Involvement of Cancer Stem Cells in Chemoresistant Relapse of Epithelial Ovarian Cancer Identified by Transcriptome Analysis

Epithelial ovarian cancer (EOC) is the most lethal gynecological malignancy. Despite the initial resection and chemotherapeutic treatment, relapse is common, which leads to poor survival rates in patients. A primary cause of recurrence is the persistence of ovarian cancer stem cells (OCSCs) with high tumorigenicity and chemoresistance. To achieve a better therapeutic response in EOC relapse, the mechanisms underlying acquired chemoresistance associated with relapse-initiating OCSCs need to be studied. Transcriptomes of both chemosensitive primary and chemoresistant relapse EOC samples were obtained from ICGC OV-AU dataset for differential expression analysis. The upregulated genes were further studied using KEGG and GO analysis. Significantly increased expression of eighteen CSC-related genes was found in chemoresistant relapse EOC groups. Upregulation of the expression in four hub genes including WNT3A, SMAD3, KLF4, and PAX6 was verified in chemoresistant relapse samples via immunohistochemistry staining, which confirmed the existence and enrichment of OCSCs in chemoresistant relapse EOC. KEGG and GO enrichment analysis in microarray expression datasets of isolated OCSCs indicated that quiescent state, increased ability of drug efflux, and enhanced response to DNA damage may have caused the chemoresistance in relapse EOC patients. These findings demonstrated a correlation between OCSCs and acquired chemoresistance and illustrated potential underlying mechanisms of OCSC-initiated relapse in EOC patients. Meanwhile, the differentially expressed genes in OCSCs may serve as novel preventive or therapeutic targets against EOC recurrence in the future.


Introduction
Epithelial ovarian cancer (EOC) is the most lethal gynecological cancer with a 5-year relative survival rate of 29% in patients at advanced stage [1]. The poor prognosis of EOC is largely attributed to relapse and chemoresistance [2,3]. More than 70% EOC patients suffered from tumor recurrence after standard treatments including optimal surgery and platinum/paclitaxel chemotherapy [4]. The majority of these patients had to receive frequent chemotherapy or radiotherapy with gradually shortened platinum-free intervals (PFI), which finally led to chemoresistance [5]. Poly ADP-ribose polymerase (PARP) inhibitors such as olaparib have been reported to increase the progress-free survival (PFS) in EOC, prolong PFI, and delay relapse [6]. However, only patients with Breast Cancer susceptibility gene 1/2 (BRCA1/2) mutations or homologous recombination deficiency (HRD), which account for around 34% and 50% of the total patients, respectively, are recommended to use these agents [3,7,8]. No therapeutic method was proved effective in treating recurrent chemoresistant EOCs. The underlying mechanisms of relapse and chemoresistance remain unclear.
Cancer stem cells (CSCs) are a subpopulation of cancer cells with the capabilities of self-renewal and differentiation. Emerging evidence has indicated CSCs' role as the seed of tumor development, relapse, and drug resistance in various types of malignancies including EOC [9,10]. CSCs in EOC were firstly purified from a patient's ascites, which displayed their tumorigenic property [11]. Other ovarian cancer stem cells (OCSCs), identified by surface markers CD44, Prominin 1 (PROM1, CD133), and Aldehyde Dehydrogenase 1 Family Member A1 (ALDH1A1), reportedly involved in EOC development, relapse, and chemoresistance [12]. The inherent characteristics of high plasticity, slow cell cycling, and efficient DNA repair empowered CSCs to survive treatment and enabled further chemoresistance [13]. Increased expression of CSC markers and transcription factors, such as CD44, KIT Proto-Oncogene (KIT, CD117), POU Class 5 Homebox 1 (POU5F1, OCT4), and NANOG, was observed in ovarian cancer cells treated by cisplatin or paclitaxel both in vitro and in vivo [14][15][16]. Higher expression level of CSC markers including ALDH1A1, CD44, and CD133 was also reported in tumor tissues collected immediately after primary chemotherapy compared with paired primary samples from EOC patients [17]. In tumors from recurrent platinum-resistant EOC patients, CD133 expression was significantly increased [17]. These findings suggested that OCSCs survived after chemotherapy, generated chemoresistance, and contributed to EOC recurrence. OCSCs are thus considered as a promising therapeutic target against chemoresistant relapse of ovarian cancer. Landen et. al reported that targeting ALDH1A1 in EOC sensitized resistant tumor cells to chemotherapy [18]. Knockdown of GLI Family Zinc Finger 2 (Gli2), a transcription factor in Hedgehog signaling pathway, increased sensitivity of EOC cells to cisplatin [17]. Notably, OCSCs were identified recently using specific surface markers, and a few subpopulations had just been reported in EOC. The mechanisms underlying acquired chemoresistance associated with relapse-initiating OCSCs are not fully understood.
To investigate the molecules and pathways involved in OCSC-induced EOC chemoresistant recurrence, we identified differentially expressed genes (DEGs) between paired chemosensitive primary and chemoresistant relapse samples from EOC patients based on transcriptomic data from nextgeneration RNA-sequencing analysis. Functional enrichment analysis using DEGs found that signaling pathways regulating pluripotency of stem cells were upregulated in chemoresistant relapse samples. Upregulation of hub gene expression, including Wnt Family Member 3A (WNT3A), SMAD Family Member 3 (SMAD3), Kruppel-Like Factor 4 (KLF4), and Paired Box 6 (PAX6), was validated using immunohistochemistry (IHC) staining in chemoresistant relapse samples. Moreover, we demonstrated that the biological processes (BPs) such as DNA damage response, drug efflux, and quiescent state were enriched in OCSCs compared with non-OCSCs, which may explain the underlying mechanisms of chemoresistance in OCSCs. Our study suggested the involvement of CSCs in EOC chemoresistant relapse using transcriptomic data from clinical tumor samples. Through transcriptome analysis, we found out relevant genes and molecular pathways potentially contributing to chemoresistance in OCSCs. These findings reinforced the role of OCSCs as a promising therapeutic target in manage-ment of chemoresistant recurrent EOC and further suggested that these DEGs might become novel preventive or therapeutic targets against OCSCs in the future.

Materials and Methods
2.1. Data Resources. The RNA sequencing data of International Cancer Genome Consortium Ovarian Cancer-Australia (ICGC OV-AU) was obtained from the ICGC Data Portal (https://dcc.icgc.org/) and European Genomephenome Archive (EGA) repository (EGAD00001000877, https://ega-archive.org/) [19]. Patients who relapse 6 months or more after initial chemotherapy are considered as chemosensitive primary cases, while patients relapse within 6 months are termed as chemoresistant primary cases. Chemoresistant relapse is only for patients who were sensitive to initial chemotherapy but failed subsequent treatment [19,20]. According to the criterion of chemosensitive primary and chemoresistant relapse, 24 of 93 patients from ICGC OV-AU dataset were included in this study. A total of 39 samples from these 24 EOC patients, including 14 chemosensitive primary samples, 12 paired chemoresistant relapse samples, and 13 unpaired chemoresistant relapse samples, were selected for further analysis. Detailed information of the samples is listed in Supplementary Table S1. A total of 8 microarray datasets, including 5 chemotherapy-related datasets and 3 CSC-related datasets (Supplementary Table S2), were downloaded from the Gene Expression Omnibus (GEO) database (https://www .ncbi.nlm.nih.gov/geo/) [21][22][23][24][25][26][27].

Principle Component Analysis (PCA).
To detect the outlying samples, the PCA of the normalized expression matrix was performed, relied on the FactoMineR package (version 2.3) [28]. The normalized expression matrix from the ICGC OV-AU dataset was established by filtering genes with low read abundance after the rlog transformation based on the R package DESeq2 (version 1.28.1) [29]. Finally, the Fac-toExtra package (version 1.0.7) was used to visualize the results of PCA [30].

Differential Expression Analysis.
After the PCA, a total of 36 samples from the ICGC OV-AU were sent for further analysis. The DESeq2 package was applied to identify DEGs between chemosensitive primary and chemoresistant relapse samples. Genes with |log 2 fold change | >1 and adjusted probability (P) value < 0.05 were considered statistically significant. The volcano plot was created by ggplot2 (version 3.3.2) [31], and the heatmap (hierarchical clustering, using Euclidean distances and the complete algorithm, scaled by gene) based on the rlog-transformed expression value of these DEGs was generated by the pheatmap package (version 1.0.12) [32]. For the microarray data from GEO database, the limma package (version 3.44.4) was used to analyze changes in gene expression [33].    Table 1). Specifically, signaling pathways regulating pluripotency of stem cells were enriched in chemoresistant relapse tumors (adjusted P value = 0.0019) (Figure 1(c) and Table 1). GO analysis found that these upregulated DEGs were mainly enriched in BPs involving organ development, such as epidermis development, skin development, gland development, and epidermal cell differentiation (Figure 1(c) and Table 2). And downregulated DEGs in chemoresistant relapse tumors were enriched in PI3K-Akt signaling pathway, ECM-receptor interaction, primary immunodeficiency, and BPs including extracellular matrix organization, and T cell activation (Figure 1(d) and Tables 1 and 2). These data demonstrated that chemoresistant relapse tumors possessed different gene expression pattern compared with chemosensitive primary samples. Further functional enrichment analysis suggested potential activation of signaling pathways regulating pluripotency of stem cells in chemoresistant relapse tumor samples, indicating the involvement of OCSCs in chemoresistant recurrence of EOC.

Identification of Crucial OCSC Genes Involved in
Chemoresistant Relapse. Having found that signaling pathways upregulated in chemoresistant relapse samples, we intended to investigate the genes involved in chemoresistant relapse of EOC. Firstly, DEGs listed in signaling pathways regulating pluripotency of stem cells were investigated. It showed that 18 DEGs in this pathway were upregulated in chemoresistant relapse class (Figure 2(a)). Among them, the expression of KLF4, a well-known stem cell marker [37], was most notably upregulated (log 2 fold change = 2:93,    Table S5). In addition, the expression levels of 4 CSC surface markers including CD44, CD117, CD133, and ALDH1A and 3 stem cell transcription factors including SRY-Box Transcription Factor 2 (SOX2), OCT4, and NANOG were also investigated. However, no significant difference was found in the expression of these 7 CSC-related genes between chemoresistant relapse tumor samples and chemosensitive primary samples (Figure 2(a)). Then, the expression levels of 18 DEGs and 7 CSCrelated genes in chemoresistant ovarian cancer cell lines and chemoresistant tumor samples were assessed for further validation. The published microarray expression datasets (GSE33482 and GSE15709 for chemoresistant tumor cells line and GSE51373, GSE28739, and GSE131978 for chemoresistant tumors) were obtained from GEO database. In dataset GSE33482 [38], the expression of WNT3, WNT3A, FGFR3, OTX1, CD117, ALDH1A1, SOX2, and NANOG was elevated in chemoresistant A2780cis line but not its parental chemosensitive line A2780, while the expression of other genes remained unchanged (Figure 2(b)). In dataset GSE15709 [21], the expression of WNT16, KLF4, FGFR3, and ALDH1A1 was remarkably downregulated in chemore-sistant Round5 A2780 line compared with the parental line A2780, while other genes showed no statistical difference at the transcriptional level (Figure 2(c)). In datasets for chemoresistant primary tumors (GSE51373 [22], GSE28739 [23], and GSE131978 [24]), none of the 25 CSC-related genes showed notable change between primary chemoresistant samples and primary chemosensitive samples Additionally, the expression of CSC-related genes was further assessed in EOC specimens via IHC staining. A total of 12 paraffin-embedded tumor tissue samples (1 pair of chemosensitive primary and chemoresistant relapse, 1 chemoresistant relapse, 5 chemosensitive primary, and 4 chemoresistant primary) were obtained from EOC patients (Supplementary Table S3). The PPI network analysis identified 4 hub genes out of 18 DEGs, including WNT3A/ WNT3, SMAD3, PAX6, and KLF4, which were selected for further validation (Figure 3(a)). Positive staining of KLF4 and SMAD3 was observed in paired chemosensitive primary tumor, while the expression of all 4 hub genes was observed in paired chemoresistant relapse tumor with a significant enhanced signal in KLF4 and SMAD3 staining (Figure 3(b)). Signal of KLF4, SMAD3, and PAX6, but not WNT3A, was detected in the unpaired chemoresistant relapse tumors (Figure 3(b)). A similar expression pattern was found in chemosensitive primary and chemoresistant primary tumors with positive staining of KLF4, SMAD3, and PAX6, but negative staining of WNT3A (Figure 3(b)). Statistical analysis demonstrated that the H-score of 4 hub genes was higher in chemoresistant relapse tumor than in   Journal of Oncology the paired chemosensitive primary tumor, which was in consistence with the results of transcriptome analysis (Figure 3(c)). No significant difference was detected in Hscore of these 4 hub genes between chemosensitive primary and chemoresistant primary samples (Figure 3(c)). These results revealed that CSC-related genes such as KLF4, PAX6, SMAD3, and WNT3A were correlated with acquired chemoresistance of EOC, while no evidence of direct correlation was found between CSC-related genes and chemoresistance in primary tumors during this study. It suggested that these aberrantly expressed chemoresistant relapse-related CSC genes could be used as potential therapeutic targets in patients suffering from EOC recurrence.

Identification of the Mechanisms Underlying
Chemoresistance of OCSCs. Recent studies reported that the chemoresistance of OCSCs may be traceable to its quiescent state, increased ability of drug efflux, and enhanced response to DNA damage, all of which might help OCSCs to survive the conventional chemotherapy [10]. To clarify the correlation between these OCSC properties and acquired chemoresistance, 3 microarray expression datasets of isolated OCSCs (GSE28799, GSE82304, and GSE33874) were obtained from GEO database. KEGG and GO enrichment analyses were performed based on DEG information from each dataset, respectively. KEGG pathways and GO BPs correlated with cell cycle, drug response, drug transport, and DNA damage response were listed specifically. KEGG analysis showed that  7 Journal of Oncology cell cycle pathway was inhibited, while pathways of platinum drug resistance, ABC transporters, and nucleotide excision repair were activated in spheroid OVCAR3 (GSE28799) [26] (Figure 4(a)). Pathways of cell cycle and base excision repair were inhibited in ALDH-high SKOV3 (GSE82304) [25], while platinum drug resistance pathways were activated (Figure 4(a)). In side population (SP) cells from fresh ascites of HGSOC patients (GSE33874) [27], none of the KEGG pathways mentioned above was significantly enriched (Figure 4(a)). GO analysis showed that BPs of cell cycle and drug response were inhibited and DNA damage response was activated in spheroid OVCAR3 (GSE28799) and ALDH-high SKOV3 (GSE82304), while no drug transport in BPs was enriched (Figure 4(b)). In SP cells (GSE33874), BP of drug response was inhibited, while BPs of cell cycle, drug transport, and DNA damage response were not enriched (Figure 4(b)). These results confirmed that quiescent state, increased ability of drug efflux, and enhanced response to DNA damage may have caused the chemoresistant recurrence in EOC patients.

Discussion
Chemoresistant recurrence accounts for most fatalities in EOC patients. The majority of patients, who were successfully treated with chemotherapy following their diagnosis, suffered from recurrence with shortening PFI, which led to chemoresistance and death eventually [10]. OCSCs were considered the primary cause of tumor relapse and chemoresistance. In present study, we revealed the enrichment of CSCs in chemoresistant relapse compared with chemosensitive primary tumors by analyzing transcriptomic data from EOC patients. CSC-related genes such as KLF4, SMAD3, and PAX6 from signaling pathways regulating pluripotency of stem cells were found upregulated in chemoresistant relapse tumors via bioinformatics analysis and IHC. No significant difference was noticed in the expression of CSC markers including CD44, CD133, and ALDH1A1 or stem cell transcriptional factors such as SOX2, OCT4, and NANOG between chemoresistant relapse and chemosensitive primary samples. By analyzing the microarray expression data of OCSCs via KEGG and GO, we further validated that quiescent state, increased ability of drug efflux, and enhanced response to DNA damage may have caused the chemoresistant recurrence and led to low survival rates in EOC patients.
Enrichment of CSCs had been reported in post chemotherapy tumors via CSC marker identification. Based on the expression of several CSC markers, such as CD44, CD24, and Epithelial cell Adhesion Molecule (EpCAM), a population of cells, characterized with increased tumorigenic, metastatic, and chemoresistant potential, was found enriched in vitro culture after the treatment of cisplatin or paclitaxel [16]. CD133 + and Stem cell antigen-1 + (Sca-1 + ) tumor-initiating cells were also detected to persist after paclitaxel and carboplatin chemotherapy in mouse models [39]. In our study, we revealed 18 pivotal CSC-related genes upregulated in chemoresistant relapse samples through transcriptomics. Several frequently used CSC identification markers including CD44, CD117, CD133, ALDH1A, SOX2, OCT4, and NANOG were also investigated between chemoresistant relapse and chemosensitive primary samples. These genes were reported to be overexpressed in and be associated with the progression of many malignant tumors such as melanoma, bladder cancer, prostate cancer, gastric cancer, and colorectal cancer [40][41][42][43][44][45]. However, the expression levels of these genes were found with no statistical difference, which may be ascribed to the phenotypic and functional heterogeneity of CSCs. For example, ALDH + CSCs tend to be more proliferative, the CD44 + /CD24 -CSCs have a disposition to be more invasive, and the Leucine Rich Repeat Containing G Protein-Coupled Receptor 5 + (LGR5 + ) cells are more quiescent and chemoresistant [46,47]. Therefore, the heterogeneity of CSC markers is one of the reasons for the inefficiency of CSC targeting. In addition, it was reported that posttranscriptional mechanisms and proteostasis might cause inconsistent expression of CSC markers between mRNA and protein levels [48]. Although we found no significant difference of these CSC markers according to the transcriptome analysis, there might be remarkable differences at the protein level due to these epigenetic regulations.
The 18 DEGs detected in this study are mainly involved in Wnt, TGF-β, and MAPK pathways, which may become potential therapeutic targets to eliminate OCSCs ( Figure 2). Moreover, 4 out of these 18 identified genes were further confirmed in resected tissues from EOC patients through transcriptome analysis and IHC, which may possess a better clinical value in developing novel therapies against recurrent EOC. The inhibition of Wnt signaling pathway was proved to suppress the CSC populations [49][50][51]. In recent years, several agents targeting the Wnt signaling pathway have been under clinical trials, including ipafricept, vantictumab, and CWP232291. Although these inhibitors may Class ICGC OV-AU Class A2780 A2780cis   10 Journal of Oncology eliminate CSCs, it could affect the physiological processes in normal cells at the same time [52,53]. The TGF-β signaling pathway maintains the homeostasis and quiescence of CSCs. Galunisertib (LY215729), one of TGF-β signaling pathway inhibitors, has been proved effective and safe for prostate cancer in a phase II study [52,54]. An increasing number of drugs targeting CSCs are on the way to clinical application, and CSC targeting therapy joint with conventional che-motherapeutic treatment may be the next step closer to the cure of cancer in the future. The enrichment of CSC signaling pathways in the chemoresistant relapse tumors was further validated in chemoresistant tumor cell line A2780cis (Figure 2(b), GSE33482). However, no OCSC enrichment was observed in another dataset using a closely related cell line Round5 A2780 (Figure 2(c), GSE15709). This may be attributed to      Journal of Oncology the difference in cell line establishment methods between the two; only 5 rounds of cisplatin treatment were performed in the latter. In addition, no OCSC enrichment or statistical difference in CSC-related gene expression was observed between chemosensitive and chemoresistant primary tumor samples assessed by transcriptome analysis and IHC (Figures 2 and 3). These findings were inconsistent with the hypothesis that CSCs are seeds of chemoresistance. Aside from complex components of tumor tissues, rare CSCs exist among them. The relatively low sensitivity of transcriptome analysis may be the reason why no CSC enrichment was observed. Steg et al. reported that the expression level of OCSC markers significantly increased in post chemotherapy recurrent chemoresistant samples but not in recurrent samples without secondary chemotherapy [17]. It suggested that chemotherapy might be one of the factors that activate OCSCs. Based on these findings, it is possible that an increasing population of OCSCs can be found in recurrent tumor tissues from patients receiving multiline chemotherapy as the treatment going on. This growing number of residual OCSCs might gradually shorten the PFI in EOC patients. Moreover, different chemoresistant signal pathways were found in SP, a special type of CSCs, compared with CSCs isolated by surface markers. Only cellular response to drug was enriched in SP, while in surface marker identified CSCs, drug response, drug transport, cell cycle, and DNA damage response were all enriched ( Figure 4). All these findings proved that chemoresistance is an inherent characteristic of OCSCs but also indicated that heterogeneity of CSCs should be taken into consideration when developing targeting strategies. Certainly, more thorough studies as well as more detailed information are needed to identify the relationship between OCSCs and chemoresistance in EOC relapse. Additionally, the inclusion of chemoresistant relapse samples from solid tumors other than ascites for comparative analysis might supply more consolidated evidence, which would minimize the bias caused by the differences in tumor microenvironment. In order to investigate the underlying mechanisms, extensive functional genomics experiments accompanied with sensitive transcriptome analysis are necessary, which will also reveal the potential specific CSC molecular targets to develop novel therapies against chemoresistant EOC recurrence.

Conclusions
In summary, the present study reported the potential activation of signaling pathways regulating pluripotency of stem cells in chemoresistant relapse tumors. The transcriptome analysis revealed upregulated CSC-related genes involved in acquired chemoresistance of EOC tumors. Our findings suggested the involvement of OCSCs in chemoresistant relapse of EOC and also indicated that these CSC DEGs correlated with acquired chemoresistance could become potential molecular targets to eliminate OCSCs in EOC. Based on the evidence, we further suggested the necessity to combine OCSC targeting therapy with conventional therapy in order to prevent chemoresistance as well as improve the prognosis of EOC patients.

Data Availability
The original contributions presented in the study are included in the article and supplementary material.

Ethical Approval
The studies involving human participants were reviewed and approved by the ethics committees of the Shanghai Tenth People's Hospital, School of Medicine, Tongji University.

Supplementary Materials
Supplementary Figure S1: the PCA score plots show a total of 39 samples in the ICGC AU-OV dataset. Three outlying samples were labelled. Supplementary Figure S2: the volcano plot of the differentially expressed genes in chemoresistant relapse samples. The threshold is |log2 fold change| > 1 and adjusted P value < 0.05. The upregulated genes are shown in red, while the downregulated genes are shown in blue. Supplementary Figure S3: immunohistochemistry images of tumors from chemosensitive primary, chemoresistant primary, and chemoresistant relapse patients. The parts circled by the black boxes are shown in