Regulation of hypoxia responses by flavin adenine dinucleotide‐dependent modulation of HIF‐1α protein stability

Abstract

Oxygen deprivation induces a range of cellular adaptive responses that enable to drive cancer progression. Here, we report that lysine‐specific demethylase 1 (LSD1) upregulates hypoxia responses by demethylating RACK1 protein, a component of hypoxia‐inducible factor (HIF) ubiquitination machinery, and consequently suppressing the oxygen‐independent degradation of HIF‐1α. This ability of LSD1 is attenuated during prolonged hypoxia, with a decrease in the cellular level of flavin adenine dinucleotide (FAD), a metabolic cofactor of LSD1, causing HIF‐1α downregulation in later stages of hypoxia. Exogenously provided FAD restores HIF‐1α stability, indicating a rate‐limiting role for FAD in LSD1‐mediated HIF‐1α regulation. Transcriptomic analyses of patient tissues show that the HIF‐1 signature is highly correlated with the expression of LSD1 target genes as well as the enzymes of FAD biosynthetic pathway in triple‐negative breast cancers, reflecting the significance of FAD‐dependent LSD1 activity in cancer progression. Together, our findings provide a new insight into HIF‐mediated hypoxia response regulation by coupling the FAD dependence of LSD1 activity to the regulation of HIF‐1α stability.

Introduction

An accumulating body of literature reports that a number of chromatin regulators, including various histone modification enzymes, are involved in different stages of tumorigenesis. Both genetic alterations and aberrant expression of chromatin regulators are known to alter the chromatin landscape and thereby affect cancer cell proliferation and/or differentiation (Dawson & Kouzarides, 2012). Recent studies show that the availability of metabolic cofactors necessary for the enzymatic activity of chromatin regulators also affects cancer progression (Gut & Verdin, 2013). Furthermore, a growing list of non‐histone substrates for histone‐modifying enzymes expands the functional scope of chromatin regulators beyond changing chromatin structure and toward regulating diverse signaling pathways involved in different cellular functions (Hamamoto et al, 2015). Given these findings, targeting of chromatin regulators has emerged as a promising anti‐cancer therapeutic strategy.

Lysine‐specific demethylase 1 (LSD1) is a histone‐modifying enzyme that removes mono‐ and di‐methyl groups from lysine 4 or lysine 9 on H3 via a flavin adenine dinucleotide (FAD)‐dependent oxidative reaction (Lan et al, 2008). Mono‐ and di‐methylation of H3K4 (H3K4me1/2) is strongly associated with transcriptional activation, whereas di‐methylation of H3K9 (H3K9me2) is involved in a repressive mode of transcription (Zhou et al, 2011). Due to the dual specificity toward H3K4me1/2 and H3K9me1/2, the effect of LSD1 on target gene expression appears to be context‐dependent. Indeed, LSD1 is known to physically associate with multiple transcriptional regulators including NuRD, Co‐REST, AR, and AML (Nakamura et al, 2002; Metzger et al, 2005; Shi et al, 2005; Wang et al, 2009b). In conjunction with specific binding partners, LSD1 regulates the transcription of distinct sets of target genes. Furthermore, LSD1 is known to regulate the methylation dynamics of non‐histone proteins. The di‐methyl group at lysine 370 on p53, which facilitates the binding of 53BP1, is erased by LSD1, causing the downregulation of p53‐responsive genes (Huang et al, 2007). In addition, the stability of DNMT, E2F, and MYPT1 proteins is regulated by the non‐histone demethylase activity of LSD1 (Wang et al, 2009a; Kontaki & Talianidis, 2010; Cho et al, 2011). These findings highlight that LSD1 has diverse physiological roles, acting not only as a chromatin modifier but also as a signaling regulator for non‐chromatin events.

Aberrant expression of LSD1 has been shown in many cancer types including blood, neuronal, prostate, lung, colorectal, bladder, pancreatic, and breast cancers (Amente et al, 2013), implicating LSD1 as a cancer biomarker. In acute myeloid leukemia (AML), a type of blood cancer, LSD1 bound to MLL supercomplex regulates the expression of the MLL‐mediated oncogenic program by modulating H3 methylation (Harris et al, 2012). In line with these observations, either genetic or pharmacological inhibition of LSD1 suppresses oncogenic stem cell characteristics (Harris et al, 2012; Schenk et al, 2012), highlighting the therapeutic potential of targeting LSD1 for AML treatment. However, despite the strong correlation of LSD1 expression with cancerous incidence and poor prognosis, the mechanism of LSD1‐mediated solid cancer progression remains elusive.

The rapid growth of solid tumors frequently generates oxygen‐deprived microenvironments due to inefficient oxygen delivery from distal blood vessels (Bristow & Hill, 2008). Chromatin regulators modulating histone acetylation or methylation appear to affect hypoxic responses by serving as transcriptional cofactors of hypoxia‐inducible factors (HIFs) (Zhong et al, 2010; Mimura et al, 2012). Moreover, histone lysine deacetylases such as SIRT1 and HDAC4 are suggested to modulate the acetylation dynamics of HIF‐1α protein (Lim et al, 2010a; Geng et al, 2011). Conversely, hypoxia can modulate the enzymatic activity of chromatin regulators per se by changing the amount of their substrates and metabolic cofactors. For instance, the lysine deacetylase activities of SIRT1 and SIRT6 are downregulated by a reduced NAD⁺/NADH ratio in hypoxic cells (Lim et al, 2010a; Zhong et al, 2010), and the enzymatic activity of JmjC domain‐containing histone lysine demethylases (KDMs), such as KDM3A, decreases in hypoxia due to the low concentration of oxygen (Xia et al, 2009). However, unlike most KDMs, LSD1 (also known as KDM1A), which lacks the JmjC domain, does not require oxygen as its co‐substrate (Lan et al, 2008), and the molecular mechanism of its involvement in the hypoxic response remains to be clearly defined.

In this study, given that LSD1 is strongly associated with solid cancer progression, which is frequently accompanied by hypoxia, we investigated the molecular functions of LSD1 in hypoxic cancer cells. First, we demonstrate that HIF‐1α is stabilized when LSD1 demethylates RACK1 protein, a key component of oxygen‐independent degradation of HIF‐1α subunits. We also provide evidence that a change in cellular FAD level is a critical factor determining HIF‐1α stability during hypoxia, as it modulates LSD1‐dependent RACK1 demethylation. Furthermore, we demonstrate the prognostic significance of the association among FAD biosynthetic enzyme expression, LSD1‐target gene expression and HIF‐1 signature gene expression in a cancer cohort. Taken together, these findings lead us to propose the biological significance of FAD metabolism and LSD1 activity in HIF‐mediated hypoxic responses and their contribution to cancer progression.

Results

The lysine demethylase activity of LSD1 is required for the accumulation of HIF‐1α protein in hypoxia

To investigate the role of LSD1 in the cell proliferation of solid cancers, we estimated the sensitivity of 50 different cancer cell lines to siRNA‐mediated depletion of LSD1 and stratified them into three groups based on their dependence on LSD1 for cell proliferation (Table EV1 and Appendix Fig S1A and B). Cells displaying more than 65% of proliferation defects by LSD1 depletion were classified as Class I, while those exhibiting < 35% of defects were grouped as Class III. The remaining cells were categorized as Class II. We then determined the gene expression profile of representative cell lines from each group by microarray analyses and identified molecular features closely correlated with the LSD1 dependence of cell proliferation. Gene set enrichment analyses (GSEA) showed that among others, “glycolysis” was particularly highly enriched in Class I compared with Class III (Figs 1A and EV1A). This result was reproducible regardless of the tissue origin of different cancer cells. In fact, when human hepatocellular carcinoma (HCC) cells analyzed for the GSEA were examined, Class I cells displayed significantly higher average glycolysis activities than Class III cells as experimentally confirmed by measuring the levels of glucose uptake and lactate production (Fig 1B). These results indicate that cells with higher LSD1 dependence tend to exhibit higher glycolytic activity, thus leading us to hypothesize that LSD1 may have an important role in the glycolysis‐dependent proliferation of cancer cells. To test this hypothesis, we first examined the effect of LSD1 knockdown (KD) on cell proliferation under hypoxic conditions in which cellular glycolytic activities are upregulated by way of metabolic adaptation. We found that depletion of LSD1 significantly impaired cell proliferation, especially under hypoxia, with a tendency of Class I cells (Huh‐1, NCI‐H596) more significantly affected than Class III cells (PLC/PRF/5, Colo‐205; Fig 1C). Next, we determined the involvement of LSD1 in the elevation of glycolysis under hypoxia in these cell lines. Consistent with its effect on cell growth, LSD1 KD resulted in a significant reduction in both glucose uptake and lactate production in hypoxia (Fig EV1B and C). These results suggest that LSD1 positively regulates glycolytic activity in hypoxia.

Figure 1. The lysine demethylase activity of LSD1 is required for hypoxic accumulation of HIF‐1α protein.

1. Gene set enrichment analysis (GSEA) showing the enrichment of glycolysis pathway in the Class I as compared with Class III. Ten cell lines from Class I (HLF, JHH‐6, Huh‐1, Malme‐3M, JHH‐4, NCI‐H1299, NCI‐H596, MIA‐PaCa‐2, SK‐HEP‐1, and SNU‐475) and nine cell lines from Class III (SNU‐739, SNU‐423, SNU‐886, Calu‐1, BxPC‐3, Capan‐2, PLC/PRF/5, AsPC‐1, and Hep3B) were analyzed (left; various tissues). HCC cells from Class I (HLF, JHH‐6, Huh‐1, JHH‐4, SK‐HEP‐1, and SNU‐475) and Class III (SNU‐739, SNU‐423, SNU‐886, PLC/PRF/5, and Hep3B) were also analyzed (right; liver only). NES: normalized enrichment score, FDR: false discovery rate.
2. Assessment of glycolytic activities in five representative cell lines from either HCC Class I (red; HLF, JHH‐6, Huh‐1, JHH‐4, SK‐HEP‐1, and SNU‐475) or Class III (blue; SNU‐739, SNU‐423, SNU‐886, PLC/PRF/5, and Hep3B). Glucose uptake (left) and lactate production (right) were measured by detecting fluorescence at Ex/Em = 530/590 and colorimetric absorbance at OD 565 nm, respectively. Values are means ± SD of biological triplicate samples. Student's t‐test was performed to assess statistical significance indicated as P‐values.
3. Effects of LSD1 silencing on the proliferation of Class I (Huh‐1, NCI‐H596) and Class III (PLC/PRF/5, Colo‐205) cells. Cells transfected with indicated siRNA were incubated for 24 h in normoxia, cultured further for 96 h in normoxia (21% O₂) or under hypoxic conditions (3% O₂), and then, relative cell viabilities were measured by MTT assays. Values are means ± SD of biological triplicate experiments. P‐values were determined by Student's t‐test.
4. HIF‐1α expression in LSD1‐depleted cells under hypoxic conditions (1% O₂, 8 h). The expression levels of indicated proteins and mRNAs were detected by immunoblotting (IB) and RT–PCR, respectively.
5. Effect of LSD1 enzymatic activity on the hypoxic (1% O₂, 4 h) expression of HIF‐1α protein. HEK293T cells ectopically expressing either wild‐type (WT) or catalytically dead mutant (K661A) of LSD1 were determined for HIF‐1α expression by IB and RT–PCR.
6. Effect of the pharmacological inhibition of LSD1 enzymatic activity on the HIF‐1α protein level in NCI‐H596 cells. Cells were pre‐treated with 100 μM clorgyline, 4 mM pargyline, or 2 mM tranylcypromine for 24 h, subsequently cultured under hypoxic conditions (1% O₂, 8 h), and then assessed for the HIF‐1α protein level by IB.

Source data are available online for this figure.

Figure EV1. Inhibition of LSD1 activity attenuates the hypoxic elevation in glycolytic activity by decreasing the HIF‐1α protein accumulation.

1. Expression of genes analyzed in Fig 1A is displayed in the heat map. Gene rows are ordered according to gene ranks. Cell lines in each column are ordered according to hierarchical clustering using Euclidean distance as a metric. The genes in core enrichment group belonging to the leading edge subset are denoted in green box.
2. Effect of LSD1 loss on the glucose uptake in Huh‐1 (Class I) and Colo‐205 (Class III) cells under hypoxic conditions (1% O₂, 24 h). Values are means ± SD of biological triplicate experiments measuring fluorescence at Ex/Em = 530/590 nm normalized by cell numbers. P‐values were determined by Student's t‐test.
3. Effect of LSD1 silencing on the lactate production in Huh‐1 (Class I) and PLC/PRF/5 (Class III) cells under hypoxic conditions (1% O₂, 24 h). Values are means ± SD of biological triplicate experiments detecting absorbance at OD 565 nm normalized by cell numbers. P‐values were determined by Student's t‐test.
4. The activity of HRE reporter in LSD1‐depleted HEK293T cells under normoxic or hypoxic (1% O₂, 24 h) conditions was determined by luciferase assays. Values are means ± SD of biological triplicate experiments. P‐values were determined by Student's t‐test.
5. Heat map for the HIF‐1 target gene expression in LSD1‐depleted Huh‐1 cells (siLSD1) and its controls (siControl) in duplicates. The HIF‐1 target genes were selected from the previous genomic studies by Benita et al (2009), Ortiz‐Barahona et al (2010), and Xia et al (2009). Genes involved in glycolysis according to KEGG pathways were grouped as “Glycolytic” while the other genes were categorized as “non‐glycolytic”.
6. The level of ectopic HIF‐1α protein was determined in normoxic HEK293T cells expressing different amount of ectopic LSD1 by IB against the indicated antibodies. In parallel, mRNA levels were determined by RT–PCR.
7. The HIF‐1α levels in Fig 1E and related two more replicate immunoblots (shown in the source data) were quantified by Multi Gauge V3.0 (FUJIFILM). The signal intensity of HIF‐1α was normalized by that of ACTB. Values are means ± SD of biological triplicates. P‐values were determined by Student's t‐test.
8. The activities of HRE reporter in normoxic HEK293T cells co‐transfected with HA‐HIF‐1α along with either wild‐type (WT) or catalytically dead mutant (K/A: K661A) form of LSD1 were determined by luciferase assays. Values are means ± SD of biological triplicates. P‐values were determined by Student's t‐test. In parallel, mRNA levels were determined by RT–PCR.

Source data are available online for this figure.

Therefore, we examined the direct role of LSD1 in the regulation of hypoxic glycolysis. HIF‐1, known as the master regulator of hypoxia responses (Keith et al, 2012), plays a pivotal role in the upregulation of glycolysis. We first measured the effect of LSD1 on the activity of the 5 × HRE (hypoxia‐responsive element) reporter carrying five tandem repeats of the HIF binding site in HEK293T cells. LSD1 KD significantly decreased HRE reporter activity both in normoxia and in hypoxia (Fig EV1D). Consistently, siRNA‐mediated depletion of LSD1 resulted in a dramatic reduction of HIF‐1α protein accumulation under hypoxia in various cell types (Fig 1D and Appendix Fig S1C and D). Similar results were observed when cells were treated with the hypoxia‐mimetic CoCl₂ (Appendix Fig S1E). By contrast, in cells depleted of LSD1 by shRNA targeting the 3′ UTR of LSD1, the hypoxic induction of HIF‐1α protein was rescued by the ectopic expression of LSD1 (Appendix Fig S1F), indicating the direct involvement of LSD1 in the regulation of HIF‐1α protein accumulation. Supporting these observations, LSD1 depletion caused a significant reduction in HIF‐1 target gene expression including those involved in glycolysis (Fig EV1E and Appendix Fig S1G). Conversely, overexpression of LSD1 increased ectopic HIF‐1α protein expression even in normoxia, without affecting its mRNA level (Fig EV1F). In accordance, ectopic expression of wild‐type LSD1 along with HIF‐1α enhanced HIF‐mediated activation of the HRE reporter in normoxia (Fig EV1H). These results raise the possibility that LSD1 has a direct function in the accumulation of HIF‐1α irrespective of oxygen. By contrast, a catalytically dead mutant of LSD1, LSD1 (K661A), failed to increase HIF‐1α protein level in hypoxic cells (Figs 1E and EV1G) or facilitate HIF transcriptional activity (Fig EV1H). LSD1 inhibitors also blocked the hypoxic accumulation of HIF‐1α (Fig 1F) and the expression of HIF‐1 target genes under hypoxia (Appendix Fig S1H) at concentrations not affecting cell viability (Appendix Fig S1I and J). Together, these results demonstrate the essential role of LSD1 and its demethylase activity in the accumulation of HIF‐1α protein and the upregulation of glycolysis in hypoxia.

LSD1 regulates HIF‐1α protein stability in a RACK1‐dependent manner

The fact that LSD1 depletion blocks HIF‐1α protein accumulation under hypoxia without affecting its mRNA expression (Figs 1 and EV1, and Appendix Fig S1) indicates that LSD1‐mediated HIF‐1α regulation is achieved at the post‐transcriptional level. We therefore examined the molecular mechanism of LSD1 action by investigating its effects on the stability of HIF‐1α protein. Inhibition of proteasomal degradation by MG132 partially restored the accumulation of HIF‐1α protein in LSD1‐depleted NCI‐H596 cells (Fig 2A). This result suggests that LSD1 is involved in suppressing the proteasomal degradation of HIF‐1α. We then examined whether LSD1 interferes with the oxygen‐dependent HIF‐1α degradation mediated by von Hippel–Lindau (VHL) protein. Ablation of VHL did not restore the HIF‐1α protein level in LSD1‐depleted NCI‐H596 cells under hypoxia (Fig 2B). In addition, decreased hypoxic expression of HIF‐1α protein was still observed in RCC4 cells lacking functional VHL, in which LSD1 function was genetically or pharmacologically blocked (Fig 2C and D). Note that the LSD1 inhibitors had little cytotoxic effects on RCC4 cells at the test concentrations (Fig EV2A and B). These results suggest that LSD1‐dependent HIF‐1α stability is not mediated by the VHL‐dependent pathway. Moreover, ectopic expression of both wild‐type HIF‐1α and a mutant that was rendered resistant to the oxygen‐induced proteasomal degradation by introducing proline‐to‐alanine substitutions at proline residues 402 and 564 (P/A) was also suppressed by LSD1 depletion (Fig 2E). These results, along with those for LSD1 overexpression in normoxia (Fig EV1F and H), indicate that LSD1 stabilizes HIF‐1α protein via an oxygen‐independent pathway that does not require VHL. Therefore, we examined whether LSD1 is involved in either CHIP‐Hsp70 or RACK1‐Hsp90 pathways, both of which are the oxygen‐independent mechanisms of HIF‐1α degradation (Liu et al, 2007; Luo et al, 2010). We found that double KD of LSD1 along with RACK1 but not with CHIP restored the hypoxia‐induced HIF‐1α protein level (Fig 2F). RACK1 KD also restored HIF‐1α protein level in the cells treated with the LSD1 inhibitor, clorgyline under hypoxia (Fig 2G). These results demonstrate that LSD1 negatively regulates the RACK1‐mediated HIF‐1α degradation pathway. Interestingly, accumulation of HIF‐2α protein, another substrate of the RACK1‐mediated degradation machinery (Liu et al, 2007), was also compromised upon LSD1 depletion in 786‐O cells, which are genetically null for VHL and HIF‐1α (Fig EV2C), whereas LSD1 overexpression increased endogenous HIF‐2α protein in normoxic HEK293T cells (Fig EV2D). Collectively, these results indicate that LSD1 promotes the accumulation of HIF‐1α and possibly HIF‐2α proteins by suppressing their oxygen‐independent degradation via the RACK1‐Hsp90 pathway.

Figure 2. LSD1 inhibits the RACK1‐dependent degradation of HIF‐1α protein.

1. Effect of blocking proteasomal activity on the HIF‐1α protein levels in LSD1‐depleted cells. NCI‐H596 cells were cultured either in normoxia (21% O₂) or under hypoxic conditions (1% O₂) with or without MG132 (10 μM) for 4 h and then subsequently subjected to measuring protein levels by IB with the indicated antibodies.
2. Effect of silencing VHL on the HIF‐1α protein level in LSD1‐depleted NCI‐H596 cells under hypoxic conditions (1% O₂, 8 h). The expression levels of indicated proteins and mRNAs were detected by IB and RT–PCR, respectively.
3. Regulation of HIF‐1α protein expression by LSD1 in RCC4 (VHL^−/−) cells under either normoxic or hypoxic (1% O₂, 8 h) conditions. Protein levels were determined by IB with the indicated antibodies.
4. Effect of LSD1 inhibitors (C: 100 μM clorgyline, P: 4 mM pargyline, or T: 1 mM tranylcypromine, 16 h) on the accumulation of HIF‐1α in normoxic RCC4 cells. Protein levels were determined by IB against the indicated antibodies.
5. Examination of the oxygen independence of the HIF‐1α protein regulation by LSD1. The levels of either wild‐type (WT) or mutant (P/A: P402A, P564A) form of Myc‐tagged HIF‐1α ectopically expressed in LSD1‐depleted HEK293T cells in normoxia were measured by IB.
6. Examination of the role of CHIP‐Hsp70 and RACK1‐Hsp90 pathways in the LSD1‐mediated HIF‐1α regulation. Hypoxic (1% O₂, 8 h) accumulation of HIF‐1α protein in NCI‐H596 cells co‐transfected with siLSD1 along with either siRACK1 or siCHIP was monitored by IB analysis. mRNA levels were determined by RT–PCR.
7. Examination of the role of CHIP‐Hsp70 and RACK1‐Hsp90 pathways in HIF‐1α regulation using an LSD1 inhibitor. Hypoxic (1% O₂, 8 h) accumulation of HIF‐1α protein in either RACK1‐ or CHIP‐depleted NCI‐H596 cells pre‐treated with or without clorgyline (100 μM, 24 h) was monitored by IB analysis. mRNA levels were determined by RT–PCR.

Source data are available online for this figure.

Figure EV2. LSD1 is necessary for the RACK1‐dependent accumulation of HIFα subunits.

1. Cytotoxic effect of different LSD1 inhibitors or camptothecin at the indicated concentration on RCC4 cells was examined by FACS analyses for Annexin V and propidium iodide doubly stained RCC4 cells that were treated with the indicated drugs (C: clorgyline, P: pargyline, T: tranylcypromine) or DMSO alone for 16 h.
2. Percentile population of apoptotic cells determined in (A). Values are means ± SD of biological triplicates.
3. HIF‐2α protein levels in LSD1‐depleted 786‐O cells either in normoxia or under hypoxic conditions (1% O₂, 8 h) were monitored by IB against the indicated antibodies. Expression of the indicated mRNAs was determined by RT–PCR.
4. The levels of ectopic HIF‐2α in normoxic HEK293T cells expressing different amounts of ectopic LSD1 were determined by IB against the indicated antibodies. mRNA levels were assessed by RT–PCR.

Source data are available online for this figure.

LSD1 specifically regulates the methylation of lysine 271 residue on RACK1

We next investigated the molecular mechanism by which LSD1 suppresses the RACK1‐Hsp90 pathway in more detail. Co‐immunoprecipitation assays using HEK293T cells ectopically expressing RACK1 and LSD1 showed that LSD1 physically associates with RACK1 (Fig 3A). Consistently, endogenous RACK1 but not CHIP was detected in the immunoprecipitation complex for LSD1 that was ectopically expressed in HEK293T cells (Fig EV3A and B). We also observed the endogenous binding of LSD1 to RACK1 by detecting LSD1 in the endogenous RACK1 complex immunoprecipitated from either HEK293T or NCI‐H596 cells (Fig 3B). In vitro binding between recombinant GST‐LSD1 and His‐RACK1 proteins indicated that the two proteins directly interact with each other without an involvement of additional components (Fig EV3C). Despite the well‐defined role of LSD1 as a transcriptional regulator, transcription of RACK1 and Hsp90 was not affected by LSD1 depletion (Fig EV3D). These results raise the possibility that LSD1 post‐transcriptionally regulates RACK1 protein via its lysine demethylase activity. Examination by immunoprecipitating Flag‐tagged RACK1 from HEK293T cells, followed by immunoblotting with anti‐pan‐methyl‐lysine antibody, indicated that RACK1 was indeed methylated in vivo and that its methylation level was downregulated by LSD1 overexpression (Fig 3C). By contrast, LSD1 KD increased methylation on RACK1 (Fig 3D). As the methylation of RACK1 protein has not been reported previously, we carried out tandem mass spectrometry analyses on RACK1 immunoprecipitated from HEK293T cells and found that RACK1 is di‐methylated at lysine 271 residue (K271; Fig 3E). Supporting this finding, substitution of the K271 residue to alanine, but not the K172 residue, another candidate methylatable residue found by mass spectrometry analysis, prominently decreased the methylation signal of ectopically expressed RACK1 (Fig 3F). A polyclonal antibody directed against K271 di‐methylated peptide (Figs 3G and EV3E) detected the di‐methylation of K271 on RACK1 (RACK1K271me2) ectopically expressed in HEK293 cells (Fig 3H). The RACK1K271me2 signal was dramatically decreased by LSD1 overexpression in HEK293T cells, whereas it was increased, either coming from ectopically expressed or endogenous RACK1, by genetic depletion or pharmacological inhibition of LSD1 (Figs 3I and J, and EV3F and G). These results strongly suggest that methylation at K271 on RACK1 might be directly regulated by LSD1. To examine whether LSD1 indeed directly demethylates RACK1K271me2, we performed an in vitro LSD1 assay employing recombinant LSD1 protein and synthetic peptides encompassing the amino acid region 265–277 on RACK1. LSD1 demethylated the RACK1 peptide carrying K271me2 (RACK1K271me2) as well as the positive control peptide carrying di‐methylation of the K4 residue of histone H3 (H3K4me2; Fig 3K). By contrast, no LSD1 activity was observed toward the peptides mutRACK1K271me2 (carrying alanine substitution mutations adjacent to K271 residue) and RACK1K172me2 (encompassing the amino acid region 166–178 on RACK1 with di‐methylation at K172 residue), indicating that the lysine demethylase activity of LSD1 is specific to RACK1K271me2 in the given sequence context. This result reinforces that RACK1K271me2 is a bona fide non‐histone substrate of the lysine demethylase activity of LSD1.

Figure 3. LSD1 demethylates RACK1 protein at K271 residue.

1. Identification of LSD1‐RACK1 interaction by reciprocal co‐immunoprecipitations of ectopically expressed proteins from HEK293T cells.
2. Identification of physical interaction between endogenous RACK1 and LSD1 proteins. The interaction was detected by immunoprecipitating endogenous RACK1 from either HEK293T or NCI‐H596 cells, followed by IB with an anti‐LSD1 antibody.
3. Identification of the LSD1‐mediated regulation of RACK1 methylation. RACK1 protein methylation was monitored by immunoprecipitating ectopically expressed RACK1 from HEK293T cells with or without LSD1 overexpression, followed by IB analysis with the indicated antibodies including anti‐pan‐methyl‐lysine antibody (Pan‐Me‐K).
4. RACK1 protein methylation in LSD1‐depleted HEK293T cells was monitored as described in (C).
5. Mass spectrometric analysis of RACK1 immunoprecipitated from HEK293T cells indicates RACK1 methylation at K271 residue.
6. Experimental confirmation of the methyl‐lysine residue of RACK1 that is targeted by LSD1. Wild‐type (WT) and mutant forms (K172A or K271A) of RACK1 proteins were ectopically expressed in HEK293T cells, immunoprecipitated using an anti‐Flag antibody and then determined for their methylation content as described in (C).
7. Dot‐blot analysis of RACK1K271 peptides carrying mono‐ (me1), di‐ (me2), or tri‐(me3) methylation at the K271 residue as well as the non‐methylated peptide (me0) shows the specificity of anti‐RACK1K271me2 antibody.
8. Di‐methylation at the K271 residue on ectopically expressed RACK1 in HEK293T cells was determined as described in (C) except using anti‐RACK1K271me2 antibody (K271me2) in place of anti‐pan‐methyl‐lysine antibody.
9. Di‐methylation at the K271 residue on ectopically expressed RACK1 in LSD1‐depleted HEK293T cells was determined as described in (H).
10. Identification of the LSD1‐mediated regulation of methylation on endogenous RACK1 protein. Methylation on RACK1 protein in LSD1‐depleted HEK293T cells was monitored by immunoprecipitating endogenous RACK1, followed by IB with the indicated antibodies.
11. In vitro lysine demethylase activity assay was performed using recombinant LSD1 protein and the indicated synthetic peptides. Please see the Materials and Methods section for the amino acid sequences of the peptides. Values are means ± SD of biological duplicates. P‐values were determined by Student's t‐test.

Source data are available online for this figure.

Figure EV3. LSD1 downregulates RACK1 protein methylation.

1. Co‐immunoprecipitation of endogenous RACK1 with ectopically expressed LSD1 from HEK293T cells.
2. Endogenous CHIP protein was undetectable in the Myc‐LSD1 complex immunoprecipitated from HEK293T cells. The indicated proteins were determined by IB.
3. GST pull‐down assay detecting the in vitro physical interaction between recombinant GST‐LSD1 and recombinant His‐RACK1. The levels of His‐RACK1 and GST‐LSD1 were visualized by IB with anti‐histidine antibody and Coomassie blue staining, respectively.
4. Expression levels of the mRNAs from indicated genes including HSP90 and RACK1 in LSD1‐depleted NCI‐H596 cells were assessed by RT–PCR. The PDK1 mRNA level was monitored as a control to determine the upregulation of hypoxia‐responsive genes.
5. Dot‐blot assay showing the blocking effect of either RACK1K271me2 or RACK1K271me0 peptide on the activity of anti‐RACK1K271me2 antibody.
6. Methylation on ectopically expressed RACK1 protein in 100 μM clorgyline‐treated HEK293T cells (16 h) was monitored as described in Fig 3H.
7. Cytotoxicity of 100 μM clorgyline in HEK293T cells (16 h) was monitored by FACS analyses for Annexin V‐stained cells, compared to the 4 μM camptothecin‐induced apoptosis of HEK293T cells.

Source data are available online for this figure.

LSD1‐mediated demethylation of RACK1K271me2 inhibits physical interaction between RACK1 and HIF‐1α, suppressing HIF‐1α protein turnover

Next, we examined the biological consequence of the LSD1‐mediated demethylation of RAKC1K271me2, that is, its effect on HIF‐1α protein stability. Because RACK1 brings the Elongin C‐containing E3 ubiquitin ligase complex to HIF‐1α, leading to its proteasomal degradation, overexpression of RACK1 is known to reduce cellular HIF‐1α protein level in hypoxia (Liu et al, 2007). However, we found that unlike the ectopic expression of wild‐type RACK1, mutant RACK1(K271A) failed to decrease HIF‐1α protein level in hypoxic cells (Fig 4A). This result suggests that K271 is a critical residue for RACK1‐mediated degradation of HIF‐1α. An in vivo ubiquitination assay employing an HA‐tagged ubiquitin expression construct showed that whereas the ectopic expression of wild‐type RACK1 enhanced HIF‐1α ubiquitination, RACK1(K271A) mutant failed to do so (Fig 4B), suggesting that K271 methylation of RACK1 plays a critical role in the ubiquitination‐mediated degradation of HIF‐1α. As K271 is adjacent to the WD6 domain on RACK1, which is necessary for its association with HIF‐1α, we determined whether LSD1 controls the physical interaction between RACK1 and HIF‐1α by regulating the methylation of RACK1K271me2. In co‐immunoprecipitation assays, we observed that depletion of LSD1 dramatically enhanced the physical interaction between wild‐type RACK1 and HIF‐1α, whereas the binding of mutant RACK1(K271A) to HIF‐1α was negligible compared to that of wild‐type RACK1 and hardly affected by LSD1 KD (Fig 4C). Similar results were observed in experiments where the LSD1 activity was inhibited by clorgyline (Fig 4D). These results suggest that LSD1‐mediated demethylation of RACK1K271me2 inhibits the binding of RACK1 to HIF‐1α. Consistently, the amount of RACK1K271me2 bound to endogenous HIF‐1α was significantly increased by LSD1 depletion in NCI‐H596 cells (Fig 4E). Together, these results indicate that RACK1K271me2 facilitates the binding of RACK1 to HIF‐1α in vivo and that the LSD1‐mediated demethylation of RACK1K271me2 attenuates this interaction, thereby suppressing the ubiquitin‐mediated degradation of HIF‐1α.

Figure 4. LSD1‐mediated demethylation at the K271 residue on RACK1 prevents the physical association between RACK1 and HIF‐1α, suppressing HIF‐1α ubiquitination.

1. Identification of the role of K271 residue of RACK1 in the RACK1‐mediated regulation of HIF‐1α protein. The levels of endogenous HIF‐1α protein under hypoxic conditions (1% O₂, 4 h) were determined in HEK293T cells ectopically expressing either wild‐type (WT) or mutant (K271A) form of RACK1 by IB with the indicated antibodies.
2. An in vivo ubiquitination assay for the RACK1K271 dependence of the ubiquitination of HIF‐1α protein. HEK293T cells ectopically expressing wild‐type (WT) or mutant (K/A: K271A) form of RACK1 were assessed for the ubiquitination level of HIF‐1α.
3. Identification of the role of LSD1 and RACK1K271 in the interaction between RACK1 and HIF‐1α proteins. The physical association between RACK1 and HIF‐1α was monitored by immunoprecipitating the wild‐type (WT) or mutant (K/A: K271A) form of RACK1 from MG132‐treated (10 μM, 4 h) HEK293T cells with or without LSD1 depletion, followed by IB with the indicated antibodies.
4. Examination of the role of LSD1 activity and RACK1K172 in RACK1–HIF‐1α interaction. Binding between RACK1 and HIF‐1α proteins was examined in HEK293T cells treated with 100 μM clorgyline (16 h) as described in (C).
5. Examination of the role of K271 di‐methylation of RACK1 in the interaction between HIF‐1α and RACK1 proteins in vivo. The interaction was determined by immunoprecipitating endogenous HIF‐1α from NCI‐H596 cells grown in the presence of MG132 (10 μM) and desferrioxamine (1 mM) for 4 h, followed by IB with the indicated antibodies including anti‐RACK1K271me2 antibody.

Source data are available online for this figure.

Decrease in FAD level attenuates the LSD1‐mediated demethylation of RACK1 during prolonged hypoxia, causing downregulation of HIF‐1α expression

The robust induction of HIF‐1α protein generally occurs in the early stages of hypoxia and then gradually diminishes in later stages in most cell types (Keith et al, 2012). The kinetics of HIF‐1α protein expression might have important biological relevance to its function in hypoxia. We wondered whether RACK1K271me2 is dynamically regulated in different physiological conditions and hypothesized that LSD1‐mediated RACK1K271me2 dynamics might be involved in the downregulation of HIF‐1α protein during prolonged hypoxia. To address this question, we first examined the dynamics of RACK1K271me2 in NCI‐H596 cells during prolonged hypoxia (Fig 5A). Compared to those both in normoxia and during the early stages of hypoxia (i.e., at 3 h), the RACK1K271me2 signal on the immunoprecipitated RACK1 protein gradually increased during late hypoxia (i.e., at 12 and 24 h). Importantly, the increase in RACK1K271me2 was inversely correlated with the reduction in cellular HIF‐1α protein level. The late hypoxia‐dependent increase in RACK1K271me2 was also observed with ectopically expressed RACK1 after 24 h of hypoxia (Fig EV4A). These observations indicate that LSD1‐mediated RACK1K271me2 dynamics might indeed be responsible for the downregulation of HIF‐1α during prolonged hypoxia. However, neither the transcript nor the protein level of LSD1 was prominently altered during the course of hypoxia (Fig 5A), suggesting that the enzymatic activity but not the expression level of LSD1 might determine the RACK1K271 methylation profile during prolonged hypoxia.

Figure 5. Decrease in cellular FAD level during prolonged hypoxia causes downregulation of the LSD1‐mediated RACK1K271me2 demethylation and thereby reduces the HIF‐1α protein level.

1. The K271 di‐methylation profile of RACK1 protein during the progression of hypoxia. The RACK1K271me2 level was monitored by immunoprecipitating endogenous RACK1 from NCI‐H596 cells harvested at different times in hypoxia (1% O₂), followed by IB with the indicated antibodies. In parallel, the expression of different proteins and mRNAs from the cells harvested after the indicated times under hypoxia (1% O₂) was also determined by IB and RT–PCR, respectively. Immunoprecipitation with anti‐IgG antibody was performed using the cells harvested at 24 h in hypoxia.
2. FAD levels in NCI‐H596 cells were measured at different time points in hypoxia (1% O₂) by determining the fluorescence of probes at Ex/Em=530/590 nm. Then, the measured FAD levels were normalized by cell numbers. Values are means ± SD of biological triplicates. *P < 0.05 by Student's t‐test (6 h: P = 0.0000000246, 12 h: P = 0.0110, 24 h: P = 0.0012).
3. Examination of the RFK dependence of the HIF‐1α protein expression in hypoxia. RFK‐ or LSD1‐depleted NCI‐H596 cells were cultured under hypoxia conditions (1% O₂, 6 h), and the expression of the indicated proteins and mRNAs was assessed by IB and RT–PCR, respectively.
4. Effect of RFK depletion on the level of RACK1K271me2 was monitored by immunoprecipitating endogenous RACK1 from NCI‐H596 cells grown in normoxia, followed by IB with the indicated antibodies. The control IgG experiment was performed using the cells transfected with a siControl construct.
5. The effect of RACK1 silencing on the HIF‐1α protein accumulation in NCI‐H596 cells at different time points in hypoxia (1% O₂) was measured by IB with the indicated antibodies.
6. Effect of LSD1 loss or RFK silencing on migration and invasion of NCI‐H596 cells in hypoxia (1% O₂, 24 h). Values are mean cell numbers per field ± SD of biological triplicates. *P < 0.01 by Student's t‐test, respectively (Migration/siLSD1: P = 0.0005, Migration/siRFK: P = 0.0006, Invasion/siLSD1: P = 0.00005, Invasion/siRFK: P = 0.0007).
7. Effect of silencing LSD1, RFK, or HIF‐1α on tube formation of HUVEC cells in hypoxia (1% O₂, 24 h). Values are mean branch numbers per field ± SD for biological triplicates. *P < 0.01 by Student's t‐test (siControl: P = 0.00011, siLSD1: P = 0.0003, siRFK: P = 0.0003, siHIF‐1α: P = 0.0001).
8. Effect of replenishing the cellular FAD pool with exogenous FAD treatment on the HIF‐1α accumulation at late hypoxia. NCI‐H596 cells, which were depleted of LSD1 and/or RFK in different combinations, were cultured under hypoxia (1% O₂, 24 h) with 0.01% sodium deoxycholate only (−FAD) or 0.01% sodium deoxycholate and 100 μM FAD in combination (+FAD) and analyzed for the expression of different proteins including HIF‐1α and mRNAs by IB and RT–PCR, respectively.
9. Effect of exogenous FAD treatment on the RACK1271m1e2 level at late hypoxia. The very same NCI‐H596 cells treated as described in (H) were analyzed for the RACK1K271me2 level by immunoprecipitating the endogenous RACK1 protein, followed by IB with antibodies for RACK1K271me2 and RACK1.

Source data are available online for this figure.

Figure EV4. Changes in cellular FAD level modulate the LSD1‐mediated demethylation of RACK1K271me2.

1. The levels of RACK1K271me2 at 0 and 24 h in hypoxia (1% O₂) were monitored by immunoprecipitating ectopic RACK1 from NCI‐H596 cells expressing Flag‐RACK, followed by IB with the indicated antibodies.
2. Expression of the indicated proteins and mRNAs in Calu‐6 cells harvested at different times in hypoxia (1% O₂) was determined by IB with the indicated antibodies and RT–PCR, respectively.
3. FAD levels in Calu‐6 cells collected at different times in hypoxia (1% O₂) were measured by detecting fluorescence probes at Ex/Em = 530/590 nm, which was normalized by cell numbers. Values are means ± SD of biological triplicates. P‐values were determined by Student's t‐test.
4. FAD level in RFK‐depleted NCI‐H596 cells in normoxia was measured by detecting fluorescence probes at Ex/Em = 530/590 nm, which was normalized by cell numbers. Values are means ± SD of biological triplicates. P‐values were determined by Student's t‐test.
5. Effects of RFK silencing by two different siRNAs targeting RFK on the HIF‐1α protein stability in NCI‐H596 cells under hypoxic conditions (1% O₂, 6 h) were examined by IB with the indicated antibodies. mRNA levels were assessed by RT–PCR.
6. Hypoxic expression (1% O₂, 24 h) of representative HIF‐1α target genes either in LSD1‐ or in RFK1‐depleted NCI‐H596 cells was measured by quantitative RT–PCR (right). Values are means ± SD of biological triplicates. *P < 0.05 and **P < 0.005 by Student's t‐test (descriptive P‐values were described in Table EV2). Knockdown levels of LSD1 and RFK in cells taken at 6 h in 1% O₂ were determined by IB with the indicated antibodies and RT–PCR (left).
7. Effects of RFK depletion on the HIF‐mediated hypoxic responses such as glycolytic lactate production in NCI‐H596 cells either in normoxia or under hypoxic conditions (1% O₂, 24 h). Values in are means ± SD of biological triplicates. P‐values were determined by Student's t‐test.
8. Effect of exogenous FAD (100 μM) treatment on the FAD level in NCI‐H596 cells under hypoxic conditions (1% O₂, 24 h) was assessed by measuring intracellular FAD levels as described in (C). The exogenous FAD was treated as described in Fig 5H. Values are means ± SD of biological triplicates. *P < 0.001 by Student's t‐test (−siLSD1/−siRFK: P = 0.0001, +siLSD1/−siRFK: P = 0.0002, −siLSD1/+siRFK: P = 0.0001, +siLSD1/+siRFK: P = 0.0063).
9. The HIF‐1α levels in Fig 5E and a replicate immunoblot (shown in the source data) were quantified by Multi Gauge V3.0 (FUJIFILM). The signal intensity of HIF‐1α was normalized with that of ACTB. Values are means ± SD of biological duplicates. P‐values were determined by Student's t‐test.

Source data are available online for this figure.

FAD is a metabolic cofactor for LSD1 demethylase activity. We asked whether the level of FAD might be a regulating factor for LSD1‐mediated demethylation of RACK1K271me2 during prolonged hypoxia. Measurement of FAD level in NCI‐H596 cells during hypoxia indicated that cellular FAD level gradually decreases as hypoxia progresses in a manner that is inversely correlated with RACK1K271me2 level (Fig 5B). A similar correlative pattern of HIF‐1α stability and FAD availability was also observed in Calu‐6 cells (Fig EV4B and C). Investigating the reason for the decrease in FAD level during prolonged hypoxia, we observed that the expression of riboflavin kinase (RFK) and flavin adenine dinucleotide synthetase 1 (FLAD1), two enzymes essential in the FAD biosynthetic pathway, gradually diminished in the cells under prolonged hypoxic conditions, in parallel with a decrease in HIF‐1α protein level (Figs 5A and EV4B). These results strongly suggest that FAD level might play a rate‐limiting role in the LSD1‐mediated regulation of HIF‐1α stability during late hypoxia. To test this possibility, we examined the effect of RKF depletion on the hypoxic expression of HIF‐1α protein. RFK depletion clearly reduced FAD level in NCI‐H596 cells (Fig EV4D). In parallel, RFK KD led to a significant reduction in the hypoxia‐induced HIF‐1α protein level, similar to LSD1 depletion (Figs 5C and EV4E). In addition, the amount of RACK1K271me2 increased in RFK‐depleted cells, reflecting the downregulation of LSD1 enzymatic activity (Fig 5D). By contrast, silencing of RACK1 increased the accumulation of HIF‐1α protein in late hypoxia (Fig 5E), suggesting the involvement of RACK1‐mediated degradation in the decline of HIF‐1α during prolonged hypoxia. We then investigated whether the RFK‐mediated FAD supply profile during hypoxia indeed parallels the HIF‐mediated response patterns in hypoxic cells. Consistent with its effect on HIF‐1α protein stability, RFK depletion significantly decreased the hypoxic expression of HIF‐1α target genes (Fig EV4F). Accordingly, RFK depletion caused defects in the HIF‐induced phenotypes such as glycolytic lactate production, cell motility, and induction of tube formation, to the levels comparable to the loss of LSD1 (Figs 5F and G, and EV4G, and Appendix Fig S2A and B). These results demonstrate that supply of sufficient FAD via the RFK‐dependent biosynthetic pathway is critical for the accumulation of HIF‐1α protein and hypoxia responses. In support of this notion, FAD supplementation experiments showed that addition of exogenous FAD to the growth medium could compensate for the deficiency in FAD level caused by RFK KD or prolonged hypoxia (1% O₂, 24 h; Fig EV4H). Further, FAD supplementation could successfully restore the HIF‐1α protein accumulation for the cells in late hypoxia in an RFK‐independent manner, while reducing the RACK1K271me2 level (Figs 5H and I, and EV4I). Importantly, the effect of exogenous FAD treatment was mitigated upon the KD of LSD1, implying that the FAD‐dependent upregulation of HIF‐1α stability was mediated by LSD1 activity. Collectively, these results indicate that cellular FAD level can act as an important rate‐limiting factor for HIF‐dependent hypoxia responses via modulation of LSD1‐dependent RACK1 methylation dynamics and HIF‐1α stability. It is also notable that attenuation of LSD1‐dependent HIF‐1α stability caused by the downregulation of FAD biosynthetic enzymes is directly responsible for the downregulation of HIF‐1α protein level in later stages of hypoxia.

HIF‐1 signature in TNBC patients is positively correlated with the enzymatic activity of LSD1 and expression of FAD biosynthetic enzymes

Given the role of LSD1 in mediating the stabilization of HIF‐1α protein in experimental models, we wondered whether the activity of LSD1 bears any significance to HIF‐1 activity in human cancers. Triple‐negative breast cancer (TNBC) is strongly associated with a particularly high HIF‐1 pathway signature that is associated with poor prognosis (Chen et al, 2014). We re‐analyzed a publically available TNBC dataset for 579 patient samples (GSE31519), of which 114 cases had clinical information available, and confirmed the intimate association of the HIF‐1 signature with poor patient outcome. However, the expression levels of LSD1 mRNA in the patient group with a high HIF‐1 signature were not significantly different from those in the group with a low HIF‐1 signature (Appendix Fig S3). Next, we re‐classified the TNBC patient samples by LSD1 activity, which was determined from the average expression of LSD1 target genes as identified by Lim et al (2010b). Overall survival analysis showed that the patient group exhibiting higher levels of LSD1 target gene expression displayed significantly poorer prognosis than the patient group with lower levels of LSD1 target gene expression (Fig 6A). This result suggests that LSD1 activity is associated with the poor prognosis of TNBC patients. Furthermore, the average expression of LSD1 signature genes from all 579 TNBC samples was strongly correlated with the HIF‐1 pathway signature (R = 0.452; Fig 6B), suggesting that LSD1 activity is significantly involved in regulating the HIF‐1 pathway in TNBC, thereby affecting cancer progression. We also found that the expression of RFK and FLAD1 was highly correlated not only with the expression of the LSD1 signature but also with HIF activity (Fig 6C and D), implying that cellular FAD production affects the expression of HIF target genes by modulating LSD1 activity. In support of this possibility, the expression of FAD biosynthetic genes was significantly correlated with the poor prognosis of TNBC patients (Fig 6E). These results collectively suggest that LSD1 may have significant clinical relevance in TNBC patients by promoting FAD‐dependent upregulation of the HIF pathway.

Figure 6. The HIF‐1 signature is positively correlated with the expression of LSD1 target genes along with the expression of FAD biosynthetic enzymes in TNBC.

1. Kaplan–Meier graphs showing a significant association of high LSD1 signature expressing group (top 30%) with shorter survival as compared to those of low LSD1 signature group (bottom 30%) in a cohort of 114 TNBC patients. The log‐rank test P‐value is shown.
2. Positive correlation between the expression of LSD1 target genes (LSD1 signature) and HIF‐1 signature in a cohort of 579 TNBC patients. P‐value was calculated by the log‐rank test. R: correlation coefficient.
3. Positive correlation between the expression of FAD biosynthetic enzymes, RFK and FLAD, and LSD1 signature in a cohort of 579 TNBC patients. P‐value was calculated by the log‐rank test. R: correlation coefficient.
4. Positive correlation between the expression of FAD biosynthetic enzymes, RFK and FLAD, and HIF‐1 signature in a cohort of 579 TNBC patients. P‐value was calculated by the log‐rank test. R: correlation coefficient.
5. Kaplan–Meier graphs showing a significant association of high RFK/FLAD1 expressing group (top 30%) with shorter survival as compared to those of low expression group (bottom 30%) in a cohort of 114 TNBC patients. The log‐rank test P‐value is shown.
6. Schematic model for the LSD1‐mediated regulation of HIF‐1α protein stability by controlling RACK1 protein methylation in hypoxia.

Discussion

In this study, we present a mechanistic insight into how the FAD‐dependent lysine demethylase LSD1 regulates HIF‐induced responses, such as the upregulation of glycolysis in hypoxia. Our findings showed that (i) the lysine demethylase activity of LSD1 is critical for HIF‐1α protein accumulation by preventing its oxygen‐independent degradation by the RACK1‐Hsp90 pathway, (ii) LSD1 is responsible for the demethylation of K271me2 on RACK1 (a component of the HIF‐1α ubiquitination machinery), resulting in the inhibition of the physical interaction of RACK1 with HIF‐1α (Fig 6F), (iii) a change in the level of FAD, a metabolic cofactor for LSD1 demethylase activity, plays a rate‐limiting role in HIF‐1α stability during prolonged hypoxia, and (iv) LSD1 activity and the expression of FAD biosynthesis enzymes are strongly correlated with HIF‐1 signature gene expression in human cancer.

The growing list of non‐histone substrates for LSD1 raises its importance beyond the role in chromatin modification. Our finding that LSD1 demethylates K271me2 of RACK1 and thereby protects HIF‐1α from the RACK1‐mediated degradation reinforces this point. Previously, Qin et al reported that HIF‐1α might be regulated by LSD1 in pancreatic cancers (Qin et al, 2014). They suggested LSD1 as a recruiting factor for histone lysine deacetylases to HIF‐1α, which consequently suppresses acetylation‐dependent HIF‐1α degradation. However, our analyses with a catalytically dead LSD1 mutant and different LSD1 inhibitors (Figs 1 and 2, and EV1 and EV2) clearly indicate a more direct role of LSD1 as an enzyme in the regulation of HIF‐1α protein stability, because the lysine demethylase activity of LSD1 was required for increasing HIF‐1α stability. Supporting this finding, regulation of HIF‐1α protein methylation by LSD1 and Set7/9 lysine methyltransferase has been recently reported (Kim et al, 2016). By contrast, our cellular and in vitro assays showed that LSD1 demethylates the di‐methylation of K271 residue on RACK1 (Fig 3). Furthermore, K271A mutation on RACK1 rendered the physical association between RACK1 and HIF‐1α irresponsive to the activity of LSD1 (Fig 4). These results reveal a novel mechanism of the non‐chromatic action of LSD1; that is, LSD1‐mediated demethylation of K271 on RACK1 inhibits the binding of RACK1 to HIF‐1α and consequently suppresses RACK1‐mediated HIF‐1α degradation. The level of HIF‐2α, another target of RACK1‐dependent degradation, was also affected by LSD1 (Fig EV2), indicating that the non‐histone lysine demethylase function of LSD1 plays a key regulatory role in the RACK1‐mediated ubiquitination pathway that is common for the degradation of both HIF‐α subunits.

Methylation of lysine residues on histones or non‐histone proteins such as p53 produces a binding motif for interacting proteins (Huang et al, 2007; Musselman et al, 2012). RACK1, being composed of seven WD domains, is a multifaceted scaffolding protein that physically interacts with several proteins including HIF‐1α (Adams et al, 2011). The lysine 271 residue on RACK1 is located at the region extended from the WD6 domain, which is the minimal region for cellular HIF‐1α binding (Liu et al, 2007). Based on our finding that the overexpression of wild‐type RACK1, but not RACK1(K271A) mutant, decreased HIF‐1α protein level (Fig 4), we postulate that K271 residue participates in the formation of the binding surface for HIF‐1α and that modification of K271 can exert crucial effects on the interaction between RACK1 and HIF‐1α. RACK1 and Hsp90 competitively bind to the PAS‐A domain located at the N‐terminal region of HIF‐1α (Liu & Semenza, 2007). Although the PAS‐A domain does not display structural similarity to defined methyl‐lysine binders, our findings suggest that RACK1 with K271me2 acquires a stronger HIF‐1α binding affinity than Hsp90, balancing subsequent reactions toward the proteasomal degradation of HIF‐1α. Future identification of a lysine methyltransferase responsible for RACK1K271 methylation will increase our understanding of the physiological role of RACK1 as a regulator of HIF‐mediated responses induced in different microenvironments.

The enzymatic activity of LSD1 requires FAD, a metabolite derived from riboflavin and ATP. The LSD1‐mediated enzymatic reaction converts FAD to a reduced form, FADH₂ (Barile et al, 2000). In this study, we found that the cellular amount of FAD decreases in prolonged hypoxia, as the expression of the rate‐limiting enzymes of FAD biosynthesis, such as RFK and FLAD1, was downregulated with the progression of hypoxia (Fig 5). Further, we observed the RACK1‐dependent diminution of HIF‐1α protein in prolonged hypoxia, which was correlated with the decrease in FAD and increase in RACK1K271me2 in the absence of a prominent change in LSD1 expression at both mRNA and protein levels. These findings indicate that cellular FAD level and FAD‐dependent LSD1 activity regulating RACK1K271me2 might be the major determinants of HIF‐1α stability during prolonged hypoxia. In line with these findings, ablation of RFK expression reduced not only the cellular amount of FAD but also HIF‐1α protein level, consequently downregulating HIF‐1 target gene expression. Furthermore, either RACK1 depletion or exogenously supplied FAD facilitated the induction of HIF‐1α protein in FAD‐depleted conditions, such as prolonged hypoxia. It is notable that the HIF‐1α level in RACK1‐depleted cells at late hypoxia was still lower than that at early hypoxia. Considering the previous findings that the upregulation of PHD3 was implicated in regulating the HIF‐1α protein stability under chronic hypoxia in some models (Appelhoff et al, 2004; Ginouves et al, 2008), we speculate that the induction of PHD3 activity and the suppression of LSD1 activity due to decrease in cellular FAD level might function in combination to downregulate the HIF‐1α protein stability during prolonged hypoxia.

FAD availability in cells can also be modulated by the conversion between FAD and FADH₂ through oxidoreductive reactions involved in key metabolic pathways, such as fatty acid β‐oxidation and tricarboxylic acid (TCA) cycle coupled with oxidative phosphorylation. Metabolic regulation of lysine methylation has emerged as an important regulatory mechanism of diverse cellular pathways (Black et al, 2012). The physiological regulation of JmjC‐containing histone lysine demethylases by the fluctuation of TCA cycle intermediates, such as α‐ketoglutarate, succinate, and fumarate, has been shown in a number of studies (Gut & Verdin, 2013). Modulation of LSD1 activity by cellular FAD availability has recently been reported by Hino et al, who presented a correlation between the physiological FAD pool and LSD1 activity in adipose differentiation (Hino et al, 2012). Our analyses of hypoxic cells reinforced the model that LSD1 activity is physiologically modulated by a change in cellular FAD amount. Taken together, our findings suggest that FAD level is an important metabolic modulator of HIF‐mediated responses, including the elevation of glycolysis. Recently, the role of LSD1 in coordinating glycolytic and mitochondrial metabolism was reported (Sakamoto et al, 2015). Indeed, their observation that LSD1 is necessary for the VHL‐independent accumulation of HIF‐1α protein is consistent with our results. It would be interesting to examine whether the role of LSD1 in mediating the metabolic shift to the glycolytic pathway can be regulated by changes in cellular FAD level in different physiological contexts.

Proteasomal degradation of HIF‐1α protein plays a pivotal role in regulating HIF signature gene expression. Notably, HIF‐1α protein level is elevated not only under a hypoxic microenvironment but also in oncogenic conditions of various cancers. Indeed, different defects in proteasome‐dependent HIF‐1α degradation pathways, such as VHL inactivation or SHARP1 overexpression, cause the elevation of HIF‐1α protein level irrespective of oxygen, showing a strong association with cancer progression (Montagner et al, 2012; Gossage et al, 2015). Considering HIF‐1α as a valuable target for cancer treatment, a few reagents have been developed to facilitate HIF‐1α degradation as a means of inactivating HIF‐mediated responses. Our results shed a light on the development of a combination therapy employing LSD1 inhibitors in conjunction with HIF‐1α inhibitors for the treatment of cancers that are highly associated with the HIF pathway. Furthermore, this study suggests the modulation of FAD production as a potential therapeutic strategy to regulate LSD1 and HIF pathway activities.

Materials and Methods

Cell culture

Cell lines obtained from American Type Culture Collection and Korean Cell Line Bank were maintained in Dulbecco's modified Eagle's medium (DMEM; Gibco BRL, Grand Island, NY) or Roswell Park Memorial Institute 1640 medium (RPMI1640; GIBCO BRL). Human umbilical vein endothelial cells (HUVECs) were maintained in endothelial cell basal medium 2 (EBM‐2, Lonza). JHH cell lines purchased from Japanese Collection of Research Bioresources Cell Bank were maintained in William's E medium (GIBCO BRL). All media were supplemented with 10% fetal bovine serum (GIBCO BRL) and 1× antibiotic–antimycotic (GIBCO BRL). Hypoxic or hypoxia‐mimetic culture conditions were constructed either with O₂/CO₂ incubator (1–3% O₂, 5% CO₂, and 92–94% N₂, MCO‐5M, Sanyo, Japan) or with cobalt chloride (1 mM CoCl₂, Sigma‐Aldrich, St. Louis, MO, USA), respectively.

Reagents, transfection, and antibodies

LSD1 inhibitors (clorgyline, pargyline, and tranylcypromine), flavin adenine dinucleotide (FAD), and desferrioxamine (DFO) were purchased from Sigma‐Aldrich. MG132 was purchased from Calbiochem. Lipofectamine Plus (Invitrogen, Carlsbad, CA, USA) and RNAiMAX (Invitrogen) reagents were employed for DNA plasmid and siRNA reverse transfection, respectively. The primary and secondary antibodies used in this study are listed in Appendix Table S1.

Plasmids

Full‐length LSD1 (BC048134) clone was purchased from OpenBiosystem (USA). HA‐HIF‐1α, HA‐HIF‐1α P402A/P564A, HA‐HIF‐2α, pEGFP‐N1‐RACK1, and HA‐ubiquitin clones were purchased from Addgene (USA). Myc/His‐tagged LSD1 and HIF‐1α were subcloned into EcoRI/KpnI and BamHI/KpnI sites of pcDNA™3.1/myc‐His(−) vector (Invitrogen), respectively. Flag‐tagged RACK1 and Myc/His‐tagged RACK1 were subcloned into HindIII/KpnI site of pcDNA3.1/C‐Flag vector (Invitrogen) and KpnI/HindIII sites of pcDNA™3.1/myc‐His(−) vector (Invitrogen), respectively. Site‐directed mutagenesis was performed using a KOD‐Plus‐Mutagenesis Kit (Toyobo, Japan) according to manufacturer's instructions. Primers used for DNA subcloning and mutagenesis are described in Appendix Table S2. Small interfering RNA (siRNA) and lentiviral‐based short hairpin RNAs (shRNAs) are listed in Appendix Table S3.

Microarray analysis

Gene expression profile was executed using an Illumina BeadChip Array platform (Illumina, San Diego, CA, USA). Using RNeasy Mini kit (QIAGEN, Valencia, CA, USA), total RNA was isolated, and then, the extracted RNA integrity was determined with Experion™ RNA StdSens (BIO‐RAD Laboratories, Hercules, CA, USA). Illumina TotalPrep RNA Amplification kit (Applied Biosystem/Ambion, Austin, TX, USA) was used to generate biotin–UTP‐labeled cRNA according to manufacturer's instructions. Labeled cRNA (750 ng) was hybridized to HumanHT‐12 v3 chips at 58°C for 16 h. After hybridizing procedures, chips were scanned using a BeadArray Reader and BeadScan software (Illumina). Data extraction and normalization were elevated by BeadStudio ver.3 software (Illumina). Gene set enrichment analysis (GSEA) was conducted using GSEA v2.1.0 software (Gene Set Enrichment Analysis; Broad Institute). The accession numbers of the microarray data in GEO are GSE79232 and GSE85345.

Cell proliferation assay

Cell viability was measured using CellTiter‐Blue^® Cell Viability Assay (Promega, Madison, WI, USA) or 3‐(4,5‐dimethylthiazol‐2‐yl)‐2,5‐diphenyltetrazolium bromide (MTT, Sigma‐Aldrich) assay. To determine LSD1 dependence, 2 × 10³ cells were transiently reverse‐transfected with 20 nM of siLSD1 (QIAGEN) or control siRNA (Bioneer, Daejeon, Korea) in 96‐well plates. After 120 h, CellTiter‐Blue^® Reagent was added, and in 2 h, fluorescence signal at Ex/Em = 530/590 nm was evaluated by fluorometer (SYNERGY/HTS, BioTek). Proliferation rate in hypoxic conditions for 96 h following 1 day after siRNA transfection was determined by MTT assay as described in Mosmann (1983).

Glucose uptake and lactate production assay

Glucose uptake and lactate production were measured with Glucose Uptake Fluorometric Assay Kit (BioVision, Milpitas, CA, USA) and EnzyChrom™ L‐Lactate Assay Kit (BioAssay Systems, Hayward, CA, USA), respectively. All assays were performed according to manufacturer's protocols. To evaluate glucose uptake levels, cells were seeded in 96‐well plates (2 × 10³), and then, after 48 h, fluorescence at Ex/Em = 530/590 nm was determined by fluorometer (BioTek). To quantify lactate production levels, cells transfected with 0.5 μM of siRNAs in 6‐well plates (3 × 10⁶) for 24 h were exposed to normoxic or hypoxic conditions. After another 24 h, supernatants were collected to assess the lactate level by optical densitometer at 565 nm (LUMINOSKAN, Thermo Scientific).

RT–PCR

Total RNA was isolated using a Trizol reagent (Invitrogen) and assessed with a ND‐1000 spectrophotometer (NanoDrop, USA). First‐strand cDNA synthesis was executed with the RevertAid™ M‐MuLV Reverse Transcriptase (Thermo Scientific) according to manufacturer's methods. RT–PCR was performed with an AccuPower PCR Master Mix (Bioneer) and GeneAmp PCR System 970. The PCR products were electrophoresed on 1% agarose gels and stained with RedSafe™ (iNtRON, Seoul, Korea). PCR products were visualized with GelDoc™ XR+ and ImageLab™ Software (BIO‐RAD). Quantitative real‐time RT–PCR was carried out using FastStart Universal SYBR Green Master (Rox) (Roche, Mannheim, Germany) on a CFX96™ Real‐Time PCR machine (BIO‐RAD). Primers used for RT–PCRs are listed in Appendix Table S4.

Luciferase assay

Luciferase assays were performed using a Dual‐Luciferase Reporter Assay System (Promega), according to manufacturer's instructions. Cells (7.5 × 10⁴) cultured in 12‐well plates for 24 h were co‐transfected with different combinations of luciferase reporter (5 × HRE; 0.2 μg), Renilla (0.05 μg) and the plasmids expressing [shGFP, shLSD1, HIF‐1α (0.5 μg), wild‐type LSD1 (0.5 μg), and LSD1K661A (0.5 μg)]. In 24 h after the transfection, cells were exposed to either normoxic or hypoxic conditions for additional 24 h. Luciferase activities were determined using luminometer (LUMINOSKAN ascent, Thermo Scientific). Data represent the signals normalized with Renilla luciferase activity.

Immunoblot and immunoprecipitation

Immunoblotting was executed according to the standard method. Cells were lysed in RIPA buffer (50 mM Tris–HCl pH 7.5, 150 mM NaCl, 10 mM EDTA, 1% Triton X‐100) with protease inhibitor cocktail (Roche) or 2× sample buffer (120 mM Tris–HCl pH 6.8, 20% glycerol, 4% SDS, 5% β‐mercaptoethanol). Immunoprecipitation was performed using ANTI‐Flag M2 (Sigma‐Aldrich) and Ezview™ Red Anti‐c‐Myc Affinity Gel (Sigma‐Aldrich). Immunoprecipitation of endogenous RACK1 was conducted using anti‐RACK1 antibody (10 μg for 10 mg of cell extracts) and Pierce™ Protein L Plus Agarose (Thermo Scientific). Equal amount of whole‐cell lysate or IP complexes were resolved by 10–15% SDS–PAGE. Blot was analyzed by horseradish peroxidase (HRP)‐conjugated antibody and enhanced chemiluminescence (ECL, Pierce). The immunoblots were developed with LAS‐4000 (Luminescent Image analyzer, FUGIFILM). The signal intensities of immunoblots were quantified with Multi Gauge V3.0 (FUJIFILM).

LC/MS

The gel bands of immunoprecipitated RACK1 from HEK293T cells were excised and in‐gel digested (Park et al, 2013). The peptide extracts were lyophilized and analyzed by LC‐MS/MS using Nano Acquity UPLC system (Waters, USA) interfaced with a LTQ Orbitrap Elite mass spectrometer (Thermo Scientific, USA), equipped with a nano‐electrospray device. The MS/MS spectra were analyzed using Proteome discover software version 1.3 with the protein sequence of RACK1 (UniProtKB‐P63244).

In vitro LSD1 demethylase assay

In vitro LSD1 demethylase assay was performed using Histone Demethylase Assay (Active Motif, Carlsbad, CA, USA) according to manufacturer's instructions. Recombinant LSD1 proteins (1 pmol) were added to synthetic peptides (~70 pmol) encompassing the residues 265–277 on RACK1 with or without the di‐methylation of K271 residue (RACK1K271me0, RACK1K271me2), RACK1K271me2 peptide carrying alanine substitutions at the amino acids adjacent the K271 (mutRACK1K271me2), or H3K4me2 peptides provided by the vendor as a positive control for the assay (H3K4me2). A synthetic peptides encompassing the 166–178 amino acid region on RACK1 with di‐methylation of K172 residue (RACK1K172me2) were also used for the assay. Reaction mixtures were incubated for 0.5 h at 37°C, and then, demethylase activity was measured with fluorometer (BioTek) by Ex/Em = 410/480 nm. The peptide sequences were as follows: RACK1K271me0, I‐I‐V‐D‐E‐L‐K‐Q‐E‐V‐I‐S‐T; RACK1K271me2, I‐I‐V‐D‐E‐L‐Kme2‐Q‐E‐V‐I‐S‐T; mutRACK1K271me2, I‐I‐V‐D‐A‐A‐Kme2‐A‐A‐V‐I‐S‐T; RACK1K172me2, V‐S‐C‐G‐W‐D‐K(me2)‐L‐V‐K‐V‐W‐N (synthesized by PEPTRON, Korea).

In vivo ubiquitination assay

For in vivo ubiquitination assays, HEK293T cells co‐transfected with HA‐tagged ubiquitin plasmid, Myc‐tagged HIF‐1α, and Flag‐tagged RACK1 (either wild‐type or K271A mutant) were treated with the 10 μM proteasome inhibitor MG132 for 4 h. Myc‐tagged HIF‐1α was precipitated from whole‐cell lysates prepared with RIPA buffer using precleared anti‐Myc affinity gel (Sigma‐Aldrich) and then resolved in a 10% SDS–PAGE gel. To detect ubiquitination pattern of HIF‐1α, Western blot analysis using anti‐HA antibody was executed.

Measurement of FAD level in cells

Intracellular flavin adenine dinucleotide (FAD) amounts were evaluated with FAD Colorimetric/Fluorometric Assay Kit (BioVision) according to manufacturer's instructions. Cells seeded in 6‐well plates (3 × 10⁶) were exposed either to normoxic or 1% hypoxic conditions for 0, 3, 6, 12, and 24 h and harvested with ice‐cold phosphate‐buffered saline (PBS). Then, perchloric acid precipitation was performed using Deproteinizing Sample Preparation Kit (BioVision). FAD amounts were assessed by fluorometer (BioTek) at Ex/Em = 530/590 nm. Then, the FAD amounts were normalized by cell numbers. For exogenous FAD supplementation, cells were seeded in 6‐well plates (3 × 10⁶). After 24 h, the cells were treated with 0.01% sodium deoxycholate for 1 min, mixed with 100 μM FAD following PBS washing, and subsequently incubated under normoxic or hypoxic conditions.

Migration/invasion assays

For the invasion assay, the upper chambers were coated with matrigel (Becton Dickinson, Vedford, MA). NCI‐H596 cells were reverse‐transfected with 0.5 μM siRNAs targeting the indicated genes. At 24 h after the transfection, the cells were shifted to hypoxia (1% O₂). At 24 h after the hypoxic incubation, the cells were harvested with 0.025% trypsin–EDTA and seeded (1 × 10⁵ cells) in inserts. At 24 h after, the inserts were fixed with 100% methanol and stained with 0.5% crystal violet. Invaded cells invaded counted manually. Migration assay was conducted as described for the invasion assay without matrigel coatings.

Tube formation assays

Human umbilical vein endothelial cells (HUVECs) were reverse‐transfected with 0.5 μM siRNAs targeting the indicated genes. At 24 h after the transfection, cells were grown in hypoxia (1% O₂) for 24 h. After the hypoxic incubation, cells harvested with 0.025% trypsin–EDTA at 24 h were seeded (2.5 × 10⁴ cells) on matrigel‐coated 96‐well plates. At 4 h after, tube formations were observed under a light microscope.

GST pull‐down assay

Recombinant GST‐tagged LSD1 (H00023028‐P01) and recombinant His‐tagged RACK1 protein were purchased from Abnova and Abcam, respectively. One μg of recombinant GST or GST‐tagged LSD1 protein was incubated with GST‐Bind Agarose (ELPIS Biotech) for overnight at 4°C, followed by the incubation with 1 μg of His‐tagged RACK1 for additional 6 h. After washing the GST pull‐down complex with RIPA buffer three times, the proteins bound to the agarose were eluted by 2× sample buffer.

Fluorescent‐activating cell sorting (FACS) analysis

Cells harvested with trypinase were washed with ice‐cold PBS containing 2% FBS twice, followed by re‐suspension in 1× Annexin‐binding buffer (Molecular Probes). Subsequently, the cells were incubated with Annexin V‐FITC and propodium iodide for 10 min. Flow cytometry analyses were performed using BD FACSVerse System.

Public microarray data analysis

To generate Kaplan–Meier survival curves and to determine correlation coefficients among LSD1 and HIF‐1 pathway signatures and RFK/FLAD1 expression, we used GSE31519 dataset carrying gene expression profiles of 579 TNBC patient tissues. Among them, dataset of 114 TNBC patients with prognosis information was extracted and analyzed for the Kaplan–Meier survival analyses. Patients were considered to have a “high” HIF‐1 (VEGFA, SLC2A1, KDM3A, PDK1, DDIT4, NDRG1, PFKFB3, PIK3CA, RORB, CREBBP, PIK3CB, HK2, EGLN1) or LSD1 (E2F1, MIK67, CCNA2, CCNF, CDC25A, CENPF, MYBL2, PLK1, SKP2) signature, or RFK/FLAD1 expression if they had average expression values of all the genes belonging to the indicated signature above the 30^th percentile. Patients with the average expression values of the same genes below the 30^th percentile were considered to have a “low” signature.

Statistical methods

Data are presented as means ± SD. Statistical significance was determined by Student's t‐test. The Kaplan–Meier survival curves and the correlation curves to determine the Pearson correlation coefficients were generated with GraphPad Prism (GraphPad Software).

Author contributions

ZX, YW, DP, KCP, and YIY conceived the project, and S‐JY, KCP, J‐AK, and YIY designed the experiments. S‐JY performed most of the molecular biology experiments and conducted GSEA analyses. S‐JY and YSP carried out the experiments to stratify cell lines according to the LSD1 dependence of cell growth. JHC carried out FACS analyses and tube formation assays. BM carried out migration and invasion assays. H‐JA carried out HIF‐2α‐related assays. JYL and JYK conducted mass spectrometry analyses. DCL and HAS helped with DNA subclonings. MK helped with statistical analyses. J‐AK analyzed public TNBC dataset. S‐JY, J‐AK, and YIY wrote the manuscript. EK, KCP, J‐AK, and YIY supervised the project.

Conflict of interest

The authors declare that they have no conflict of interest.

Supporting information

Appendix

Expanded View Figures PDF

Table EV1

Table EV2

Source Data for Expanded View and Appendix

Review Process File

Source Data for Figure 1

Source Data for Figure 2

Source Data for Figure 3

Source Data for Figure 4

Source Data for Figure 5

The EMBO Journal (2017) 36: 1011–1028