Distinct molecular signature of recurrent ovarian tumor cells isolated from the ascites of advanced-stage serous ovarian cancer patients

Seventypercent of ovarian cancer patientsdiedue toconsecutive episodesof recurrences resulting from the re-growth of ovarian tumor cells resistant to conventional chemotherapies. In an effort to identify chemoresistance mechanisms, we compared the expression of genes in tumor cells isolated from the ascites of advanced-stage serous ovarian cancer patients prior to (chemonaive,CN) andafter chemotherapy treatments (chemoresistant/recurrent, CR). A novel, recently publishedmethodwas used for the isolation of tumor cells from the ascites of CN and CR patients. Illumina HT-12 platform was used to assess the differential expressionof genes (DEGs) between the isolated tumor cells from the ascites ofCNandCRpatients. The identification of DEGswas achieved by comparing the genetic signatures of CN versusCR samples by amean expression ratio (fold change) of 2 and P < 0.05. Validation of selected genes was performed by quantitative Real Time Polymerase Chain Reaction (qRT-PCR). The dominant canonical pathways in the CR versus CN tumor cells were determined by Ingenuity Pathway Analysis. Gene expression analysis revealed differential expression of 414 genes, with 179 genes up regulated and 235 down regulated in theCR group. There were significant differences in gene expressions encoding for proteins involved with cancer stem cells, cell-cell adhesion, embryonic development, tumor suppression, immune surveillance, retinoic acid and energymetabolism in tumor cells isolated from CR compared to CN patients. Pathway analysis revealed that changes in cell cycle pathways, prominently those involved with mitosis and polo-like kinase (PLK1), G2/M DNA damage and proteins linked with cell cycle checkpoint regulation associated with chemoresistance. This preliminary molecular profiling, on a small number of patient samples, suggests an important discrimination of genes in the isolated tumor cells derived from the ascites of CN andCRpatients. This type of study on a larger cohort of samples may have important clinical implications for the development of therapeutic strategies to overcome chemoresistance and associated recurrences in ovarian cancer patients.


INTRODUCTION
Ovarian cancer represents 3% of all the new cancer cases in American women, but accounts for 5% of all the cancerrelated deaths [1].This discrepancy occurs due to the resistance of ovarian cancer patients to current chemotherapy regimens resulting in the deaths of 70% of the patients within the first five years of diagnosis [2].The vast majority of ovarian cancer patients diagnosed with an advanced-stage disease undergo debulking surgery followed by adjuvant chemotherapy consisting of a platinum agent (typically carboplatin) alone or in combination with a taxane (paclitaxel) [2].Initially, seventy percent of the women respond to this therapy, but unfortunately the majority of these patients eventually relapse due to drug-resistant recurrent disease and die due to peritoneal metastasis [3].Disappointingly, the five year survival period of these patients has remained unchanged and as low as $30% for the last thirty years [4].
Metastasis in ovarian cancer is unique as it is usually localized within the peritoneal cavity and derives directly from the ovaries and/or the fallopian tubes to the adjacent organs (extraovarian pelvic organs, colon, bladder and liver) and/or by the attachment of exfoliated cancer cells which survive as cellular aggregates and are carried by the peritoneal tumor fluid (ascites) to surrounding organs in the peritoneal cavity [5][6][7][8].The presence of ascites is associated with a poor prognosis [4,9].Microscopic inspection of ascites display a complex heterogeneous picture of cellular environment constituting single cells, floating multicellular aggregates of non-adherent cells, cancer-associated fibroblasts, myeloid cells, activated mesothelial cells and cancer stem cells (CSCs) [10][11][12].Extensive seeding of these floating cells on the uterus, sigmoid colon and omentum is frequently encountered in advanced-stage and recurrent patients and ultimately leads to disruption of major organs and eventually death [7].During the course of recurrence, cells within the small microscopic residual cellular aggregates release soluble pro-angiogenic mediators, which diffuse out from the tumor population and bind to endothelial cells of mature blood vessels leading to angiogenesis, which sets the tumor expansion and recurrence in motion [13].Hence, the presence of floating cellular aggregates, commonly known as spheroids, in the ascites of ovarian cancer patients is strongly associated with recurrence, and there is an urgent need to study these spheroids in the ascites in order to establish the mechanisms of recurrence.
Mechanisms underlying the development of resistance to platinum-based agents have been well characterized in other cancers and include DNA repair mechanisms, altered cellular transport of the drugs, increased antioxidant production, and reduction of apoptosis [14][15][16].Elevated gene expression affecting cellular transport, DNA repair, apoptosis, cell-extracellular matrix and cell-cell adhesion has been observed in ovarian cancer patient's samples resistant to platinum-based therapy [17][18][19].Taxanes were originally used as an alternative to platinum based agents in order to overcome platinum resistance in patients [20].The development of taxane resistance has also been well studied and characterized [21,22].Typical mechanisms of paclitaxel resistance involve alteration in drug transport, altered expression of microtubule proteins, expression of taxane metabolizing proteins and altered cell signaling resulting in reduced apoptosis [22][23][24][25].The roles of some of these factors in cancer patients in response to taxane treatment (e.g.altered expression of class III b-tubulin, reduced apoptosis conferred by survivin expression and metabolism of taxanes by cytochrome P450 reductase) have also been implicated in clinical samples [23][24][25].
It is still not known if the mechanisms of resistance to 'combination chemotherapy' are a combined response of tumor cells to single agents or if a novel mechanism of resistance exists that is different from the resistance mechanisms observed with single agents.In recent studies, we have demonstrated ovarian CSCs to be involved with resistance to both platinum and taxanebased chemotherapies [26,27].We and others have also shown recurrent ovarian tumors to be enriched with CSCs and mediators of pathways that regulate CSCs, suggesting that CSCs may contribute to the development of recurrence [10,11,28,29].
To date most of the research conducted to understand the chemoresistance mechanisms have used cancer cell lines and few data are available on the relevance of these studies on the potential mechanisms of chemoresistance in clinical samples [30,31].In recent studies, gene expression profiles of chemonaive (primary tumors obtained during debulking surgery) and post chemotherapy tumors have been analyzed to determine the molecular signature/s associated with chemoresistance [17,32].These studies, although important, used tumor sections, which are likely to present a complex molecular profile of not only chemoresistant tumor cells but also associated stroma and infiltrated cells.As a consequence, these mixed genetic profiles may not truly represent the associated pathways regulating chemoresistance in tumor cells; hence, such results may misrepresent the targets proposed for future therapeutic interventions.Isolated tumor cells that survive chemotherapy treatments in patients are likely to experience changes in gene expression allowing them to withstand the selective pressure of the drugs.These phenotypically changed tumor cells are likely to exhibit a molecular signature associated with chemoresistance when compared to the gene expression profile of isolated tumor cells before chemotherapy treatment.
In the present study, we have used our recently described novel separation technique to isolate tumor cells from the ascites of advanced-stage CN and CR serous ovarian cancer patients [10].Ascites samples were collected from patients (not matched) at the time of surgery prior to chemotherapy treatment (CN) and at different time points during recurrence (CR).In this preliminary study, changes in gene expression associated with chemoresistance and recurrence were analyzed on a small set of unmatched CN (n ¼ 4) and CR (n ¼ 4) samples by a microarray gene expression profiling method.This study aimed to (i) identify the molecular signature associated with the isolated ascites-derived tumor cells of CR patients; (ii) provide novel information about specific genes that regulate the chemoresistant/recurrent phenotype of ascites-derived CR tumor cells, and (iii) provide an insight into cellular pathways that regulate chemoresistance in ascites-derived CR ovarian tumor cells.Our data identified novel genes and associated pathways which may have clinical relevance in designing therapeutic interventions for ovarian cancer patients.To our knowledge this is the first study, which has demonstrated a distinct molecular profile of isolated tumor cells from the ascites of CR patients.

Patient Recruitment
Ascites were collected from patients diagnosed with advanced-stage serous ovarian adenocarcinoma after obtaining written informed consent under protocols approved by the Research and Human Ethics Committee (HEC # 09/09) of The Royal Women's Hospital, Melbourne, Australia.The histopathological diagnosis, tumor grades and stages were determined by independent staff pathologists as part of the clinical diagnosis.Ascites (As) samples (500 ml-2L) were obtained during surgery from patients with primary carcinoma (n ¼ 7), and at the time of recurrence (n ¼ 6) (Table 1).Apart from As22 which was collected twice from the same patient (As22C and D) during sequential ascites removal within a month, other samples were from individual cases.CN patients (n ¼ 7) did not receive any chemotherapy.All CR patients (n ¼ 6) were diagnosed with a recurrent disease within 1-8 months after completion of their first line of chemotherapy treatment (Table 1).These patients had partial response to the first and subsequent lines of chemotherapy.The che-motherapy agents administered to patients, and the number of chemotherapy cycles varied from patient to patient and are indicated in Table 1.In CR group, ascites was collected from patients at recurrence after the patients have received the cycles of chemotherapy described in Table 1.

Isolation of tumor cells from the ascites of ovarian cancer patients
Tumor cells from ascites were separated using the method described previously [10].Briefly, cells were collected from ascites by centrifugation and cleared of red blood cells by hypotonic shock.The remaining cells were then cultured on 6-well low attachment plates for 3-4 days and both adherent (non-tumorigenic, stromal cells) and nonadherent (tumorigenic) cells were screened for fibroblast surface protein (FSP), cancer antigen 125 (CA125) and cytokeratin 7 (CK7) by flow cytometry to assess the purity of each fraction [10].The non-adherent epithelial tumorigenic population rich in CK7 and CA125 and lacking FSP and vimentin was processed further for microarray analysis.

Flow cytometry analysis
The flow cytometry method has been described previously [33].All data were analysed using Cell Quest software (Becton-Dickinson, Bedford, MA, USA).Results are expressed as mean intensity of fluorescence (MIF).

Immunofluorescence analysis
Immunofluorescence analysis was performed as described previously [33].Images were captured using the Leica TCS SP2 laser, and viewed on a HP workstation using the Leica microsystems TCS SP2 software.

RNA extraction and microarray analysis
Isolated tumor cells were homogenised in TRIzol.RNA extracts >3 mg from clinical samples (Table 1) were outsourced to Australian Genome Research Facility (AGRF) Melbourne, Australia for microarray processing.The Illumina (Sentrix Human HT12v4) platform with 47,232 probes was used as described before [34,35].Briefly, the Agilent BioAnalyser 2100 was used to determine the quality and integrity of the RNA using the NanoChip method (Agilent Technologies, USA).A total of 500 ng of RNA was labeled using the Total Prep RNA amplification kit (Ambion, USA).1.5 ug of labeled cRNA was prepared for hybridisation to the Sentrix Human-HT12 Beadchip by preparing a probe cocktail that included GEX-HYB Hybridisation Buffer (Illumina).A total hybridisation volume of 30 ml was loaded into a single array on the Sentrix Human-HT12 Beadchip.The chip was hybridised at 58C for 16 h in an oven with a rocking platform and washed using the appropriate protocols as outlined in the Illumina manual (http://support.illumina.com/documents/MyIllumina/3466bf71-78bd-4842-8bfc-393a45d11874/WGGEX_Direct_Hybridization_Assay_Guide_11322355_A.pdf), and then coupled with Cy3 and scanned in the Illumina iScan scanner.The scanner operating software, GenomeStudio was used to process measured signal intensities into a text file for analysis.

Analysis of the microarray data
This was performed by AGRF.Raw signal intensity data from Illumina HT-12 slides (www.illumina.com)were background corrected within GenomeStudio.Mean signal intensities were calculated per sample.Individual signal intensities were floored to a value 10% of the mean signal intensity of the array, i.e. to a value of 40.This step was performed to eliminate negative expression values where original, non-background subtracted intensities were within the background range.Data were log2 transformed, and quantile normalized in Partek Genomics Suite 6.5 (www.Partek.com).Only probes with a coefficient of variation of more than 5% were further considered (n ¼ 17,163).A t-test assuming unequal variance was performed, with a bootstrap multiple testing correction (200 randomizations).Gene ontology enrichment analysis was performed in Partek (Genomics Suite 6.5), using either the default gene ontology categories or the KEGG GO database (www.genome.jp/kegg/).Enrichment Fisher exact p-value was calculated based on the number of genes in the provided gene list in relation to the number of genes in gene groups in the genome annotation file.The enrichment score was calculated as negative antilog of that p-value.The larger the enrichment score, the higher the enrichment of that functional group in the provided gene list.
The identification of DEGs was achieved by comparing the genetic signatures of CN versus CR samples and was defined by a mean expression ratio (fold change or FC) of 2 and a P-value cut-off of 0.05.The gene expression data was analyzed using the established two-dimensional hierarchical clusters of genes in the form of a color-coded Heat Map.A Principal Component Analysis (PCA) generated by Partek Genomics Suite 6.6 was used to translate the data into a three-dimensional image whereby the dimensionality of the data set was reduced to a sphere-like representation while maximizing the discrimination between the groups [36].

Ingenuity pathway analysis
The molecular interactions of the canonical pathways between the tumor cells of CN and CR ovarian cancer patients was established by correlating the results obtained from the gene expression data with genes whose biological functions are known in the literature using Ingenuity Pathway Analysis (http://www.ingenuity.com).The CRassociated DEGs list containing the 414 unique genes was filtered using the criteria of FC > 2 and P < 0.05.Ingenuity recognizes the Illumina identifiers and generated common pathways or molecular connections between the CR and CN isolated ascites tumor cells.Representations of the molecular relationship between CN and CR ascites tumor cells were generated based on FC > 2 and P < 0.05 using Benjamini-Hochberg multiple resting correction values.

Quantitative Real Time Polymerase Chain Reaction (qRT-PCR) Analysis
Validation of selected genes identified by microarray analysis was performed by qRT-PCR.RNA was extracted from ascites-derived tumor cells using TriZol from the six different CN and CR patients described in Table 1.Extracted RNA was quantified using the NanoDrop-2000 Spectrophotometer (NanoDrop Technologies INC).RNA quality and integrity were verified by agarose gel electrophoresis.Complementary DNA (cDNA) was synthesized from 500 ng of RNA using the High Capacity cDNA Transcription Kit (Applied Biosystmes, CA, USA) as per manufacturer's instructions.PCR reactions were performed in triplicate with negative controls where water was used in place of reverse-transcribed template included for each primer pair to exclude PCR amplification of contaminating DNA.The primers used for qRT-PCR are summarized in Table 2.A PCR product was amplified for each set of primers, purified using the QIAquick Ò Gel Extraction Kit (Qiagen), quantified using the NanoDrop-2000 Spectrophometer and verified by size by agarose gel electrophoresis.The PCR product was then diluted from 500 fg to 0.5 fg and was used as a standard for quantitative analysis.Absolute quantification of PCR products was performed as described previously [37].

Morphology of cells collected from the ascites of ovarian cancer patients
Ascites cells derived from both CN (n ¼ 4) and CR patients (n ¼ 3) were assessed by phase contrast microscopy after seeding on low attachment plates for 24 h.Two distinct populations of cells were observed: (i) multicellular aggregates (spheroids) that floated as three-dimensional structures in the growth medium without attachment (Figure 1A), and (ii) spindle shaped fibroblast-like single cells that adhered to the low attachment plates (Figure 1C).
Morphological assessment of the spheroids revealed a three dimensional cluster of cells loosely compacted together and surrounded by layers of cells (Figure 1A).In general spheroids were in the form of loose aggregates of small clusters of cells with a defined outer rim (Figure 1A).After 24 h on tissue culture plastic, most spheroids attached to the plates and there was a transformation from a three dimensional structure to flattened cellular clusters containing several layers of cells adherent on top of each other (Figure 1B).The periphery of the spheroids showed thin elongated cells moving out of the spheroid whereas cells towards the centre were more rounded in structure.As the cells moved away from the centre, cell-cell contact was reduced resulting in the slow disaggregation of the spheroid (Figure 1B).On the other hand, single cells attached to the plastic as elongated spindle-like cells having a fibroblast-like morphology (Figure 1D).

Assessment of cell surface markers by flow cytometry
Single cells and spheroids (dispersed by trypsinization) were characterized by the cell surface expression of FSP, CA125, and CK7 by flow cytometry.High expression of CA125 and CK7 was observed in the cells dispersed from spheroids, while no expression of FSP was evident (Figure 1E).On the other hand, the single cells were positive for FSP and no expression of CA125 and CK7 was evident (Figure 1F).

Analysis by immunofluorescence
We next analyzed the expression and localization of vimentin in ascites samples by immunofluorescence.Consistent with the flow cytometry results, no expression of vimentin was observed in the cells within the spheroids (Figures 2B-C).Vimentin expression was only evident in cells protruding out of spheroids (Figures 2B-C).On the other hand, single cells demonstrated strong expression of vimentin (Figure 2E-F).Vimentin expression was evident in both cytoplasm and membrane of single cells (Figures 2E-F).

Heat Map and Principal Component analysis
Eight ascites-derived spheroid tumor samples were used to determine the gene expression changes in CN and CR tumor cells derived from the ascites of ovarian cancer patients.A pool of 47,232 probes was used to assess the expression of 18,009 genes.To analyse the variation in gene expression between the CN and CR groups, two dimensional colour-coded Heat Maps were generated using the filtering criteria of FC > 2 and P < 0.05 (Figure 3).The genes represented in yellow are up regulated, while those represented in red are down regulated in the respective CN or CR groups.The Heat Map demonstrated a distinct separation of genes in CN and CR tumor groups.Genes which were up regulated in the CR group were down regulated in the CN group and vice versa, suggesting that these two groups are distinctly different from each other (Figure 3A).This was supported by the PCA mapping of 83.4% obtained between the CN and CR spheroid tumor samples (Figure 3B).These observations were consistent with PC1 values of 70.1% along the x-axis which indicates significant genetic variation between the CN and CR groups, while the small PC2 variation of 6.66% indicates small variation of samples within the same groups.Changes in gene expression associated with chemoresistance and recurrence in isolated ascites tumor cells A total of 414 genes were identified to be differentially expressed between the CN and CR groups.Of these 190 probes representing 179 genes were up regulated and 243 probes representing 235 genes were down regulated in the CR spheroid tumor samples.The 20 most up regulated or down regulated genes are presented in Tables 3 and 4, respectively.Of the 20 up regulated genes, 4 genes were unknown (Table 3).Three genes, LAMA3, CPN8 and PAX8, were identified twice with different probes.Genes involved with cell-cell adhesion such as CLDN16, CLDN1, LAMA3 [38][39][40] and inhibition of metastasis and angiogenesis, ADAMTS9 [41] topped the list (Table 3).This was followed by genes involved in embryonic development such as UCA1 [42], TGFbR3 [43] and PAX8 [44].
Of the twenty most down regulated proteins, Serpin A3 involved with inflammatory reactions topped the list [49] (Table 4).The majority of the proteins in the twenty most down regulated list were associated with tumor infiltration and endothelial cells as well as host immunity.These included HCLS1 [50], SOX 18 [51], HOXB5 [52], HLA-DRB4 [53], LY9 [54], IF116 [55], CD99 [56], and BCL11A [57].Tumor suppressor genes such as FBP1 [58], RASSF2 [59] were also included in the top twenty list of down regulated genes.In addition, the cell cycle regulator CDKN3 [60], inhibitor of cellular migration FGD3 [61], and regulator of cytokinesis KIF20A [62] were also in the top 20 down regulated genes.However, HOXB2, which has previously been shown as a regulator in breast tumorigenesis was lost in CR ascites tumor cells [63].In addition GLIPR2, a regulator of fibrosis and EMT [64], RGS2, a regulator of G protein signalling 2 involved with cellular stress [65] PSRC1, proline/serine-rich coiled-coil 1 which functions as a microtubule destabilizing protein and controls mitotic progression [66] and PMP22, peripheral myelin protein 22 shown to be involved with modulation of alpha6 integrin in human endometrium [67] was down regulated in CR tumors.The differential distribution of genes in the isolated CR compared to CN tumor cells obtained from the ascites of ovarian cancer patients is demonstrated in Figure 4.The microarray expression data has been uploaded as Supplementary Table 1.Involved in the transport of monocarboxylate across the plasma membrane.It is postulated that it has function in the disposition of drugs [48] Gene ontology enrichment analysis identified three major functional types of DEGs in the CR group.These functions included cellular components (e.g.chromosomal regulation, mitosis, cytoskeletal organization, etc), biological processes (e.g.regulation of nuclear division, organelle localization, mitoses, cytokinesis, cell cycle, cellular component assembly, etc) and molecular functions (e.g.microtubule motor activity, gene transcriptional repressor activity, etc).This data has been provided in Supplementary Table 1.

Validation of candidate genes by qRT-PCR
We next selected 6 genes from the most up regulated and 3 genes from the down regulated candidates to validate using qRT-PCR.Most of the selected genes had previously been associated with ovarian cancer progression, except ADAMTS9 which had not been linked to ovarian cancer.Twelve samples (CN ¼ 6 and CR ¼ 6) (Table 1) were used to validate the microarray gene expression changes in CN and CR tumor cells isolated from the ascites of ovarian cancer patients (Figure 5).
There was a significant difference in 5 out of the 6 candidate up regulated genes selected from the microarray.These genes were involved with cell-cell adhesion such as CLDN16 (P < 0.05), CLDN1 (P < 0.01), the tumor suppressor gene ADAMTS9 (P < 0.05) involved with inhibition of metastasis and angiogenesis and genes involved in embryonic development such as TGFbR3  The absolute gene expression value of six genes up regulated and (B) three down regulated in CR compared to CN samples.RNA from CN (n ¼ 6) and CR (n ¼ 6) was extracted, cDNA was prepared and q-PCR for CLDN-1, CLDN-16, ADAMTS9, PROM1, PAX8, TGFBR3, Serpin A3, BCL11A and CRABP2 was performed and the resultant mRNA levels were quantified from the standards prepared as described in the Material and Methods section.The experiments were performed on six independent CN and CR samples in triplicate.Significant intergroup variations are indicated by Ã P < 0.05, ÃÃ P < 0.01.
(P < 0.05) and PAX8 (P < 0.01) (Figure 5A).However, PROM1, also known as the common ovarian cancer stem cell marker CD133, was not significantly enhanced in the CR group compared to the CN group in a set of 6 samples analysed although the trend of enhanced expression was evident.Although all three down regulated genes selected for validation, SerpinA3, BCL11A and CRABP2, did not show significant differences between the CN and CR groups, the decreasing trend in CR compared to CN group was evident (Figure 5B).

Changes in gene expression in As22D compared to As22C
Ascites 22C was obtained from a chemoresistant and recurring patient who had received sequential doses of chemotherapy listed in Table 1 (carboplatin and paclitaxel) (3 cycles), doxorubicin (4 cycles), AMG386 topotecan (2 cycles), cyclophosphamide (2 cycles).Ascites 22D was drained from the same patient within a month after the patient had received an additional cycle of cyclophosamide (number of chemotherapy cycles received ¼ carboplatin and paclitaxel (3 cycles), doxorubicin (4 cycles), AMG386 topotecan (2 cycles), cyclophosphamide (3 cycles).
Even though both samples overlapped each other in PCA plot (Figure 3B), some genetic differences was evident in the Heat Map (Figure 3A).On further investigation, 16 differentially expressed genes (8 up and down regulated in both cases) with a fold-change of >2-fold was observed in As22D compared to As22C.The genes are listed in Tables 5  and 6.

Ingenuity pathway analysis
To elucidate the underlying biological significance of DEGs identified by the microarray analysis between the CN and CR groups, Ingenuity Pathway Analysis was performed on the gene set identified using FC > 2 and P < 0.05 criteria.This was performed to identify the canonical pathways that uniquely regulate the chemoresistant phenotype of ascites tumor cells.With a data set of 414 DEGs (179 up regulated and 235 down regulated), 172 canonical pathways were identified.The gene content of the CR-associated 414 DEGs list corresponding to 433 probes was used for the generation of the biological networks using the Analyze Networks algorithm.The genes listed within the 414 DEG list were connected based upon their known relationship or functions associated with CR-dependent genes in the Ingenuity Pathways Knowledge Base.These networks were unique for CR-associated DEGs and are described in Supplementary Table 2.The majority of the pathways identified by Ingenuity Pathway Analysis were implicated in cancer, cellular growth and proliferation, cellular development, mitosis/cell cycle regulation, cellular assembly and organization.The top ten pathways are described in Table 7.Of these, the pathway regulated by the PLK1 associated with Cell Cycle and DNA damage topped the list (Table 8).The principal genes involved in these pathways CDC25c, CCNB1 and PLK1 were all down regulated in CR tumors compared to CN tumors (Table 8).Potential involvement of PLK1, CDC25c and CCNB1 in response to chemotherapy (DNA damage response) leading to G2-M cell cycle arrest and initiation of mitosis are depicted in Figures 6A and B.

DISCUSSION
To our knowledge, this is the first study which describes a genome wide microarray transcriptional profiling analysis on a small set of isolated tumor cells obtained from the ascites of advanced-stage CN and CR serous ovarian cancer patients.This was performed to specifically identify CR-associated genes and pathways that could be involved in the recurrence and subsequent progression of serous ovarian cancer.By combining our novel method of isolating ovarian tumor cells from the ascites of ovarian cancer patients with gene microarray analysis we were able to compare the gene expression profiles between CN and CR tumor cells without any back ground noise from the associated stromal cellular component.Using the criterion FC > 2 and P < 0.05, this study demonstrated a unique gene expression profile of isolated tumor cells prior to and after chemotherapy as evidenced  by the large number of DEGs (n ¼ 414), of which 179 were up regulated and 235 were down regulated.The Heat Map and PCA analyses suggested a distinct separation with little overlap among the CR and CN associated genes.Overall, CR tumors displayed a unique up regulation of genes with functional relevance to cell-cell adhesion and tight junctions, embryonic development, cancer stem cells, tumor suppressor and genes involved in calcium and phospholipid signaling.Claudins 1 (CLDN1) and 16 (CLDN16), which topped the up regulated list of genes, showed significant differences by validation at the mRNA level, are cell-cell It is an early trigger for G2/M transition.It phosphorylates and activates CDC25C, a phosphatase that dephosphorylates and activates the cyclinB/CDC2 complex for entry of cells in G2-M phase [111,112] Potential adhesion proteins that form the backbone of the apicolateral tight junctions and are crucial for epidermal barrier function [68].These proteins are also essential for maintaining epithelial cell polarity [69].In ovarian cancer, high expression of CLDNS 3, 4, 5 and 7 have been reported to be involved with tumorigenesis [70][71][72].Over expression of CLDN4 has also been shown to contribute to platinum resistance in ovarian cancer [73].Even though significantly enhanced expression of CLDN1 has been demonstrated in ovarian carcinoma effusions and is associated with poor survival [74], enhanced expression of CLDN1 and CLDN16 in CR compared to CN ascites-derived tumor cells has not been reported before.
Changes in gene expression profiles and corresponding epigenetic changes have been observed in cancer cells and embryonic stem cells [75,76].Hence, it was not surprising to note high expression of several embryonic stem cell genes in CR tumor cells, suggesting that therapy resistance programming of CR tumor cells may rely on the functional attributes of genes required for embryogenesis.Of the genes involved with embryogenesis, ADAMTS9 was most up regulated (FC > 22) [77].ADAMTS9 has been described as a tumor suppressor gene in a number of cancers including oesophageal and nasopharyngeal carcinoma [78].In addition, ADAMTS9 is also involved with the cleavage of proteoglycans such as versican and aggrecan [79].Versican and aggrecan are important stromal components of ovarian carcinomas and have been shown to be involved with metastatic dissemination and angiogenesis of ovarian cancer cells [80].Enhanced ADAMTS9 expression in CR tumors may indicate low versican expression, due to higher proteolytic cleavage, resulting in the suppression of the stromal component and concomitant reduction in ovarian tumor cell dissemination outside the peritoneal environment [81].This is consistent with our previous studies where we have demonstrated a relatively low stromal component in CR compared to CN ascites tumor cells [10].Increased versican in the carcinoma stroma associates with poorer outcomes, possibly due to the facilitated migration of carcinoma cells away from the tumor [80,81], and versican has also been shown to mediate mesenchymal epithelial transition to facilitate metastasis of breast carcinoma cells in the lungs [82][83][84].
We also demonstrate significantly higher expression of other embryonic and tumor suppressor genes such as PAX8 [44], TGFbR3 [85], and UCA1 [42] in CR compared to CN tumor cells.Of these three genes, significantly enhanced expression of PAX8 and TGFbR3 was validated by qRT-PCR in CR ascites-derived tumor cells compared to CN tumor cells.PAX8 is a lineage-restricted transcription factor that plays an essential role in the organogenesis of the Mullerian duct [86].In the reproductive tract, PAX8 expression is restricted to secretory cells of the fallopian tube epithelium [87], which recent reports have suggested as the cell of origin of serous ovarian cancer [88].As such, PAX8 has been shown to be overexpressed in ovarian cancers [87] and amplified in 16% of primary ovarian serous tumors [89].Selective suppression of PAX8 expression has been shown to induce apoptotic cell death in ovarian cancer cell lines, suggesting that PAX8 is required for the proliferation of ovarian cancer cells [89].Contrary to the expression of PAX-8, the expression of TGFbR3 has been shown to be down-regulated in the majority of ovarian carcinomas and this was shown to be progressive with increasing tumor grade [90].TGFbR3 expression had been shown to have a significant inhibitory effect in ovarian cancer invasiveness, migration and the levels of MMP-2 and MMP-9 by promoting tumor suppressor effects of inhibin [90], as well as antagonizing the signals received by TGFb [91,92].We have previously demonstrated that tumor cells from the ascites of CN and CR patients lack MMP-2 and MMP-9 expression [10].This is also consistent with the immotile and non-aggressive nature of recurrent ovarian cancer where tumor growth is localized within the peritoneum microenvironment and is more dependent on dissemination by landing onto peritoneal organs rather than aggressive invasion through the vasculature [7].As the expression of TGFRb3 is maintained by epigenetic transcriptional changes [90] it can be contemplated that CR tumor cells may undergo genetic reprogramming event under repetitive DNA damage repair processes resulting in cells with relatively high expression of TGFRb3 [93,94].However, the level of expression of the soluble form of TGFbR3, generated by ectodomain shedding of the cell surface receptor, yet remains to be investigated in the context of CN and CR tumors.
Ovarian cancer has been classified as a stem cell disease [95][96][97].A recent study has presented a stem-like classification of high-grade serous tumors with poor patients' survival [98].PROM1 (CD133) a transmembrane glycoprotein has been defined as a marker for ovarian cancer stem cells [99], and the expression of PROM1 in ovarian tumors has been associated with poor prognosis [100] and is directly regulated by epigenetic modification [101].In our study, these observations were supported by the significantly enhanced expression of PROM1 in CR (FC > 9.30) compared to CN tumors by gene microarray.However, validation of PROM1 in CR versus CN samples did not gain significance even though an increasing trend in CR samples was evident.This was probably due to the small number of samples tested for this study.We have also previously demonstrated enhanced expression of PROM1 and other CSC markers in ovarian cancer cell lines and tumor cells isolated from the ascites of ovarian cancer patients in response to in vitro chemotherapy treatment [26,27].In addition, we have also demonstrated the emergence of CSC-like phenotype in mouse-xenograft models on intraperitoneal administration of chemotherapy after inoculation of ovarian cancer cells [27,29,97,102].These findings are consistent with the microarray data provided in Supplementary Table 1 which shows significant enhanced expression of PDGFRb and JAG1 among the top 30 genes in CR (FC > 4.80 and 4.30) compared to CN tumor cells.PDGFRb and JAG1 have been associated with epithelial mesenchymal transition and CSCs in pancreatic cancer [103].Thus, further studies on recurrent ovarian tumors in preclinical and clinical settings are needed to understand how CSC markers such as PROM1, PDGFRb and JAG1 signals to support stem cell maintenance and tumor progression in the ascites microenvironment.
A unique down regulation of genes with functional relevance to protease inhibition, immunosuppressive tumor microenvironment, and energy and retinoic acid metabolism, regulator of cell cycle and mediator of antiinflammatory reactions was observed in CR tumors.SerpinA3 (FC > À26) topped the list among the down regulated proteins.SerpinA3, also known as alpha 1-antichymotrypsin is a member of the serpin super family, which inhibits the activity of certain proteases, such as cathepsin G in neutrophils and chymases in mast cells, by cleaving them into a different conformation [104].Down regulation of SerpinA3 expression may protect CR tumor cells from damage caused by neutrophil-and mast cell-associated proteolytic activities.Besides SerpinA3, other down regulated proteins such as HCLS1 [105], SOX18 [106], LY96 [also known as myeloid differentiation factor -2 (MD-2)] [107], HLA-DRB4 [53], and BCL11A [57] have been shown to be involved with lymphoangiogenesis, presentation of class II molecules to antigen presenting dendritic cells, and LPS signalling through Toll-like receptor 4. Down regulation of these molecules in CR tumor cells suggests that genes that regulate the activities of cytotoxic T cells, antigen presenting dendritic cells and differentiation of precursor endothelial cells associated with infiltrating lymphocytes are dominantly suppressed, suggesting that this mechanisms is used by CR tumors to escape host immune surveillance.These observations are consistent with the significant down regulation of CD99, a stromal factor expressed on cancer associated fibroblasts and stromal lymphocytes [108] again supporting reduced stromal/infiltrating T cell component in CR tumors as described before [10].
A particularly interesting observation in the CR gene list was the loss of CRABP2 which previously has been shown to be over expressed in serous ovarian tumors [109].Loss of CRBP2 in CR tumors may indicate concomitant loss of vitamin A metabolism and retinoic acid receptor signalling required for differentiation of ascites tumor cells.Loss of CRBP2 expression has been shown to be associated with decreased disease-free survival rates in Head and Neck Squamous Cell Carcinoma [110].
Interestingly, the two sequential chemotherapy samples obtained from the same CR patient (CR tumors-As22C and As22D) displayed close interaction in the PCA plot suggesting somewhat similar gene expression makeup.However, on further investigation 16 differentially genes were identified in these samples.Among these, genes like APOF [111], KCNH6 [112], KYNU [113], LRAP [114], MAGT1 [115], MAL [116], MGC26356 and MRP63 were up regulated by 2.7 to 5.3-folds in As22D compared to As22C.These up regulated proteins are involved with lipid transfer, voltage gated potassium channels, biosynthesis of NAD cofactor, magnesium transport and T-cell mediated immune functions.The down regulated genes included ADCY3 [117], ASS1 [118], C11orf63, C3ORF34 [119], IL-8 [120], LCN2 [121], LRG1 [122] and MT1A [123] were down regulated by 2.3 to7.5-folds in As22D compared to As22C.Among the down regulated proteins IL-8 has been shown to have tumor promoting effect in ovarian cancer [120,124].It is not clear if the changes in the above genes is due to administration of a single dose of cyclophosmide to the patient or is it due to the de novo changes in the tumor during progression of the disease.
In order to identify 'dominant' pathways that may regulate the chemoresistant and chemonaive phenotype of ascites-derived tumor cells, we used Ingenuity Pathway Analysis.Of these, the mitotic pathways regulated by Polo-like kinase (PLK1), cell cycle: G2/M Damage checkpoint regulation and role of checkpoint proteins in cell cycle checkpoint control topped the list.PLKs are serine/ threonine kinases, which are over expressed in many cancers and serve as biomarkers and a target for cancer therapy [125,126].PLK1 is expressed only in dividing cells from G2 onward and is degraded at the end of mitosis [125].PLK1 in combination with CDC25c and CCNB1 regulates mitotic entry, spindle formation and cytokinesis in human cells [126].PLK1 also regulates G2-DNA damage checkpoint and is required for checkpoint recovery following checkpoint inactivation; that is when the damage is completely repaired and the cells restart the cell cycle [126].PLK1 has been shown as an independent prognostic marker for ovarian cancer patients [127] and PLK2 as an epigenetic determinant of chemosensitivity and clinical outcomes in ovarian cancer [128].
In conclusion, we have demonstrated for the first time a unique and novel pattern of gene expression in isolated CR tumors compared to CN tumors obtained from the ascites of ovarian cancer patients.This preliminary study on a small sample size has utilised a systematic approach of studying only the tumor cells isolated from the ascites of CR and CN patients.We demonstrate an underlying lineage-specific relationship between high-grade serous ovarian carcinomas and ascites-derived tumor cells by the expression of lineage-specific essential gene PAX8.Unlike most other carcinomas, which dedifferentiate during progression and recurrence, recurrent ovarian tumors in the ascites microenvironment exist as epithelial cells in the form of spheroids [6,10].In that setting, up regulation of genes associated with cell-cell adhesion, embryonic stem cells, CSCs and genes antagonistic for migration and invasion (TGFBR3) defines the true characteristics of CR tumors in the ascites microenvironment.Concomitant down regulation of genes involved with host immune surveillance provides an immunosuppressive environment that limits the host immune system to fight the tumor.This preliminary study, builds the framework of future studies, which will focus on particular genes and pathways of interest that may have therapeutic potential in reducing ascites-associated recurrences in ovarian cancer patients.

Figure 1 .
Figure 1.Morphological features and expression of surface markers of ascites-derived spheroids and single cells.(A) Spheroids and (C) single cells were seeded on low attachment plates immediately after collection.Morphological features of (B) spheroids and (D) single cells on tissue culture plastic after 24 h following seeding.Images were assessed by a phase contrast microscope.Magnification was 100x, scale bar ¼ 50 mm.The images are representative of n ¼ 14 samples.(E-F) Flow cytometric assessment of purified cells from the ascites of CN (n ¼ 7) and CR (n ¼ 7) ovarian cancer patients was performed by incubating the cells with either control IgG or relevant primary antibodies against the respective antigens followed by incubation with secondary phycoerythrin conjugated antibody.Results are representative of (n ¼ 14) independent samples.The filled histogram in each figure represents control IgG, black lines indicate protein expression in respective cells.

Figure 3 .
Figure 3. Differential expression of genes (DEGs) in CN and CR samples.(A) The Heat Map analysis of genes was performed at P < 0.05 and Fold Change >2.The gene expression analysis is based on 433 probes representing 414 DEGs.DEGs in the CN and CR samples are displayed in a color-coded matrix which is divided into four large rectangular boxes representing the up regulated (yellow) or down regulated (red) genes.The columns along the x-axis (from right to left), are the clinical samples [CN n ¼ 4, followed by CR (n ¼ 4)].The intensity of colors are indicative of how positively (yellow) or negatively (red) these 414 genes are detected by 433 probes.(B): PCA analysis of genes at P < 0.05 and fold change >2.The PCA which contributes to the overall differences in the CN and CR groups is noted at the top of the graph.Variation between the CN (grey) and CR (red) groups are indicated by spheres.

Figure 2 .
Figure 2. Expression of vimentin in spheroids and single cells.(A-F) Purified spheroids and single cells were evaluated for the expression and localization of vimentin by immunofluorescence using mouse monoclonal antibody (green) as described in the Methods and Materials.Cellular staining was visualized using the secondary Alexa 488 (green) fluorescent labeled antibody, and nuclear staining was detected by DAPI (blue).A and D, cells were only stained with DAPI (blue), B and E, cells were stained with anti-vimentin followed by Alexa 488 (green), C and F overlay of DAPI and vimentin staining.Magnification was 200x; scale bar ¼ 50 mm.

Figure 4 .Figure 5 .
Figure 4. Differential distribution of genes in the isolated CR compared to CN tumor cells obtained from the ascites of ovarian cancer patients.Cellular functions are assigned to genes which were evaluated at P < 0.05 and Fold Change (FC) >2.

Figure 6 .
Figure 6.Potential involvement of PLK1, CDC25c and CCNB1 in response to chemotherapy treatment in ascites-derived tumor cells.(A) In response to chemotherapy treatment ATM, ATR, chk1 and chk2 kinases are activated which causes nuclear exclusion of CDC25c in residual ascites-derived tumor cells.This results in mitotic arrest in preparation for DNA repair.PLK1 in that case is inactivated by active ATM or ATR response.(B) After DNA repair checkpoint kinases are silenced, phosphatases can reverse the inhibition enforced by checkpoint kinases so that PLK1 can be activated with concomitant activation of CCNB1-CDK1 complex.This results in the mitotic entry of cells with repaired DNA.

Table 1 .
Description of serous ovarian cancer patients recruited for this study

Table 2 .
Human oligonucleotide primer sequences for quantitative real-time PCR

Table 5 .
8 up regulated genes in Ass22C compared to AS22D tumor cells (FC > 2) The protein has been localized to the endoplasmic reticulum of T-cells and is a candidate linker protein in T-cell signal transduction.In addition, this proteolipid is localized in compact myelin of cells in the nervous system and has been implicated in myelin biogenesis and/or function.The protein plays a role in the formation, stabilization and maintenance of glycosphingolipid-enriched membrane microdomains.

Table 7 .
Top ten Ingenuity Pathway analysis

Table 8 .
Expression of common genes listed in the top three Ingenuity Pathway Analysis