Metabolic Plasticity in Ovarian Cancer Stem Cells

Ovarian Cancer is the fifth most common cancer in females and remains the most lethal gynecological malignancy as most patients are diagnosed at late stages of the disease. Despite initial responses to therapy, recurrence of chemo-resistant disease is common. The presence of residual cancer stem cells (CSCs) with the unique ability to adapt to several metabolic and signaling pathways represents a major challenge in developing novel targeted therapies. The objective of this study is to investigate the transcripts of putative ovarian cancer stem cell (OCSC) markers in correlation with transcripts of receptors, transporters, and enzymes of the energy generating metabolic pathways involved in high grade serous ovarian cancer (HGSOC). We conducted correlative analysis in data downloaded from The Cancer Genome Atlas (TCGA), studies of experimental OCSCs and their parental lines from Gene Expression Omnibus (GEO), and Cancer Cell Line Encyclopedia (CCLE). We found positive correlations between the transcripts of OCSC markers, specifically CD44, and glycolytic markers. TCGA datasets revealed that NOTCH1, CD133, CD44, CD24, and ALDH1A1, positively and significantly correlated with tricarboxylic acid cycle (TCA) enzymes. OVCAR3-OCSCs (cancer stem cells derived from a well-established epithelial ovarian cancer cell line) exhibited enrichment of the electron transport chain (ETC) mainly in complexes I, III, IV, and V, further supporting reliance on the oxidative phosphorylation (OXPHOS) phenotype. OVCAR3-OCSCs also exhibited significant increase in CD36, ACACA, SCD, and CPT1A, with CD44, CD133, and ALDH1A1 exhibiting positive correlations with lipid metabolic enzymes. TCGA data show positive correlations between OCSC markers and glutamine metabolism enzymes, whereas in OCSC experimental models of GSE64999, GSE28799, and CCLE, the number of positive and negative correlations observed was significantly lower and was different between model systems. Appropriate integration and validation of data model systems with those in patients’ specimens is needed not only to bridge our knowledge gap of metabolic programing of OCSCs, but also in designing novel strategies to target the metabolic plasticity of dormant, resistant, and CSCs.


Introduction
Ovarian cancer (OvCa) is the fifth most common cancer in females and remains the most lethal gynecologic malignancy in the present day [1]. The five-year survival rate of women diagnosed with early localized disease is over 90%, but drops precipitously when diagnosed at stages III or IV [1]. The standard of care clinical management for OvCa includes debulking surgery followed by adjuvant chemotherapy or neoadjuvant chemotherapy followed by surgery [2]. Despite initial responses to chemotherapy, recurrence of chemo-resistant disease is encountered in almost 75% of  Next, we analyzed data of publicly available datasets from studies that compared OCSCs with their parental cells. In a study that characterized OCSCs from OVCAR3 cell line, GSE28799 [41], we found that, consistent with the original report, CD44, CD24, CD117/KIT, CD133/PROM1, ALDH1A1, and SOX2 were significantly upregulated in OVCAR3 spheroid-derived stem cells compared to their parental OVCAR3 cells. No significant changes were detected in OCT4/POU5F1 and NANOG, however, NOTCH1 expression was significantly down-regulated in spheroid-derived OCSCs compared to parental OVCAR3 cells (Figure 2A). In another study GSE64999 [42], CSC-related features of established serially diluted spheroid cells were examined in vitro under serum-containing and CSC culture conditions (will Cancers 2020, 12, 1267 5 of 27 be referred to thereafter as differentiated and undifferentiated OCSCs, respectively). In this study [42], differentiated spheroids cultured in the presence of serum-containing media underwent epithelial differentiation with epithelial-like morphology and reduction of stem cell markers ALDH1A1 and SOX2 compared to parental undifferentiated spheroids cultured in stem cell media. Further analysis of the associated GSE64999 dataset confirmed the significant upregulation of ALDH1A1 and SOX2 but not the other putative OCSCs' markers ( Figure 2B). Significance was determined using multiple t-test and Holm-Sidak method with each row analyzed individually with p < 0.05, without assuming a consistent SD.

Correlation Between OCSCs Markers and Glycolysis
Glycolysis is an oxygen-independent metabolic pathway that occurs in the cytosol, generating ATP from the conversion of glucose into pyruvate. Glycolytic metabolic reprogramming is critical for the maintenance of CSCs and is associated with cancer progression and chemo-resistance [48]. OCSCs share a similar pattern of glycolytic events with CSCs in cancers of the brain, breast, lung, liver, and bone, that they significantly increase their glucose uptake and lactate production compared with their non-CSC counterparts [49][50][51].
We first compared the expression of the enzymes involved in glycolysis in OCSCs and their parental cells in both GSE28799 and GSE64999. We found that the glucose transporter SLC2A1, Hexokinase 1 (HK1), and pyruvate dehydrogenase kinase 1 (PDK1) were significantly upregulated in OVCAR3 spheroid-derived OCSCs compared to their parental OVCAR3 cells ( Figure 3A). In contrast, there was no significant difference in glycolytic enzymes between undifferentiated and differentiated spheroids in GSE64999 ( Figure 3B). There was a general trend of positive correlations between OCSC markers and glycolytic enzymes in OVCAR3-spheroids OCSCs compared to their parental OVCAR3. CD44 and CD24 exhibited significant positive correlation with HK1 and phosphoglycerate kinase 1 (PGK1). NOTCH1 exhibited significant positive correlation with HK2. None of the other stemness markers exhibited significant positive correlation with glucose transporters or glycolysis ( Figure 3C and Table S1). Consistently, OCSC markers in undifferentiated spheroids in GSE64999, did not exhibit a glycolytic enrichment. CD44 positively correlated with SLC2A6, while CD24 negatively correlated with PHGDH. CD117/KIT exhibited significant negative correlations with SLC2A1 and SLC2A5. CD133/PROM1 exhibited significant positive correlation with HK2, PGK1, PDK1, and enolase1 (ENO1). ALDH1A1 and SOX2 positively correlated with SLC2A3, whereas POU5F1/OCT4 and NANOG positively correlated with SLC2A2. However, NOTCH1 did not exhibit any significant correlation with glycolytic enzymes or glycolysis ( Figure 3D and Table S2).
Analysis of patients' tumors, from TCGA data revealed that CD44 and NOTCH1 were the top CSC markers whose expression exhibited positive correlations with glucose transporters and glycolytic enzymes ( Figure 3E and Table 1). NOTCH1 positively correlated with six glucose Significance was determined using multiple t-test and Holm-Sidak method with each row analyzed individually with p < 0.05, without assuming a consistent SD.

Correlation between OCSCs Markers and Glycolysis
Glycolysis is an oxygen-independent metabolic pathway that occurs in the cytosol, generating ATP from the conversion of glucose into pyruvate. Glycolytic metabolic reprogramming is critical for the maintenance of CSCs and is associated with cancer progression and chemo-resistance [48]. OCSCs share a similar pattern of glycolytic events with CSCs in cancers of the brain, breast, lung, liver, and bone, that they significantly increase their glucose uptake and lactate production compared with their non-CSC counterparts [49][50][51].
We first compared the expression of the enzymes involved in glycolysis in OCSCs and their parental cells in both GSE28799 and GSE64999. We found that the glucose transporter SLC2A1, Hexokinase 1 (HK1), and pyruvate dehydrogenase kinase 1 (PDK1) were significantly upregulated in OVCAR3 spheroid-derived OCSCs compared to their parental OVCAR3 cells ( Figure 3A). In contrast, there was no significant difference in glycolytic enzymes between undifferentiated and differentiated spheroids in GSE64999 ( Figure 3B). There was a general trend of positive correlations between OCSC markers and glycolytic enzymes in OVCAR3-spheroids OCSCs compared to their parental OVCAR3. CD44 and CD24 exhibited significant positive correlation with HK1 and phosphoglycerate kinase 1 (PGK1). NOTCH1 exhibited significant positive correlation with HK2. None of the other stemness markers exhibited significant positive correlation with glucose transporters or glycolysis ( Figure 3C and Table S1). Consistently, OCSC markers in undifferentiated spheroids in GSE64999, did not exhibit a glycolytic enrichment. CD44 positively correlated with SLC2A6, while CD24 negatively correlated with PHGDH. CD117/KIT exhibited significant negative correlations with SLC2A1 and SLC2A5. CD133/PROM1 exhibited significant positive correlation with HK2, PGK1, PDK1, and enolase1 (ENO1). ALDH1A1 and SOX2 positively correlated with SLC2A3, whereas POU5F1/OCT4 and NANOG positively correlated with SLC2A2. However, NOTCH1 did not exhibit any significant correlation with glycolytic enzymes or glycolysis ( Figure 3D and Table S2).  their parental cells in GSE28799, and undifferentiated vs. differentiated OCSCs in GSE64999, respectively. Significance was determined using multiple t-test and Holm-Sidak method with each row analyzed individually with p < 0.05, without assuming a consistent SD. (C-E) Bar graphs represent the prevalence of positive (red) and negative (blue) correlations of the indicated markers in OVCAR3-spheroids and the glycolytic enzymes in Supplement Tables S1-2 and Table 1. * indicates the number of significant correlations either positive or negative.

Correlation Between OCSCs Markers and TCA Cycle
The TCA cycle is a hub for the integration of multiple catabolic and anabolic pathways, as glycolysis, gluconeogenesis, mitochondrial electron transport chain, fatty acid, and cholesterol synthesis as well as glutamine metabolism. TCA cycle generates metabolic intermediates that are not only critical for anabolic and catabolic pathways and redox homeostasis, but are also implicated in the regulation of transformation, carcinogenesis, inflammation, and immunity [22,[56][57][58]. The expression and function of the TCA enzymes and metabolites in cancer cells in general and OvCa in particular are unfolding; however, in OCSCs, these are not yet unraveled. Therefore, we sought to determine the association of TCA enzymes with OCSC model systems and in correlation with OCSC markers in patients' specimens and model systems. We further analyzed the expression of the transcripts of TCA cycle enzymes in OVCAR3-stem cells and their parental controls in GSE28799. We found that aconitase 1 (ACO1), isocitrate dehydrogenase (IDH1), IDH3A, succinate-CoA ligase GDPforming subunit beta (SUCLG2), and malate dehydrogenase 2 (MDH2) were significantly upregulated in OVCAR3-stem cells, whereas malic enzyme 2 (ME2) was significantly downregulated ( Figure 4A). In GSE64999, no significant differences were found between the transcripts of TCA cycle enzymes between differentiated and undifferentiated OCSCs ( Figure 4B). In GSE28799, there was a correlation between the expression of the enzyme transcripts and OCSC markers in OVCAR3-stem cells revealed a trend towards positive correlations with TCA enzymes. NANOG positively correlated with ten enzymes with significance only with CS. ALDH1A1 exhibited positive though insignificant correlations with nine enzymes. Significance was determined using multiple t-test and Holm-Sidak method with each row analyzed individually with p < 0.05, without assuming a consistent SD. (C-E) Bar graphs represent the prevalence of positive (red) and negative (blue) correlations of the indicated markers in OVCAR3-spheroids and the glycolytic enzymes in Supplement Tables S1-S2 and Table 1. * indicates the number of significant correlations either positive or negative.
Analysis of patients' tumors, from TCGA data revealed that CD44 and NOTCH1 were the top CSC markers whose expression exhibited positive correlations with glucose transporters and glycolytic enzymes ( Figure 3E and Table 1). NOTCH1 positively correlated with six glucose transporters (SLC2A1-6), HK1 and HK2, the rate limiting enzymes in glycolysis, and nine out of thirteen mapped enzymes involved in glycolysis. Similarly, CD44 was significantly positively correlated with three out of the six glucose transporters, hexokinases as well as nine out of thirteen enzymes involved in glycolysis. CD117/KIT was positively correlated with SLC2A1-4, HK2, and two out of thirteen mapped glycolytic enzymes, 6-phosphofructo-2-kinase/fructose-2,6-biphosphatase 3 (PFKFB3) and phosphoglycerate dehydrogenase (PHGDH); CD133/PROM1 correlated with HK1 and HK2, and two out of the thirteen enzymes in glycolysis. ALDH1A1 correlated with HK1, SLC2A1, and seven out of the thirteen enzymes involved in glycolysis. SOX2 correlated positively with SLC2A3 and four out of the thirteen enzymes in glycolysis. Other cell stemness markers as CD24, OCT4, and NANOG did not significantly correlate with glucose transporters, but positively correlated with only two of the glycolytic enzymes.
Taken together, data from GSE28799 and TCGA show positive correlations between the transcripts of OCSC markers and glycolytic markers, whereas in GSE64999, fewer positive correlations were observed with a tendency towards a negative correlation with glycolytic markers. These discrepancies may be due to technical variabilities in experimental model systems using different cell lines under diverse cell growth media; whereas in patients' tumor samples, the correlations are influenced by tumor heterogeneity and environmental cues. The high glucose and supplemental growth factors and serum present in culture media, combined with normoxic conditions and other pharmacologic inhibitors as Rho-associated protein kinase (ROCK) inhibitor used in the OCSC model in GSE64999 [42] do not represent the glucose and oxygen deprived conditions resulting from dysfunctional vasculature and lack of perfusion that characterizes the OvCa tumor microenvironment (TME) [52]. Specifically, OvCa cells are highly sensitive to oxygen conditions; it is known that to survive hypoxic conditions, OCSCs are forced to upregulate their stem-like properties and behave more aggressively when brought back to higher oxygen environments [53]. Hypoxia also induced a decrease in OXPHOS and fatty-acid desaturation [52]. Further limitations that should be considered are that the in vitro culture models lack vital extracellular matrix components including immune cells, stromal cells, and structural matrices, all of which are known to influence OCSC expression, growth, and differentiation. The absence of biophysical properties including interstitial flow, oxygen partial pressure, and surrounding environmental stiffness as well as the lack of biochemical cues are all factors that must be considered when comparing cell culture systems to in vivo CSC properties [54]. In CCLE databases, the correlations represent the behavior of OvCa cell lines under optimal growth conditions. The expression of CD44 positively and significantly correlated with most glycolytic phenotype in OCSC model systems, patients' tumors as well as non-stem OvCa cells. This finding is consistent with earlier reports that CD44 is crucial for the regulation of glycolytic metabolism [55].

Correlation between OCSCs Markers and TCA Cycle
The TCA cycle is a hub for the integration of multiple catabolic and anabolic pathways, as glycolysis, gluconeogenesis, mitochondrial electron transport chain, fatty acid, and cholesterol synthesis as well as glutamine metabolism. TCA cycle generates metabolic intermediates that are not only critical for anabolic and catabolic pathways and redox homeostasis, but are also implicated in the regulation of transformation, carcinogenesis, inflammation, and immunity [22,[56][57][58]. The expression and function of the TCA enzymes and metabolites in cancer cells in general and OvCa in particular are unfolding; however, in OCSCs, these are not yet unraveled. Therefore, we sought to determine the association of TCA enzymes with OCSC model systems and in correlation with OCSC markers in patients' specimens and model systems. We further analyzed the expression of the transcripts of TCA cycle enzymes in OVCAR3-stem cells and their parental controls in GSE28799. We found that aconitase 1 (ACO1), isocitrate dehydrogenase (IDH1), IDH3A, succinate-CoA ligase GDP-forming subunit beta (SUCLG2), and malate dehydrogenase 2 (MDH2) were significantly upregulated in OVCAR3-stem cells, whereas malic enzyme 2 (ME2) was significantly downregulated ( Figure 4A). In GSE64999, no significant differences were found between the transcripts of TCA cycle enzymes between differentiated and undifferentiated OCSCs ( Figure 4B). In GSE28799, there was a correlation between the expression of the enzyme transcripts and OCSC markers in OVCAR3-stem cells revealed a trend towards positive correlations with TCA enzymes. NANOG positively correlated with ten enzymes with significance only with CS. ALDH1A1 exhibited positive though insignificant correlations with nine enzymes. CD117/KIT positively correlated with six enzymes with significance only with IDH2 and MDH2. Other factors exhibited positive though insignificant correlations with TCA enzymes. CD44 and CD24 exhibited similar five positive and six negative correlation patterns with significant negative correlation with ME2. However, significant negative correlations were observed between CD133 with IDH3B, and SOX2 with IDH2. Neither POU5F1/OCT4 nor NOTCH1 bear significant correlation with key TCA enzymes ( Figure 4C and Table S3). Correlating the expression of CSC transcripts with those transcripts of TCA in undifferentiated and differentiated spheroids in GSE64999 revealed a more negative trend. CD44 positively correlated with five enzymes with significance only with ACO1. ALDH1A1 and SOX2 exhibited identical five positive and six negative profiles with significant positive correlation with SUCLG1. POU5F1/OCT4 positively correlated with four enzymes, of them significance was noted with citrate synthase (CS), IDH3B, and SUCLG1. NANOG negatively correlated with seven enzymes with significant negative correlation with MDH2; while NOTCH1 negatively correlated with six enzymes with significance with ACO1. CD24, CD117, and CD133 were all not found to be significantly correlated with enzymes involved in the TCA cycle ( Figure 4D and Table S4).
correlate with any enzyme in the TCA cycle with a trend towards negative correlations with six enzymes ( Figure S3).
These data indicate that the reliance of OCSCs cells on the TCA is demonstrated in TCGA data by the positive correlations between OCSC markers and TCA enzymes. Spheroids from characterized OVCAR3 not only exhibited significant upregulation of most of the TCA enzymes but also exhibited positive correlations with OCSC markers and TCA enzymes suggestive of their reliance on TCA cycle. The discrepancies between the two OCSC model systems further highlight the influence of environmental cues on the expression of their metabolic programing and plasticity.
Conversely, OCSC-like spheroids relied on anaerobic glycolysis and the PPP with decreased reliance on the TCA cycle compared to their parental cells [59]. Another study demonstrated that in SKOV3 and lung A549 model systems, cell populations with high telomerase activity, exhibited significantly higher ability to form spheroids (and hence CSC phenotype), and exhibited enhanced glycolysis, OXPHOS, and increased mitochondrial mass [60]. Consistently, a subset of cancer cells with stem like properties, called side population (SP) cells, were identified in OvCa and other cancer types and exhibited higher glycolytic activity than non-SP cells. The percentage of SP cells significantly increased in glucose-rich conditions. Paradoxically, glucose deprivation or the presence of a glycolytic inhibitor, 3-BrOP, significantly decreased the number of SP cells and decreased their tumor forming ability in mice xenografts [61]. Glucose-induced CSC-like SP proliferation was mediated through an ATP-dependent suppression of AMPK and activation of the AKT pathway [61].  Supplement Tables  S3-S4 and Table 2. * indicate the number of significant correlations either positive or negative. Analysis and correlation of the OCSCs' markers with TCA enzymes from TCGA datasets ( Figure 4E and Table 2) revealed that NOTCH1 positively correlated with all TCA enzymes with significant correlations with all enzymes except SUCLG1/2. Similarly, CD133 positively correlated with all eleven TCA enzymes with significance observed with CS and ACO1. CD44 positively and significantly correlated with ten out of the eleven enzymes namely CS, ACO1/2, IDH1/2/3A, SUCLG1/2, MDH2, and ME2, while CD24 positively correlated with ten out of the eleven enzymes with significant correlations with CS, ACO2, CD117, SUCLG1, and ME2. ALDH1A1 positively correlated with nine out of the eleven enzymes with significance observed with CS, ACO1, IDH1, and SUCLG1. POU5F1/OCT4 positively correlated with eight out of the eleven enzymes with significance observed with ACO1/2, SUCLG2, and ME2. CD117/KIT positively correlated with seven out of eleven enzymes with positive significance observed with CS, ACO1, and IDH2 and significant negative correlation with MDH2. SOX2 positively correlated with seven enzymes with significance observed only with ACO1, whereas NANOG positively correlated with six enzymes with significance observed with CS.
Correlation of OvCa cell line data acquired from CCLE demonstrated a generalized trend of negative correlation between OCSC markers and TCA cycle enzymes. Although CD117/KIT positively correlated with all enzymes, no significance was noted. SOX2 positively correlated with eight enzymes, of them, only CS and IDH3A were significant. ALDH1A1 and POU5F1/OCT4 negatively correlated with nine enzymes with significant negative correlation with CS2 and ME2 as well as IDH3A, respectively. Similarly, CD133/PROM1 negatively correlated with nine enzymes with significance only with ACO1 and SUCLG2. Consistently, CD44 negatively correlated with eight enzymes with significance observed with ACO2 and MDH2; whereas NANOG negatively correlated with five enzymes with significant negative correlation with SUCLG2. NOTCH1 did not significantly correlate with any enzyme in the TCA cycle with a trend towards negative correlations with six enzymes ( Figure S3).
These data indicate that the reliance of OCSCs cells on the TCA is demonstrated in TCGA data by the positive correlations between OCSC markers and TCA enzymes. Spheroids from characterized OVCAR3 not only exhibited significant upregulation of most of the TCA enzymes but also exhibited positive correlations with OCSC markers and TCA enzymes suggestive of their reliance on TCA cycle. The discrepancies between the two OCSC model systems further highlight the influence of environmental cues on the expression of their metabolic programing and plasticity.
Conversely, OCSC-like spheroids relied on anaerobic glycolysis and the PPP with decreased reliance on the TCA cycle compared to their parental cells [59]. Another study demonstrated that in SKOV3 and lung A549 model systems, cell populations with high telomerase activity, exhibited significantly higher ability to form spheroids (and hence CSC phenotype), and exhibited enhanced glycolysis, OXPHOS, and increased mitochondrial mass [60]. Consistently, a subset of cancer cells with stem like properties, called side population (SP) cells, were identified in OvCa and other cancer types and exhibited higher glycolytic activity than non-SP cells. The percentage of SP cells significantly increased in glucose-rich conditions. Paradoxically, glucose deprivation or the presence of a glycolytic inhibitor, 3-BrOP, significantly decreased the number of SP cells and decreased their tumor forming ability in mice xenografts [61]. Glucose-induced CSC-like SP proliferation was mediated through an ATP-dependent suppression of AMPK and activation of the AKT pathway [61].

OXPHOS in Ovarian Cancer Stem Cells
OXPHOS is a process by which NADH and FADH 2 generated in TCA cycle transfer electrons to the mitochondrial ETC and occurs via several redox reactions that take place in the inner mitochondrial membrane (IMM). These reactions facilitate the generation of an electrochemical proton (H + ) gradient, which subsequently drives the synthesis of energy rich adenosine triphosphate (ATP) by ATP synthases [62].
CSCs may exhibit highly glycolytic or OXPHOS phenotypes with plasticity of metabolic switch between phenotypes depending not only on cancer type and environmental cues, but also upon glucose starvation or OXPHOS blockade [55,63]. For example, CD44 + /CD117 + OCSCs isolated from patients' ascitic fluid exhibited enhanced glucose uptake with heightened OXPHOS. Upon glucose starvation, these OCSCs underwent complete quiescence and down-regulated most metabolic activities, while maintaining an OXPHOS profile [63]. In further support of this, cellular bio-energetic profiling of established and patient-derived OvCa cell lines showed that chemo-sensitive cancer cell lines displayed a glycolytic phenotype while their chemo-resistant counterparts developed an adaptive switch between OXPHOS and glycolysis [64]. Integrated proteomic, metabolomic, and bioenergetic analyses of OVTOKO cells growing in 3D spheroids and expressing high ALDH1A1 levels, showed a high-OXPHOS subtype relying on OXPHOS, supported by glutamine and FAO and a low-OXPHOS subtype that was mainly glycolytic [65].
Comparative analysis of the expression of ETC complexes in OVCAR3-OCSCs in GSE28799 revealed significant upregulation of most of the transcripts of complex I (NADH:ubiquinone oxidoreductases, NDUFA Figure 5A), with inconsistent changes of enzymes of complex II (succinate dehydrogenases, SDHA-D, Figure 5B). Significant increase was observed in six out of eight transcripts of complex III enzymes (ubiquinol: cytochrome C oxidoreductases, UQCR) in OVCAR3-OCSCs compared to parental cells ( Figure 5C), but only in four out of thirteen transcripts of complex IV enzymes (cytochrome C oxidases, COX4I1, COX6B1, COX7A1, and COX7C, Figure 5D). Whereas, OVCAR3-OCSCs exhibited significant increase in all the transcripts of complex V enzymes (ATP synthases) compared to parental cells ( Figure 5E). In GSE64999, undifferentiated OCSCs exhibited significant upregulation only in one complex I enzyme, NDUFA2, compared to differentiated OCSCs ( Figure 5F) with no significant changes in complex II enzymes ( Figure 5G). In addition, undifferentiated OCSCs exhibited an upregulation trend of complex III enzyme transcripts with significance observed in three out of the eight enzymes investigated ( Figure 5H), whereas for complex IV, they exhibited significant upregulation of COX7C with an upregulation trend in the other transcripts ( Figure 5I). Concordantly, the transcripts of ATP synthases were upregulated in undifferentiated OCSCs with significance observed only in ATP5E ( Figure 5J). Together, this data indicates that OCSC spheroids exhibit enrichment of the ETC mainly in complexes I, III, IV, and V, further supporting reliance on the OXPHOS phenotype. Interestingly, correlations between the expression of OCSC transcripts and the transcripts of the key ETC enzymes revealed that in OVCAR3-OCSCs in GSE28799, there was a trend of positive correlations with complex I enzymes with NOTCH1, NANOG, and ALDH1A1 exhibiting significant correlations with the transcripts of one of the twelve enzymes. However, there were inconsistent positive and negative correlations between the transcripts of OCSC markers and the transcripts of complexes II, III, IV, and V ( Figure 6A and Table S5). In contrast, in GSE64999 the transcripts of OCSC markers in undifferentiated OCSCs exhibited inconsistent positive and negative correlations with transcripts of complexes I, III, IV, and V, with CD24 and CD44 exhibiting positive though insignificant correlations with complex II enzymes ( Figure 6B and Table S6). Interestingly, the correlation patterns of the transcripts of OCSC markers and ETC enzymes in patients' tumors in TCGA data ( Figure 6C and Table 3) phenocopied those of undifferentiated OCSCs. These data suggest that OCSC markers do not correlate with those of ETC enzymes in the model systems or in patients' tumors. The inconsistencies in the positive and negative correlations and their significance may be due to the existence of multiple isoforms for ETC enzymes with redundant activity, or compensation of the activity by mitochondrially encoded ETC enzymes.
Of note, is that analysis of OvCa cell line data curated from CCLE also revealed similar inconsistent pattern of positive and negative correlations with ETC enzymes ( Figure S4). CD44 exhibited negative correlations with all investigated enzymes of complexes I and V with significant negative correlations with NDUFS1 and NDUFS2 of complex I. CD44 also exhibited negative correlations with four out of nine enzymes in complex III with significant negative correlation with UQCRC2. In addition, CD44 exhibited negative though insignificant correlation with two out of four complex II enzymes and four out of thirteen complex IV enzymes. SOX2 exhibited positive correlations with all ETC complexes with significant positive correlations with NDUFS3 and NDUFV1 of complex I, CYSC of complex III, and ATP5J of complex V. POU5F1/OCT4 negatively correlated with most of the enzymes of complexes I-V with significant negative correlations with NDUFA10, NDUFAF4, and NDUFAS1-3 of complex I, SDHC of complex II, UQCRC1 and UQCRH of complex III, COX5A of complex IV as well as ATP5D of complex V. However, POU5F1/OCT4 only positively correlated with COX7A1 of complex IV. PROM1/CD133 exhibited a trend of a negative correlations with complexes II, III, and V with significant negative correlation with NDUFA10 of complex I and UQCR10 of complex III as well as significant positive correlation with COX4I1 and COX6B1 of complex IV. KIT/CD117 exhibited positive correlation with complex I, III, and V, and a negative correlation with complex IV, with a significant positive correlation only with NDUFS1 of complex I. ALDH1A1 exhibited negative correlation with complex I and II enzymes with significant negative correlation with NDUFA10 of complex I, and COX6A2 and COX7A1 of complex IV. However, ALDH1A1 positively correlated with COX6A1. Furthermore, NANOG exhibited positive correlation with complexes I, II, and V, with significant positive correlation only with NDUFA11, SDHC, and ATP5J. Finally, NOTCH1 exhibited negative correlations with complexes II, IV, and V, with significant negative correlation only with COX4I2 and COX6B1 as well as ATP5C1. Together, data from CCLE further imply that the expression of the transcripts of OCSC markers CD44, POU5F1/OCT4, and NOTCH1 exhibit most of the significant negative correlations.
Cancers 2020, 12, x 15 of 31 V, with CD24 and CD44 exhibiting positive though insignificant correlations with complex II enzymes ( Figure 6B and Table S6). Interestingly, the correlation patterns of the transcripts of OCSC markers and ETC enzymes in patients' tumors in TCGA data ( Figure 6C and Table 3) phenocopied those of undifferentiated OCSCs. These data suggest that OCSC markers do not correlate with those of ETC enzymes in the model systems or in patients' tumors. The inconsistencies in the positive and negative correlations and their significance may be due to the existence of multiple isoforms for ETC enzymes with redundant activity, or compensation of the activity by mitochondrially encoded ETC enzymes. Of note, is that analysis of OvCa cell line data curated from CCLE also revealed similar inconsistent pattern of positive and negative correlations with ETC enzymes ( Figure S4). CD44 exhibited negative correlations with all investigated enzymes of complexes I and V with significant negative correlations with NDUFS1 and NDUFS2 of complex I. CD44 also exhibited negative

Correlation of OCSC Markers With Lipid Metabolism
Lipid-associated pathways are essential for the maintenance of CSCs [66]. Lipid metabolism is inherently bound to the glucose and amino acid metabolic pathways in order to meet the increasing bio-energetic needs of CSCs. Specifically, epithelial OCSCs utilize OXPHOS and FAO to overcome glucose starvation, and this metabolic trait confers resistance to chemotherapy [64]. It has been shown that when CD44 + CD117 + CSCs were isolated from OvCa primary culture, and detached from a monolayer into a suspension state, they reprogramed their metabolism from glycolysis to TCA cycle with active lipid metabolism compared to adherent cultures [67]. Consistently, ALDH + CD133 + OCSCs and spheroids have increased levels of unsaturated lipids compared to non-CSCs in monolayers [68]; an effect that was inhibited by inhibition of lipid desaturases that significantly reduced stemness and eliminated sphere formation in vitro and tumor initiation in vivo [68]. Moreover, overexpression of fatty acid synthase (FASN), a key enzyme in lipogenesis correlated with poor disease outcome [69]. Accordingly, the inhibition of FASN reversed platinum resistance in resistant OvCa cells [70].
Analysis of OVCAR3-OCSCs transcripts in GSE28799 dataset, revealed that compared to parental OVCAR3, OVCAR3-OCSCs exhibited significant increase in the transcripts of fatty acid transporter (CD36), acetyl CoA carboxylase A (ACACA) that catalyzes the first committed step in fatty acid (FA) synthesis from acetyl CoA to malonyl CoA [71], stearoyl-CoA desaturase (SCD), that is also involved in FA biosynthesis, mainly unsaturated FAs, as well as carnitine palmitoyltransferase 1A (CPT1A) that catalyzes the first committed step in FAO [72,73] (Figure 7A). Analysis of the transcripts in undifferentiated vs. differentiated OCSC models in GSE64999 dataset revealed a significant

Correlation of OCSC Markers With Lipid Metabolism
Lipid-associated pathways are essential for the maintenance of CSCs [66]. Lipid metabolism is inherently bound to the glucose and amino acid metabolic pathways in order to meet the increasing bio-energetic needs of CSCs. Specifically, epithelial OCSCs utilize OXPHOS and FAO to overcome glucose starvation, and this metabolic trait confers resistance to chemotherapy [64]. It has been shown that when CD44 + CD117 + CSCs were isolated from OvCa primary culture, and detached from a monolayer into a suspension state, they reprogramed their metabolism from glycolysis to TCA cycle with active lipid metabolism compared to adherent cultures [67]. Consistently, ALDH + CD133 + OCSCs and spheroids have increased levels of unsaturated lipids compared to non-CSCs in monolayers [68]; an effect that was inhibited by inhibition of lipid desaturases that significantly reduced stemness and eliminated sphere formation in vitro and tumor initiation in vivo [68]. Moreover, overexpression of fatty acid synthase (FASN), a key enzyme in lipogenesis correlated with poor disease outcome [69]. Accordingly, the inhibition of FASN reversed platinum resistance in resistant OvCa cells [70].
Analysis of OVCAR3-OCSCs transcripts in GSE28799 dataset, revealed that compared to parental OVCAR3, OVCAR3-OCSCs exhibited significant increase in the transcripts of fatty acid transporter (CD36), acetyl CoA carboxylase A (ACACA) that catalyzes the first committed step in fatty acid (FA) synthesis from acetyl CoA to malonyl CoA [71], stearoyl-CoA desaturase (SCD), that is also involved in FA biosynthesis, mainly unsaturated FAs, as well as carnitine palmitoyltransferase 1A (CPT1A) that catalyzes the first committed step in FAO [72,73] (Figure 7A). Analysis of the transcripts in undifferentiated vs. differentiated OCSC models in GSE64999 dataset revealed a significant increase in SCD in the undifferentiated OCSCs with an increase, though insignificant, in the enzymes involved in FA synthesis and oxidation as well as CD36 ( Figure 7B). This data indicates that OCSCs exhibit a lipid metabolic phenotype that is consistent with the OvCa in the lipid-rich peritoneal TME.
Correlation of the putative OCSC markers and lipid transporters and enzymes in OVCAR3-OCSCs in GSE28799 dataset revealed that NOTCH1, NANOG, OCT4, ALDH1A1, and CD117 exhibited more positive, though insignificant, correlations with lipid metabolism markers than CD44, CD24, CD133, or SOX2. The latter markers exhibited a trend towards negative correlations, with CD44 and CD24 exhibiting significant negative correlations with fatty acid-binding protein 4 (FABP4) ( Figure 7C and Table S7). In GSE64999, CD44, CD133, and ALDH1A1 exhibited more positive correlations with lipid metabolic enzymes with significance between CD44 and CPT1B as well as CD133 and CPT1A. OCT4 exhibited significant positive correlation with CD36, but negatively correlated with ACACA. NANOG was negatively correlated with ACACA ( Figure 7D and Table S8). TCGA data indicated significant positive correlation between NOTCH1 and CD44, with lipid metabolic markers, followed by CD133, ALDH1A1, and CD117 with significant positive correlations exhibited by SOX2 ( Figure 7E and Table 4). Conversely, in CCLE datasets, CD44 negatively correlated with ACACB, CPT1A, and CPT2. CD117 positively correlated with ACACB, CPT2, FASN, and SCD1. CD133 positively correlated with CPT1A. ALDH1A1 was negatively correlated with ACACB. SOX2 was positively correlated with CPT1A and CPT2 and negatively correlated with FABP4. OCT4 was negatively correlated with FABP4, ATP citrate lyase (ACLY), ACACA, FASN, and SCD1 and positively correlated with CPT1A/B. NANOG was positively correlated with CPT2. NOTCH1 was not significantly correlated with the enzymes investigated ( Figure S5). These data further support the distinct metabolic phenotypes and "metabolic plasticity" of OCSCs as the acquisition of the OCSC phenotype is associated with enrichment of a lipid metabolic signature. These may explain the ability of OCSCs to survive in the lipid-rich peritoneal TME leading to recurrence after optimal or suboptimal surgery and chemotherapy.   Supplement Tables S7-8 and Table 4. * indicate the number of significant correlations either positive or negative.

Glutamine/Glutamate Metabolism in OCSCs
Glutamine metabolism provides the carbon and amino-nitrogen that are necessary for biosynthesis of amino acids, nucleotides, and lipids [74][75][76]. As an anabolic process, glutaminolysis promotes the production of macromolecules with lower energetic potential [77]. Furthermore, glutamine enters cancer cells through the alanine, serine, cysteine transporter 2 (ASCT2; also known as SLC1A5) and is subsequently hydrolyzed to glutamate and ammonia by glutaminase (GLS). Glutamate has dual roles: while it can be combined with cysteine and glycine to form the reduced glutathione (GSH), a major antioxidant that regulates oxidative stress [78], it can also be converted into αketoglutarate (αKG) by glutamate dehydrogenases (GLUD) to provide TCA cycle intermediates and, hence, energy production [79]. Furthermore, OCSCs like other CSCs exploit glutamine metabolism for therapeutic resistance. Glutaminolysis significantly correlated with poor survival of OvCa patients [80]. Additionally, glutamine and glutamate concentrations were higher in spheroids than in adherent OCSCs [65]. Targeting glutamine metabolism using a pan-transaminase inhibitor hindered the growth of spheroids from ovarian clear cell carcinomas with inhibition of the mTOR pathway [65], suggesting a promising therapeutic strategy.
Analysis of the expression of the enzymes involved in glutamine metabolism in OCSCs and their parental cells in GSE28799 revealed that the glutamine transporter SLC1A1 and the enzyme GLS, were significantly upregulated in OVCAR3 spheroid-derived OCSCs compared to their parental OVCAR3 cells ( Figure 8A). In contrast, there was no significant difference in glutamine metabolism enzymes between undifferentiated and differentiated spheroids in GSE64999 ( Figure 8B). In both  Supplement Tables S7-S8 and Table 4. * indicate the number of significant correlations either positive or negative.

Glutamine/Glutamate Metabolism in OCSCs
Glutamine metabolism provides the carbon and amino-nitrogen that are necessary for biosynthesis of amino acids, nucleotides, and lipids [74][75][76]. As an anabolic process, glutaminolysis promotes the production of macromolecules with lower energetic potential [77]. Furthermore, glutamine enters cancer cells through the alanine, serine, cysteine transporter 2 (ASCT2; also known as SLC1A5) and is subsequently hydrolyzed to glutamate and ammonia by glutaminase (GLS). Glutamate has dual roles: while it can be combined with cysteine and glycine to form the reduced glutathione (GSH), a major antioxidant that regulates oxidative stress [78], it can also be converted into α-ketoglutarate (αKG) by glutamate dehydrogenases (GLUD) to provide TCA cycle intermediates and, hence, energy production [79]. Furthermore, OCSCs like other CSCs exploit glutamine metabolism for therapeutic resistance. Glutaminolysis significantly correlated with poor survival of OvCa patients [80]. Additionally, glutamine and glutamate concentrations were higher in spheroids than in adherent OCSCs [65]. Targeting glutamine metabolism using a pan-transaminase inhibitor hindered the growth of spheroids from ovarian clear cell carcinomas with inhibition of the mTOR pathway [65], suggesting a promising therapeutic strategy.
Analysis of the expression of the enzymes involved in glutamine metabolism in OCSCs and their parental cells in GSE28799 revealed that the glutamine transporter SLC1A1 and the enzyme GLS, were significantly upregulated in OVCAR3 spheroid-derived OCSCs compared to their parental OVCAR3 cells ( Figure 8A). In contrast, there was no significant difference in glutamine metabolism enzymes between undifferentiated and differentiated spheroids in GSE64999 ( Figure 8B). In both datasets, very few correlations between stem cell markers and glutamine transporters were noted. In GSE28799, both CD24 and CD44 positively correlated with SLC1A1, while CD133 positively correlated with GLS and negatively correlated with GLUD2. ALDH1A1 positively correlated with SLC1A2 while OCT4 positively correlated with SLC1A3 ( Figure 8C and Table S9). In GSE64999, CD133 positively correlated with GLS. CD24 negatively correlated with GLUD1. ALDH1A1 and SOX2 both negatively correlated with SLC1A6. NOTCH1 negatively correlated with SLC1A1 ( Figure 8D and Table S10).  Supplement Tables S9-10 and Table 5. * indicate the number of significant correlations either positive or negative.
While our findings are based on OCSC transcript levels, it is necessary to clarify that the transcript level and proteome levels may not necessarily correlate due to post transcriptional modifications, including splice variation. Alternative splicing accounts for the major difference between the number of protein-coding genes and the number of proteins that are ultimately produced from translation [81]. Through the various splicing mechanisms, as constitutive splicing, mutually exclusive exons, cassette alternative exon, alternative 3′ or 5′ splice site, and intron retention, splicing plays a key role in protein diversification. In effect, by allowing several functionally distinctive proteins to be encoded by the same original gene, splicing does not allow for the direct conversion of the transcriptomic signature into the proteomic signature [81]. Moreover, due to intraclonal heterogeneity, it is challenging to assign a specific set of stemness markers that define the different subsets of CSCs [82]. Such heterogeneity includes quiescent, precancerous, primary, migrating, chemo-resistant, and radio-resistant CSCs [83]. In OCSCs specifically, multiple subtypes have been identified including proliferating CSCs expressing OCT4, ALDH1/2, CD44, and LGR5/Ki67 +  Tables S9-S10 and Table 5. * indicate the number of significant correlations either positive or negative. Further analysis of patients' tumors from TCGA data indicated that CD44, ALDH1A1, and NOTCH1 exhibited significantly positive correlations with enzymes involved in glutamine metabolism. CD44 positively correlated with SLC1A1-5, GLS, and GLUD1/2. Similarly, ALDH1A1 exhibited significant positive correlation with four out of the seven glutamine transporters, along with GLS and GLUD1/2. NOTCH1 significantly positively correlated with five out of the seven glutamine transporters, along with GLS and GLUD1/2. CD133 positively correlated with five out of the seven glutamine transporters, while CD117 correlated with three out of seven glutamine transporters, as well as with GLS and GLUD1/2. SOX2 and OCT4 positively correlated with three and two glutamine transporters, respectively. OCT4 also positively correlated with GLS and GLUD1/2. CD24 positively correlated with one glutamine transporter, as well as SLC1A1, but no other significant correlations with other glutamine transporters. NANOG was not positively correlated with any of the glutamine transporters yet was correlated with GLS and GLUD2 ( Figure 8E and Table 5). Of note is that transcriptomic profiling of data of the OvCa cell lines in CCLE demonstrated that the transcripts of CD44 and OCT4 both positively correlated with SLC1A1 and GLS; whereas OCT4 negatively correlated with SLC1A2 and SLC1A5. CD117 positively correlated with SLC1A3, while CD133 positively correlated with SLC1A5. Additionally, ALDH1A1 negatively correlated with SLC1A7, while NANOG negatively correlated with SLC1A1 ( Figure S6). In summary, data from TCGA show positive correlations between the transcripts of OCSC markers and glutamine metabolism enzymes, whereas in OCSC experimental models of GSE64999, GSE28799, and CCLE, the number of positive and negative correlations observed was significantly lower implying that glutamine metabolism is indispensable for OCSCs and OvCa cells grown in normal conditions, whereas in their complex TME, OvCa cells utilize glutamine metabolism in an anaplerotic reaction for generating energy and metabolic intermediates to fulfill their increasing demands.
While our findings are based on OCSC transcript levels, it is necessary to clarify that the transcript level and proteome levels may not necessarily correlate due to post transcriptional modifications, including splice variation. Alternative splicing accounts for the major difference between the number of protein-coding genes and the number of proteins that are ultimately produced from translation [81]. Through the various splicing mechanisms, as constitutive splicing, mutually exclusive exons, cassette alternative exon, alternative 3 or 5 splice site, and intron retention, splicing plays a key role in protein diversification. In effect, by allowing several functionally distinctive proteins to be encoded by the same original gene, splicing does not allow for the direct conversion of the transcriptomic signature into the proteomic signature [81]. Moreover, due to intra-clonal heterogeneity, it is challenging to assign a specific set of stemness markers that define the different subsets of CSCs [82]. Such heterogeneity includes quiescent, precancerous, primary, migrating, chemo-resistant, and radio-resistant CSCs [83]. In OCSCs specifically, multiple subtypes have been identified including proliferating CSCs expressing OCT4, ALDH1/2, CD44, and LGR5/ Ki67 + markers and non-proliferating CSCs expressing SSEA4 + /Ki67 − or ALDH1/2 + /Ki67 − markers. Other subsets of OCSCs include CD133 + and CD44 + /CD117 + all of which can be influenced by tumor microenvironmental cues, upregulation of signaling pathways, and tumor grade, making it difficult to confine OCSCs to a standardized list of markers [4,82].

Conclusions
In this study, we sought to investigate the predictive significance of the gene expression of OCSCs in correlation with transcripts of metabolic pathways that are upregulated in HGSOC, that could inform about the metabolic plasticity of tumors, predict the behavior of recurrent or chemo-resistant disease, and ultimately may guide on therapeutics targeting the enriched pathways in combination with standard of care therapy. Comparison of transcripts of OCSC markers with those of enzymes involved in metabolic pathways in OCSC spheroid model systems with patients' data revealed several distinctive differences in the expression of the stem cell markers. Correlations with energy metabolic pathways, not only between the two model systems investigated, but also with patients' tumors and cumulative behavior of OvCa cell data from CCLE were elucidated. While these differences between spheroid model systems could be attributed to technical variability in establishing and maintaining of OCSCs as spheroids, nevertheless, they highlight the significance of OvCa heterogeneity, and the influence of micro-environmental cues on metabolic plasticity of OCSCs that may influence their survival and self-renewal capabilities.
It is noteworthy that the limitation of the present study is the paucity of data from established model systems of OCSCs with adequate biological replica that can be used for systematic biological, metabolic as well as transcriptomic analyses to bridge the gap of metabolic programing in OCSCs.
Supplementary Materials: The following are available online at http://www.mdpi.com/2072-6694/12/5/1267/s1, Figure S1: Pearson's correlation between OvCa stem cell (OCSC) markers. Figure S2: Correlations of OCSC markers with glucose transporters and glycolysis enzymes in OvCa cell lines' data curated from CCLE. Figure  S3: Correlations of OVSC markers with TCA cycle enzymes in OvCa cell lines' data curated from CCLE. Red indicates positive correlations. Figure S4: Pearson Correlation between the expression of OCSC transcripts and the transcripts of the key enzymes involved in ETC in OvCa cell lines' data curated from CCLE. Figure S5: Correlations of OCSC markers with fatty acid transporters and lipid metabolism enzymes in OvCa cell lines' data curated from CCLE. Figure S6: Correlations of OCSC markers with glutamine/glutamine transporters and metabolic enzymes in OvCa cell lines' data curated from CCLE. Table S1: Pearson Correlation between the expression of CSC transcripts and the transcripts of glucose transporters and the key enzymes involved in glycolysis in OVCAR3 spheroids stem cells curated from GSE28799. Table S2: Pearson Correlation between the expression of CSC transcripts and the transcripts of glucose transporters and the key enzymes involved in glycolysis in undifferentiated spheroids stem cells curated from GSE64999. Table S3: Pearson Correlation between the expression of CSC transcripts and the transcripts of the key TCA enzymes in OVCAR3 stem cells from GSE28799. Table S4: Pearson Correlation between the expression of CSC transcripts and the transcripts of the key TCA enzymes in undifferentiated OCSCs from GSE64999. Table S5: Pearson Correlation between the expression of CSC transcripts and the transcripts of the key enzymes involved in the Electron Transport Chain in OvCa cell lines curated from GSE28799. Table S6: Pearson Correlation between the expression of CSC transcripts and the transcripts of the key enzymes involved in the Electron Transport Chain in OvCa cell lines curated from GSE64999. Table S7: Pearson Correlation between OCSC markers and lipid transporters and key enzymes in lipid metabolism in GSE28799. Table S8: Pearson Correlation between OCSC markers and lipid transporters and key enzymes in lipid metabolism in GSE64999. Table S9: Pearson Correlation between OCSC markers and glutamate transporters and key enzymes of glutamate/glutamine metabolism in GSE28799. Table S10: Pearson Correlation between OCSC markers and glutamate transporters and key enzymes in glutamate/glutamine metabolism in GSE64999.