Role of the Microenvironment in Ovarian Cancer Stem Cell Maintenance

Despite recent progresses in cancer therapy and increased knowledge in cancer biology, ovarian cancer remains a challenging condition. Among the latest concepts developed in cancer biology, cancer stem cells and the role of microenvironment in tumor progression seem to be related. Indeed, cancer stem cells have been described in several solid tumors including ovarian cancers. These particular cells have the ability to self-renew and reconstitute a heterogeneous tumor. They are characterized by specific surface markers and display resistance to therapeutic regimens. During development, specific molecular cues from the tumor microenvironment can play a role in maintaining and expanding stemness of cancer cells. The tumor stroma contains several compartments: cellular component, cytokine network, and extracellular matrix. These different compartments interact to form a permissive niche for the cancer stem cells. Understanding the molecular cues underlying this crosstalk will allow the design of new therapeutic regimens targeting the niche. In this paper, we will discuss the mechanisms implicated in the interaction between ovarian cancer stem cells and their microenvironment.


Introduction
Ovarian cancer remains a challenging condition for both clinicians and scientists. Indeed, it oen presents as an advanced metastatic disease; however most patients are treated with a combination of major debulking surgeries and chemotherapy to achieve complete cytoreduction (no tumor residue) [1]. e clinical course of patients with no residue at the end of the treatment remains unpredictable with a group of early recurrence (refractory patients) [2]. e clinical trials of targeted therapies (trastuzumab, imatinib, etc.) as well as dose intensi�cations or use of several agents have failed to signi�cantly improve outcomes [3][4][5][6]. Finally, procedures such as intraperitoneal chemotherapy or hyperthermic intraoperative chemotherapy have only a slight effect on prognosis with signi�cant increase in overall morbidity [7]. e biology of ovarian cancers also has striking features; over the last decade the heterogeneity of ovarian cancers among and within subtypes has been illustrated by transcriptomic and genetic pro�ling [8]. Many authors have presented prognosis signatures without a clear translation to the clinical setting [9]. Recently, a broad study by e Cancer Genome Atlas (�CGA) has demonstrated among other �ndings that serous ovarian adenocarcinoma could be clustered in 4 different subtypes without being able to relay them to prognosis [10]. e mutational spectrum of ovarian cancers seems to be limited with most genetic events happening at the copy number variation level. Metastatic lesions have a genetic pro�le different to primary lesions, again re�ecting tumor heterogeneity [11]. However the speci�c biological features responsible for recurrences have not been clearly identi�ed.
Recently, the concept of cancer stem cells (CSCs) has emerged as an alternative to the clonal theory of tumor evolution. Indeed among the heterogeneous populations constituting a tumor, a small proportion of cells (0.01% to 0.1%) have properties that mimic to certain extent normal stem cell biology: (i) self-renewal with asymmetric and symmetric cell division; (ii) recapitulation of the tumor heterogeneity in immune-suppressed mice; (iii) ability to undergo serial passages in vitro and in vivo due to unlimited division potential [12]. e role and biology of ovarian cancer stem cells have been already illustrated in other comprehensive reviews [13,14]. e tumor is now perceived as a complex structure where the tumor cells closely interact with the stroma, which provides protumoral and prometastatic cues [15]. Our group has demonstrated the role of mesenchymal stem cells in transferring multidrug resistance protein (MDR) or inducing a prometastatic phenotype of ovarian cancer cells [16,17]. us, microenvironment might have a real role in the biology of ovarian cancer stem cells (OCSCs).
Here, we review the data about ovarian cancer stem cells and their interaction with the tumoral microenvironment. Understanding the molecular cues responsible for the crosstalk between the tumor and its stroma might help us design new therapeutic strategies aiming at disrupting speci�c prostemness tumor-stroma interaction rather than targeting tumor cells alone.

Ovarian Cancer Stem Cells
Genetic changes in regular stem cells might give rise to OCSCs [18,19]. As the exact origin of ovarian cancer is still debated (ovarian surface epithelium versus fallopian tube) and its complexity is not limited to one subtype, characterization and de�nition of OCSCs have been really challenging. Besides, OCSCs can display different states (quiescent or proliferative) depending on the microenvironment and the cellular stresses such as chemotherapy which makes it more di�cult to gather a unique de�nition [20,21]. Currently surface markers or a particular phenotype (side population) are used to identify OCSCs. e most commonly described marker is CD133. Different authors showed that CD133 + from cell lines or primary xenogras had greater capacity to initiate tumors than CD133 − [22,23]. OCSCs were more comprehensively characterized by the combination of CD133 and the stem cells marker aldehyde dehydrogenase (ALDH) [24,25]. Finally previously described CSCs markers CD44 and CD117 were used to characterize OCSCs. Cancer stem cells have the increased ability to be grown in 3D anchorageindependent culture setup as spheres (Figures 1(a) and 1(b)). e formation of primary and/or secondary sphere is currently routinely used to enrich and/or quantify the stem cell population [26]. e other striking feature of OCSCs is their chemoresistance and thus their potential role in residual and recurrent disease even if this has not been yet clinically demonstrated [22,27,28]. Indeed in ovarian cancer, CD44 + CD117 + spheroids were resistant to chemotherapy and were able to initiate and propagate tumors in mice [22]. Similarly Luo et al. described that chemoresitsant CD117 + cells isolated from xenogras displayed phenotypic feature of cancer stem cells such as serial transplantation and asymmetric division [29]. Recently, Gao et al. described that CD24 + population expressed increased level of some stem cells genes such as Nestin, -catenin, Bmi-1, Oct4, Oct3/4, Notch1, and Notch4 compared to CD24 − and displayed quiescence, chemoresistance, and tumor initiation [20].
One of the challenges is to determine the hierarchy of the different markers described. In mammary gland, a hierarchy of stem cells is described using the different breast cancer stem cells markers [30]. Such hierarchy is essential to understand and identify the factors which regulate CSCs selfrenewal versus proliferation and differentiation. Recently, Burgos-Ojeda et al. proposed a hierarchy for the OCSCs where they hypothesized that a common ovarian cancer stem cell can undergo asymmetric division to give rise to two different early OCSC progenitors ALDH + CD133 + (CD24 +/− ) or CD44 + CD117 + (CD24 +/− ). Each of these early OCSC progenitors can then produce intermediate progenitor cells by asymmetric division which could produce more differentiated tumors [14].
e molecular drivers of the hierarchy can represent potentially important therapeutic targets. Several studies described correlations between OCSC markers and patient prognosis. Zhang et al. reported a poor prognosis associated with CD133 expression from a tumor bank of over 400 ovarian cancers [31]. More recently, Steg et al. showed the presence of CD133, ALDH1A1, and CD44 at low number in primary samples. e same markers were increased on a sample collected aer chemotherapy and reduced back to initial level in recurrent tumor samples, suggesting their role in recurrence [32]. Using all these stem cells markers as target in a clinical trial seems to be the logical next step. Unfortunately, while limited numbers of tumors express CD133 (34-40%) or CD177 (30-40%), CD24 and CD44 are expressed in numerous tumors but targeting these cells in vivo is limited by wide expression of these molecules in normal tissues [22,25,33].
Stemness relies on a very precise equilibrium between the stem cells and the components of the niche. Recently, stem cells and their niches were identi�ed in mammalian tissue such as the nervous system, muscle satellite cells, and spermatogonial stem cells [34,35]. Many studies described how the niche and the stem cell interact through tissue-speci�c molecular signaling in maintaining stemness and inducing expansion of the stem cell population. For example, recent studies have clearly demonstrated the role of the endothelial niche in the expansion and maintenance of stemness of hematopoietic stem cells [36]. Molecular cues from stromal cells or the extracellular matrix will provide the signaling to maintain and expand the stem cell phenotype.
e constitution of the tumor stroma brings another level of complexity. While for the sake of clarity we have separated different elements, most of them remain closely related and dependent.

Stromal Environment and Stemness
Cancer lesions are oen perceived as never healing wounds with an in�ammatory microenvironment. e in�ltrating in�ammatory cells include tumor promoting and tumor killing subtypes. Much molecular signaling can be hijacked by cancer cells and enhance tumorigenesis and progression toward a metastatic phenotype [37]. e stromal in�ammatory reaction constitutes an environment containing many bioreactive molecules such as proliferative and survival signaling (EGF, FGF, HGF, IGF, or IL-6) that could enable CSCs maintenance and expansion. Moreover, many of these cytokines have been implicated in the occurrence of epithelial to mesenchymal transition (EMT) in many tumor types [38]. Recent lines of evidences have linked EMT phenotype to stemness [39,40]. erefore, we could assume that an in�ammatory microenvironment will enable a subfraction of the tumor cells to gain/maintain a mesenchymal phenotype permissive to maintenance of stemness. e tumor cells also participate to the in�ammatory stroma as the upregulation of in�ammatory molecules has been documented in the literature [11]. Ovarian cancers overexpress LL-37 (leucine, leucine 37) which is a member of the cathelicidin family of antimicrobial polypeptides. While LL-37 does not act directly on ovarian cancer cells, it attracts mesenchymal stem cells (MSCs) into ovarian tumor xenogras. MSC treated by LL-37 secreted increased amount of IL-1 receptor antagonist, IL-6, IL-10, CCL5, and VEGF and had a proangiogenic effect [41]. Long et al. demonstrated that CD133 + cells had increased expression of the chemokine CCL5 and its receptors, CCR1, CCR3, and CCR5, compared to CD133 − [42]. CCR5 mediated nuclear factor B (NF-B) dependent MMP9 secretion. ese studies demonstrate the complex crosstalk relying on multiple cytokines and resulting in a permissive niche that will provide all molecular cues for maintenance and expansion of ovarian cancer stem cells.
e tumor stroma constitutes a hypoxic microenvironment before the appropriate signaling cues are able to induce neoangiogenesis. Hypoxia maintains and even upgrades stem cell characteristics [43,44]. Under anaerobic conditions, glycolysis is favored and only a small amount of the pyruvate will be directed toward the mitochondria. e glycolytic metabolism activates tumor suppressor genes and oncogenes such as p53, RAS, or MYC. ese oncogenes can activate HIF1 and HIF2 associated with activation of pluripotency marker genes such as OCT4, SOX3, and KLF4 [45]. Liang et al. studied the effect of hypoxia on ovarian cancer stem cells [46]. ey demonstrated increased ability for OCC to form spheres and colonies. CD44 displayed higher level of the stemness transcription factors OCT3/4 and Sox2 when cultured in hypoxic condition. In another study, CD44 and CD133 expression was increased through the Sox2 and OCT3/4 regulation in two different ovarian cancer cell lines (ES-2 and OVCAR3) [47].

Cellular Elements of the Stroma
e in�ammatory stroma attracts other cell types such as mesenchymal stem cells and endothelial cells [48]. e protumoral and premetastatic roles of both cell types have been widely described in the literature. However few studies point out their interaction with cancer stem cells. ere are many models of cellular interaction mediated by direct contact (tunneling nanotubes, synapses, trogocytosis) or microparticle mediation (Figures 1(c) to 1(f)). eir role in the maintenance of stemness remains to be clearly established. Mitsui et al. described increased expression of the CD133 + and sphere formation when Yolk sac ovarian cancer stem cells were cocultured with peritoneal MSCs [49]. e CD133 + cells displayed increased migration and invasion in culture with the MSCs. e crosstalk in this study seemed mediated through the SDF1/CXCR4 axis. In a more comprehensive approach, McLean et al. showed that cancer-associated MSCs (CA-MSC) had greater ability to increase tumor growth compared to normal MSCs [15]. ey demonstrated that the CA-MSCs had abnormal BMP production. Treatment with recombinant BMP2 had the ability to increase ovarian cancer cell line stem cell population as de�ned by ALDH and CD133 + (up to 60%). Similarly the treatment of primary derived spheres with BMP2 also induced a 3.2-to 4.4-fold increase of the ALDH + population.
As illustrated above, there is a strong interaction between cancer cells and the different elements of the niche. is crosstalk has a strong role in tumor biology as it participates to the plasticity of the tumor cells. Abelson et al. used the human embryonic stem cell derived experimental platform [50]. ey isolated different clones from a single clear cell ovarian tumor. ey showed that while some clones were not able to grow in a classical xenogra model their injection in a hESC-derived teratoma produced a tumor recapitulating the different cell populations of the primary tumor. More interestingly, they demonstrated that the microenvironment could switch the non-stem-cell EPCAM + CD44 − population to a stem cell EPCAM + CD44 + population.
e microenvironment-dependent phenotypic plasticity has great therapeutic implication. is could result in failure of treatments targeting a single stable self-renewing clone. One option might then be to use multimodal approach to balance the equilibrium between the self-renewing and rapidly proliferating populations. Maintaining a permanent low level of self-renewing cell population will allow having a chronic disease rather than a rapidly lethal tumor. Many questions need to be answered such as whether self-renewal is a durable state rather than a dynamic niche dependent which is supported by many �ndings in the literature [51,52]. is might also be dependent on tumor type as, for example, the Morrison groups demonstrated that nonputative stem cell population could give rise to a tumor in xenogra models when the microenvironment was modulated [53]. ese �ndings were however obtained in the melanoma model and mice malignant peripheral nerve sheath tumors both originating from the neural crest.
e role of endothelial cells in cancer stemness has not been yet clearly identi�ed. Shank et al. have studied the action of metformin on ovarian cancer stem cells [54]. ey demonstrated that metformin reduced ALDH + CSC in vitro and in vivo and inhibited the growth of ovarian tumor spheres. One of the action of metformin resulted in decreased microvascular density consistent with previous studies [55,56]. e data in the literature demonstrates that CSCs are highly angiogenic [25,57] and that endothelial cells participating to neoangiogenesis provide essential growth factors for OCSCs [58].

Cytokines Environment and Global Crosstalk
Cytokines play an essential role in intercellular communications as described above. Many of them regulate stem cell phenotype in a variety of contexts ranging from normal development to neoplasia. Cao et al. showed that TGF which is highly secreted in the ovarian cancer microenvironment induces tissue transglutaminase (TG2) expression and its enzymatic activity [59]. e treatment by TGF induced spheroid formation enabling peritoneal dissemination. ey demonstrated that TG2 was responsible for an EMT-mediated increase of the CD44 + CD117 + population. Interestingly the effect of TGF was greater when the ovarian cancer cells were cultured on �bronectin once again demonstrating the additive role of the different component of the microenvironment.
Among the different cytokines c-kit's role as a stem cell factor has already been described and plays a particular role. Indeed c-kit-mediated pathways are activated in cancer [60,61]. Ovarian tumor abnormal expression of c-kit has been associated with poor prognosis. Chau et al. demonstrated increased expression of c-kit aer enrichment for OCSC [62]. ey showed that c-kit knockdown inhibited sphere formation. ey displayed that hypoxia increased ckit expression which in turn induced overexpression of the ABC drug ABCG2 transporter through the Wnt/ -catenin pathway, leading to chemoresistance in OCSC. One of the interesting �ndings of this group is that multiple rounds of chemotherapy seemed to enrich for OCSC harboring a high chemoresistance pro�le.
e crosstalk between cancer cells and the microenvironment has been illustrated in many other contexts. In a study by Ko et al., the authors showed a poor prognosis of tumors with HOXA9 expression [63]. In vitro, HOXA9 was not able to induce autonomous tumor cell growth. However ovarian cancer HOXA9 expression induced a cancer-associated �broblasts phenotype for the peritoneal �broblasts which stimulated OCC and endothelial cell growth. HOXA9 activated the transcription of TGF 2 which acted in a paracrine manner on peritoneal �broblasts which in turn upregulated the protumoral panel of cytokines (CXCL12, IL-6, and VEGF-A) expression. is study illustrates the promotion of a permissive microenvironment which will provide the optimal ground for tumor growth. In accordance with these data, Alvero et al. demonstrated a very intricate relation between OCSC and the microenvironment [64]. ey demonstrated that ovarian cancer stem cells participated to blood vessels and acquired markers of endothelial cell such as CD34 and VE cadherin. Interestingly, the process was not relying on VEGF but IKK /NF b. While these data need to be con-�rmed by more functional studies demonstrating the ability of these cells to act as endothelial cells, the participation of the ovarian cancer stem cells to blood vessels beyond underlying their crosstalk with the endothelium could suggest resistance pathways to anti-VEGF-based targeted therapies.

Epithelial to Mesenchymal Transition (EMT)
Ovarian cancer represents a heterogeneous group of tumors with distinct clinical features, genetic alterations, and tumor behaviors. e phenomenon of EMT has been widely studied in ovarian cancer. e authors have suggested that OCC can undergo EMT to detach and MET to develop a metastatic nodule. ese data should therefore be investigated in the context of OCSCs. Indeed targeting EMT has been suggested, as a potential treatment [65]. Several studies in breast cancer have demonstrated that EMT induced an increase in the CSCs population de�ned as CD44 + CD24 − [66][67][68]. However, in 2012, Sarrio et al. indicated that a mesenchymallike phenotype did not correlate with the acquisition of global stem cells/progenitors characteristics in breast cancer [69]. Supporting these �ndings Celi�-Terrassa et al. showed that the acquisition of mesenchymal features (correlated with the loss of their epithelial properties) by cancer cells occurred at the expense of their self-renewal potential, in prostate and bladder cancer [70]. e body of data suggests that stemness might be a plastic phenotype that could depend both on tumor type and global stromal context.
In ovarian cancer, only a few studies have focused on the link between EMT and OCSCs. Recently, Jiang et al. demonstrated that the ovarian cancer cell lines displayed a side population with mesenchymal traits and typical mesenchymal genes. Inhibition of EMT process by Snail1 silencing decreased this side population occurrence and affected its invasive capacity and tumorigenicity in vivo [71]. Dahl Steffensen et al. established a correlation between the percentage of epithelial OCSCs and survival in early-stage ovarian cancer (FIGO I/II) in a cohort of 117 patients [72]. Concordant with �ndings in other epithelial cancers [70], Yin et al. demonstrated the ability of the OCSCs to generate peritoneal metastasis in an in vivo model. Furthermore they showed that TWIST-1 (a major transcription factor implicated in EMT) is constitutively degraded by the proteasome in OCSCs [73]. ey suggest that OCSCs could be a source of ovarian cancer metastasis through balance of EMT/MET.

Other Tumors
Ovarian cancer is not the only one to be maintained by a subpopulation of cells that display stem cell properties, mediate metastasis, and contribute to treatment resistance. A similar hierarchy governs many solids tumors, including breast [74], pancreas [75], glioblastoma [76], and prostate [77]. ey are de�ned by different cell surface markers and characterized by speci�c phenotypic traits.
Several markers have been proposed in the literature to identify CSCs in many human cancers, but to date there is still no gold standard to de�ne CSCs, leading to the hypothesis that the CSC phenotype might be dynamically switched [78]. Compared with the hematopoietic tumors, the properties of CSCs in solid tumors remained relatively unde�ned until recently. e �rst solid CSCs were identi�ed in breast cancer by Al-Hajj et al. in 2003 with two surface markers CD44 + /CD24 −/ [74]. Further characterization of breast CSCs was established using ALDH1 [79], mammosphere assay [80], and transplantation into immunode�cient mice [81]. For instance, CD133 has been described to be one of the most recurrent CSC markers in a number of solid malignancies, including brain tumor [ carcinoma [83], hepatocellular carcinoma [84], colorectal cancer [85,86], and lung cancer [87]. All those CSCs are regulated by, and in turn regulate, cells within the tumor microenvironment. Recently, an emerging area of research supports that CSCs may promote tumor angiogenesis. As mentioned in glioma CSCs by Bao et al., the VEGF expression in CD133 + cells was 10-20-folds upregulated, combined with a signi�cant increase in vascular density demonstrated by CD31 staining [57]. Furthermore, they described that therapy with VEGF antibody (bevacizumab) could reduce CSC-induced vascular endothelial cell migration and tube formation. ey �nally demonstrated, in vivo, that bevacizumab inhibited speci�cally the tumor growth of CSC-derived xenogra. Other studies support this �nding that CSCs contribute to tumor vascular development in glioma [88,89]. Cytokines produced by endothelial cells directly regulate CSCs contributing to their maintenance and their proliferation [90,91]. is cross-regulation between CSCs and endothelial cells seems to be common in various solid tumor such as breast cancer [92], colon cancer [93], or brain tumor [94].
CSCs are also known to play a role in metastatic disease. Indeed, in breast cancer cells, CSCs can go through EMT via activation of Hedgehog (Hh), Wnt, Notch, or TGF (transforming growth factor-) leading to the upregulation of a group of transcriptional factors that drive EMT, resulting in the transformation of epithelial-like CSCs into cells with aggressive mesenchymal-like phenotypes [95]. All these pathways are induced by extracellular factors related to tumor microenvironment such as matrix metalloproteinase (MMP) family proteins [96]. e involvement of CSCs mesenchymal transition in metastatic spread was described in many tumors including head and neck squamous cell carcinoma [97], colorectal cancer [98], prostate cancer [99], and pancreatic cancer [100].
Inside their complex microenvironment, CSCs are also in close interaction with MSCs or tumor-associated macrophage [101]; however their precise interaction remains to be elucidated.

Conclusion
e recent discovery of cancer stem cells in solid tumors mimicking leukemia has added another level of complexity to tumor heterogeneity. Indeed, the tumor now appears to be constituted by different cell populations harboring different phenotypes. Moreover, data presented above argue for a tremendous plasticity induced not only by clonal evolution but also by the interaction between the cancer cells and their microenvironment. It is difficult today to have a clear BioMed Research International 7 perception of the essential molecular hubs as the number of studies per disease is still limited and many studies have addressed few molecules rather than broad pathways ( Figure  2). Accumulating evidence reveals that the composition of tumor microenvironments may de�ne CSCs role throughout the different steps of carcinogenesis.
ese �ndings have several consequences for patients� management. Indeed, as the tumor is now perceived as a dynamic structure, new factors (stem cell fraction, presence of stromal elements, and immune in�ltrate) might be useful to predict prognosis. More importantly, we might have to consider absence of chemosensitivity rather than chemoresistance. Indeed, in the clonal theory of tumor evolution, upon treatment clones were able to develop resistance. In a more global approach, we could consider that at a time point some tumoral cells might not be sensitive to chemotherapy protected by their stemness and/or their interaction with the tumor microenvironment. Obviously, clonal and stem cell theories are not mutually exclusive, and under selective pressure the tumor plasticity could shi through clonal selection.
erefore, there is a great need to gain a comprehensive understanding of the networks governing tumor plasticity, in particular the interaction between the stem cell compartment and the stroma. is will drive the design of new therapeutic approaches disrupting the tumor-stroma interaction to reduce tumoral plasticity.