Abstract
Methylation of lysine and arginine residues on histone tails affects chromatin structure and gene transcription1, 2, 3. Tri- and dimethylation of lysine 9 on histone H3 (H3K9me3/me2) is required for the binding of the repressive protein HP1 and is associated with heterochromatin formation and transcriptional repression in a variety of species4, 5, 6. H3K9me3 has long been regarded as a 'permanent' epigenetic mark7, 8. In a search for proteins and complexes interacting with H3K9me3, we identified the protein GASC1 (gene amplified in squamous cell carcinoma 1)9, which belongs to the JMJD2 (jumonji domain containing 2) subfamily of the jumonji family, and is also known as JMJD2C10. Here we show that three members of this subfamily of proteins demethylate H3K9me3/me2 in vitro through a hydroxylation reaction requiring iron and 
-ketoglutarate as cofactors. Furthermore, we demonstrate that ectopic expression of GASC1 or other JMJD2 members markedly decreases H3K9me3/me2 levels, increases H3K9me1 levels, delocalizes HP1 and reduces heterochromatin in vivo. Previously, GASC1 was found to be amplified in several cell lines derived from oesophageal squamous carcinomas9, 11, 12, and in agreement with a contribution of GASC1 to tumour development, inhibition of GASC1 expression decreases cell proliferation. Thus, in addition to identifying GASC1 as a histone trimethyl demethylase, we suggest a model for how this enzyme might be involved in cancer development, and propose it as a target for anti-cancer therapy.
Methylation of lysine 9 on histone H3 is associated with epigenetically silenced chromatin1, 2, 3. As documented by studies of SUV39H1 and SUV39H2 knockout mice13, loss of the H3K9me3 mark results in genomic instability and predisposes to cancer. Hence, enzymes capable of reversing this particular mark have long been sought, although their existence has been questioned. The latter view has been reinforced by the fact that H3K9me3 is required for the establishment and maintenance of heterochromatin, a "very stable and heritable chromatin state"8. In the present study we show that the jumonji protein GASC1 (ref. 9) and its homologues JMJD2A and JMJD2B demethylate H3K9me3/me2.
In an effort to identify proteins that interact with H3K9me3, we performed affinity purification of HeLa nuclear proteins using H3 peptides either non-methylated or trimethylated at lysine 9. Several proteins were specifically enriched on H3K9me3 as compared to non-methylated H3K9 (Supplementary Fig. S1). Among the proteins identified was the jumonji protein GASC1 (Supplementary Fig. S2). Owing to the presence of a jumonji N domain (JmjN) and a jumonji C domain (JmjC), the protein is also denoted as JMJD2C10 (Supplementary Fig. S2). Of particular interest, JmjC domain proteins have been suggested to be histone demethylases14, and just before the completion of this study the JmjC domain protein FBXL11 was shown to demethylate H3K36me2/me1 (ref. 15).
Because we found GASC1 to be associated with H3K9me3, we were intrigued by the idea that GASC1 might be involved in demethylating this epigenetic mark. To test this possibility we purified full-length, His-tagged human GASC1 from baculovirus-infected insect cells (Supplementary Fig. S3). Recombinant GASC1 was further purified by size-exclusion chromatography, and the highly purified GASC1 was tested for demethylase activity, by incubation with histones. As shown in Fig. 1a, GASC1 very efficiently reduced levels of H3K9me3. Similarly, GASC1 demethylated H3K9me3 on nucleosomes, the physiologically relevant template (Supplementary Fig. S4a).
Figure 1: GASC1 demethylates H3K9me3/me2.
a, Fractions from size-exclusion chromatography of recombinant GASC1 were incubated with histones for 30 min at 37 °C and demethylation activity was assayed by immunoblotting. b, Demethylation activity in selected fractions assayed by immunoblotting. c, d, H3K9me3 peptides were incubated with or without GASC1 and the material was analysed by mass spectrometry (c) and immunoblotting (d). e, Demethylation assay using varying amounts of wild-type and mutant GASC1 on bulk histones. f, Predicted structure of GASC1 in complex with H3K9me3 tail (1–11) and its cofactors; the lysine H3K9 methyl groups are marked in yellow. Left panel shows an overview; right panel shows a magnification of the iron-binding residues. -KG, -ketoglutarate.
To analyse further the specificity of the demethylase activity of GASC1, we incubated fractions from size-exclusion chromatography of GASC1 with histones and evaluated the methylation status of various epigenetic marks by immunoblotting (Fig. 1b and Supplementary Fig. S4b). Here, H3K9me3 and H3K9me2 from histones were completely removed in the fractions with the highest concentrations of GASC1 (Fig. 1b, fractions 9–12). A concomitant increase in the level of H3K9me1 was also observed, consistent with GASC1 converting H3K9me3 to H3K9me2 and then to H3K9me1. In contrast, the levels of other tested epigenetic marks including H3K4me3, H3K4me2, H3K27me3 and H4K20me3 were unaffected by GASC1 treatment (Fig. 1b and Supplementary Fig. S4b).
Notably, H3K9me2 appeared to be increased when histones were incubated with low to moderate amounts of GASC1 (Fig. 1b, fractions 6, 8, 14 and 16). The reason for the apparent increase in the H3K9me2 mark could be due to a steady-state level of this mark being reached in the initial phase of the demethylation reaction. Thus, it can be proposed that although some H3K9me2 is lost through GASC1 demethylation, additional H3K9me2 will be added to the global pool by GASC1 demethylation of H3K9me3, at least as long as significant levels of H3K9me3 are present. To confirm this assumption, we performed additional studies in which synthetic H3 peptide substrates methylated at K9 or K27 were incubated with varying amounts of GASC1. These experiments showed that even low concentrations of GASC1 could effectively remove both H3K9me3 and H3K9me2, whereas other tested methylated lysines (including H3K9me1) were unaffected (Supplementary Fig. S5).
Consistent with the results in Fig. 1a, b, the demethylase activity of GASC1 varies in a concentration-dependent manner and can be abolished by heat denaturation of GASC1 (Supplementary Fig. S4b), suggesting that it is an enzyme-dependent reaction. We also performed an additional in vitro demethylation study, where pure synthetic H3K9me3 peptide was incubated with GASC1. After 30 min incubation almost all H3K9me3 was converted to H3K9me2 and H3K9me1 (Supplementary Fig. S4c).
To demonstrate formally that GASC1 demethylates H3K9me3 and H3K9me2, mass spectrometry was performed on a synthetic H3K9me3 peptide incubated alone or in the presence of GASC1. This analysis revealed that most H3K9me3 peptide (approximately 80%) was converted to H3K9me2 and further to H3K9me1 after incubation with GASC1 (Fig. 1c, d).
Next, we considered whether the close GASC1 homologues JMJD2A and JMJD2B (Supplementary Fig. S2) could also demethylate H3K9me3/me2 in vitro. We cloned full-length human JMJD2A and JMJD2B and produced recombinant proteins from baculovirus-infected cells (Supplementary Fig.S3). Indeed, both homologues can demethylate H3K9me3/me2 (Supplementary Fig. S6).
To obtain an insight into the mechanism of the demethylation reaction, we modelled human GASC1 onto the structure of factor inhibiting HIF1 (FIH), the only jumonji protein for which the three-dimensional structure has been resolved16, 17. In agreement with the alignment (Supplementary Figs S7 and S8), the modelling indicated that residues H190, E192 and H288 form an essential part of the iron-binding groove of GASC1 (Fig. 1f). On the basis of this in silico model we predicted that mutating H190 and E192 would be sufficient to abrogate the iron-binding ability of GASC1 and thus also to inhibit its demethylation activity.
Although wild-type GASC1 has a robust demethylase activity, mutant GASC1 has no detectable demethylase activity (Fig. 1e), thereby indicating that the putative iron-binding residues are crucial to the demethylase activity of GASC1.
Dioxygenases belonging to the cupin superfamily are dependent upon Fe(ii) and -ketoglutarate16, 18. In addition, some dioxygenases have the additional requirement of ascorbate for full catalytic activity15, 19, 20.
To test whether GASC1-mediated demethylation would fit the reaction mechanism described above and depicted in Fig. 2a, we first tested the importance of the putative cofactors for the demethylation reaction.
Figure 2: Reaction mechanism for GASC1 demethylation.
a, 1: Fe(ii) is bound by residues His 190, Glu 192 and His 288. 2: -Ketoglutarate is bound. 3: Iron binds oxygen. 4: Oxidative decarboxylation of -ketoglutarate produces carbon dioxide, succinate and ferryl (Fe(iv) = O). 5: Ferryl hydroxylation of methylated K9 releases formaldehyde. 6: The histone tail and succinate leave the GASC1 molecule. b, Demethylation assay using a synthetic H3K9me3 peptide as substrate, in the presence or absence of cofactors. AA, ascorbic acid. KG, ketoglurate. c, Demethylation assay using histones as substrates in the presence of varying amounts of N-oxalylglycine. d, Formaldehyde produced by incubating a fixed amount of GASC1 (40 
g) with varying amounts of H3K9me3 peptide or the non-methylated peptide, measured using a demethylation–FDH-coupled assay28. The y axis shows absorbance values at 340 nm.
In the presence of all its cofactors GASC1 demethylates H3K9me3 in vitro, whereas GASC1 incubated with EDTA (chelating iron) or in the absence of its cofactors was significantly less efficient (Fig. 2b). The ability of the demethylation reaction to occur in vitro without the addition of cofactors is probably due to the co-purification of cofactors (for example, iron and -ketoglutarate) with recombinant GASC1. For this reason we sought to establish further the importance of these cofactors by testing the ability of N-oxalylglycine, CoCl2 and NiSO4 to inhibit GASC1-mediated demethylation of H3K9me3. These compounds are all able to inhibit the demethylation reaction effectively (Fig. 2c and Supplementary Fig. S9). N-oxalylglycine is an -ketoglutarate analogue and presumably inhibits the activity of GASC1 by displacing -ketoglutarate from the iron-binding residues of GASC1. Similarly, inhibition of GASC1 by CoCl2 and NiSO4 probably involves dislocation of Fe(ii) from the iron-binding site of GASC1 by cobalt(ii) or nickel(ii) ions. The requirement of ascorbic acid, -ketoglutarate and iron in the demethylation reaction and the inhibition of the reaction by nickel and cobalt salts and N-oxalylglycine strongly indicate that GASC1 is a dioxygenase.
Moreover, in agreement with the proposed reaction scheme, incubation of GASC1 with its substrate released formaldehyde (Fig. 2d). Taken together, these results demonstrate that GASC1 directly demethylates H3K9me3 and H3K9me2 in vitro through a hydroxylation reaction requiring iron and -ketoglutarate, producing formaldehyde and H3K9me1.
Having established that GASC1 can demethylate H3K9me3 and H3K9me2 in vitro, we next considered whether GASC1 could modulate heterochromatin formation/maintenance in vivo. Ectopic expression of GASC1 in human and murine fibroblasts caused an efficient decrease of H3K9me3 (Fig. S10a,b) and an increase of H3K9me1 (Supplementary Fig. S10a).
The decrease in H3K9me3/me2 levels and increase of H3K9me1 was also apparent in U2OS cells transfected with wild-type GASC1, but not with the putative Fe(ii)-binding mutant (Fig. 3a,b), further confirming the crucial importance of these predicted Fe(ii)-coordinating residues. In contrast, levels of H3K4me3 and H3K27me3 were not affected (Supplementary Fig. S11).
Figure 3: Ectopic expression of GASC1 leads to loss of H3K9me3/me2 in vivo.
a, Demethylation of H3K9me3/me2 by wild-type but not mutant GASC1 in vivo as measured by immunofluorescence. White arrowheads indicate transfected cells. b, Effect of ectopic expression of haemagglutinin (HA)-tagged GASC1 and mutant on H3K9me3 in vivo on histones. c, d, Ectopic expression of JMJD2A and JMJD2B leads to demethylation of H3K9me3 and H3K9me2 and an increase in H3K9me1 levels in U2OS cells. e, Confocal microscopy of mouse embryo fibroblasts transfected with HA-tagged GASC1. HP1-
is delocalized in cells overexpressing GASC1. f, Loss of chromatin-associated HP1- and HP1-
in HEK293 cells expressing tetracycline (Tet)-inducible Myc-tagged GASC1. Total indicates total lysate, whereas chromatin indicates the chromatin fraction.
As observed for GASC1, ectopic expression of JMJD2A and JMJD2B led to a significant reduction of H3K9me3 and H3K9me2 levels, with a concomitant increase in H3K9me1 (Fig. 3c, d). These results suggest that GASC1 belongs to a family of histone H3K9 demethylases.
Heterochromatin formation and maintenance requires the presence of H3K9me3/me2 and HP1 binding5, 6, 8. Because the ectopic expression of GASC1 leads to a global reduction of H3K9me3/me2 levels, we tested whether increased amounts of GASC1 affected HP1 binding and localization. As predicted, HP1- was delocalized in NIH 3T3 cells overexpressing GASC1 (Fig. 3e). Similarly, ectopic expression of the GASC1 homologues JMJD2A and JMJD2B led to delocalization of HP1- (Supplementary Fig. S12). We further validated these results by demonstrating decreased levels of HP1- and HP1- associated with chromatin in GASC1-overexpressing cells (Fig. 3f). Taken together, these results suggest that GASC1 might have a physiologically important role in controlling heterochromatin formation and maintenance.
The strict control of heterochromatin formation and maintenance is critical for both proper biological function and genomic integrity. For instance, centromeric heterochromatin formation is essential for the correct segregation of chromosomes during mitosis21. Similarly, it has been demonstrated that the deletion of Suv39h1 and Suv39h2 genes in mice, the products of which are responsible for di- and trimethylation of H3K9 in heterochromatin, leads to chromosomal instability and collaborates with oncogenes in inducing mouse lymphomas13. Therefore, aberrant activity of GASC1 or its homologues could potentially lead to genomic instability and consequently cancer. Indeed, GASC1 was originally identified as a gene frequently amplified in oesophageal squamous cell carcinoma9, 12, 22 and GASC1 is overexpressed in various cancer types containing chromosomal abberations11.
To begin to elucidate the physiological function of GASC1 in normal and cancer cells, we tested the relative levels of GASC1 mRNA and the H3K9me3 mark in two oesophageal carcinoma cell lines that have been reported to contain several copies of the GASC1 gene (KYSE150 and KYSE450)9. The human cell lines U2OS, 293 and TIG3 were included for comparison. As shown in Fig. 4a, GASC1 levels are indeed increased 3–5-fold in the oesophageal carcinoma cell lines as compared to other carcinoma cell lines and normal human fibroblasts. However, the increased expression of GASC1 did not lead to a global decrease in H3K9me3 levels (Fig. 4b). These results demonstrate that the increased levels of GASC1 are not sufficient to lead to global demethylation of H3K9me3, and may imply that GASC1 expressed at physiological levels regulates the H3K9me3 levels locally.
Figure 4: Inhibition of GASC1 leads to decreased proliferation.
a, GASC1 expression in various cell lines determined by real-time quantitative PCR (RT-qPCR). b, H3K9me3 status in various cell lines. c, KYSE150 and U2OS cells were transfected with shRNA targeting GASC1 or control shRNA. The efficiency of the shRNA interference was evaluated by RT-qPCR. d, e, H3K9me3 status and growth curves of KYSE150 and U2OS cells after shRNA expression. f, g, Data from ref. 29 (f) and ref. 30 (g) re-analysed to show expression levels of JMJD2 proteins in prostate tumours (PC), normal prostate tissues (NP) and unmatched lymph node metastases (LNM). Bars indicate medians; P-values of Mann–Whitney U-test are provided; see Supplementary Information for further details. Error bars on qPCR experiments and growth curves indicate standard deviations and standard error of the mean, respectively.
High resolution image and legend (107K)To test whether GASC1 can contribute to tumour cell maintenance, we inhibited the expression of GASC1 in KYSE150 and (as a control) U2OS cells by short hairpin RNA (shRNA). The efficiency of inhibition of GASC1 expression was similar in the two cell lines (Fig. 4c), and consistent with the results reported above, we did not observe changes in the global levels of H3K9me3 (Fig. 4d). Notably, however, inhibition of GASC1 expression caused a significant reduction of proliferation in both KYSE150 and U2OS cells (Fig. 4e). Although the inhibition of cell proliferation was more pronounced in KYSE150 cells, this finding indicates that GASC1 is required for cell proliferation, also of cells in which GASC1 levels are normal. Therefore, future studies are required to understand whether GASC1 contributes to tumour development.
Meanwhile, to obtain further support for a potential involvement of GASC1 in cancer, we searched the Oncomine database23 for differential GASC1 expression in normal versus tumour tissue. The expression of GASC1 and its homologues JMJD2A and JMJD2B were significantly increased in prostate cancers relative to normal tissue (Fig. 4f, g). These results suggest that increased expression of all three JMJD2 proteins might contribute to the development of human cancer possibly by demethylation of H3K9me3/me2.
Di- and trimethylated H3K9 is required for HP1 binding, essential for heterochromatin formation and is associated with transcriptional repression. Moreover, these epigenetic marks are increased on specific genes in senescence, a key mechanism guarding cells against cancer24, 25, 26, 27. It can therefore be speculated that amplification of GASC1 may reduce the ability of cells to become senescent, or lead to de-repression of otherwise silenced oncogenes. Accordingly, inhibition of GASC1-demethylating activity could potentially constitute a new anti-neoplastic therapeutic modality.
Using several independent lines of evidence we have shown that GASC1 and its homologues JMJD2A and JMJD2B demethylate the repressive histone H3K9me3/me2 marks both in vitro and in vivo. Our findings demonstrate that histone trimethylation is a reversible modification. This finding may potentially have far-reaching implications for human disease, notably cancer. Further studies will be required to unravel the biology of GASC1 and the jumonji protein family and shed light on their possible role in epigenetic signalling, transcriptional regulation and human disease.
Methods
Demethylation assay
Bulk histones or synthetic histone peptides were incubated with purified His-tagged GASC1 in demethylation buffer (50 mM Tris pH 8.0, 50 mM KCl, 10 mM MgCl2, 1 mM -ketoglutarate, 40 M FeSO4, 2 mM ascorbic acid) at 37 °C. When using nucleosomes as substrate MgCl2 was omitted in the reaction. In a typical reaction, either 6 g of bulk histones or 2 g of modified histone peptides were incubated with 20 g GASC1 in a volume of 100 l for 30 min. Reaction mixtures were analysed by either western blotting using specific antibodies, or by formaldehyde release assays.
Formaldehyde release assay
Formaldehyde release assays were performed essentially as described28. All reactions were performed in a total volume of 200 l per reaction in a quartz cuvette. In short, recombinant GASC1 (typically 40 g) was incubated in 150 l demethylation buffer (see above) in the presence of 2 mM NAD+ and 0.2 U formaldehyde dehydrogenase (FDH) at 37 °C for 5 min. Then the reaction was started by adding the substrates (histone peptides). The absorbance at 340 nm was measured with 0.5-min intervals (15 min total) using a Genesys 10UV Thermospectronic spectrophotometer at 37 °C.
GASC1 demethylation of H3K9me3 peptide
Six micrograms of recombinant GASC1 was incubated with 3 g H3K9me3 peptide in FDH buffer in a final volume of 90 l for 30 min at 37 °C. Urea was added to a final concentration of 4 M and the mixture was incubated at 20 °C for 15 min. An equal volume of 1% trifluoroacetic acid (TFA) was added and the sample loaded on a reversed-phase mini C8 column packed in a 100-l tip (column volume of 20 l). After washing in 1% TFA, the bound peptide was eluted in 20 l (30% methanol, 25% formic acid). One-third of the eluted material was analysed by SDS–polyacrylamide gel electrophoresis (PAGE) and western blotting using first anti-H3K9me3, followed by anti-biotin antibody. The rest of the material was analysed by mass spectrometry.
Mass spectrometry analysis
One-third of the eluate was injected in 1% TFA using an Agilent 1100 Nano HPLC onto a C18 column (Reprosil-Pur C18-AQ 3 m; Dr. Maisch GmbH) packed into a spray emitter (75 m internal diameter, 8 m opening, 70 mm length; New Objectives). Peptides were eluted in a gradient from buffer A (5% acetonitrile and 0.5% acetic acid) to buffer B (acetonitrile and 0.5% acetic acid) going from 0% to 20% in 10 min at 300 nl min-1. Spectra were recorded on a LTQ-FT mass spectrometer (Thermoelectron).
Additional materials and methods, including methods for cloning procedures and immunostaining, are described in the Supplementary Methods.
