Glycolysis, via NADH‐dependent dimerisation of CtBPs, regulates hypoxia‐induced expression of CAIX and stem‐like breast cancer cell survival

Adaptive responses to hypoxia are mediated by the hypoxia‐inducible factor (HIF) family of transcription factors. These responses include the upregulation of glycolysis to maintain ATP production. This also generates acidic metabolites, which require HIF‐induced carbonic anhydrase IX (CAIX) for their neutralisation. C‐terminal binding proteins (CtBPs) are coregulators of gene transcription and couple glycolysis with gene transcription due to their regulation by the glycolytic coenzyme NADH. Here, we find that experimental manipulation of glycolysis and CtBP function in breast cancer cells through multiple complementary approaches supports a hypothesis whereby the expression of known HIF‐inducible genes, and CAIX in particular, adapts to available glucose in the microenvironment through a mechanism involving CtBPs. This novel pathway promotes the survival of stem cell‐like cancer (SCLC) cells in hypoxia.

Tumour hypoxia is a key element of cancer biology. It correlates with resistance to radiotherapy and chemotherapy and increased metastatic spread. Tumour hypoxia is caused by avascular tumour growth and irregular blood flow, but also by the increased metabolic activity and proliferation of tumour cells or cells present in the tumour microenvironment [1,2]. It is estimated that more than 40% of breast cancers contain areas of hypoxia or anoxia [3,4].
Adaptive responses to hypoxia are mediated by the heterodimeric transcription factors hypoxia-inducible factor 1 (HIF-1) and HIF-2. In physiological oxygen conditions, the HIF-a subunit is hydroxylated at conserved proline and asparagine residues by oxygendependent prolyl hydroxylases (PHDs) and by factor inhibiting HIF (FIH) [5][6][7]. This allows for von Hippel-Lindau protein binding and subsequent proteasome-mediated degradation [8,9]. Consequently, under low oxygen pressure, the a subunit accumulates and translocates to the nucleus where it binds the b subunit to form an activating complex with the coactivator complex CBP/p300 at the hypoxia-responsive elements of target genes [6,[10][11][12]. HIF target genes contribute to aggressive tumour phenotypes by promoting increased invasiveness [13], angiogenesis [14], immune evasion [15] and pH regulation [16].
Hypoxia-inducible factor 1 is also an important modulator of tumour cell metabolism as its target genes promote glycolysis by maintaining glycolytic flux required for the synthesis of ATP [17,18] and metabolites for the synthesis of macromolecules such as nucleic acids [19]. Other HIF-1 targets suppress mitochondrial metabolism and the formation of reactive oxygen species [20,21]. Overall, this response enables tumour cell survival by enabling oxygen-independent energy generation and anabolic metabolism required by highly proliferating cells. Increased HIF-1 activity and the consequent driven changes in metabolism are therefore a major contributor to aggressive tumour growth, and the expression in tumours of increased levels of HIF-1a and the products of its key target genes relevant to glycolytic metabolism, such as GLUT1 and CAIX, are known markers of poor patient survival [22,23].
The magnitude of the oxygen-regulated HIF response is also modulated by metabolites in addition to oxygen that control the activity of the PHD enzymes [50,51]. Indeed, glycolysis itself can promote HIF-1 activity in an apparent feed-forward mechanism that may help match the magnitude of the HIF response with the availability of glucose and its metabolites in the hypoxic microenvironment [50]. However, less well understood is whether glycolysis may affect the expression of known HIF-inducible genes, to fine-tune the hypoxic response to meet the demands of the metabolic microenvironment. Here, therefore, we have addressed this question using breast cancer cell models, focussing on CtBPs as potential regulators of such a pathway, and CAIX [52] as a known key HIF-responsive gene that is essential for the maintenance of pH homeostasis in highly glycolytic cells.

Analysis of mRNA and protein
Western blotting was performed as described in [54]. Blots showing the effects of glycolysis or CtBPs on CAIX protein are representatives of three independent experiments. The following primary antibodies were applied to detect protein expression: mouse anti-human CtBP1 (clone E12) (Santa Cruz Biotechnology, Dallas, TX, USA), mouse anti-human CtBP2 (#612044. BD Transduction Laboratories, Franklin Lakes, NJ, USA), goat anti-human CtBP2 (E16; required for detection of C-terminally his-tagged CtBP2) (Santa Cruz Biotechnology), goat anti-human CAIX (AF2188. R&D systems, Minneapolis, MN, USA), mouse anti-human V5 tag (clone E10/V4RR) (Thermo Fisher Scientific, Waltham, MA, USA) and rabbit-anti human actin (A5060. Sigma-Aldrich). Bands were quantified using ImageJ, background-subtracted and normalised to actin; CAIX abundance following hypoxia or DMOG was set as 100% and abundance following further manipulation of glycolysis or CtBPs expressed relative to this. RNA from frozen cell pellets was extracted using the ReliaPrep TM RNA Cell Miniprep System Kit (Promega), following the manufacturer's instructions. cDNA was synthesised from RNA concentrations ranging between 250 and 1000 ng. MMLV reverse transcriptase, oligo dT15 primer and dNTPs were all from Promega. Reverse transcription was performed at 37°C for 1 h, followed by 5 min at 94°C. CAIX mRNA expression was determined by Taq Data were analysed using the comparative C T method relative to ACTB. For all experiments, graphs show result in the absence of either hypoxia or DMOG in light orange, and following hypoxia or DMOG in dark blue, with fills of the bars further indicative of the treatments shown on the graph axis.

Statistical analysis
GRAPHPAD PRISM version 6.01 (GraphPad, La Jolla, CA, USA) was used for statistical analysis. For RT-qPCR, statistical significance was calculated using one-way ANOVA and Fisher's LSD test. Unless stated otherwise, all quantitative data are n = 3 biological replicates, for qPCR, each of these being derived from technical duplicates, error bars = SEM. *P < 0.05, **P < 0.01, ***P < 0.001.

Regulation of the expression of known hypoxia and HIF-inducible genes by glycolysis
In order to establish a model system to examine the effect of glycolysis on the induction of hypoxia-responsive genes, we first performed a time course analysis for the induction of CAIX mRNA in three breast cancer cell lines following transfer to a 1% oxygen hypoxic culture environment (Fig. 1A). In two of the lines, MCF-7 and MDA-MB-231, reproducible significant induction was observed, and these were selected for further analysis. Induction was maximal after 17 hours and was consistently of substantially greater magnitude in MCF-7 than MDA-MB-231. To establish a role of HIF-1a in this induction of CAIX, we transfected MCF-7 cells with HIF1A siRNA prior to the hypoxic insult (Fig. 1B). The siRNA caused a significant, albeit incomplete, knockdown of the HIF1A transcript, and a significant, 27%, reduction in the abundance on CAIX mRNA in hypoxic cells. DMOG is a cell-permeable inhibitor of 2-oxoglutarate-dependent dioxygenases including prolyl-4-hydoxylases, and thus prevents HIF-1a from being targeted for VHLdependent ubiquitination and degradation in the presence of oxygen [57,58]. The DMOG-induced expression of CAIX was similarly muted by HIF1A siRNA. Thus, whilst both hypoxia and DMOG may potentially have effects on gene transcription independently of HIF-1a through effects on other targets, an equivalent proportion of the effects of both interventions on CAIX expression is HIF1a-dependent. We cannot determine from these experiments the extent to which HIF-1a-independent pathways may be involved, as the knockdown of HIF1A by the siRNA was incomplete. However, prior studies that included the use of HIF1A knockout cells have demonstrated a high degree of dependence of CAIX expression on HIF-1a, and, unlike most other known HIF-inducible genes, CAIX is not effectively induced by HIF-2a [59][60][61].
We next set out to determine the effect on hypoxiainduced gene expression of partially inhibiting glycolysis at the point of exposure to low oxygen, mimicking an acute and simultaneous reduction loss of bloodderived oxygen and glucose due to restriction of blood supply to a tumour. This was achieved by the addition of 10 mM 2-DG, a competitive inhibitor of hexokinase II [62], to the 25 mM glucose-containing culture  (Fig. 1C). In MCF-7 cells, hypoxia induced a 30.6-fold induction of CAIX mRNA compared with untreated controls, whereas in the presence of 2-DG, levels of CAIX under hypoxia were significantly 63.7% lower than when glycolysis was not inhibited. This effect of 2-DG was selective for CAIX, as hypoxia-induced increases in PDK1, EGLN3, FAM162A, TMEM45A or INSIG2 mRNAs were not inhibited by 2-DG, and indeed, PDK1 induction was significantly increased. In the MDA-MB-231 cell line, the magnitude of CAIX induction by hypoxia was less than that in MCF-7 (12.3 fold) and its abundance under hypoxia was not decreased by 2-DG. 2-DG did enhance the induction of PDK1 by hypoxia in MDA-MB-231, and the other mRNAs studied were either not affected or increased.
As an alternative approach for the suppression of glycolysis in the breast cancer cell lines, we have recently described a model in which cells are adapted to culture in media containing 10 mM fructose, rather than 25 mM glucose. Cells uptake and retain fructose less effectively than they do glucose, and therefore, whilst fructose enters the glycolytic pathway and enables the proliferation of adapted cells at comparable rates to the glucose-cultured counterparts, their rates of glycolysis are markedly reduced [53]. Again, in MCF-7 cells, restriction of glycolysis resulted in a substantial reduction in the hypoxia-induced increase in CAIX mRNA (Fig. 1D). In this model, the induction of all the other hypoxia-inducible genes was also decreased in the less glycolytic cells; this difference compared with the 2-DG model may either be due to the different methods used to suppress glycolysis or be the fact that the 2-DG was only added at the point when hypoxia was induced, whereas the fructose-cultured cells had adapted to reduced glycolysis over a period of> 30 days.
As hypoxia-induced CAIX mRNA expression was attenuated by inhibiting glycolysis in MCF-7 in both models, we also examined the effects of glycolysis on CAIX protein expression under hypoxia. In MCF-7 cells, the changes in CAIX protein essentially mirrored those of the mRNA in both the 2-DG (Fig. 1Ei) and fructose (Fig. 1Eii) models. In MDA-MB-231 cells, 2-DG did reduce hypoxia-induced CAIX protein induction, although no effect on the mRNA has been observed. This suggests that CAIX protein abundance in these cells may also be regulated though another mechanism, such as post-transcriptional control.
To determine whether these observed effects of glycolysis on hypoxia-induced expression of CAIX and other known HIF targets in the MCF-7 model were dependent on hypoxia per se, we used DMOG as an alternative intervention that induces a response that includes increased cellular HIF activity. In MCF-7 cells, DMOG caused a 211-fold increase in the abundance of CAIX mRNA, whereas in the presence of 2-DG, levels of CAIX following DMOG treatment were significantly 56.7% lower than when glycolysis was not inhibited (Fig. 2A). Of the other genes examined, the DMOG-induced upregulation of EGLN3, FAM162A and TMEM45A mRNAs was also inhibited by 2-DG, and the remainder was unaffected. In the fructose model of glycolytic restriction, DMOG-induced CAIX mRNA expression was also repressed in the less glycolytic cells, as was the expression of PDK1, EGLN3 and TMEM45A (Fig. 2B). Western blotting confirmed that DMOG induces an increase in CAIX protein expression and that this increase is suppressed by inhibiting glycolysis in both the 2-DG (Fig. 2Ci) and fructose (Fig. 2Cii) models.

Regulation of the expression of known hypoxia and HIF-inducible genes by CtBPs
Having established that glycolysis is required for effective induction of CAIX mRNA and protein in the MCF-7 model system, we set out to determine the role of CtBPs in the response. Firstly, we silenced CtBP1 and CtBP2 expression using an siRNA that targets the mRNA of both genes, and found that, similar to the effects of glycolytic restriction, this resulted in a significant 47.0% decrease in the abundance of CAIX mRNA in hypoxic MCF-7 cells (Fig. 3A). The expression of PDK1 and INSIG2 mRNAs was also reduced. MTOB is a metabolite in the methionine salvage pathway that is known to bind CtBPs and inhibit their ability to regulate gene expression [63]. Treatment of MCF-7 cells with MTOB resulted in an 87.8% reduction in the abundance of CAIX mRNA in hypoxic MCF-7 cells (Fig. 3B). Compared with CtBP1/2 siRNA, MTOB was less selective in its effects on CAIX, as it reduced the induction of all the known HIF target genes examined, although the effect was greatest on CAIX.
To complement the CtBP inhibition experiments, we also examined the effect of CtBP overexpression on the induction of known HIF-inducible genes. CtBP2 was used for these studies as it contains a nuclear localisation sequence whereas CtBP1 does not [64]. In glycolytic MCF-7 cells cultured in standard 25 mM glucose-containing media and exposed to hypoxia, the expression of CAIX mRNA in cells overexpressing CtBP2 was 213% higher than in cells containing the empty vector (Fig. 3Ci). A mutated version of CtBP2 (CtBP2 G189A ) that is compromised in its ability to bind Of the other known HIF-inducible genes studied, there was a trend for CtBP2 overexpression to modestly increase their expression in hypoxic cells, though for wild-type CtBP2 this was only significant for TMEM45A. In the cells in which glycolysis was restricted with culture in fructose, the overall expression of known HIF-inducible genes was reduced compared with glycolytic cells (Fig. 3Cii) as shown previously, but again, there was a clear positive effect of wild-type CtBP2 on the expression of CAIX mRNA, as well as mRNAs for EGLN3, FAM162A, TMEM45A and INSIG2. Notably, the effect of the G189A mutation on this activity of CtBP2 was more pronounced in these less glycolytic cells.
Western blotting for CAIX protein confirmed that the effects of CtBPs on CAIX mRNA abundance also resulted in concomitant changes in the protein in the siRNA experiments (Fig. 3Di) and MTOB experiments (Fig. 3Dii). CtBP2 overexpression only resulted in a modest increase in CAIX protein (Fig. 3Diii), suggesting that the enhanced increase in CAIX mRNA was not fully translated into increased protein synthesis. Immunoblotting for CtBPs confirmed the expected effects of the siRNA and overexpression plasmids on CtBP protein abundance in these experiments.
We then performed a comparable series of experiments using DMOG, as opposed to hypoxia, and again examined the role of CtBPs in the transcriptional response to this intervention. In the presence of DMOG, CtBP1/2 siRNA significantly reduced the abundance of CAIX mRNA by 29% (Fig. 4A). PDK1 and EGLN3 mRNA were also suppressed, whereas TMEM45A expression was upregulated. Similar to the hypoxia experiments, MTOB also caused a large 82.6% reduction in CAIX mRNA abundance in DMOG-treated cells (Fig. 4B), and the expression of all the other known HIF-inducible genes was significantly negatively affected. CtBP2-overexpressing cells cultured in high glucose media and DMOG showed 355% higher expression of CAIX mRNA than vector control cells (Fig. 4Ci), and again, the G189A mutant exhibited intermediate levels of expression. CtBP2 overexpression also enhanced the expression of the majority of the other known HIF-inducible genes analysed, albeit to a lesser extent in most cases. In the less glycolytic cells cultured in fructose-containing medium, the overexpression of CtBP2 has no effect on DMOG-induced CAIX mRNA expression, and  effects on the other known HIF-inducible genes (Fig. 4Cii).

Effect of CtBP dimerisation on hypoxia-induced expression of CAIX, and the abundance of stem cell-like cancer cells
Sustained hypoxia over several days can increase the occurrence of stem cell-like cancer (SCLC) cells in a HIF-dependent manner [65], and furthermore, the induction of CAIX appears to be an important determinant of this effect [66]. We therefore wished to determine the role of CtBPs in SCLC cells under hypoxia. Furthermore, given the evidence using the CtBP2 G189A mutant, which suggested a role for NADH-dependent CtBP dimerisation in the regulation of CAIX expression, a function that is regulated by glycolysis, we specifically set out to further test this aspect of CtBP regulation in the control of CAIX expression and SCLC cell biology. We have previously developed an approach of microinjection of fragments of CtBP proteins to act as dominant negative inhibitors of specific functions [45,46], including CtBP DD , which encompasses the dehydrogenase/dimerisation domain and thus competes for dimerisation of endogenous CtBP dimers [46]. We therefore constructed an inducible system where the expression of V5-tagged CtBP DD could be switched on by doxycycline in stably transduced MCF-7 cells. A control system, CtBP DDM , which carries two point mutations in the dimerisation domain that inhibits dimerisation [46,67] and thus renders CtBP DD ineffective, was also generated. Figure 5A shows the effects of inhibition of CtBP dimerisation after 18 hours of mild hypoxia (3% O 2 ffi 3 kPa) that are commonly observed in tissues [68]. Induction of CtBP DD by doxycycline resulted in a significant, up to 56.1%, inhibition of CAIX expression in hypoxic cells, having a comparable effect to 2-DG. CtBP DDM had no effect, demonstrating that the effect was due to the specific effects of CtBP DD on CtBP dimerisation, and none of the other known HIFinducible genes were significantly downregulated by CtBP DD , demonstrating that the effect is selective to CAIX. Figure 5B shows induction of the CtBP DD and CtBP DDM proteins, and Fig. 5C shows that the effects on CAIX mRNA are reflected in the abundance of CAIX protein. We then performed mammosphereforming assays [69] to assess the prevalence of SCLC cells in the population. These experiments were performed after only 24-h induction of CtBP DD followed by 18 h by hypoxia to limit the possibility of any observed effects being to secondary effects of CtBP inhibition. Compared with the previously published reports using~72 h hypoxia [65,66], under these short-term assay conditions there was no significant enrichment of SCLC cells in the population by hypoxia (Fig. 5D). The inactive control CtBP DDM had no negative effect on mammosphere formation when induced by doxycycline under any conditions, and likewise, the CtBP dimerisation inhibitor CtBP DD had no effect when induced in normoxic cells. However, in hypoxic cells, after addition of the inducing agent, CtBP DD -expressing cells demonstrated a significant, 29.7%, reduction of mammosphere-forming potential compared with noninduced cells. Thus, within the relatively short time they were exposed to mild hypoxia, at least a proportion of the SCLC cells in the population had acquired a dependency on CtBP dimerisation for their survival. These data are therefore consistent with a role of CtBP dimerisation in controlling the expression of HIF target genes such as CAIX that are required for the selective survival of SCLC cells in a hypoxic microenvironment.

Discussion
Here, we have focussed on the effects of glycolysis, and the CtBP family of glycolysis-sensing transcriptional regulators, on the hypoxia-induced expression of CAIX and other known HIF-inducible genes. Through HIFs, hypoxia induces a broad-ranging transcriptional response to adapt cells to their altered microenvironment, including the upregulation of genes associated with glycolysis to facilitate continued cell growth in the absence of oxygen [17][18][19]. However, in a microenvironment such as a tumour, which may have both an underdeveloped blood supply, and contain large numbers of highly metabolic cells, glucose availability can also be reduced [70,71], potentially limiting the cell ability to increase rates of glycolytic flux. Under these circumstances, a modified response may be required; for example, this increased expression of proteins such as CAIX, which are required in highly glycolytic cells [52], would be an unnecessary waste of metabolic resource. Our experiments using 2-DG to inhibit glycolysis in breast cancer cell lines support this hypothesis, with the data from MCF-7 demonstrating that this is, at least in part, controlled through expression of the mRNA. This response was selective for CAIX, demonstrating it is not through global regulation of HIF activity. These findings agree with the prior work of Natsuizaka et al who studied the effect of shortterm low glucose on the hypoxia response in a panel of cancer cell lines and found it to enhance the induction of some genes, whilst the induction of others, notably CAIX, was suppressed [72]. In a separate study, glucose deprivation also abolished the induction of CAIX protein by hypoxia in fibrosarcoma and pharyngeal carcinoma models [73]. Our model of chronic adaptation to reduced glycolysis differs from acute inhibition of glycolysis in both our 2-DG experiments and the above published studies in that the cells extensively reprogram their gene expression [53] prior to the hypoxic insult. In this model, the hypoxic response of all the genes was reduced somewhat, but the effect on CAIX was markedly the most substantial. Both acute inhibition and chronic inhibition of glycolysis also suppress the induction of CAIX mRNA and protein when cells were exposed to DMOG, demonstrating that the mechanism involved is not explicitly dependent upon the cells being hypoxic. None of the studies cited above provide a mechanism for the effects of glycolysis and CAIX expression. In order to test whether CtBPs are required for the glycolysis-dependent regulation of CAIX, we applied different approaches to alter CtBP expression and function. We utilised siRNAs and CtBP2 overexpression to alter CtBP expression levels. Furthermore, NADH-dependent CtBP function was more selectively inhibited by the pharmacological CtBP inhibitor MTOB and an inducible CtBP dimerisation inhibitor miniprotein. We found that in hypoxic MCF-7-and in DMOG-treated MCF-7, both downregulated total CtBP protein expression and impaired NADH-dependent CtBP function caused a reduction in hypoxia-induced CAIX expression, whereas CtBP2 overexpression increased CAIX gene and protein expression in a manner dependent upon its ability to bind NADH. These results strongly support a glycolysis-dependent regulation of CAIX gene and protein expression that is dependent on CtBPs. Induction of CAIX expression is strictly regulated by HIFs and SP1 but can be modulated by MAPK and PI3K signalling in a cell type-specific manner [59,74,75]. The impact of pathways such as this which modulate the HIF response is clearly demonstrated by immunohistochemical studies of protein expression in tumour samples, where the expression of CAIX, whilst clinically significant in terms of predicting patient outcome, may only weakly correlate with the expression of HIF proteins, and does not correlate with the products of other HIF-responsive genes [76]. Our demonstration of a glycolysis-NADH-CtBP-dependent pathway in the regulation of the hypoxia-induced CAIX expression thus advances the understanding of the mechanisms underlying important clinical observations such as these. As CtBPs are recruited to promoters by DNA binding transcriptional regulators, our data support a model whereby HIFs cooperate on gene promoters with transcription factors that are interaction partners of CtBPs to render some HIF target genes responsive to glycolysis. However, validation of this model will require further in-depth mechanistic analysis and the exclusion of other possible explanations, which could include an indirect effect of CtBPs that replicates the effect of glycolysis, rather than transducing it, or an effect on HIF subunits themselves that result in a promoter-selective effect. Supporting a direct effect of CtBPs on hypoxia-responsive promoters, a recent study [77] reanalysed the ChIP-seq and gene expression data of Di et al [40] in CtBP-manipulated MCF-7 cells to identify genes whose expression were both regulated by CtBPs and had CtBPs associated with their promoters, and, using gene ontogeny analysis, found the strongest enriched biological pathway to be 'response to hypoxia'. Overall therefore, our study may support a shift of paradigm away from CtBPs solely acting as corepressors of transcription towards a context-dependent coregulator of transcription, as there is a growing body of literature highlighting the activating role of CtBPs in the coregulation of their target genes [33][34][35][36][37].
It is known that CtBPs promote the formation of SCLC cells [35,40,49] and CAIX has also been clearly demonstrated to be an important marker of cells with a stem cell-like phenotype [66,78]. In this light, we investigated the role of glycolysis-CtBP-NADH axis in the formation of SCLC cells in MCF-7 breast cancer cells. We found that inhibition of CtBP dimerisation using the inducible miniprotein reduced the mammosphereforming potential in hypoxia but not in normoxia. These data are consistent with our recent work that metformin, a pharmacological inhibitor of complex I of the mitochondrial electron transport chain, sensitised breast SCLC cells to CtBP inhibition [49]. The data we have shown here, combined with the findings from Ledakie et al whose work also focussed on the MCF-7 cell system [66], support a model whereby the regulation of CAIX expression by CtBPs may be a key determinant of this effect in hypoxic cells.
In conclusion, we have shown for the first time that CtBPs are a functional link between hypoxia-induced glycolysis and hypoxia-induced expression of the HIF target gene CAIX. It will be of great interest to dissect the mechanisms underlying this link, such as any potential functional or physical interactions between HIF-1, CtBPs and other transcriptional regulators at the CAIX promoter. Additionally, this work strengthens the understanding of CtBPs as an important oncogene in epithelial cancers and supports the rationale for the further development of CtBPs as potential drug targets for cancer therapy [46,79,80].