Lysine Demethylase LSD1 Coordinates Glycolytic and Mitochondrial Metabolism in Hepatocellular Carcinoma Cells

Patterns of gene expression are maintained and often reprogrammed by the combination of transcription factors and epigenetic factors involved in modifications of DNAs and histones, leading to conversion of cellular phenotypes. In particular, alterations of these epigenetic marks are hallmarks of many types of cancer cells (1–3). As another notable feature, proliferative cancer cells are thought to rely on energy production from the glycolytic pathway, even under high oxygen conditions: the so-called aerobic glycolysis (4–6). This metabolic remodeling may be interpreted as an adaptation to the hypoxic microenvironment where the cancer cells originally reside (7). Recent reports have highlighted that metabolic enzymes involved in such process are manipulated in cancer cells (8). Mitochondrial function is also modified such that it supports the production of biomacromolecules and reactive oxygen species rather than ATP (9). These lines of evidence support the notion that metabolic alteration in cancer cells is not merely a consequence of impaired cellular functions by transformation, but rather an ordered reprogramming of energy flow that fuels the accelerated cell growth. Epigenetic plasticity has been discussed as an underlying mechanism for metabolic reprogramming (10). In addition, recent studies on isocitrate dehydrogenase mutations in various cancers defined that misguided metabolic flow leads to the impaired activities of epigenetic factors (11). Thus metabolism–epigenome crosstalk may profoundly contribute to the dysregulated gene expression in cancer (12, 13). However, little is known about how specific epigenetic factors control cancer cell metabolism.

Lysine-specific demethylase-1 (LSD1, also known as KDM1A) is a flavin-dependent amine oxidase, which, in general, suppresses gene expression by removing the methyl group from mono- and dimethylated histone H3 at lysine 4 (H3K4; ref. 14). LSD1 knockout mice die early in development (15), and LSD1-null embryonic stem cells showed impaired viability (16), suggesting that LSD1 plays a crucial role in cell functions. Several studies showed that overexpression of LSD1 drives cell proliferation in various cancers (17–20). We have recently found that LSD1 suppresses mitochondrial respiration and maintains energy storage in murine adipocytes under the obese condition (21). Therefore, it is fascinating to test whether LSD1 facilitates the metabolic reprogramming in cancer cells.

HepG2 and Huh-7 cells from human hepatocellular carcinomas (HCC) were grown in DMEM (Sigma), supplemented with 10% (v/v) heat-inactivated FBS. Human telomerase-immortalized hepatic NeHepLxHT cells were cultured in DMEM/Nutrient Mixture F-12 Ham (Sigma), supplemented with 10% (v/v) heat-inactivated FBS, dexamethasone, insulin, and G-418. MDA-MB-231 cells from breast cancer were cultured in RPMI1640 (Sigma), supplemented with 10% (v/v) heat-inactivated FBS. All cell lines were purchased from ATCC or JCRB Cell Bank, and were authenticated by the providers by short-tandem repeat analysis or isoenzyme analysis. All cell lines were thawed within a short term after receipt, and were used at early passages. We did not observe any morphologic changes or altered growth rates during our maintenance culture. For the knockdown experiments, specific siRNAs were introduced into the cells using RNAiMAX reagent (Invitrogen) when they were approximately 50% confluent. Most experiments were done at 96 hours after siRNA introduction, and were carried out at 72 hours after tranylcypromine treatment, unless otherwise described. For hypoxic conditions, cells were incubated with 200 μmol/L cobalt chloride (CoCl₂) or 1 mmol/L DMOG for 12 hours. Information on the siRNAs used in this study is listed in Supplementary Table S1. For cell-cycle analyses, cells were fixed with ethanol, stained with propidium iodide, and were subjected to FACS analysis.

Animal experiments were conducted in accordance with the guidelines of the Animal Care and Use Committee of Kumamoto University (Kumamoto, Japan). Control siRNA or LSD1 siRNA-1–treated HepG2 cells were dissociated, and resuspended in Matrigel/HBSS (BD Biosciences) mixture. Twenty-four hours after siRNA introduction, 5 × 10⁶ control or LSD1-KD cells were injected into right and left flanks, respectively, of 6-week-old SCID mice (FOX CHASE SCID C.B-17/lcr-scid/scidJcl). Three weeks after transplantation, tumors were harvested, and collected tumors were weighed and dissected for RNA analysis.

Thirty-eight patients with HCC, who had undergone tumor resection at the National Cancer Center Hospital (Tokyo, Japan), between May 2003 and December 2005, were enrolled in this study. As described previously (22), histologic classification was assessed according to the World Health Organization Histological Classification of Tumors. The Ethics Committee of the National Cancer Center approved this study, with written informed consent from all patients.

Immunohistochemistry (IHC) for GLUT1 and LSD1 was performed using a polymer-based method with Envision +Dual Link System-horseradish peroxidase (DK-2600 Glostrup; Dako). Sources and dilutions of primary antibodies were as follows: anti-GLUT1 (rabbit polyclonal; ab15309, Abcam; 1:200) and anti-LSD1 (rabbit polyclonal; ab17721, Abcam; 1:200). Formalin-fixed, paraffin-embedded serial tissue sections (4 μm) were placed on silane-coated slides for IHC. Antigen retrieval was carried out by heating in a target retrieval solution (Tris/EDTA buffer, pH 9; Dako Cytomation) for GLUT1 and in a 0.01 mol/L citrate buffer (pH 6.0) for LSD1 at 121°C for 10 minutes in a pressure cooker. The other procedures used in IHC were similar to our previous report (22).

Immunoreactivities of GLUT1 were defined as follows: 0+, no membrane staining; 1+, membrane staining, lower than the intensity of the membrane in red blood cells; 2+, membrane staining, equivalent or higher than the intensity of red blood cells membrane within the same section. Immunoreactivities of LSD1 were defined as follows: 0+, no nuclear staining; 1+, nuclear staining, equivalent to the intensity of the normal hepatocyte epithelium (NHE); 2+, nuclear staining, higher than the intensity of the NHE within the same section. The IHC score of 1+ and 2+ of GLUT1 and 2+ of LSD1 were defined as positive for expression of each protein. The percentage of positive area of GLUT1 was then defined as the percentage of positively membrane stained area (IHC scores of 1+ and 2+) divided by the total membrane area of the tumor cells. When there was a positive area, it was defined as positive case of GLUT1. The positive index of LSD1 was assessed by counting the positive cells in randomly selected three 40× high-power fields (more than 300 tumor cells). Finally, immunoreactivities of GLUT1 were positive in 13 (34.2%) of the 38 cases. Correlations between positive case of GLUT1 and positive index of LSD1 were evaluated by the Student t test.

Data are presented as means ± SD. Statistical analyses were performed by the two-tailed Student t test.

To determine the involvement of LSD1 in metabolic pathways, live monitoring using an extracellular flux analyzer was performed in human HepG2 cells, which are well-established hepatocellular carcinoma (HCC) cells. In this study, LSD1 knockdown(KD) was achieved using two different siRNAs, which conferred limited influence on the cell cycle and growth under the current experimental setting (Supplementary Fig. S1A–S1C). Compared with control cells, the oxygen consumption rate (OCR) increased in LSD1-KD cells (Fig. 1A, left), while the loss of LSD1 decreased the extracellular acidification rate (ECAR), an index of glycolytic activity (Fig. 1A, middle). The increased OCR/ECAR ratio suggested that LSD1 depletion converted energy production from glycolysis to mitochondrial respiration in the cells (Fig. 1A, right). Real-time monitoring of glycolytic activities was then carried out under glucose starvation and subsequent readdition (Fig. 1B). The glucose-dependent increase in the ECAR was significantly blocked by LSD1 depletion (Fig. 1B, left). Glucose exposure induced a rapid shift from mitochondrial respiration to glycolysis in control cells, whereas this metabolic shift was attenuated in LSD1-KD cells (Fig. 1B, right and Supplementary Fig. S1D).

To further examine glucose utilization under the LSD1-depleted conditions, glucose uptake was measured using a fluorescent glucose analog, 2-[N-(7-Nitrobenz-2-Oxa-1,3-Diazol-4-yl)Amino]-2-Deoxy-D-Glucose (2-NBDG). Flow cytometric analysis revealed a significant decrease of 2-NBDG uptake in the LSD1-KD cells (Fig. 1C and Supplementary Fig. S1E). On the other hand, fluorescent staining of mitochondria with JC-1 dye revealed that LSD1 inhibition, either by siRNAs or by the enzymatic inhibitor tranylcypromine, significantly increased both mitochondrial mass (FL1) and membrane potential (FL2; Fig. 1D and Supplementary Fig. S1F), which corresponded with the active oxygen consumption in the LSD1-KD cells. Similar results were obtained in another human HCC cell line, Huh-7 (Supplementary Fig. S4A and S4B).

The metabolic landscapes in HepG2 cells were then characterized, using a capillary electrophoresis time-of-flight mass spectrometry (CE-TOFMS)-based metabolomics analysis (Fig. 1E and Supplementary Fig. S2; Supplementary Table S2). The loss of LSD1 significantly reduced the products from the early steps of glycolysis, such as glucose-6-phosphate and fructose-6-phosphate, together with a relative decrease in lactate levels. Products of pentose phosphate pathway branching from glycolysis were also dramatically reduced, which is consistent with the limited glucose utilization under LSD1-KD. On the other hand, acetyl-CoA was elevated to a detectable level, which might have contributed to the enhanced mitochondrial respiration by LSD1-KD. Collectively, LSD1 inhibition activates mitochondrial respiration and represses glycolysis, suggesting that LSD1 controls energy flow in HCC cells.

We performed expression microarray analyses to clarify the role of LSD1 in metabolic gene regulation. Gene set enrichment analysis revealed that the gene set of “glycolysis and gluconeogenesis” was ranked as significantly affected by LSD1 depletion (Fig. 2A and Supplementary Fig. S3). The data were validated using quantitative RT-PCR in HepG2 cells (Fig. 2B and C) and Huh-7 cells (Supplementary Fig. S4C and S4D). Importantly, among this gene set, glycolytic genes were downregulated by LSD1-KD (Fig. 2B), whereas genes encoding the rate-limiting enzymes of gluconeogenesis were upregulated (Fig. 2C). In addition, genes involved in mitochondrial fatty acid β-oxidation were also induced in the LSD1-KD cells (Fig. 2A and D and Supplementary Fig. S4C). In Fig. 1E, these gene expression profiles are shown with the metabolomics data [downregulated (blue) and upregulated genes (red) by LSD1-KD]. To further validate the importance of LSD1 in glycolytic regulation, we produced HepG2 cells that overexpress LSD1, and observed a significant increase of glycolytic gene expression (Supplementary Fig. S6). Thus, LSD1 regulates metabolic genes that link glucose uptake to the glycolytic pathway in HCC cells. We also tested a breast cancer cell line, MDA-MB-231, and observed that the important glycolytic genes were downregulated by LSD1-KD (Supplementary Fig. S7).

As glucose uptake was significantly reduced by LSD1-KD in HCC cells (Fig. 1C), we postulated that glucose transporter proteins had been decreased. To verify this, we examined the mRNA expression of well-characterized glucose transporter genes GLUT1-4 (also known as SLC2A1–4) under LSD1-KD. We found that both the mRNA and protein levels of GLUT1 were decreased by LSD1-KD (Fig. 3A and B and Supplementary Fig. S4D). GLUT1 has been reported to be highly expressed and rate limiting for glucose transport in HCC cells (23). Consistent with this report, glucose uptake was strikingly impaired in GLUT1-KD HepG2 cells (Fig. 3C and Supplementary Fig. S5A and S5B).

Recent reports have shown that LSD1 is overexpressed in various cancers (18, 20, 24–26). To test the importance of LSD1 in glucose metabolism in human HCC, we conducted immunohistochemical studies of HCC tissues from 38 patients. Representative images are shown in Fig. 3D: LSD1 and GLUT1 were not detected in case 1, while both were positively stained in case 2. As summarized in Fig. 3E, the index of the LSD1 immunoreactivities was significantly high in the GLUT1 positively stained tissues (34.2% of the examined cases), indicating a close correlation between LSD1 and GLUT1 expression in clinical specimens. To test whether the metabolic function of LSD1 correlates with the cancerous state, similar experiments were repeated in telomerase-immortalized NeHepLxHT cells of human hepatic origin (27). Expression of LSD1 mRNAs was remarkably higher in HCC cells, compared with normal liver and NeHepLxHT cells (Supplementary Fig. S8A). Interestingly, the loss of LSD1 significantly increased glucose uptake in NeHepLxHT cells (Supplementary Fig. S8B and S8C), with the concomitant upregulation of GLUT1 and glycolytic genes (Supplementary Fig. S8D). These results suggest that LSD1 is uniquely involved in the glucose metabolism of cancer cells.

LSD1 was originally identified as an H3K4 demethylase involved in gene repression (14), and some reports have shown its alternative role in H3K9 demethylation (17). To test the epigenetic regulation of metabolic genes by LSD1, chromatin immunoprecipitation (ChIP) analyses were performed in HepG2 cells. LSD1 inhibition induced PGC-1α, which is involved in transcriptional control of mitochondrial oxidative metabolism, with a significant enrichment of transcriptionally active marks, such as methylated H3K4 (Fig. 4A–C). In addition, the gene encoding long chain specific acyl-CoA dehydrogenase (LCAD, also known as ACADL), which is involved in fatty acid oxidation, was enriched with methylated H3K4 in the LSD1-KD cells (Fig. 4C, right). ChIP analyses further showed that both PGC-1α and LCAD gene promoters were directly bound by LSD1 (Fig. 4D). Thus, the results show that LSD1 epigenetically represses genes for mitochondrial metabolism and fatty acid oxidation by H3K4 demethylation. On the other hand, modified histone H3, including methylated H3K4 and H3K9 on GLUT1 and hexokinase-2 (HK2) gene promoters, did not markedly change in LSD1-depleted cells (Fig. 5A).

As glucose metabolism genes did not exhibit notable changes in histone modifications after LSD1-KD, we reasoned that transcriptional control of these genes had been affected. Hypoxia-inducible facter-1α (HIF1α) plays a central role in glycolytic gene regulation both in normal and cancer cells (28). As the majority of LSD1-regulated glucose metabolism genes were also the targets of HIF1α, we hypothesized that LSD1 might functionally interact with HIF1α (24, (25). To address this, we first examined the HIF1α enrichment in the regulatory regions of glucose metabolism genes. Under aerobic condition, HIF1α is hydroxylated by HIF prolyl hydroxylases (PHD), and the hydroxylated HIF1α is ubiquitinated by von Hippel-Lindau (VHL) protein followed by the proteasomal degradation. Upon hypoxic stress, O₂-dependent activities of PHDs are attenuated thereby HIF1α protein accumulates (29). Under CoCl₂ treatment, which inhibits O₂-dependent PHD activity (30, 31), we observed HIF1α binding at the vicinity of the putative hypoxia-responsive element (HRE at site 3) of the GLUT1 gene (Fig. 5B, gray bars in the top panel). Interestingly, HIF1α occupancy in the predicted HRE decreased to approximately 40% in LSD1-depleted cells (orange bars). A similar reduction in HIF1α binding by LSD1-KD was found at the transcription start site (exon 1) of HK2 (Supplementary Fig. S9A). LSD1 also appeared to be present on the promoter region of GLUT1 (Fig. 5B, bottom panel) and HK2, further suggesting the coregulation of these genes by LSD1 and HIF1α (Supplementary Fig. S9A). The expression of metabolic genes under the combination of LSD1 depletion and CoCl₂ treatment was then assayed. In accordance with the reduced HIF1α binding, LSD1 depletion abolished the induction of GLUT1 and most of the glycolytic genes by hypoxic stress (Fig. 5C, genes shown in red). In addition, hypoxic stress-induced suppression of metabolic genes (e.g., PGC-1α) was also ablated by LSD1-KD (Fig. 5C, genes shown in blue). Moreover, depletion of HIF1α and LSD1 exhibited analogous effects on the expression of glucose metabolism genes under normoxia, suggesting that these factors cooperate under both normoxia and hypoxia (Supplementary Fig. S9B and S9C). Thus, LSD1 regulates glycolytic genes through the HIF1α-mediated pathway in HCC cells.

To further explore the regulation of HIF1α function by LSD1, we examined the protein expression of HIF1α. Notably, we found that LSD1-KD significantly decreased HIF1α protein, as well as GLUT1, in the presence of CoCl₂ in HepG2 and Huh-7 cells (Fig. 6A and Supplementary Fig. S9D, shown quantitatively in the graphs). We also obtained similar data under hypoxia (1% O₂) (Fig. 6B). The LSD1-HIF1α-GLUT1 connection was further verified by immunostaining analyses of a mixed culture that contained control and LSD1-KD cells. Under CoCl₂ treatment, both nuclear LSD1 and HIF1α were enriched in the control cells, whereas HIF1α expression clearly decreased in the LSD1-depleted cells (Fig. 6C and D). Consistent with this, GLUT1 protein was reduced in the LSD1-KD cells, compared with the control cells (Fig. 6E and F), indicating the coexistence of the LSD1-HIF1α-GLUT1 at the single-cell level.

We then examined whether LSD1 transcriptionally or posttranscriptionally controls HIF1α protein in cultured HCC cells. The HIF1α mRNA level was not affected by either LSD1 depletion or by CoCl₂ addition (Supplementary Fig. S9E). It has been reported that activated PI3K–AKT–mTOR signaling is implicated in the enhanced HIF1α translation in cancer cells (32, 33), and that the phosphorylation of S6K and 4E-BP1 is the major downstream effector of the mTOR signaling, which leads to protein synthesis (34). However, we did not find any reduction in their phosphorylation level by LSD1-KD (Supplementary Fig. S9F). Thus, transcriptional and translational pathways did not seem to be responsible for the decrease of HIF1α by LSD1-KD.

To clarify whether HIF1α protein is reduced by LSD1-KD via the ubiquitin proteasome degradation, we treated the cells with a proteasome inhibitor MG-132 or a PHD inhibitor dimethyloxalylglycine (DMOG; Supplementary Fig. S9G). The treatment with CoCl₂, or with DMOG to a lesser extent, stabilized the HIF1α protein with absence of the accumulated hydroxyl-form. In contrast, proteasome inhibition clearly augmented the hydroxyl-HIF1α protein with detectable smeared bands of high molecular weight (>120 kDa; Supplementary Fig. S9G), indicating that HIF1α was degraded via the PHD-mediated proteasomal pathway in LSD1-KD cells as well as control cells. Furthermore, we checked the ubiquitination of HIF1α protein in FLAG-tagged HIF1α introduced HepG2 cells (Supplementary Fig. S9H). Polyubiquitinated HIF1α proteins accumulated under proteasome inhibition in LSD1-KD cells as well as in control cells. Our repeated experiments showed that (i) HIF1α was destabilized via the ubiquitin proteasome pathway in LSD1-KD cells and (ii) DMOG treatment did not reverse the effect of LSD1-KD on HIF1α.

Recently, Qin and colleagues demonstrated that LSD1 contributes to the stabilization of HIF1α protein through facilitating its deacetylation by HDAC2 in pancreatic cancer cells (35). We therefore tested whether this model could also be applied to HCC cells in our study. However, we did not observe a physical interaction of HIF1α protein with either LSD1 or HDAC1/2 in HepG2 cells (Supplementary Fig. S10A). Accordingly, LSD1 depletion did not affect the acetylation status of HIF1α (Supplementary Fig. S10B). In contrast, interaction between LSD1 and HDAC1/2 could be clearly observed (Supplementary Fig. S10A), consistent with previous studies (36, 37). These findings suggest that HIF1α is not the target of LSD1/HDAC-mediated deacetylation in HCC cells.

Collectively, these results suggest that LSD1 protects HIF1α from proteasomal degradation at least in part through a hydroxylation-independent mechanism.

To gain insight into the in vivo role of LSD1 in HCC development, we transplanted siRNA-introduced HepG2 cells subcutaneously into SCID mice. Twenty-four hours after siRNA transfection, control and LSD1-KD HepG2 cells were injected into left and right flanks of each mouse, respectively, and then the mice were kept for 3 weeks until sacrifice. Interestingly, in 6 of 9 mice, tumors derived from LSD1-KD cells were markedly smaller than the controls, with approximately 36% reduction in the cumulative tumor weight (Fig. 7A and B). In 3 mice, we observed dramatic reduction in tumor size by LSD1-KD. In these tumors, many of the glycolytic genes under regulation by LSD1 in vitro showed decreased expression, compared with the control tumors from the same mice (Fig. 7C, left). On the other hand, in the mice where LSD1-KD did not reduce the tumor size, there was no obvious difference in the glycolytic gene expression between the control and LSD1-KD (Fig. 7C, right). These results suggest that LSD1 supports the HCC tumor engraftment and/or growth. Next, we administrated tranylcypromine, a chemical inhibitor of LSD1, in mice after the xenograft tumors had been expanded (Supplementary Fig. S11B). We observed that tranylcypromine-treated tumors tended to show smaller size compared with the controls but without statistical significance (Supplementary Fig. S11C).

Taken together, these results show that LSD1 coordinately maintains the increased glucose uptake, followed by glycolysis and the suppression of mitochondrial respiration in cancer cells, thereby leading to the typical cancer metabolism known as the Warburg effect (shown schematically in Fig. 7D).

In the current study, we demonstrated that LSD1 coordinates energy production through activating glycolytic pathway and suppressing the mitochondrial respiration in HCC cells. Of great interest, LSD1 repressed mitochondrial metabolism genes via the H3K4 demethylating activity, while glycolytic genes were induced, at least in part, via the LSD1-mediated HIF1α transactivation. Orchestration of the glycolytic shift by LSD1 is well reflected on the metabolomic profile of LSD1-KD cells, in which the early glycolytic products and pentose phosphate pathway intermediates were clearly decreased compared with the control. An increase of acetyl-CoA level in LSD1-KD cells was in good agreement with the upregulation of fatty acid oxidation genes. Acetyl-CoA is metabolized by citrate synthase in the mitochondria into citrate that inhibits PFK activity thus may contribute to the reduced glycolysis. However, we did not find a change in the citrate level by LSD1-KD (Supplementary Fig. S2), suggesting that elevated acetyl-CoA was utilized to fuel respiratory chain activity as demonstrated by the extracellular flux analysis. Our metabolomic data together with extracellular flux, glucose uptake, and mitochondrial potential data clearly define the contribution of LSD1 in the maintenance of glycolytic activation and mitochondrial suppression in HCC cells.

Coordinated expression of metabolic genes is crucial for the adaptive metabolic remodeling in response to severe environmental conditions. Recent work by Duteil and colleagues demonstrated that LSD1 promotes the oxidative respiration in white adipocytes in response to strong catabolic stimuli such as β-adrenergic activation (38), whereas we have previously shown that LSD1 suppresses the mitochondrial respiration under adipogenic and lipogenic conditions (21), suggesting that LSD1 translates environmental fluctuations to diverse metabolic outcomes. In the current study, even though LSD1 depletion resulted in the impaired glucose utilization in HCC cells in vitro, we did not observe a clear effect on cell growth and cell-cycle progression. As sufficient amount of nutrients and oxygen are present under normal culture condition, LSD1-depleted cancer cells might have used alternative pathways to supply energy sources. Interestingly, we demonstrated that LSD1 depletion led to the size reduction of xenograft tumors. As we employed transient siRNA introduction before the transplant, it is likely that LSD1 expression had been reduced only for the first 3 to 4 days. This raises an intriguing possibility that LSD1 was required for the cells to adapt to hypoxic and nutrient-poor condition. Of note, both LSD1-depleted xenograft tumors and HepG2 cells in vitro exhibited a reduced expression of VEGFA, a HIF1α-responsive gene that is a key factor for the angiogenesis and tumor growth (Fig. 7C and Supplementary Fig. S11A; ref. 39). Together, these lines of evidence suggest that LSD1 plays an essential role in the metabolic reprogramming of cancer cells under fluctuating environment.

Our data indicate that LSD1 maintains the HIF1α-mediated transcriptome in two HCC lines, HepG2 and Huh-7. At the same time, we observed that a subset of glycolytic genes was differentially regulated by LSD1 (Supplementary Fig. S4). Previous report has identified, using integrative genomic approach, that HIF1α responsive genes varied across cell types, and some but not all glycolytic genes were the universal targets (40). As most of the LSD1-target glycolytic genes were coregulated by HIF1α (Supplementary Fig. S9C), it is possible that the target gene preference could be attributed to the responsiveness to HIF1α. In addition, the maintenance of glycolytic pathway by LSD1 was not found in nontransformed cells (Supplementary Fig. S8B and S8D). The difference in LSD1 expression level may contribute to this selectivity, as HCC cells showed markedly higher expression compared with nontransformed cells and the normal liver (Supplementary Fig. S8A). This possibility is backed up by our data that the forced expression of LSD1 resulted in the increase of glycolytic gene expression (Supplementary Fig. S6). The difference between cancer and nontransformed cells also suggests that activation of glucose utilization by LSD1 may require additional oncogenic signals. Phosphorylation signaling cascades mediated by PI3K and MAPK regulate the levels of HIF1α (41). Consistently, enhanced HIF1α activity has been observed in HCC in both human and mice (33, 42). In addition, LSD1 interacts with Myc, which is also an important regulator of glucose flux, and activates the Myc target genes (43), implying that PI3K/MAPK and LSD1/Myc pathways may contribute to the oncogenic energy utilization. Moreover, p53 is a negative regulator of glycolytic enzyme phosphoglycerate mutase (PGAM; ref. 44) and glucose transporters (GLUT1 and GLUT4; ref. 45), and was reported to be inactivated through the LSD1-mediated demethylation (46). Considering the dual role of LSD1 in glycolysis and mitochondrial respiration, these evidences raise an intriguing possibility that LSD1 plays a pivotal role in regulating cancer cell metabolism in response to oncogenic and antioncogenic signals.

We found in the current study that LSD1 is required for maintaining the HIF1α protein level, thus contributing to the glycolytic gene expression. Our data indicate that LSD1 protects HIF1α from proteasomal degradation. The functional association of LSD1 with HIF1α was enhanced under hypoxic stress, suggesting that LSD1 facilitates the canonical HIF1α stabilization pathway. However, we did not find a change in the hydroxylation level of HIF1α nor physical interactions of LSD1 with PHDs (data not shown). Direct interpretation of this is that LSD1 affects the ubiquitination and/or proteasomal degradation of HIF1α at least in part through a hydroxylation-independent mechanism. Intriguingly, we observed that the ratio of conjugated to unconjugated forms of HIF1α was higher in LSD1-KD cells, compared with control cells, owing in part to the decrease of unconjugated HIF1α in the LSD1-KD (Supplementary Fig. S9H). Thus LSD1 may protect HIF1α from ubiquitination and the subsequent proteasomal degradation. LSD1 has been reported to remove the methyl group from some nonhistone proteins either to augment or to repress their functions (47). Known or unknown proteins involved in HIF1α stabilization might be modulated by LSD1 through such mechanisms. In summary, the current study demonstrated that LSD1 maintains the glycolytic gene expression in cancer cells without affecting the major chromatin marks, but through the modulation of a transcriptional activator.

Our current study provides evidence that LSD1 functions as a pivotal regulator of cancer cell metabolism, and highlights the epigenetic plasticity of the cellular metabolic state under LSD1 inhibition. Moreover, recent studies demonstrated that LSD1 has an essential role in the maintenance of pluripotency in embryonic stem cells and cancer stem cells (48, 49), and that such highly proliferative cells obtain energy preferentially by glycolysis (50). Therefore, the LSD1-mediated metabolic switch may be generally crucial for cancer metabolism, as well as for proliferative capacity.

No potential conflicts of interest were disclosed.

Conception and design: A. Sakamoto, S. Hino, M. Nakao

Development of methodology: A. Sakamoto, S. Hino, M. Nakao

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Sakamoto, S. Hino, K. Nagaoka, K. Anan, R. Takase, H. Matsumori, Y. Kanai

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Sakamoto, S. Hino, K. Anan, H. Ojima, K. Arita, M. Nakao

Writing, review, and/or revision of the manuscript: A. Sakamoto, S. Hino, Y. Kanai, M. Nakao

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Nakao

Study supervision: K. Arita, M. Nakao

This work was supported by a Grant-in-Aid for Scientific Research on Priority Areas and on Innovative Areas (3307) from the Ministry of Education, Culture, Sports, Science and Technology of Japan, by the Japan Science and Technology Agency (CREST), and by a grant from The Uehara Memorial Foundation (M. Nakao). This work was also supported by a Grant-in-Aid for Scientific Research from Japan Society for the Promotion of Science and by the Nakatomi Foundation (S. Hino).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.