Novel Histone Demethylase LSD1 Inhibitors Selectively Target Cancer Cells with Pluripotent Stem Cell Properties

Histone methylation is a major covalent modification of histones that epigenetically regulates cell-specific gene expression patterns (1, 2). While methylations of histone H3 at lysines 9 (H3K9) and 27 (H3K27) suppress gene expression, the methylations of lysine 4 (K4) in histone H3 (H3K4) usually associate with actively transcribed genes (1–3). Histone methylation is dynamically controlled by specific histone methyltransferases and demethylases (3). Specifically, the methylations at H3K4 are primarily catalyzed by histone methyltransferase complexes composed of the members of MLL (mixed lineage leukemia) SET-domain methyltransferases, ASH2, WDR5, and RBBP5 (2, 4). The methyl groups in H3K4 are removed by histone demethylases LSD1 (also called KDM1, AOF2, or BHC110), LSD2, and JARID1A-1D (1, 3, 5, 6).

Lysine-specific demethylase 1 (LSD1) belongs to the flavin adenine dinucleotide (FAD)-dependent amine oxidase family and specifically catalyzes the demethylation of di- and monomethylated H3K4 through amine oxidation (1, 5, 7). In contrast to the JARID1 family of Jumonji C (JmjC) domain-containing demethylases that remove the methyl group from tri-, di-, and monomethylated H3K4 (1), demethylation by LSD1 requires a protonated nitrogen in the methylated histone, precluding it from removing the methyl group from trimethylated H3K4 (1, 5, 8). LSD1 is highly conserved in high eukaryotes, and null mutation of LSD1 genes in the mouse causes embryonic lethality (9). However, its pathophysiologic function remains unclear (9, 10). Because the catalytic domain of LSD1 shares significant similarity with other members of the amine oxidase family, most investigation on LSD1 function involves the use of nonselective amine oxidase inhibitors, originally developed against 2 major isoforms of monoamine oxidases, MAO-A and MAO-B, and act through the irreversible modification of the covalently bound FAD at high concentrations (millimoles; refs. 5, 8, 11–15). However, these monoamine oxidase inhibitors induce substantial toxicity in vivo, causing difficulty and confusion for interpretation of LSD1 in vivo function.

Cancer stem cells are considered as the origin of various heterogeneous cancer populations due to their pluripotent or multipotent stem cell property (16–19). Development of therapeutic drugs that target cancer stem cells is an unmet medical demand, as these cells appear to be more resistant to conventional chemo- or radiotherapy. They also often act as the source for metastasis or recurring drug-resistant cancers after treatment (16, 19, 20). Teratoma and teratocarcinoma are pluripotent germ cell tumors caused by abnormal development of embryonic stem (ES) cells (21–23). Other pluripotent tumors include embryonic carcinomas, seminomas, choriocarcinomas, and tumors of yolk sac. These tumor cells often display pluripotent stem cell properties, express stem cell markers Oct4 and Sox2, and are capable of differentiating into various tissue types (21–24). It is well established that ES cells have distinct patterns of histone methylations and other epigenetic modifications for their maintenance and self-renewal (25). Reprogramming of somatic cells into the induced pluripotent stem (iPS) cells by expression of Oct4 and Sox2 is associated with dramatic rearrangement of histone methylation (25, 26). However, it remains unclear the exact role of various histone methylases and demethylases in defining the pluripotency of ES or cancer stem cells.

The docking template structure of LSD1 was derived from the crystal structure of LSD1 bound to the substrate-like peptide (PDB code: 2v1d; ref. 27). Docking was done using the latest version AutoDock 4.0 (28). The illustrated structures were made by PyMOL (29).

F9, NCCIT, NTERA-2, HeLa, 293, NIH3T3, and mouse ES cells were obtained from American Type Culture Collection within 6 months of the submission. All cells were maintained in Dulbecco's Modified Eagle's Medium with 10% FBS (30). F9 cells were grown on Petri dishes coated with 0.1% gelatin, whereas mouse ES cells were cultured on irradiated mouse embryonic fibroblasts (30, 31). All cells were authenticated in Nevada Cancer Institute by specific markers such as Oct4, Sox2, p53, and p16^Ink4a and by their cell morphology and used within 3 months.

Human LSD1 and mouse LSD2 were expressed in bacteria and purified as glutathione S-transferase (GST) fusion proteins (5, 30). Human LSD1 and JARID1A proteins were also obtained from BPS Biosciences. The recombinant proteins were used for demethylation assays using either the dimethylated H3K4 peptide or isolated histones from the nuclei of HeLa cells as reported (refs. 5, 32, 33; see Supplementary Materials and Methods). The conversion efficiency of the dimethylated (Me2) H3K4 peptide to mono- (Me1) and nonmethylated (Me0) H3K4 is calculated using the integrated product peak areas in MALDI-TOF (matrix-assisted laser desorption/ionization–time-of-flight) mass spectrometry (Applied Biosystems 4800 Plus) and the following formula:

Area (Me0) × 2 + area (Me1)/[area (Me0) + area (Me1) + area (Me2)] × 2

The in vitro IC₅₀ of each CBB compound for LSD1 inhibition was calculated using the conversion efficiency and the GraphPad Prism5 software (5, 34).

CBB1002, CBB1003, and CBB1007 were incubated with F9 cells for 2 hours. The extensively washed cells were lysed immediately and the lysates were mixed with 4 volume of acetone at −20°C for 6 hours. The precipitated proteins/macromolecules were removed by centrifugation, and the supernatant was lyophilized and dried pellets were dissolved in methanol. The presence of CBB compounds was analyzed by mass spectrometry using the pure compounds and the lysates from untreated cells as m/z controls.

To understand the function of LSD1, we developed novel chemical compounds that specifically inhibit LSD1 demethylase. As a template for new LSD1 inhibitors, we used the protein crystal structure of LSD1 and a substrate-like peptide inhibitor, which was derived from the N-terminal 21-amino acid residues of histone H3 peptide in which lysine 4 (K4) is replaced by methionine (Fig. 1A, H3K4M) that binds to LSD1 with high binding affinity (K_i = 0.05 μmol/L; refs. 14, 27, 35, 36). The positively charged residues (Arg2 and Arg8) of the peptide establish favorable electrostatic interactions with a cluster of negatively charged residues on LSD1 surface that involve Asp375, Glu379, Asp553, Asp555, Asp556, Asp557, and Glu559 (Fig. 1A). The funneled channel that accesses to FAD is blocked by the peptide inhibitor. On the basis of the structural features of LSD1 active site, especially the highly acidic properties of the surface around the active site, we designed a nonpeptide chemical scaffold that binds to LSD1 with a similar noncovalent binding mode to that of the peptide inhibitor (Fig. 1B). The guanidinium groups of the small molecules, which mimic the arginines, form strong hydrogen bonds with the negatively charged residues of LSD1. A small compound library composed of total 9 small molecules (CBB1001–CBB1009) based on the original design (Fig. 1B–D) was synthesized (Fig. 1D).

The synthesis of the individual LSD1 inhibitor used conventional solution chemistry as shown in scheme 1 using CBB1001 as an example (Fig. 1C). Monomethyl isophthalate was converted into the corresponding bromide 3 (51% yield) by a 2-step sequence including selective reduction of the acid group in 2 followed by treatment with N-bromosuccinimide and triphenylphosphane. A condensation of bromide 3 with 1-[N,N′-bis(tert-butoxycarbonyl)amidino]piperazine 4 afforded the key intermediate 5 which was hydrolyzed to produce intermediate 6. Intermediate 11 was initiated by treatment of methyl 4-cyano-benzoate with lithium bis(trimethylsilyl)amide, followed by protection with di-tert-butyldicarbonate to produce ester 8, which was converted to acid and condensed with 2-(trimethylsilyl)ethyl piperazine-1-carboxylate to produce fragment 11 (83% yield). The 2-(trimethylsilyl)ethyl carbamate protecting group in 11 was removed by the action of tetra-n-butylammonium fluoride, and the resulting free amine 12 was condensed with acid 6 to provide the corresponding coupling product. This was then treated with trifluoroacetic acid to effect a global deprotection to furnish CBB1001 (77% yield). The preparation of additional 8 LSD1 inhibitory compounds (CBB1002–CBB1009) was conducted according to synthetic procedures identical to that of CBB1001 except modifications at the R position (Fig. 1B–D).

To test the potency of the compounds, the demethylation of a synthetic histone H3 amino-terminal substrate peptide that contains the dimethylated K4 by the recombinant human LSD1 protein was established (Fig. 2A). The LSD1 protein displayed an intrinsic demethylase activity on this substrate in vitro, yielding the intermediate monomethylated and final unmethylated H3K4 peptides (5), which can be separated, resolved, and semiquantified by mass spectrometry (Fig. 2B). Analysis of CBB compounds at various dose–response concentrations revealed that most of CBB compounds could effectively inhibit LSD1 activity, with CBB1002, CBB1003, and CBB1007 displayed the best half maximal inhibitory concentration (IC₅₀) at 11.16, 10.54, 5.27 μmol/L, respectively (Fig. 2B, C, and H).

We also used methylated histones as LSD1 substrates to analyze the effects of CBB compounds (5). As reported (5), LSD1 efficiently demethylated both mono- and dimethylated H3K4 in vitro but did not have detectable effects on trimethylated H3K4 and dimethylated H3K9 (Fig. 2D and E). At 10 μmol/L, all CBB compounds except CBB1001 and CBB1009 significantly blocked the demethylase activity of LSD1 on mono- and dimethylated H3K4 but not trimethylated H3K4 or di-methylated H3K9 (Fig. 2D and E). Analysis of inhibitory concentrations of representative CBB1002, CBB1003, and CBB1007 suggests that their IC₅₀ values are similar to the assays by mass spectrometry, with IC₅₀ value of CBB1002 and CBB1003 about 5 to 10 μmol/L, and CBB1007 about 1 to 5 μmol/L (Fig. 2E).

We also examined the effects of CBB compounds on other histone demethylases, such as LSD2 and JARID1A. LSD2 is a paralogue of LSD1 which shares 38% of identical amino acid residues and 56% of amino acid similarity with LSD1 protein in the conserved catalytic carboxy domain (6, 32). It can demethylate both mono- and dimethylated H3K4. JARID1A is a Jumonji domain-containing histone demethylase that specifically removes tri-, di-, and monomethylated H3K4 (1, 33). Mass spectrometry–based demethylation assays revealed that CBB compounds did not inhibit LSD2 and JARID1A demethylase activities (Fig. 2F and G), suggesting that CBB compounds specifically inhibit LSD1 but not other demethylases.

LSD1 is a demethylase for mono- and dimethylated H3K4 and loss of LSD1 in vivo causes the accumulation of these methylated forms of H3K4 (5, 11, 15). To determine the in vivo effects of new LSD1 inhibitors, we treated mouse F9 embryonic teratocarcinoma cells with CBB compounds and then monitored the accumulation of methylated H3K4 (Fig. 3A). Notably, while CBB1001, CBB1002, CBB1005, and CBB1009 did not have significant effects at 10 μmol/L, CBB1003, CBB1004, and CBB1006 to CBB1008 caused a reproducible increase of mono- and dimethylated H3K4 after treatment (Fig. 3A) but had very little effects on trimethylated H3K4 and dimethylated H3K9, suggesting that they are specific for LSD1 but not other histone demethylases. Analysis of the effects of 2 representative compounds CBB1003 and CBB1007 at various concentrations in F9 cells indicated that the IC₅₀ values of CBB1003 and CBB1007 are about 5 to 10 and 1 to 5 μmol/L, respectively (Fig. 3B).

Loss of LSD1 can cause the activation of epigenetically suppressed genes, such as CHRM4/M4-ArchR and SCN3A, in cultured cells (5), as a consequence of increased levels of H3K4 methylation. To investigate whether CBB compounds can inhibit LSD1 and consequently induce epigenetically suppressed gene expression, we monitored the activation of CHRM4 and SCN3A genes by quantitative reverse transcriptase PCR (RT-PCR). While CBB1001, CBB1002, CBB1005, and CBB1009 could not activate the expression of these genes at 10 μmol/L (Fig. 3C), treatment of F9 cells with CBB1003, CBB1004, and CBB1006 to CBB1008 led to the activation of CHRM4 and SCN3A expression (Fig. 3C). Notably, we can detect the activation of CHRM4 and SCN3A expression as low as 0.5 to 1 μmol/L of CBB1007 and 5 to 10 μmol/L of CBB1003 (Fig. 3D). Quantitative real-time RT-PCR analysis at various dose–response concentrations of CBB1003 and CBB1007 revealed that the IC₅₀ vale of CBB1003 or CBB1007 is 8.45 or 3.74 μmol/L, respectively (Fig. 4A and B). We also found that LSD1 inhibitors can markedly induce the expression of genes for differentiation (Fig. 4A), such as FOXA2 (37), suggesting that inhibition of LSD1 may induce pluripotent F9 cells to differentiate.

LSD1 is highly conserved in high eukaryotes, but it can only demethylate di- and monomethylated H3K4 (5), whereas the demethylases of JARID1 family (1A-1D) that contain the Jumonji C (JmjC) domain can remove tri-, di-, and monomethylated H3K4 (3). In previous reports, loss of LSD1 by siRNA-mediated ablation in many cultured cells usually did not cause any obvious inhibition of cell growth (5, 15), presumably due to the presence of other H3K4 demethylases (JARID1A-1D) in the same cell that may functionally substitute for the activity of LSD1 to demethylate H3K4.

Strikingly, we have noticed that treatment of CBB1003 and CBB1007 led to significant growth inhibition of mouse embryonic teratocarcinoma F9 cells (Fig. 5A–E), as analyzed by cell doublings, MTT proliferation assays, and bromodeoxyuridine (BrdUrd) incorporation. The growth-inhibitory effects of CBB1003 and CBB1007 correlate with their ability to induce mono- and dimethylations of H3K4 and the activation of epigenetically suppressed gene expression (Figs. 3–5), suggesting that the growth inhibition is a consequence of specific LSD1 inhibition in F9 cells. Because CBB1002 can effectively inhibit LSD1 activity in vitro with similar IC₅₀ to that of CBB1003 and CBB1007, but it could not inhibit F9 cell growth and alter H3K4 methylation or gene expression in vivo, it is likely that only CBB1003 and CBB1007 can enter the cell whereas CBB1002 cannot. To test this possibility, we examined whether F9 cells can retain CBB1002, CBB1003, and CBB1007 after exposing to the compounds for 2 hours. We found that while CBB1002 is not detectable, both CBB1003 and CBB1007 can be readily found in F9 cells by mass spectrometry (Fig. 4C). These results suggest that modification at the R position in the CBB chemical scaffold (Fig. 1C and D) affects the cellular permeability of these compounds.

The inhibitory effects of CBB compounds on F9 cells also prompted us to examine whether LSD1 compounds can inhibit other cancer or normal cells. A screen of various cultured cell lines revealed that, consistent with previous reports (5, 15), many cultured cancer or established somatic cell lines, such as HeLa, 293, or NIH3T3 (Fig. 5G and H), did not display any significant growth inhibition after the treatment of CBB1003 or CBB1007, even at relatively high concentrations (100–200 μmol/L).

The mouse F9 cell is a pluripotent ES cell–like teratocarcinoma cell because it expresses pluripotent stem cell markers Oct4, Sox2, and Lin28. It also retains stem cell-like properties such as rapid spherical growth and the ability to differentiate (21, 24). Inoculation of F9 cells into the immunodeficient mouse can induce teratocarcinomas which can differentiate into a wide range of tissue types (24). However, HeLa, 293, and NIH3T3 cells do not express stem cell markers Oct4 and Sox2 and are incapable of differentiation and hence are considered non-stem cell lineage. To determine whether human teratocarcinoma cells or similar pluripotent germ cell tumors are also sensitive to LSD1 inhibitors, human pluripotent mediastinal mixed germ NCCIT cell, an intermediate cell between seminoma and embryonic carcinoma, and pluripotent human testicular embryonic carcinoma NTERA-2 cell were examined. We found that the growth of both NCCIT and NTERA-2 cells are highly sensitive to LSD1 inhibitors (Fig. 5G and H). Because embryonic carcinoma and teratocarcinoma are derived from pluripotent ES cells, we also tested whether mouse ES cells are sensitive to LSD1 compounds. Treatment of mouse ES cells with CBB1003 and CBB1007 also led to substantial inhibition of the spherical growth of ES cells (Fig. 6D). These studies indicate that embryonic carcinomas, teratocarcinoma, seminomas, possibly other stem cell–derived cells of pluripotent germ cell tumors (21) and ES cells are particularly sensitive to LSD1 inhibitors, suggesting that LSD1 plays an essential function in these pluripotent cancer cells or ES cells.

To confirm the critical role of LSD1 in pluripotent cancer cells and the specificity of LSD1 inhibitors, we analyzed the effects of LSD1 ablation by siRNA in F9 and HeLa cells. After ablating LSD1 expression with a specific siRNA targeted at an mRNA region that is identical between mouse and human LSD1 genes, only F9 cells displayed marked inhibition of cell growth and BrdUrd incorporation but HeLa cells did not (Figs. 6A–C and 5F). We also ablated LSD2 by siRNA to determine whether it has any effects on the growth of these cells. While LSD1 demethylates the methylated H3K4 in certain promoter regions (1), LSD2 has a distinct function to demethylate mono- and dimethylated H3K4 in the highly transcribed coding region enriched with trimethylated histone H3 at lysine 36 (H3K36; ref. 6). We found that ablation of LSD2 by specific siRNAs caused the inhibition of HeLa cell proliferation but not that of F9 cells (Fig. 6A–C). These observations suggest that LSD1 compounds are specifically target at LSD1 but not LSD2.

To determine the mechanism by which pluripotent F9, NCCIT, and NTERA-2 cancer cells are selectively sensitive to LSD1 inhibition or inactivation, LSD1 protein levels were compared between these cells and HeLa and 293 cells. Our analysis revealed that F9, NCCIT, and NTERA-2 cells express high levels of LSD1 protein whereas the expression of LSD1 is much lower in HeLa and 293 cells (Fig. 6E). The elevated levels of LSD1 are associated with the expression of pluripotent stem cell markers Oct4 and Sox2 (Fig. 6E), which are not expressed in HeLa or 293 cells. In contrast, LSD2 is relatively highly expressed in HeLa, 293, and NTERA-2 but is very low in F9 and NCCIT cells (Fig. 6E). Our observation suggests that the elevated levels of LSD1 may render F9, NCCIT, and NTERA-2 cancer cells more dependent on LSD1, which may consequently confer the selective sensitivity of these pluripotent cancer cells toward the loss of LSD1. In addition, we also found that inactivation of LSD1 by treatment of LSD1 inhibitors or siRNAs caused the significant downregulation of Sox2 and Oct4 expression (Fig. 6F), whereas the expression of other proteins, such as CUL1 or actin, was not altered. Thus, by regulating the stem cell–specific expression of Sox2 and Oct4 or probably other essential stem cell genes, LSD1 plays a unique and essential role in the proliferation of many stem cell–like pluripotent cancer cells.

We also wondered whether LSD1 serves as a good protein marker for human teratocarcinomas, embryonic carcinomas, seminomas, and other pluripotent germ cell tumors in situ (21–23). By screening human seminoma tissue microarrays, we found that while normal human testicular tissues surrounding the tumors contain very low levels of LSD1 (Fig. 7A), immunostaining of tumor microarrays revealed that 6 of 6 human testicular seminomas (100%) displayed high levels of LSD1 protein (Fig. 7B). These seminomas are likely to be pluripotent, as they also express stem cell marker Oct4 (Fig. 7B; refs. 21–23). The marked elevation of LSD1 protein levels in human testicular seminomas is consistent with our observation that pluripotent F9, NCCIT, and NTERA-2 cancer cells also contain high levels of LSD1, Oct4, and Sox2 proteins and that these cells are highly sensitive to LSD1 inhibitors.

In this report, we have developed a new class of chemical compounds that specifically inhibit histone demethylase LSD1. Unlike previous monoamine oxidase inhibitors which form a covalent chemical bond between FAD and the compounds and thus require high structural constraints of the compounds (13, 14, 38), the new LSD1 inhibitors specifically interact with LSD1 and inhibit its activity without forming a covalent bond (Fig. 1). The simplicity of compound synthesis and the low spatial requirement of compound interaction with LSD1 allow more flexible, efficient, and rapid chemical synthesis schemes for further diversified modifications to identify better compounds with high affinity, specificity, cell permeability, and stability. The development of our LSD1 inhibitory compounds thus represents a new strategy to inhibit the activity of LSD1 to modulate the epigenetic regulation of gene expression.

The specificity of new LSD1 inhibitors provides us a unique tool to explore the biologic function of LSD1. By using these compounds as a probe, we unexpectedly found that inhibition of LSD1 selectively blocked the growth of pluripotent teratocarcinoma, embryonic carcinoma, seminoma, and ES cells, likely due to their highly elevated levels of LSD1 expression (Figs. 5–7). Many tissue-specific cancer stem cells, such as breast, ovarian, and lung cancer stem cells, also express Oct4 and Sox2 (39–41). Whether these cancer stem cells are sensitive to our LSD1 inhibitors remains to be tested. One additional use of our novel LSD1 inhibitors is to remove teratomas/embryonic carcinomas during stem cell–based therapy. One major problem in stem/iPS cell–based therapy is the formation of embryonic carcinomas, teratomas, or teratocarcinomas by incomplete differentiation of ES/iPS cells in the organs of recipients (42). Removal of these tumors by LSD1 inhibitors may ensure the successful application of stem cell–based therapy. Our work thus suggests that LSD1 and methylation at H3K4 serve as novel targets for growth inhibition of stem cell–like tumors for cancer therapy. While epigenetic regulators constitute some appealing targets for therapeutic inhibition, this study reveals the first target for which cancer stem cell selectivity may be achievable.

No potential conflicts of interest were disclosed.

This work was supported by grants from Hong Kong Polytechnic University (PolyU 5636/08M), Hong Kong Research Grants Council (PolyU 5638/07M), Fong Shu Fook Tong Foundation and Joyce M. Kuok Foundation to T. Ye, and funds from Shenzhen Bureau of Science, Technology and Information (JC200903160367A, ED200609150042A, and ZD200806180051A), National Natural Science Foundation of China (NSFC30971616), and NIH (R01CA989550) to H. Zhang. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

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.