Objective
The objective of the study was to evaluate the biological validity of ovarian cancer (OVCA) screening and early detection efforts and to characterize signaling pathways associated with human cancer metastasis and patient survival.
Study Design
Using genome-wide expression profiling and deoxyribonucleic acid sequencing, we compared pelvic and matched extrapelvic implants from 30 patients with advanced-stage OVCA for expression of molecular signaling pathways and p53 gene mutations. Differentially expressed pathways were further evaluated in a series of primary or early-stage vs metastatic or recurrent cancer samples from 389 ovarian, prostate, and oral cancer patients. Metastasis pathways were also evaluated for associations with survival in 9 independent clinicogenomic datasets from 1691 ovarian, breast, colon, brain, and lung cancer and leukemia patients. The inhibitory effects of 1 pathway (transforming growth factor [TGF]-WNT) on in vitro OVCA cell migration were studied.
Results
Pelvic and extrapelvic OVCA implants demonstrated similar patterns of signaling pathway expression and identical p53 mutations. However, we identified 3 molecular pathways/cellular processes that were differentially expressed between pelvic and extrapelvic OVCA samples and between primary/early-stage and metastatic/advanced or recurrent ovarian, oral, and prostate cancers. Furthermore, their expression was associated with overall survival from ovarian cancer ( P = .006), colon cancer (1 pathway at P = .005), and leukemia ( P = .05). Artesunate-induced TGF-WNT pathway inhibition impaired OVCA cell migration.
Conclusion
Advanced-stage OVCA has a unifocal origin in the pelvis. Molecular pathways associated with extrapelvic OVCA spread are also associated with metastasis from other human cancers and with overall patient survival. Such pathways represent appealing therapeutic targets for patients with metastatic disease.
Ovarian cancer (OVCA) has the highest mortality rate of all gynecological cancers, with an estimated 22,280 cases of OVCA diagnosed in 2012 and 15,500 women dying from the disease. Most OVCA patients are diagnosed at an advanced stage with disseminated intraperitoneal metastases, such that the majority succumb to the disease within 5 years. Although significant efforts have focused on strategies to detect OVCA at an earlier, more curable stage, these efforts are based on the assumption that advanced-stage OVCA originates from a single focus in the ovarian epithelium and that extrapelvic, abdominal implants result from disease spread from the primary lesion along peritoneal surfaces, so-called unifocal disease. In contrast, proponents of a multifocal, field-effect phenomenon support a view that advanced-stage epithelial OVCA does not begin as a single focus in the pelvis but rather appears at multiple sites on the peritoneal surface throughout the pelvis and abdomen, somewhat simultaneously. Advocates of a multifocal theory of OVCA continue to question the value of efforts directed toward development of screening technologies that detect localized (early-stage) OVCA before it has spread outside the pelvis.
In this study, we sought to define the focal origin of OVCA. We evaluated the pathoetiologic relationship between primary pelvic and matched extrapelvic implants via genome-wide ribonucleic acid (RNA)-based expression data and p53 deoxyribonucleic acid (DNA) mutational analysis in matched samples from 30 patients with advanced-stage OVCA. Our hypothesis is based on the assumption that, if multiple pelvic and extrapelvic OVCA lesions develop in a simultaneous fashion (a multifocal origin), then the activation of molecular signaling pathways that occurred during carcinogenesis in each individual lesion (and particularly between pelvic and extrapelvic disease) is likely unique. Moreover, a multifocal origin would cast doubt on the likely success of screening efforts focused on disease originating in the pelvis. Such conclusions would have significant public health implications.
In contrast, if OVCA has a unifocal origin, then the pelvic and extrapelvic implants should express similar signaling pathways and have similar p53 mutational patterns. Furthermore, in the face of a unifocal origin, any differences in expression profiles between pelvic and extrapelvic OVCA implants would represent processes associated with metastatic spread that may be also common to other cancer types, may influence clinical outcome, and, importantly, may serve as valid therapeutic targets for patients with metastatic cancer.
Materials and Methods
Pelvic OVCA samples and matched, nonconfluent, extrapelvic implants were obtained from 30 patients who had provided written informed consent to the Moffitt Cancer Center Institutional Total Cancer Care (TCC) protocol ( www.moffitt.org ), prior to undergoing primary cytoreductive surgery for advanced-stage serous epithelial OVCA. The study was carried out with approval from the University of South Florida Institutional Review Board.
A pelvic sample was resected from the ovarian tissue, which, in the opinion of the surgeon, most likely represented the primary site in the pelvis. From each patient, a matched, nonconfluent extrapelvic implant was identified and collected. Samples were flash frozen in liquid nitrogen within 10 minutes of surgical resection and stored at –80°C. A histopathological review was performed to confirm the diagnosis, and samples were macrodissected to ensure greater than 70% tumor content. Total RNA and genomic DNA were extracted from each sample.
Normal ovarian surface epithelium (NOSE) samples were obtained from patients who had provided written informed consent to the TCC protocol and had undergone oophorectomy at the Moffitt Cancer Center for nonmalignant disease, not associated with the ovary. Immediately after surgical resection, the surface epithelium was gently scraped from the surface and immediately subjected to RNA isolation. To ensure sufficient quantities of RNA, NOSE RNA samples were pooled in groups of 3 or 4 to produce a minimum RNA quantity of 50 ng. As a result of such pooling, 49 normal ovaries were analyzed on 12 Affymetrix GeneChip assays (Santa Clara, CA).
Approximately 30 mg of tissue was used for each RNA and DNA extraction. Tissues were pulverized in BioPulverizer H tubes (Bio101) using a Mini-Beadbeater (Biospec Products, Bartlesville, OK). Total RNA was collected using the QIAGEN RNeasy minikit (Valencia, CA) according to the manufacturer’s instructions. RNA quality was checked on an Agilent Bioanalyzer (Palo Alto, CA) to assess the quality of RNA via the 28S:18S ribosomal RNAs. Genomic DNA was isolated using the QIAGEN QIAamp DNA minikit according to the manufacturer’s instructions.
For microarray analysis, 10 μg of total RNA was used to develop the targets for Affymetrix microarray analysis, and probes were prepared according to the manufacturer’s instructions. Briefly, biotin-labeled complementary RNA was produced by in vitro transcription, fragmented, and hybridized to the customized human Affymetrix HuRSTA gene chips (HuRSTA-2a520709). Expression values were calculated using the robust multiarray average algorithm implemented in Bioconductor extensions ( http://www.bioconductor.org ) to the R statistical programming environment.
A Student t test was used to identify differentially expressed genes in comparisons among NOSE, pelvic, and extrapelvic sample genomic data. For each comparison, the 12 NOSE samples were grouped together. Pelvic and extrapelvic genomic profiles were analyzed as groups (pelvic as one group, extrapelvic as another) and as individual pairs (comparisons of matched pelvic/extrapelvic pairs from the same patient). As such, the following comparisons were made: (1) grouped NOSE vs grouped pelvic implants, (2) grouped NOSE vs grouped extrapelvic implants, (3) grouped pelvic vs grouped extrapelvic implant, (4) grouped NOSE vs individual pelvic implants, (5) grouped NOSE vs individual extrapelvic implants, and (6) individual pelvic vs individual matched extrapelvic samples from the same patient. For each of the comparisons, differentially expressed genes were analyzed using GeneGO MetaCore software (St Joseph, MI) to identify represented biological pathways.
Identified pathways were further evaluated for differential representation in 4 publically available gene expression datasets encompassing 389 patient samples including: (1) OVCA (n = 12; 4 early- and 8 advanced-stage), GEO accession number GSE14407 , U133Plus gene chip; (2) oral cancer (n = 27; 22 primary lesions, 5 metastases), GEO accession GSE2280 , U133A gene chip; (3) prostate cancer (n = 271; 196 primary lesions, 75 metastases), GEO accession GSE6919 , U95 gene chip; and (4) prostate cancer (n = 79; 40 nonrecurrent, 39 recurrent lesions), GEO accession GSE25136 , U133A gene chip (by Student t test, gene cutoff P < .01).
Principal component analysis (PCA) was performed using Evince software ( evince.umbio.com ) for Figure 1 . Log-rank tests were used to test associations between pathway expression (using a median PCA score value cutoff) and overall survival within 9 publically available datasets comprising 1691 patient samples, including cancers of the ovary, which included 4 datasets (Australian dataset [n = 218 GSE9891 ], Moffitt Cancer Center (MCC) dataset [n = 142], MD Anderson dataset [n = 53 GSE18520 ], and The Cancer Genome Atlas (TCGA) dataset [n = 497]) as well as brain (n = 182 GSE13041 ), breast (n = 187 GSE2990 ), colon (n = 177 GSE17538 ), lung (n = 58 TCGA), and blood (leukemia, n = 182 TCGA). All survival analyses were performed using the R program.
For sequence analysis of p53, exons 5-8 of p53 from primary lesions and distal metastases separated by noninvolved tissue were analyzed for primary sequence mutation patterns. Genomic DNA (100 ng) was used in PCR amplification reactions essentially as described previously using the following primers: exon 5, sense 5′-TTCCTCTTCCTACAGTACTC-3′, antisense 5′-GCAACCAGCCCTG-TCGTCTC-3′; exon 6, sense 5′-ACCATGAGCGCTGCTCAGAT-3′, antisense 5′-AGTT-GCAAACCAGACGTCAG-3′; exon 7, sense 5′-GTGTTGTCTCCTAGGTTCGC-3′, antisense 5′-CAAGTGGCTCCTGACCTGGA-3′; and exon 8, sense 5′-CCTATCCTGAGTA-GTGGTAA-3′, antisense 5′-TGAATCTGAGGCATAACTGC-3′.
Amplifications were performed using an Eppendorf Mastercycler thermocycler in 50 μL reaction volumes (100 ng genomic DNA, 1 U Taq DNA polymerase [Invitrogen, Carlsbad, CA], 1.5 mM MgCl 2 , 0.2 mM deoxynucleotide triphosphates, and 0.2 μM primer mix) by standard protocols. Briefly, samples were held at 95°C for 10 minutes followed by 30 cycles of the following: 95°C for 50 seconds, annealing temperature at 56°C or 60°C, depending on the primers, for 90 seconds, and an elongation step at 72°C for 90 seconds. After cycling, samples were held at 72°C for 10 minutes and cooled to 4°C. PCR products were purified using the Purelink PCR purification kit (QIAGEN) and evaluated using 4% agarose gels. Sequencing was performed on an Applied Biosystem’s AB3130 genetic analysis system using BigDye 3.1 dye terminator chemistry (Foster City, CA) according to the manufacturer’s instructions. Comparative sequence analysis of p53 exons was performed using DNAStar, Lasergene 8 software (Madison, WI).
The effects of pathway inhibition on OVCA cell metastatic properties were investigated using the in vitro scratch assay. HeyA8 OVCA cells were a gift from Dr Patricia Kruk (Department of Pathology, College of Medicine, University of South Florida, Tampa, FL). Cells were maintained in RPMI 1640 medium (Invitrogen) supplemented with 10% fetal bovine serum (FBS; Fisher Scientific, Pittsburgh, PA), 1% sodium pyruvate, 1% penicillin/streptomycin (Cellgro, Manassas, VA), and 1% nonessential amino acids (HyClone, Hudson, NH). Monolayers, 75-80% confluent, were cultured in serum-free media for 4 hours and then mechanically disrupted to create a wound using a 1 mL pipette tip. Culture plates were washed twice with serum-free media to remove floating cells and then incubated with media containing 10% FBS and either vehicle (dimethylsulfoxide [DMSO]) or drug. The DMSO concentration was maintained below 0.5% so as not to influence cell growth or migration. The underside of the culture plate by the wound area was marked with a Sharpie for reference, and wounds were imaged by phase-contrast microscopy on days 0, 1, and 2.
Results
Comparison of overall expression patterns
PCA modeling was used to assess the overall similarities in gene expression among NOSE, pelvic, and extrapelvic samples. PCA generates a set of vectors (termed first principal component [PC1], second principal component [PC2], etc) that summarize the overall genome-wide expression patterns for a sample. Each principal component provides a summary measure for genes that share certain expression characteristics. Comparing PCA values enables a global assessment of how similar or different samples are at a genome-wide level. The 2 first principal components for all samples are shown in Figure 1 . PC1, which explained 35.4% of the variation, separated most of the NOSE samples from the primary pelvic and the extrapelvic samples.
Comparison of pathway expression in NOSE, pelvic, and extrapelvic OVCAs
We performed grouped comparisons of NOSE, pelvic, and extrapelvic genomic data. At a significance of P < .01 (Bonferroni adjusted), 970 probe sets representing 71 signaling pathways ( P < .05) were identified when the grouped NOSE expression data were compared with the grouped primary pelvic sample data, and 1075 probe sets representing 143 signaling pathways were identified when the grouped NOSE expression data were compared with the grouped extrapelvic implant expression data. Importantly, the 60 of 71 signaling pathways (85%) present in primary pelvic samples were also represented in extrapelvic implants. At this level of significance, no probe sets were found to be differentially expressed between the grouped primary pelvic and extrapelvic samples.
When the grouped NOSE dataset was analyzed against the individual pelvic primary samples (n = 30) and the individual extrapelvic implants (n = 30), an average of 7392 and 7772 probe sets, respectively, demonstrated differential expression (greater than 2-fold). In contrast, an average of 1463 probe sets was differentially expressed between individual pelvic and matched extrapelvic implants from the same patient. Consistently, these data suggest significant similarity between the primary pelvic and matched extrapelvic implants ( Appendix ; Supplementary Table 1 ).
Mutational analysis of p53
Exons 5-8 of the p53 gene were examined in primary pelvic and matched extrapelvic implants ( Supplementary Table 2 ). A total of 13 nucleotide mutations were found in 11 of 30 primary pelvic samples. A mutation in exon 5 was found in 1 primary pelvic, whereas 3 primary pelvic lesions had a mutation in exon 6, 7 pelvic lesions had a mutation in exon 7, and 2 pelvic lesions had a mutation in exon 8. The majority of identified mutations were missense (9 of 13); however, 1 sample showed a frame shift mutation resulting from a deletion in codon 151 of exon 5, 1 sample showed a nonsense mutation in codon 294 of exon 8, and 2 samples displayed silent mutations. In every case, the p53 mutation identified in the primary pelvic was also present in the matched extrapelvic implant.
Pathways associated with metastasis influence clinical outcome
We also sought to identify pathways present in extrapelvic samples that were not present in pelvic samples (termed candidate metastasis pathways [CMPs]). We adopted 2 statistical approaches to this: comparisons of data grouped together and individual patient-matched samples. Five CMPs demonstrated differential expression using both approaches; that is, they were present in extrapelvic samples but not in pelvic samples when data were compared both in grouped analyses (81 total pathways; Supplementary Table 3 ) and in 15 or more of 30 (50%) of the patients for whom individual comparisons were made between matched pelvic and extrapelvic samples (24 pathways total; Supplementary Table 4 ).
These 5 CMPs included the following: (1) chemokines and cell adhesion (chemokines/cell adhesion pathway), (2) transforming growth factor (TGF)-beta and cytoskeletal remodeling (TGF-WNT/cytoskeleton remodeling pathway), (3) histamine signaling in dendritic cells and immune response (histamine signaling/immune response pathway), (4) Toll-like receptor (TLR) signaling pathways and immune response (TLR pathway), and (5) protein folding, membrane trafficking, and signal transduction of G-alpha (i) heterotrimeric G-protein (G-alpha pathway).
To further explore the validity of these 5 CMPs, we evaluated each in 4 publically available external gene expression datasets from primary or early-stage cancers vs metastatic/advanced or recurrent cancer. Pathways associated with metastatic, advanced-stage, or recurrent disease included the following: (1) TGF-WNT/cytoskeleton remodeling pathway ( P < .0001) and chemokines/cell adhesion pathway ( P < .001) for ovarian cancer ( GSE14407 ); (2) TGF-WNT/cytoskeleton remodeling ( P < .001) for oral cavity ( GSE2280 ); and (3) TGF-WNT/cytoskeleton remodeling ( GSE6919 ; P < .001), chemokines/cell adhesion ( GSE6919 ; P < .001), histamine signaling/immune response ( GSE6919 ; P = .016), TGF-WNT/cytoskeleton remodeling ( GSE6919 ; P < .001), and chemokines/cell adhesion ( GSE6919 ; P < .001) for prostate cancer. Based on their representation in the external datasets, TGF-WNT/cytoskeleton remodeling, chemokines/cell adhesion, and histamine signaling/immune response pathways were defined as metastasis pathways from our initial list of 5 CMPs.
To further explore the clinical relevance of the 3 metastasis pathways, we evaluated associations (log-rank P values) between pathway expression (quantified by PCA modeling) and overall survival in 1691 patients from a series of 9 external clinicogenomic datasets. Expression of the TGF-WNT/cytoskeleton remodeling pathway ( Figure 2 , A) was associated with survival from OVCA (n = 218, P = .006, Figure 2 , B), colon cancer (n = 177, P = .004, Figure 2 , C), and leukemia (n = 182, P = .047, Figure 2 , D). The chemokines/cell adhesion pathway ( Figure 3 , A) was associated with survival from colon cancer (n = 177, P = .005, Figure 3 , B), and the histamine signaling/immune response pathway ( Figure 4 , A) was associated with survival from OVCA (n = 142, P < .001, Figure 4 , B) and colon cancer (n = 177, P = .02, Figure 4 , C).
Inhibition of the TGF-WNT/cytoskeleton remodeling pathway prevents cell migration
In light of the TFG-WNT/cytoskeleton remodeling pathway expression associations identified in the previous text and reports of its influence on metastatic activity in other cancer types, we performed functional studies to evaluate the effect of this pathway on OVCA cellular metastatic characteristics, specifically the influence of inhibition of this pathway using artesunate on OVCA cell migratory ability. Inhibition of TGF-WNT signaling using 25 μM or 50 μM artesunate decreased HeyA8 OVCA cell proliferation by approximately 42% and 64%, respectively, and impaired the ability of the cells to migrate into the denuded area (data not shown and Figure 5 ). In contrast, cells cultured in media containing DMSO vehicle completely filled in the gap within 2 days ( Figure 5 ).
Results
Comparison of overall expression patterns
PCA modeling was used to assess the overall similarities in gene expression among NOSE, pelvic, and extrapelvic samples. PCA generates a set of vectors (termed first principal component [PC1], second principal component [PC2], etc) that summarize the overall genome-wide expression patterns for a sample. Each principal component provides a summary measure for genes that share certain expression characteristics. Comparing PCA values enables a global assessment of how similar or different samples are at a genome-wide level. The 2 first principal components for all samples are shown in Figure 1 . PC1, which explained 35.4% of the variation, separated most of the NOSE samples from the primary pelvic and the extrapelvic samples.
Comparison of pathway expression in NOSE, pelvic, and extrapelvic OVCAs
We performed grouped comparisons of NOSE, pelvic, and extrapelvic genomic data. At a significance of P < .01 (Bonferroni adjusted), 970 probe sets representing 71 signaling pathways ( P < .05) were identified when the grouped NOSE expression data were compared with the grouped primary pelvic sample data, and 1075 probe sets representing 143 signaling pathways were identified when the grouped NOSE expression data were compared with the grouped extrapelvic implant expression data. Importantly, the 60 of 71 signaling pathways (85%) present in primary pelvic samples were also represented in extrapelvic implants. At this level of significance, no probe sets were found to be differentially expressed between the grouped primary pelvic and extrapelvic samples.
When the grouped NOSE dataset was analyzed against the individual pelvic primary samples (n = 30) and the individual extrapelvic implants (n = 30), an average of 7392 and 7772 probe sets, respectively, demonstrated differential expression (greater than 2-fold). In contrast, an average of 1463 probe sets was differentially expressed between individual pelvic and matched extrapelvic implants from the same patient. Consistently, these data suggest significant similarity between the primary pelvic and matched extrapelvic implants ( Appendix ; Supplementary Table 1 ).
Mutational analysis of p53
Exons 5-8 of the p53 gene were examined in primary pelvic and matched extrapelvic implants ( Supplementary Table 2 ). A total of 13 nucleotide mutations were found in 11 of 30 primary pelvic samples. A mutation in exon 5 was found in 1 primary pelvic, whereas 3 primary pelvic lesions had a mutation in exon 6, 7 pelvic lesions had a mutation in exon 7, and 2 pelvic lesions had a mutation in exon 8. The majority of identified mutations were missense (9 of 13); however, 1 sample showed a frame shift mutation resulting from a deletion in codon 151 of exon 5, 1 sample showed a nonsense mutation in codon 294 of exon 8, and 2 samples displayed silent mutations. In every case, the p53 mutation identified in the primary pelvic was also present in the matched extrapelvic implant.
Pathways associated with metastasis influence clinical outcome
We also sought to identify pathways present in extrapelvic samples that were not present in pelvic samples (termed candidate metastasis pathways [CMPs]). We adopted 2 statistical approaches to this: comparisons of data grouped together and individual patient-matched samples. Five CMPs demonstrated differential expression using both approaches; that is, they were present in extrapelvic samples but not in pelvic samples when data were compared both in grouped analyses (81 total pathways; Supplementary Table 3 ) and in 15 or more of 30 (50%) of the patients for whom individual comparisons were made between matched pelvic and extrapelvic samples (24 pathways total; Supplementary Table 4 ).
These 5 CMPs included the following: (1) chemokines and cell adhesion (chemokines/cell adhesion pathway), (2) transforming growth factor (TGF)-beta and cytoskeletal remodeling (TGF-WNT/cytoskeleton remodeling pathway), (3) histamine signaling in dendritic cells and immune response (histamine signaling/immune response pathway), (4) Toll-like receptor (TLR) signaling pathways and immune response (TLR pathway), and (5) protein folding, membrane trafficking, and signal transduction of G-alpha (i) heterotrimeric G-protein (G-alpha pathway).
To further explore the validity of these 5 CMPs, we evaluated each in 4 publically available external gene expression datasets from primary or early-stage cancers vs metastatic/advanced or recurrent cancer. Pathways associated with metastatic, advanced-stage, or recurrent disease included the following: (1) TGF-WNT/cytoskeleton remodeling pathway ( P < .0001) and chemokines/cell adhesion pathway ( P < .001) for ovarian cancer ( GSE14407 ); (2) TGF-WNT/cytoskeleton remodeling ( P < .001) for oral cavity ( GSE2280 ); and (3) TGF-WNT/cytoskeleton remodeling ( GSE6919 ; P < .001), chemokines/cell adhesion ( GSE6919 ; P < .001), histamine signaling/immune response ( GSE6919 ; P = .016), TGF-WNT/cytoskeleton remodeling ( GSE6919 ; P < .001), and chemokines/cell adhesion ( GSE6919 ; P < .001) for prostate cancer. Based on their representation in the external datasets, TGF-WNT/cytoskeleton remodeling, chemokines/cell adhesion, and histamine signaling/immune response pathways were defined as metastasis pathways from our initial list of 5 CMPs.
To further explore the clinical relevance of the 3 metastasis pathways, we evaluated associations (log-rank P values) between pathway expression (quantified by PCA modeling) and overall survival in 1691 patients from a series of 9 external clinicogenomic datasets. Expression of the TGF-WNT/cytoskeleton remodeling pathway ( Figure 2 , A) was associated with survival from OVCA (n = 218, P = .006, Figure 2 , B), colon cancer (n = 177, P = .004, Figure 2 , C), and leukemia (n = 182, P = .047, Figure 2 , D). The chemokines/cell adhesion pathway ( Figure 3 , A) was associated with survival from colon cancer (n = 177, P = .005, Figure 3 , B), and the histamine signaling/immune response pathway ( Figure 4 , A) was associated with survival from OVCA (n = 142, P < .001, Figure 4 , B) and colon cancer (n = 177, P = .02, Figure 4 , C).
