 1 to 3. The final score consisted of the product of the immunopositive score (averaged over the five areas) and the intensity score (averaged over the five areas). Statistical analysis was accomplished using Student's t test.

Gene expression differences between primary chemosensitive and chemoresistant tumors. The clinicopathologic features of the samples used in this investigation are presented in Table 1 (primary cancers) and Table 2 (postchemotherapy cancers). Tumors in both chemosensitive and chemoresistant groups were predominantly advanced stage, high grade (grade 2 or 3), and of serous histology. All patients were treated with platinum-based chemotherapy. For the purposes of this investigation, all chemoresistant tumors were from patients with persistent or recurrent disease present within 6 months of the initiation of first-line platinum-based combination chemotherapy. The median disease-free interval for the chemosensitive samples was 21 months (range, 13-41 months). The median age for the chemoresistant group was 61 (range, 35-77), and the median age for the chemosensitive group was 57 (range, 44-78).

Table 1.

Clinicopathologic features of primary ovarian cancer cases

Case no.Age*HistologyStageGradeChemotherapyDFE
Primary chemoresistant       
    R1 49 Clear cell IIIC Carboplatin/paclitaxel 
    R2 61 Endometrioid IIIC Carboplatin/paclitaxel 
    R3 61 Serous IV Carboplatin/paclitaxel 
    R4 49 Serous IIIC Carboplatin/paclitaxel 
    R5 77 Serous IIIC NA Carboplatin/paclitaxel 
    R6 49 Serous IIIC Carboplatin/paclitaxel 
    R7 55 Serous IV Carboplatin/paclitaxel 
    R8 56 Serous IIIC 2-3 Carboplatin/paclitaxel 
    R9 61 Serous IIIC Carboplatin/paclitaxel 
    R10 73 Serous IV Carboplatin/paclitaxel 
    R11 78 Serous IIIC Carboplatin/paclitaxel 
    R12 71 Serous IV Carboplatin/paclitaxel 
    R13 52 Serous IV Carboplatin/paclitaxel 
    R14 64 Serous IIIA Carboplatin/paclitaxel 
    R15 49 Serous IIIC Carboplatin/paclitaxel 
    R16 47 Serous IV Carboplatin/paclitaxel 
    R17 66 Serous IIIC Cisplatin/cyclophosphamide 
    R18 69 Serous IV Cisplatin/cyclophosphamide 
    R19 44 Serous IV HD§ cisplatin/cyclophosphamide 
    R20 35 Serous IV Cisplatin/cyclophosphamide 
    R21 67 Serous IIIC 2-3 Cisplatin/cyclophosphamide 
    R22 63 Endometrioid IIIC Carboplatin/paclitaxel 
    R23 76 Serous IIIC Carboplatin 
    R24 65 Serous IIIC Carboplatin/paclitaxel 
Primary chemosensitive       
    S1 51 Serous IV Carboplatin/paclitaxel 35 
    S2 45 Endometrioid IIIC 2-3 HD Carboplatin/paclitaxel 41 
    S3 65 Mixed IIIC Carboplatin/paclitaxel 13 
    S4 77 Serous IIIC 2-3 Carboplatin/paclitaxel 31 
    S5 53 Serous IIIC Carboplatin/paclitaxel 32 
    S6 71 Serous IIIC Carboplatin/paclitaxel 30 
    S7 55 Serous IIIC Carboplatin/paclitaxel 19 
    S8 78 Endometrioid IIIC 2-3 Carboplatin/paclitaxel 24 
    S9 46 Endometrioid IIIC Carboplatin/paclitaxel 21 
    S10 69 Serous IIIC Carboplatin/paclitaxel 14 
    S11 54 Serous IV Carboplatin/docetaxel 18 
    S12 53 Serous IIIC Carboplatin/paclitaxel 18 
    S13 77 Serous IIIB Carboplatin/paclitaxel 21 
    S14 56 Serous IIC Carboplatin/paclitaxel 16 
    S15 44 Serous IIIC Carboplatin/paclitaxel 16 
    S16 59 Carcinoma IIIC Carboplatin/paclitaxel 16 
    S17 44 Serous IIIC NA Carboplatin/paclitaxel 13 
    S18 69 Serous IIIC Carboplatin/paclitaxel 21 
    S19 70 Carcinoma IIIC Carboplatin/paclitaxel 36 
    S20 74 Serous IIIC Carboplatin/paclitaxel 21 
    S21 57 Serous IIIC Carboplatin/paclitaxel 18 
Case no.Age*HistologyStageGradeChemotherapyDFE
Primary chemoresistant       
    R1 49 Clear cell IIIC Carboplatin/paclitaxel 
    R2 61 Endometrioid IIIC Carboplatin/paclitaxel 
    R3 61 Serous IV Carboplatin/paclitaxel 
    R4 49 Serous IIIC Carboplatin/paclitaxel 
    R5 77 Serous IIIC NA Carboplatin/paclitaxel 
    R6 49 Serous IIIC Carboplatin/paclitaxel 
    R7 55 Serous IV Carboplatin/paclitaxel 
    R8 56 Serous IIIC 2-3 Carboplatin/paclitaxel 
    R9 61 Serous IIIC Carboplatin/paclitaxel 
    R10 73 Serous IV Carboplatin/paclitaxel 
    R11 78 Serous IIIC Carboplatin/paclitaxel 
    R12 71 Serous IV Carboplatin/paclitaxel 
    R13 52 Serous IV Carboplatin/paclitaxel 
    R14 64 Serous IIIA Carboplatin/paclitaxel 
    R15 49 Serous IIIC Carboplatin/paclitaxel 
    R16 47 Serous IV Carboplatin/paclitaxel 
    R17 66 Serous IIIC Cisplatin/cyclophosphamide 
    R18 69 Serous IV Cisplatin/cyclophosphamide 
    R19 44 Serous IV HD§ cisplatin/cyclophosphamide 
    R20 35 Serous IV Cisplatin/cyclophosphamide 
    R21 67 Serous IIIC 2-3 Cisplatin/cyclophosphamide 
    R22 63 Endometrioid IIIC Carboplatin/paclitaxel 
    R23 76 Serous IIIC Carboplatin 
    R24 65 Serous IIIC Carboplatin/paclitaxel 
Primary chemosensitive       
    S1 51 Serous IV Carboplatin/paclitaxel 35 
    S2 45 Endometrioid IIIC 2-3 HD Carboplatin/paclitaxel 41 
    S3 65 Mixed IIIC Carboplatin/paclitaxel 13 
    S4 77 Serous IIIC 2-3 Carboplatin/paclitaxel 31 
    S5 53 Serous IIIC Carboplatin/paclitaxel 32 
    S6 71 Serous IIIC Carboplatin/paclitaxel 30 
    S7 55 Serous IIIC Carboplatin/paclitaxel 19 
    S8 78 Endometrioid IIIC 2-3 Carboplatin/paclitaxel 24 
    S9 46 Endometrioid IIIC Carboplatin/paclitaxel 21 
    S10 69 Serous IIIC Carboplatin/paclitaxel 14 
    S11 54 Serous IV Carboplatin/docetaxel 18 
    S12 53 Serous IIIC Carboplatin/paclitaxel 18 
    S13 77 Serous IIIB Carboplatin/paclitaxel 21 
    S14 56 Serous IIC Carboplatin/paclitaxel 16 
    S15 44 Serous IIIC Carboplatin/paclitaxel 16 
    S16 59 Carcinoma IIIC Carboplatin/paclitaxel 16 
    S17 44 Serous IIIC NA Carboplatin/paclitaxel 13 
    S18 69 Serous IIIC Carboplatin/paclitaxel 21 
    S19 70 Carcinoma IIIC Carboplatin/paclitaxel 36 
    S20 74 Serous IIIC Carboplatin/paclitaxel 21 
    S21 57 Serous IIIC Carboplatin/paclitaxel 18 
*

Age at time of surgery.

Disease-free interval following completion of chemotherapy (months).

Not available.

§

High-dose chemotherapy with bone marrow transplant.

Table 2.

Clinicopathologic features of postchemotherapy ovarian cancer cases

Case no.Age*HistologyStageGradeClinical hxChemotherapy
PC1 57 Serous IIIC NC/ID Carboplatin/paclitaxel × 3 
PC2 74 Serous IIIB NC/ID Carboplatin/paclitaxel × 3 
PC3 59 Serous IIIC NC/ID Carboplatin/paclitaxel × 3 
PC4 77 Serous IIIC NC/ID Carboplatin/paclitaxel × 6 
PC5 73 Serous IIIC NC/ID Carboplatin/paclitaxel × 6 
PC6 68 Undifferentiated IV NC/ID Carboplatin/paclitaxel × 6 
PC7 22 Serous IIIC NC/ID Carboplatin/paclitaxel × 6 
PC8 71 Serous IV NC/ID Carboplatin/paclitaxel × 6 
PC9 71 Serous IIIC NC/ID Carboplatin/paclitaxel × 6 
PC10 45 Endometrioid IIIC PSL HD§ Carboplatin/paclitaxel 
PC11 56 Serous IIIC PSL Carboplatin/paclitaxel × 6 
PC12 58 mixed IIIC PSL Carboplatin/paclitaxel × 6 
PC13 66 Serous IIIC PSL Carboplatin/paclitaxel × 6 
PC14 42 Serous IIIC PSL Carboplatin/paclitaxel × 6 
PC15 28 Serous IIIC PSL Carboplatin/paclitaxel × 6 
Case no.Age*HistologyStageGradeClinical hxChemotherapy
PC1 57 Serous IIIC NC/ID Carboplatin/paclitaxel × 3 
PC2 74 Serous IIIB NC/ID Carboplatin/paclitaxel × 3 
PC3 59 Serous IIIC NC/ID Carboplatin/paclitaxel × 3 
PC4 77 Serous IIIC NC/ID Carboplatin/paclitaxel × 6 
PC5 73 Serous IIIC NC/ID Carboplatin/paclitaxel × 6 
PC6 68 Undifferentiated IV NC/ID Carboplatin/paclitaxel × 6 
PC7 22 Serous IIIC NC/ID Carboplatin/paclitaxel × 6 
PC8 71 Serous IV NC/ID Carboplatin/paclitaxel × 6 
PC9 71 Serous IIIC NC/ID Carboplatin/paclitaxel × 6 
PC10 45 Endometrioid IIIC PSL HD§ Carboplatin/paclitaxel 
PC11 56 Serous IIIC PSL Carboplatin/paclitaxel × 6 
PC12 58 mixed IIIC PSL Carboplatin/paclitaxel × 6 
PC13 66 Serous IIIC PSL Carboplatin/paclitaxel × 6 
PC14 42 Serous IIIC PSL Carboplatin/paclitaxel × 6 
PC15 28 Serous IIIC PSL Carboplatin/paclitaxel × 6 
*

Age at time of surgery.

Neoadjuvant chemotherapy (as indicated) followed by interval debulking surgery.

Positive second-look surgery.

§

High-dose chemotherapy as indicated with bone marrow transplant.

The comparison of gene expression profiles of 21 primary chemosensitive and 24 primary chemoresistant ovarian cancers revealed 85 transcripts with expression levels significantly different between the two groups (P < 0.001; Table 3). The difference in geometric mean expression levels for all of these transcripts was ≤2-fold. The ability of the nine most significantly differentially expressed genes (P < 0.0001) to predict clinical response was tested using a leave-one-out prediction model. This analysis revealed an accuracy of 77.8% in correctly classifying refractory and responsive tumors. After 5,000 random permutations, the likelihood of these nine genes differentiating the two groups with equal or higher accuracy by chance (i.e., the null hypothesis) was calculated as P = 0.018. These data showed that, although statistically significant differences at the mRNA level existed between chemosensitive and chemoresistant primary tumors, the magnitude of these differences was modest in primary tumor samples obtained before chemotherapy.

Table 3.

Genes differentially expressed between primary chemoresistant and primary chemosensitive ovarian cancers (P < 0.001)

IMAGE clone*UniGeneGeneDescriptionFold differenceP
366971 Hs.156346 TOP2A Topoisomerase (DNA) IIα170-kDa 1.81 0.0008 
810263 Hs.335798 RHPN2 Rhophilin, Rho GTPase-binding protein 2 1.70 0.00006 
435303 Hs.397426 KIAA4146 KIAA1416 protein 1.58 0.0006 
753464 Hs.33540 LOC389677 Similar to RIKEN cDNA 3000004N20 1.57 0.00004 
742581 Hs.42173 C6orf107 Chromosome 6 open reading frame 107 1.57 0.00002 
450653 Hs.84063 BCL2L11 BCL2-like 11 (apoptosis facilitator) 1.57 0.0009 
129345 Hs.173946 PAPD1 PAP-associated domain containing 1 1.57 0.00004 
589967 Hs.301431 ZNF71 Zinc finger protein 71 (Cos26) 1.56 0.0002 
725223 Hs.119563 PSME4 Proteasome (prosome, macropain) activator subunit 4 1.55 0.0008 
452963 Hs.321390 CUGBP1 CUG triplet repeat, RNA-binding protein 1 1.54 0.0003 
71622 Hs.496511 PRKCI Protein kinase C, ι 1.52 0.0007 
757383 Hs.59236 ANKRD27 Ankyrin repeat domain 27 (VPS9 domain) 1.52 0.0001 
220658 Hs.440394 MSH2 MutS homologue 2, colon cancer, nonpolyposis type 1 (E. coli) 1.49 0.0003 
1470530 Hs.435788 NCOA6 Nuclear receptor coactivator 6 1.47 0.0004 
261567 Hs.25812 NBS1 Nijmegen breakage syndrome 1 (nibrin) 1.47 0.00006 
592928 Hs.173001 KIAA1221 KIAA1221 protein 1.47 0.00003 
839048 Hs.156682 IGSF4 Immunoglobulin superfamily, member 4 1.47 0.0003 
773203 Hs.43627 SOX12 SRY (sex-determining region Y) box 12 1.42 0.0009 
452423 Hs.326392 SOS1 Son of sevenless homologue 1 (Drosophila) 1.42 0.00006 
306568 Hs.415997 COL6A1 Collagen, type VI, α1 1.42 0.00002 
855563 Hs.306251 ERBB3 v-erb-b2 erythroblastic leukemia viral oncogene homologue 3 1.42 0.0009 
128159 Hs.170472 TPR Translocated promoter region (to activated MET oncogene) 1.41 0.0002 
343490 Hs.381189 CBX3 Chromobox homologue 3 (HP1γ homologue, Drosophila) 1.41 0.0004 
51737 Hs.437224 RBBP8 Retinoblastoma-binding protein 8 1.38 0.00003 
247240 Hs.8258 C19orf13 Chromosome 19 open reading frame 13 1.38 0.0005 
950667 Hs.36761 HRASLS HRAS-like suppressor 1.37 0.0001 
264117 Hs.343475 CTSD Cathepsin D (lysosomal aspartyl protease) 1.35 0.0006 
741790 Hs.7942 AFTIPHILIN Aftiphilin protein 1.34 0.0004 
490414 Hs.386198 EML4 Echinoderm microtubule-associated protein-like 4 1.34 0.001 
230235 Hs.215766 GTPBP4 GTP-binding protein 4 1.34 0.0001 
853066 Hs.5719 CNAP1 Chromosome condensation-related SMC-associated protein 1 1.33 0.0002 
950092 Hs.405144 SFRS3 Splicing factor, arginine/serine–rich 3 1.32 0.0002 
859857 Hs.442787 ZNF148 Zinc finger protein 148 (pHZ-52) 1.32 0.0007 
447569 Hs.282901 RNPC2 RNA-binding region (RNP1, RRM) containing 2 1.32 0.0001 
344942 Hs.109299 PPFIA3 Protein tyrosine phosphatase, interacting protein (liprin), α3 1.32 0.0009 
767419 Hs.274351 ZDHHC9 Zinc finger, DHHC domain-containing 9 1.32 0.0006 
882434 Hs.362996 KIAA0779 KIAA0779 protein 1.31 0.00005 
283329 Hs.252451 SEMA3A Semaphorin 3A 1.31 0.0007 
359184 Hs.439523 PRKR Protein kinase, IFN-inducible 1.30 0.0006 
180156 Hs.123654 PCF11 Pre-mRNA cleavage complex II protein Pcf11 1.30 0.0002 
180156 Hs.128959  EST 1.30 0.0002 
740707 Hs.269902 KIAA0494 KIAA0494 protein 1.28 0.0006 
884892 Hs.7838 MKRN1 Makorin, ring finger protein, 1 1.27 0.00002 
826135 Hs.367811 STK38 Serine/threonine kinase 38 1.26 0.0007 
838359 Hs.77890 GUCY1B3 Guanylate cyclase 1, soluble, β3 1.26 0.00008 
878174 Hs.388164 FADS2 Fatty acid desaturase 2 1.26 0.0006 
837864 Hs.94262 DD5 Progestin-induced protein 1.26 0.0008 
753914 Hs.512235 ITPR2 Inositol 1,4,5-triphosphate receptor, type 2 1.26 0.0006 
825197 Hs.131168 SEP1 Strand-exchange protein 1 1.25 0.0006 
815772 Hs.369284 C20orf6 Chromosome 20 open reading frame 6 1.24 0.0009 
754654 Hs.118964 p66α P66 α 1.23 0.0002 
384018 Hs.411300 WBP4 WW domain-binding protein 4 (formin-binding protein 21) 1.23 0.00002 
320834 Hs.386404 UBE4B Ubiquitination factor E4B (UFD2 homologue, yeast) 1.22 0.0007 
73596 Hs.35086 USP1 Ubiquitin-specific protease 1 1.19 0.0003 
743727 Hs.389638 FLJ41501 Clone BRTHA2006975 0.86 0.0006 
277134 Hs.93836 CIP98 CASK-interacting protein CIP98 0.85 0.0009 
433289 Hs.117331 TREML1 Triggering receptor expressed on myeloid cells-like 1 0.84 0.0009 
811999 Hs.6455 RUVBL2 RuvB-like 2 (E. coli) 0.83 0.0003 
1292893 Hs.125785 LOC149018 Hypothetical LOC149018, mRNA 0.83 0.0009 
813735 Hs.410314 PCDH16 Protocadherin 16 dachsous-like (Drosophila) 0.81 0.0006 
814271 Hs.18885 CGI-116 CGI-116 protein 0.79 0.0004 
399563 Hs.120332  EST 0.78 0.0004 
814129 Hs.283716 MSCP Mitochondrial solute carrier protein 0.77 0.00002 
490484 Hs.356349 ZNF145 Zinc finger protein 145 (Kruppel-like, expressed in PML) 0.77 0.0007 
47043 Hs.154138 CHI3L2 Chitinase 3–like 2 0.73 0.001 
470035 Hs.14060 PROK1 Prokineticin 1 0.71 0.0007 
2571195 Hs.406683 RPS15 Ribosomal protein S15 0.71 0.0003 
303109 Hs.123464 P2RY5 Purinergic receptor P2Y, G-protein coupled, 5 0.70 0.0005 
431231 Hs.381870 EFEMP2 Epidermal growth factor–containing fibulin-like ECM protein 2 0.70 0.0006 
280950 Hs.422340 SRI Sorcin 0.66 0.0001 
453183 Hs.301302 SCAM-1 Vinexin β (SH3-containing adaptor molecule-1) 0.66 0.0004 
490556 Hs.26518 TM4SF7 Transmembrane 4 superfamily member 7 0.63 0.0001 
502499 Hs.75835 PMM1 Phosphomannomutase 1 0.61 0.0001 
344720 Hs.81994 GYPC Glycophorin C (Gerbich blood group) 0.60 0.0003 
2559389 Hs.150833 C4A Complement component 4A 0.58 0.0008 
2784073 Hs.322431 NEUROD2 Neurogenic differentiation 2 0.54 0.0005 
898305 Hs.439671 NBL1 Neuroblastoma, suppression of tumorigenicity 1 0.54 0.00002 
756372 Hs.37682 RARRES2 Retinoic acid receptor responder (tazarotene-induced) 2 0.52 0.0003 
1474174 Hs.367877 MMP2 Matrix metalloproteinase-2 (72-kDa type IV collagenase) 0.50 0.00004 
462939 Hs.526933  EST 0.48 0.001 
823851 Hs.469463 AEBP1 AE-binding protein 1 0.41 0.0007 
IMAGE clone*UniGeneGeneDescriptionFold differenceP
366971 Hs.156346 TOP2A Topoisomerase (DNA) IIα170-kDa 1.81 0.0008 
810263 Hs.335798 RHPN2 Rhophilin, Rho GTPase-binding protein 2 1.70 0.00006 
435303 Hs.397426 KIAA4146 KIAA1416 protein 1.58 0.0006 
753464 Hs.33540 LOC389677 Similar to RIKEN cDNA 3000004N20 1.57 0.00004 
742581 Hs.42173 C6orf107 Chromosome 6 open reading frame 107 1.57 0.00002 
450653 Hs.84063 BCL2L11 BCL2-like 11 (apoptosis facilitator) 1.57 0.0009 
129345 Hs.173946 PAPD1 PAP-associated domain containing 1 1.57 0.00004 
589967 Hs.301431 ZNF71 Zinc finger protein 71 (Cos26) 1.56 0.0002 
725223 Hs.119563 PSME4 Proteasome (prosome, macropain) activator subunit 4 1.55 0.0008 
452963 Hs.321390 CUGBP1 CUG triplet repeat, RNA-binding protein 1 1.54 0.0003 
71622 Hs.496511 PRKCI Protein kinase C, ι 1.52 0.0007 
757383 Hs.59236 ANKRD27 Ankyrin repeat domain 27 (VPS9 domain) 1.52 0.0001 
220658 Hs.440394 MSH2 MutS homologue 2, colon cancer, nonpolyposis type 1 (E. coli) 1.49 0.0003 
1470530 Hs.435788 NCOA6 Nuclear receptor coactivator 6 1.47 0.0004 
261567 Hs.25812 NBS1 Nijmegen breakage syndrome 1 (nibrin) 1.47 0.00006 
592928 Hs.173001 KIAA1221 KIAA1221 protein 1.47 0.00003 
839048 Hs.156682 IGSF4 Immunoglobulin superfamily, member 4 1.47 0.0003 
773203 Hs.43627 SOX12 SRY (sex-determining region Y) box 12 1.42 0.0009 
452423 Hs.326392 SOS1 Son of sevenless homologue 1 (Drosophila) 1.42 0.00006 
306568 Hs.415997 COL6A1 Collagen, type VI, α1 1.42 0.00002 
855563 Hs.306251 ERBB3 v-erb-b2 erythroblastic leukemia viral oncogene homologue 3 1.42 0.0009 
128159 Hs.170472 TPR Translocated promoter region (to activated MET oncogene) 1.41 0.0002 
343490 Hs.381189 CBX3 Chromobox homologue 3 (HP1γ homologue, Drosophila) 1.41 0.0004 
51737 Hs.437224 RBBP8 Retinoblastoma-binding protein 8 1.38 0.00003 
247240 Hs.8258 C19orf13 Chromosome 19 open reading frame 13 1.38 0.0005 
950667 Hs.36761 HRASLS HRAS-like suppressor 1.37 0.0001 
264117 Hs.343475 CTSD Cathepsin D (lysosomal aspartyl protease) 1.35 0.0006 
741790 Hs.7942 AFTIPHILIN Aftiphilin protein 1.34 0.0004 
490414 Hs.386198 EML4 Echinoderm microtubule-associated protein-like 4 1.34 0.001 
230235 Hs.215766 GTPBP4 GTP-binding protein 4 1.34 0.0001 
853066 Hs.5719 CNAP1 Chromosome condensation-related SMC-associated protein 1 1.33 0.0002 
950092 Hs.405144 SFRS3 Splicing factor, arginine/serine–rich 3 1.32 0.0002 
859857 Hs.442787 ZNF148 Zinc finger protein 148 (pHZ-52) 1.32 0.0007 
447569 Hs.282901 RNPC2 RNA-binding region (RNP1, RRM) containing 2 1.32 0.0001 
344942 Hs.109299 PPFIA3 Protein tyrosine phosphatase, interacting protein (liprin), α3 1.32 0.0009 
767419 Hs.274351 ZDHHC9 Zinc finger, DHHC domain-containing 9 1.32 0.0006 
882434 Hs.362996 KIAA0779 KIAA0779 protein 1.31 0.00005 
283329 Hs.252451 SEMA3A Semaphorin 3A 1.31 0.0007 
359184 Hs.439523 PRKR Protein kinase, IFN-inducible 1.30 0.0006 
180156 Hs.123654 PCF11 Pre-mRNA cleavage complex II protein Pcf11 1.30 0.0002 
180156 Hs.128959  EST 1.30 0.0002 
740707 Hs.269902 KIAA0494 KIAA0494 protein 1.28 0.0006 
884892 Hs.7838 MKRN1 Makorin, ring finger protein, 1 1.27 0.00002 
826135 Hs.367811 STK38 Serine/threonine kinase 38 1.26 0.0007 
838359 Hs.77890 GUCY1B3 Guanylate cyclase 1, soluble, β3 1.26 0.00008 
878174 Hs.388164 FADS2 Fatty acid desaturase 2 1.26 0.0006 
837864 Hs.94262 DD5 Progestin-induced protein 1.26 0.0008 
753914 Hs.512235 ITPR2 Inositol 1,4,5-triphosphate receptor, type 2 1.26 0.0006 
825197 Hs.131168 SEP1 Strand-exchange protein 1 1.25 0.0006 
815772 Hs.369284 C20orf6 Chromosome 20 open reading frame 6 1.24 0.0009 
754654 Hs.118964 p66α P66 α 1.23 0.0002 
384018 Hs.411300 WBP4 WW domain-binding protein 4 (formin-binding protein 21) 1.23 0.00002 
320834 Hs.386404 UBE4B Ubiquitination factor E4B (UFD2 homologue, yeast) 1.22 0.0007 
73596 Hs.35086 USP1 Ubiquitin-specific protease 1 1.19 0.0003 
743727 Hs.389638 FLJ41501 Clone BRTHA2006975 0.86 0.0006 
277134 Hs.93836 CIP98 CASK-interacting protein CIP98 0.85 0.0009 
433289 Hs.117331 TREML1 Triggering receptor expressed on myeloid cells-like 1 0.84 0.0009 
811999 Hs.6455 RUVBL2 RuvB-like 2 (E. coli) 0.83 0.0003 
1292893 Hs.125785 LOC149018 Hypothetical LOC149018, mRNA 0.83 0.0009 
813735 Hs.410314 PCDH16 Protocadherin 16 dachsous-like (Drosophila) 0.81 0.0006 
814271 Hs.18885 CGI-116 CGI-116 protein 0.79 0.0004 
399563 Hs.120332  EST 0.78 0.0004 
814129 Hs.283716 MSCP Mitochondrial solute carrier protein 0.77 0.00002 
490484 Hs.356349 ZNF145 Zinc finger protein 145 (Kruppel-like, expressed in PML) 0.77 0.0007 
47043 Hs.154138 CHI3L2 Chitinase 3–like 2 0.73 0.001 
470035 Hs.14060 PROK1 Prokineticin 1 0.71 0.0007 
2571195 Hs.406683 RPS15 Ribosomal protein S15 0.71 0.0003 
303109 Hs.123464 P2RY5 Purinergic receptor P2Y, G-protein coupled, 5 0.70 0.0005 
431231 Hs.381870 EFEMP2 Epidermal growth factor–containing fibulin-like ECM protein 2 0.70 0.0006 
280950 Hs.422340 SRI Sorcin 0.66 0.0001 
453183 Hs.301302 SCAM-1 Vinexin β (SH3-containing adaptor molecule-1) 0.66 0.0004 
490556 Hs.26518 TM4SF7 Transmembrane 4 superfamily member 7 0.63 0.0001 
502499 Hs.75835 PMM1 Phosphomannomutase 1 0.61 0.0001 
344720 Hs.81994 GYPC Glycophorin C (Gerbich blood group) 0.60 0.0003 
2559389 Hs.150833 C4A Complement component 4A 0.58 0.0008 
2784073 Hs.322431 NEUROD2 Neurogenic differentiation 2 0.54 0.0005 
898305 Hs.439671 NBL1 Neuroblastoma, suppression of tumorigenicity 1 0.54 0.00002 
756372 Hs.37682 RARRES2 Retinoic acid receptor responder (tazarotene-induced) 2 0.52 0.0003 
1474174 Hs.367877 MMP2 Matrix metalloproteinase-2 (72-kDa type IV collagenase) 0.50 0.00004 
462939 Hs.526933  EST 0.48 0.001 
823851 Hs.469463 AEBP1 AE-binding protein 1 0.41 0.0007 
*

IMAGE Integrated Molecular Analysis of Genomes and their Expression Consortium clone (http://madb.nci.nih.gov/CR_query.shtml).

Fold difference in geometric means of chemosensitive tumors (numerator) compared with chemoresistant tumors (denominator).

To determine whether the inclusion of a few tumors of nonserous histologies (i.e., clear cell or endometrioid) or comparison of tumors that did not all receive taxane-containing chemotherapy regimens biased the results, an unsupervised clustering analysis was also done. The resulting dendrogram revealed no segregation based on tumor histology, tumor grade, or type of platinum-based chemotherapy received. These supplementary data (Fig. S1) may be viewed at http://www.healthsystem.virginia.edu/internet/obgyn/documents/Supplemental-figures.pdf.

Semiquantitative immunohistochemical analyses were used to determine if the gene expression differences between primary chemoresistant and chemosensitive ovarian cancers were associated with significant protein expression differences. The protein product of the CTSD gene was selected for this purpose because of the availability of a commercial antibody and prior studies implicating this protease in the pathogenesis of several cancers, including ovarian and breast (16, 17). Using a semiquantitative immunohistochemical scoring system, chemosensitive samples displayed significantly higher CTSD protein expression than the chemoresistant samples (Fig. 1A). Because CTSD expression correlates with high proliferation states in several tumors (1719), the expression of Ki-67 and PCNA was also analyzed in this set of tumors using semiquantitative immunohistochemistry (Fig. 1B and C). Both markers showed significantly higher expression in the chemosensitive primary tumors. These results suggest that the higher expression of CTSD in the chemosensitive tumors correlates with a higher proliferative state that may in turn render them more sensitive to cytotoxic chemotherapy. Notably, CTSD expression was also significantly higher in the primary chemosensitive group compared with the postchemotherapy group (two-tailed t test; P = 0.0008), consistent with the hypothesis that the postchemotherapy tumors are enriched for chemoresistant clones.

Fig. 1.

Semiquantitative immunohistochemical analyses of CTSD, Ki-67, and PCNA expression. Representative photomicrographs of primary chemosensitive (A) and primary chemoresistant (B) ovarian cancers immunostained for CTSD at ×40 magnification. Results of immunohistochemical scoring for CTSD (C), Ki-67 (D), and PCNA (E). Columns, mean score; bars, SD.

Fig. 1.

Semiquantitative immunohistochemical analyses of CTSD, Ki-67, and PCNA expression. Representative photomicrographs of primary chemosensitive (A) and primary chemoresistant (B) ovarian cancers immunostained for CTSD at ×40 magnification. Results of immunohistochemical scoring for CTSD (C), Ki-67 (D), and PCNA (E). Columns, mean score; bars, SD.

Close modal

Gene expression differences between primary and postchemotherapy tumors. The overview of these comparisons is depicted in Fig. 2A. This analysis revealed that 759 genes differentiated the postchemotherapy samples from the primary chemosensitive tumors (P < 0.001). In contrast, only 229 genes were differentially expressed between postchemotherapy and primary chemoresistant groups (P < 0.001), suggesting smaller differences between the molecular profiles of the latter two groups. A comparison of the two gene lists revealed 178 genes that were common to both and thus represented those genes that were differentially expressed between postchemotherapy samples and primary tumors irrespective of intrinsic chemoresistance. The magnitude and direction of change in the expression levels of these 178 genes were very similar in the two comparisons (postchemotherapy versus primary chemoresistant and postchemotherapy versus primary chemosensitive; Fig. 2B). A partial list of these genes and their expression levels in primary tumors and postchemotherapy samples is presented in Fig. 3A. Notably, genes encoding several oxidizing enzymes, including ADH1B, ADH1C, and ALDH2, showed higher expression in the postchemotherapy samples. Also showing higher expression in postchemotherapy samples were DOC1, CAV1, DUSP1, ITM2A, DCN, and KLF4, all of which have been reported to be down-regulated in ovarian cancer when compared with normal ovary (2022). In contrast, TOP1, TOP2A, and ZWINT have been reported to be overexpressed in ovarian cancer (23, 24) but were lower expressed in the postchemotherapy samples compared with the primary tumors.

Fig. 2.

Gene expression differences between postchemotherapy and primary chemosensitive or primary chemoresistant tumors. A, an overview of the number of differentially expressed genes (P < 0.001), with the top circle representing postchemotherapy versus primary chemosensitive samples, the bottom circle representing postchemotherapy versus primary chemoresistant samples, and the overlap region representing genes differentially expressed between postchemotherapy samples and primary tumors irrespective of intrinsic chemosensitivity. B, changes in the magnitude and direction of the 178 genes that discriminated the postchemotherapy samples from all primary tumors (P < 0.001). For each gene, the fold difference between postchemotherapy and primary chemoresistant tumors (X axis) is plotted against the fold difference between postchemotherapy and primary chemosensitive tumors (Y axis). Values less than 1.0 reflect higher expression in the primary tumors. r, Pearson correlation coefficient.

Fig. 2.

Gene expression differences between postchemotherapy and primary chemosensitive or primary chemoresistant tumors. A, an overview of the number of differentially expressed genes (P < 0.001), with the top circle representing postchemotherapy versus primary chemosensitive samples, the bottom circle representing postchemotherapy versus primary chemoresistant samples, and the overlap region representing genes differentially expressed between postchemotherapy samples and primary tumors irrespective of intrinsic chemosensitivity. B, changes in the magnitude and direction of the 178 genes that discriminated the postchemotherapy samples from all primary tumors (P < 0.001). For each gene, the fold difference between postchemotherapy and primary chemoresistant tumors (X axis) is plotted against the fold difference between postchemotherapy and primary chemosensitive tumors (Y axis). Values less than 1.0 reflect higher expression in the primary tumors. r, Pearson correlation coefficient.

Close modal
Fig. 3.

Specific genes with statistically significant differential expression between postchemotherapy and primary tumors (P < 0.001). The corresponding expression values from normal postmenopausal ovaries are shown for comparison. A, genes that were differentially expressed between postchemotherapy tumors and both groups of primary tumors. The top 50 genes are shown; the full list may be found in the Supplementary Data. B, genes that were differentially higher expressed by ≥2-fold (n = 41) in the postchemotherapy compared with primary chemosensitive tumors. Genes encoding ECM-related proteins are shown in bold type. C, genes that were differentially higher expressed by ≥2-fold (n = 10) in primary chemosensitive compared with postchemotherapy tumors. D, genes differentially expressed between the postchemotherapy tumors and the primary chemoresistant tumors.

Fig. 3.

Specific genes with statistically significant differential expression between postchemotherapy and primary tumors (P < 0.001). The corresponding expression values from normal postmenopausal ovaries are shown for comparison. A, genes that were differentially expressed between postchemotherapy tumors and both groups of primary tumors. The top 50 genes are shown; the full list may be found in the Supplementary Data. B, genes that were differentially higher expressed by ≥2-fold (n = 41) in the postchemotherapy compared with primary chemosensitive tumors. Genes encoding ECM-related proteins are shown in bold type. C, genes that were differentially higher expressed by ≥2-fold (n = 10) in primary chemosensitive compared with postchemotherapy tumors. D, genes differentially expressed between the postchemotherapy tumors and the primary chemoresistant tumors.

Close modal

Only 51 genes uniquely differentiated the primary chemoresistant samples from the postchemotherapy group. In contrast, 581 genes uniquely differentiated the primary chemosensitive samples from the postchemotherapy group (Fig. 2A). In addition, the latter comparison resulted in higher magnitude differences in expression as indicated by 51 of the 581 genes showing ≥2-fold difference in their geometric mean expression. The comparison of the primary chemoresistant group with the postchemotherapy samples revealed 13 genes that showed at least a 1.5-fold difference in mean expression levels (Fig. 3D). Furthermore, only one gene, SPP1 (also known as osteopontin), had a >2-fold change in expression and was higher expressed in the primary chemoresistant group. This gene has been implicated previously in ovarian cancer (25) and has been proposed to represent a diagnostic biomarker for ovarian cancer (26). Interestingly, the postchemotherapy samples expressed higher levels of CDKN1C (p57KIP2) and ADAMTS1, both of which have been shown to function as negative regulators of proliferation (27, 28). However, these levels were still far lower than those observed in normal postmenopausal ovarian samples (Fig. 3D).

In the comparison of the postchemotherapy tumors with the primary chemosensitive samples, 41 genes showed at least 2-fold higher expression and 10 genes showed at least a 2-fold lower expression in the postchemotherapy tumors (Fig. 3B and 3C). When the expression levels of these 51 differentially expressed genes in normal postmenopausal ovary were graphed along with the tumor expression levels, an interesting and unexpected pattern emerged. The expression profile of the postchemotherapy samples resembled that of normal ovarian tissue to a much greater degree than that of the primary chemosensitive tumors. These data are consistent with a model in which the postchemotherapy samples show a partial “return to normal” or “low proliferative state” molecular expression profile with respect to this set of genes. One notable exception to the overall similarities between postchemotherapy and normal ovarian gene expression pattern was in the expression of CYR61. This gene has been implicated in angiogenesis (29) and chemoresistance (30) and was expressed at a significantly higher level in postchemotherapy samples compared with both primary chemosensitive and normal ovarian samples (Fig. 3A). In addition, several of the genes higher expressed in the postchemotherapy group were noted to be components of the extracellular matrix (ECM) or involved in its remodeling. This impression was more formally investigated by using EASE software (15) to analyze the biological categories within this gene list. This analysis confirmed the statistically significant (P < 0.05 after Bonferroni correction) overrepresentation of genes involved in ECM among the genes differentiating the postchemotherapy and primary chemosensitive tumors. One hypothesis that may be derived from this observation is that stromal-epithelial interactions or the ECM per se may be involved in acquired chemoresistance in ovarian cancer.

These data suggest that gene expression patterns in primary (pretreatment) ovarian cancers can discriminate intrinsically chemoresistant from chemosensitive tumors. An accurate assessment of predictors of response to chemotherapy is best accomplished through a prospective clinical trial involving adequate numbers of patients. However, conducting such a trial requires the identification of potential molecular predictors of response in ovarian cancer. The data derived from screening >40,000 transcripts provide candidate targets for such an evaluation.

Although highly statistically significant, the magnitude of the mean expression differences between chemosensitive and chemoresistant groups were modest for the discriminating gene set. The observed magnitude of expression differences is likely to be an underestimation for three reasons that are not mutually exclusive. First, according to the Goldie-Coldman hypothesis, chemoresistance is believed to result from a clonal selection process driven by the acquisition of drug resistance mutations (31). Thus, in primary tumors, only a small percentage of the cells are likely to possess a chemoresistant phenotype and the associated molecular changes. This results in a “dilution” of the observed gene expression differences when such tumors are compared with chemosensitive cancers. Second, the comparison of gene expression profiles in primary tumors evaluates only intrinsic chemoresistance. It is likely that a substantial component of clinical chemoresistance is biologically acquired and is therefore only manifested following exposure to chemotherapeutic agents. In support of this hypothesis, the greatest differences in gene expression were observed between postchemotherapy samples and primary chemosensitive tumors. Finally, as is the case with most investigations involving cDNA microarrays, the primary expression data are in the form of logarithmic intensity ratios. Secondary data, such as average expression levels for genes within a group, are derived by calculating the geometric rather than the arithmetic means of logarithmic intensity ratios, resulting in smaller values and smaller apparent differences.

The higher expression of CTSD in chemosensitive tumors, as determined by immunohistochemistry, shows that small differences in geometric mean expression as determined by cDNA microarrays may be associated with substantially greater differences in protein expression. Although one previous report failed to show prognostic value of CTSD expression in ovarian cancer (32), most other studies show low CTSD expression to be an adverse prognostic indicator in ovarian cancer (16, 33, 34). In view of our findings, the prognostic value of CTSD may be related to its higher expression in intrinsically chemosensitive tumors. Consistent with this hypothesis, CTSD has been implicated in p53-depenedent apoptosis following DNA damage induced by drugs and γ-irradiation (35). Further investigations are needed to evaluate a possible causal relationship between these observations and to better define the relationship between CTSD and chemosensitivity. The chemosensitive tumors also showed higher expression of the proliferative markers PCNA and Ki-67. In agreement with this observation, a previous morphologic study found a highly significant correlation between proliferation and mitotic indices and the presence of apoptotic bodies in primary ovarian cancers (36). Thus, higher rate of proliferation may contribute to the chemosensitive nature of these tumors by predisposing them to undergo apoptosis following chemotherapy.

This investigation also revealed substantially different gene expression between primary ovarian cancers and tumor samples obtained following chemotherapy. The postchemotherapy tumors are difficult to classify, based on customary clinical criteria, as either chemoresistant (cancer progression on chemotherapy or recurrence within 6 months of completing chemotherapy) or chemosensitive (disease-free interval of at least 12 months) as they fit neither clinical criterion. They all eventually became resistant to chemotherapy. Our assertion is that these tumor samples represent a state of enrichment in chemoresistant clones, as these are tumors that have survived three to six cycles of chemotherapy. A reasonable hypothesis is that the gene expression profile of these postchemotherapy samples is likely to include molecular changes associated with acquired chemoresistance, as these samples were obtained within a few weeks of completing three to six cycles of chemotherapy. Consistent with this hypothesis, fewer and smaller magnitude gene expression differences were observed between postchemotherapy and primary chemoresistant samples. However, the data also suggest that intrinsic and acquired chemoresistance are likely to manifest through nonoverlapping molecular pathways (Fig. 4). This was evident by the lack of significant overlap (some genes are part of both lists) between the gene list differentiating primary chemosensitive and chemoresistant groups and the list differentiating each group from the postchemotherapy samples. In comparing the primary tumors with postchemotherapy samples, three separate lists of differentially expressed genes were generated. Two lists identified genes that uniquely differentiated primary chemoresistant and chemosensitive tumors from the postchemotherapy samples and one that included genes that discriminated the latter from both former groups. All three lists contain genes that have been implicated previously in tumorigenesis and provide targets for prospective investigations of acquired chemoresistance in ovarian cancers.

Fig. 4.

Diagrammatic depiction of chemoresistance in primary and postchemotherapy samples. Smaller differences in gene expression observed between primary chemosensitive and chemoresistant tumors are likely the result of low relative abundance of intrinsically chemoresistant clones as predicted by the Goldie-Coleman hypothesis. Chemotherapy results in a reduction in the number of chemosensitive and an enrichment of chemoresistant clones in the postchemotherapy samples. In addition, chemotherapy is likely to induce additional genetic changes contributing to acquired chemoresistance. The combination of these effects is likely to be responsible for the robust differences in gene expression observed between primary and postchemotherapy samples.

Fig. 4.

Diagrammatic depiction of chemoresistance in primary and postchemotherapy samples. Smaller differences in gene expression observed between primary chemosensitive and chemoresistant tumors are likely the result of low relative abundance of intrinsically chemoresistant clones as predicted by the Goldie-Coleman hypothesis. Chemotherapy results in a reduction in the number of chemosensitive and an enrichment of chemoresistant clones in the postchemotherapy samples. In addition, chemotherapy is likely to induce additional genetic changes contributing to acquired chemoresistance. The combination of these effects is likely to be responsible for the robust differences in gene expression observed between primary and postchemotherapy samples.

Close modal

Finally, certain patterns of expression deserve special consideration. First, several ECM-related genes revealed differential expression between postchemotherapy and primary tumors, with all being higher expressed the postchemotherapy tumors. This functional category of genes was significantly overrepresented in the comparison between postchemotherapy and primary tumors even after rigorous statistical correction for multiple comparisons. Such a finding is consistent with a recent study of ovarian cancer chemoresistance in vitro, where a group of several ECM components was found to be up-regulated in platinum-resistant cells (37). These genes included DCN and COL6A3, two of the ECM-related genes that were found to be higher expressed in the postchemotherapy group in this study.

Another member of this group found to be overexpressed in postchemotherapy tumors was SPARC (also known as osteonectin), a matricellular glycoprotein involved in angiogenesis, cell adhesion, and ECM turnover (38). This gene is also up-regulated following chemotherapy in breast cancer (39) and has antiproliferative and tumor suppressor function in ovarian (40, 41) and breast (42) cancer cells. Furthermore, SPARC stimulates matrix metalloproteinase-2 expression in other tissues (43, 44) and may account for the observed higher expression of matrix metalloproteinase-2 in the postchemotherapy tumors. Another related ECM gene, SPARCL1 (also known as hevin, SC1, and MAST-9), was significantly higher expressed in the postchemotherapy samples compared with both groups of primary tumors, is down-regulated in several cancers, and has a negative effect on cell proliferation (45).

The higher expression of these and other antiproliferative genes (e.g., KLF4 and CAV1; refs. 21, 46) in the postchemotherapy samples, as well as similarities in the gene expression profiles of the postchemotherapy tumors and normal postmenopausal ovaries, support the concept that a decreased proliferative state may be involved in the development of acquired chemoresistance. Furthermore, given that the primary chemoresistant tumors exhibited significantly lower Ki-67, PCNA, and CTSD protein expression compared with the chemosensitive samples, decreased proliferation may also be a contributing feature to intrinsic chemoresistance. Senescent or slow-growing cells may be more tolerant of cytotoxic chemotherapy, thus allowing more time for selection of advantageous mutations and development of resistant clones. Consistent with such a hypothesis, the majority of postchemotherapy samples exhibit a decreased mitotic index compared with the prechemotherapy sample of the same tumor (47).

This investigation represents an initial effort to discover potentially important molecular mediators of intrinsic and acquired chemoresistance in ovarian cancer in vivo. It is critical to further investigate the molecular basis for chemoresistance prospectively in a large cohort with prechemotherapy and postchemotherapy sampling from the same patients. This investigation provides preliminary data that may inform such future studies and perhaps clinical protocols.

Grant support: NIH grant U01 CA88175 (J. Boyd) and Gynecologic Cancer Foundation, National Cancer Institute Gynecologic Oncology Fellowship Program (A.A. Jazaeri and K.K. Zorn).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).

We thank Drs. Edison T. Liu and William J. Hoskins for their vital roles in obtaining the Director's Challenge grant from the National Cancer Institute that made this work possible.

1
Chi DS, Sabbatini P. Advanced ovarian cancer.
Curr Treat Options Oncol
2000
;
1
:
139
–46.
2
Salom E, Almeida Z, Mirhashemi R. Management of recurrent ovarian cancer: evidence-based decisions.
Curr Opin Oncol
2002
;
14
:
519
–27.
3
McGuire WP, Markman M. Primary ovarian cancer chemotherapy: current standards of care.
Br J Cancer
2003
;
89
:
S3
–8.
4
Markman M, Hoskins W. Responses to salvage chemotherapy in ovarian cancer: a critical need for precise definitions of the treated population [editorial].
J Clin Oncol
1992
;
10
:
513
–4.
5
Alberts DS. Treatment of refractory and recurrent ovarian cancer.
Semin Oncol
1999
;
26
:
8
–14.
6
Spriggs D. Optimal sequencing in the treatment of recurrent ovarian cancer.
Gynecol Oncol
2003
;
90
:
S39
–44.
7
Johnson SW, Ozols RF, Hamilton TC. Mechanisms of drug resistance in ovarian cancer.
Cancer
1993
;
71
:
644
–9.
8
Aebi S, Kurdihaidar B, Gordon R, et al. Loss of DNA mismatch repair in acquired resistance to cisplatin.
Cancer Res
1996
;
56
:
3087
–90.
9
Coukos G, Rubin SC. Chemotherapy resistance in ovarian cancer: new molecular perspectives.
Obstet Gynecol
1998
;
91
:
783
–92.
10
Vasey PA. Resistance to chemotherapy in advanced ovarian cancer: mechanisms and current strategies.
Br J Cancer
2003
;
89
:
S23
–8.
11
Agarwal R, Kaye SB. Ovarian cancer: strategies for overcoming resistance to chemotherapy.
Nat Rev Cancer
2003
;
3
:
502
–16.
12
Skipper HE. Kinetics of mammary tumor cell growth and implications for therapy.
Cancer
1971
;
28
:
1479
–99.
13
Norton L, Simon R, Brereton HD, Bogden AE. Predicting the course of Gompertzian growth.
Nature
1976
;
264
:
542
–5.
14
Van Gelder RN, von Zastrow ME, Yool A, Dement WC, Barchas JD, Eberwine JH. Amplified RNA synthesized from limited quantities of heterogeneous cDNA.
Proc Natl Acad Sci U S A
1990
;
87
:
1663
–7.
15
Hosack DA, Dennis G, Sherman BT, Lane HC, Lempicki RA. Identifying biological themes within lists of genes with EASE.
Genome Biol
2003
;
4
:
R70
.
16
Losch A, Schindl M, Kohlberger P, et al. Cathepsin D in ovarian cancer: prognostic value and correlation with p53 expression and microvessel density.
Gynecol Oncol
2004
;
92
:
545
–52.
17
Ioachim E, Tsanou E, Briasoulis E, et al. Clinicopathological study of the expression of hsp27, pS2, cathepsin D and metallothionein in primary invasive breast cancer.
Breast
2003
;
12
:
111
–9.
18
Glondu M, Liaudet-Coopman E, Derocq D, Platet N, Rochefort H, Garcia M. Down-regulation of cathepsin-D expression by antisense gene transfer inhibits tumor growth and experimental lung metastasis of human breast cancer cells.
Oncogene
2002
;
21
:
5127
–34.
19
Ioachim E, Charchanti A, Stavropoulos N, Athanassiou E, Bafa M, Agnantis NJ. Expression of cathepsin D in urothelial carcinoma of the urinary bladder: an immunohistochemical study including correlations with extracellular matrix components, CD44, p53, Rb, c-erbB-2 and the proliferation indices.
Anticancer Res
2002
;
22
:
3383
–8.
20
Mok SC, Wong KK, Chan RK, et al. Molecular cloning of differentially expressed genes in human epithelial ovarian cancer.
Gynecol Oncol
1994
;
52
:
247
–52.
21
Wiechen K, Diatchenko L, Agoulnik A, et al. Caveolin-1 is down-regulated in human ovarian carcinoma and acts as a candidate tumor suppressor gene.
Am J Pathol
2001
;
159
:
1635
–43.
22
Manzano RG, Montuenga LM, Dayton M, et al. CL100 expression is down-regulated in advanced epithelial ovarian cancer and its re-expression decreases its malignant potential.
Oncogene
2002
;
21
:
4435
–7.
23
van der Zee AG, Hollema H, de Jong S, et al. P-glycoprotein expression and DNA topoisomerase I and II activity in benign tumors of the ovary and in malignant tumors of the ovary, before and after platinum/cyclophosphamide chemotherapy.
Cancer Res
1991
;
51
:
5915
–20.
24
Jazaeri AA, Yee CJ, Sotiriou C, Brantley KR, Boyd J, Liu ET. Gene expression profiles of BRCA1-linked, BRCA2-linked, and sporadic ovarian cancers.
J Natl Cancer Inst
2002
;
94
:
990
–1000.
25
Tiniakos DG, Yu H, Liapis H. Osteopontin expression in ovarian carcinomas and tumors of low malignant potential (LMP).
Hum Pathol
1998
;
29
:
1250
–4.
26
Kim JH, Skates SJ, Uede T, et al. Osteopontin as a potential diagnostic biomarker for ovarian cancer.
J Am Med Assoc
2002
;
287
:
1671
–9.
27
Zhang P, Liegeois NJ, Wong C, et al. Altered cell differentiation and proliferation in mice lacking p57KIP2 indicates a role in Beckwith-Wiedemann syndrome.
Nature
1997
;
387
:
151
–8.
28
Luque A, Carpizo DR, Iruela-Arispe ML. ADAMTS1/METH1 inhibits endothelial cell proliferation by direct binding and sequestration of VEGF165.
J Biol Chem
2003
;
278
:
23656
–65.
29
Babic AM, Kireeva ML, Kolesnikova TV, Lau LF. CYR61, a product of a growth factor-inducible immediate early gene, promotes angiogenesis and tumor growth.
Proc Natl Acad Sci U S A
1998
;
95
:
6355
–60.
30
Wittig R, Nessling M, Will RD, et al. Candidate genes for cross-resistance against DNA-damaging drugs.
Cancer Res
2002
;
62
:
6698
–705.
31
Goldie JH, Coldman AJ. A mathematic model for relating the drug sensitivity of tumors to their spontaneous mutation rate.
Cancer Treat Rep
1979
;
63
:
1727
–33.
32
Ferrandina G, Scambia G, Fagotti A, et al. Immunoradiometric and immunohistochemical analysis of cathepsin D in ovarian cancer: lack of association with clinical outcome.
Br J Cancer
1998
;
78
:
1645
–52.
33
Scambia G, Panici PB, Ferrandina G, et al. Clinical significance of cathepsin D in primary ovarian cancer.
Eur J Cancer
1994
;
30A
:
935
–40.
34
Baekelandt M, Holm R, Trope CG, Nesland JM, Kristensen GB. The significance of metastasis-related factors cathepsin-D and nm23 in advanced ovarian cancer.
Ann Oncol
1999
;
10
:
1335
–41.
35
Wu GS, Saftig P, Peters C, El-Deiry WS. Potential role for cathepsin D in p53-dependent tumor suppression and chemosensitivity.
Oncogene
1998
;
16
:
2177
–83.
36
Brustmann H. Apoptotic bodies as a morphological feature in serous ovarian carcinoma: correlation with nuclear grade, Ki-67 and mitotic indices.
Pathol Res Pract
2002
;
198
:
85
–90.
37
Sherman-Baust CA, Weeraratna AT, Rangel LB, et al. Remodeling of the extracellular matrix through overexpression of collagen VI contributes to cisplatin resistance in ovarian cancer cells.
Cancer Cell
2003
;
3
:
377
–86.
38
Bradshaw AD, Sage EH. SPARC, a matricellular protein that functions in cellular differentiation and tissue response to injury.
J Clin Invest
2001
;
107
:
1047
–54.
39
Korn EL, McShane LM, Troendle JF, Rosenwald A, Simon R. Identifying pre-post chemotherapy differences in gene expression in breast tumours: a statistical method appropriate for this aim.
Br J Cancer
2002
;
83
:
1093
–6.
40
Mok SC, Chan WY, Wong KK, Muto MG, Berkowitz RS. SPARC, an extracellular matrix protein with tumor-suppressing activity in human ovarian epithelial cells.
Oncogene
1996
;
12
:
1895
–901.
41
Yiu GK, Chan WY, Ng SW, et al. SPARC (secreted protein acidic and rich in cysteine) induces apoptosis in ovarian cancer cells.
Am J Pathol
2001
;
159
:
609
–22.
42
Dhanesuan N, Sharp JA, Blick T, Price JT, Thompson EW. Doxycycline-inducible expression of SPARC/osteonectin/BM40 in MDA-MB-231 human breast cancer cells results in growth inhibition.
Breast Cancer Res Treat
2002
;
75
:
73
–85.
43
Gilles C, Bassuk JA, Pulyaeva H, Sage EH, Foidart JM, Thompson EW. SPARC/osteonectin induces matrix metalloproteinase 2 activation in human breast cancer cell lines.
Cancer Res
1998
;
58
:
5529
–36.
44
Fujita T, Shiba H, Sakata M, Uchida Y, Nakamura S, Kurihara H. SPARC stimulates the synthesis of OPG/OCIF, MMP-2 and DNA in human periodontal ligament cells.
J Oral Pathol Med
2002
;
31
:
345
–52.
45
Claeskens A, Ongenae N, Neefs JM, et al. Hevin is down-regulated in many cancers and is a negative regulator of cell growth and proliferation.
Br J Cancer
2000
;
82
:
1123
–30.
46
Chen X, Whitney EM, Gao SY, Yang VW. Transcriptional profiling of Kruppel-like factor 4 reveals a function in cell cycle regulation and epithelial differentiation.
J Mol Biol
2003
;
326
:
665
–77.
47
McCluggage WG, Lyness RW, Atkinson RJ, et al. Morphological effects of chemotherapy on ovarian carcinoma.
J Clin Pathol
2002
;
55
:
27
–31.