KDM8, a H3K36me2 histone demethylase that acts in the cyclin A1 coding region to regulate cancer cell proliferation

Abstract

Localized chromatin modifications of histone tails play an important role in regulating gene transcription, and aberration of these processes leads to carcinogenesis. Methylated histone lysine residues, a key player in chromatin remodeling, are demethylated by the JmjC class of enzymes. Here we show that JMJD5 (now renamed KDM8), a JmjC family member, demethylates H3K36me2 and is required for cell cycle progression. Chromatin immunoprecipitation assays applied to human genome tiling arrays in conjunction with RNA microarray revealed that KDM8 occupies the coding region of cyclin A1 and directly regulates transcription. Mechanistic analyses showed that KDM8 functioned as a transcriptional activator by inhibiting HDAC recruitment via demethylation of H3K36me2, an epigenetic repressive mark. Tumor array experiments revealed KDM8 is overexpressed in several types of cancer. In addition, loss-of-function studies in MCF7 cells leads to cell cycle arrest. These studies identified KDM8 as an important cell cycle regulator.

Regulation of gene expression through posttranslational modification of the core histones has increasingly shown to be of great importance, particularly in a cancer setting. Among the multiple types of histone modifications, histone methylation, once considered irreversible, has quickly emerged to become a key epigenetic mark in regulating many critical cellular functions. The recent discovery of histone demethylases has shed light on the reversibility of this chromatin mark and its effects on gene expression. Studies exploring the JmjC (Jumonji C domain)–containing proteins, a new class of histone demethylases (1–4), primarily identified their enzymatic activity at the promoters of specific target genes (5, 6).

The JmjC domain–containing gene family encodes a wide range of the eukaryotic genome and is conserved in species spanning from yeast to humans. Currently, most family members classified as histone demethylases contain known histone-binding domains such as PHD and Tudor domains (7). JMJD5 (renamed KDM8) is a member of this extensive protein family that lacks recognizable histone-binding domains and remains largely unexplored. Although one study speculated that KDM8 acts as a potential tumor suppressor gene based on retrovirus insertional mutagenesis (8), no biological and molecular characterizations were described in the report.

We extensively examine and provide evidence that KDM8 possesses H3K36me2 demethylase activity and has the ability to regulate cyclin A1 transcription in MCF7 breast cancer cells. We found that KDM8 is recruited to cyclin A1 coding region bound H3K36me2 and demethylates this mark, resulting in increased transcriptional activity. This finding is a departure from previous studies that showed that the majority of histone demethylases exert their epigenetic effects at the promoters of genes. Additionally, we describe overexpression of KDM8 in breast cancer tumors as well as its requirement for MCF7 cell cycle progression.

Results

JMJD5/KDM8 Is a H3K36me2 Demethylase.

Our initial experiments explored the potential histone demethylase activity of KDM8. To this end, WT KDM8 and KDM8-H321A (JmjC domain mutant)–inducible cell lines were generated using MCF7 breast cancer cells. By analogy to other JmjC histone demethylases, the H321A mutation is predicted to knock out KDM8 enzymatic activity through disruption of its ability to bind the ferrous ion and thus its putative demethylase activity. Immunofluorescence as well as immunoblotting assays were performed using methylation-specific antibodies to screen for probable histone residue substrates. Immunofluorescence assay revealed that 80% of KDM8-overexpressing cells exhibited a substantial decrease in H3K36me2 staining. In contrast, a significant increase of H3K36me2 methylation was detected in approximately 60% of KDM8-H321A–expressing cells (Figs. 1 A and B). The result also indicates that KDM8-H321A may act as a dominant negative for H3K36me2 demethylation. Significantly, the overall level of neither H3K36me1 nor H3K36me3 methylation was altered, indicating the specificity of the enzyme. Immunoblot analysis of acid-extracted histones from the inducible cell lines further verified that overexpression of KDM8, but not the enzymatically inactive H321A mutant, demethylated H3K36me2 in vivo (Fig. 1C). The in vivo specificity of KDM8 toward H3K36me2 was further demonstrated by blotting with other commercially available antibodies against methylated lysine or arginines of histones H3 and H4, including anti-H3K36me1 and anti-H3K36me3; none revealed a decreased intensity in the KDM8-overexpressing cell line (Fig. 1C and Fig. S1A). Additionally, MS analysis using a truncated recombinant KDM8 protein consisting of the JmjC domain (101-C) or its equivalent H321A mutant with H3K36me2 peptide was performed. These results revealed that WT 101-C truncated protein demethylates H3K36me2 directly, as indicated by the 14-Da shift from the original peptide mass. In contrast, no mass shift was observed when catalytically inactive KDM8 101-C H321A mutant was used (Fig. 1D).

Fig. 1.

KDM8 H3K36me2 demethylation in MCF7 breast cancer cells. (A) KDM8 and H321A, an inactive enzymatic mutant, were induced to express as Flag fusion proteins. Indirect immunofluorescence with antibodies against Flag (red staining) and methylated H3K36me1, H3K36me2, or H3K36me3 (green staining) was used to analyze in vivo substrate specificity of KDM8. DAPI staining (blue) indicates nuclei location in each field. Cells overexpressing KDM8 (arrows) showed significant loss of H3K36me2 staining, which was dependent on the active JmjC domain, and not observed in the H321A overexpressed cells. (B) Quantitative analysis of Dox-induced KDM8/H321A MCF7 H3K36me2 stained cells from 10 random fields. Eighty percent of induced KDM8 cells showed diminished H3K36me2, whereas 60% of H321A-induced cells exhibited increases in H3K36me2 staining. Percentage of cells that exhibited no change in H3K36me2 is not shown. (C) Histones extracted from Dox-induced KDM8/H321A MCF7 were analyzed by Western blotting with antibodies against H3K36me2 and H3K9me3. A decreased signal in H3K36me2 was observed in KDM8 overexpressed cells (lane 2 vs. lane 1), which was dependent on the active JmjC domain (inactive mutant lanes 3–4). (D) MS analysis of observed in vitro KDM8 demethylase activity. Reactions using H3K36me2 peptide combined with either GST KDM8 101-C WT or GST-KDM8 101-C H321A show 14 Da shift in WT reactions only (asterisk).

JMJD5/KDM8 Expression Is Critical for MCF7 Cancer Cell Proliferation.

The biological functions and clinical significance of KDM8 in the context of breast cancer were next investigated. First, we compared the expression of KDM8 in a panel of breast cancer cell lines to that of primary human mammary epithelial cells (HMECs) by immunoblot analysis with a custom-generated rabbit polyclonal antisera specific to KDM8. Although KDM8 protein expression was very low in HMECs, all of the breast cancer cell lines, including MCF7, exhibited significantly higher expression (Fig. 2A). To extend our results, immunohistochemistry (IHC) was conducted using a human tissue microarray containing 40 individual breast cancer samples and patient-matched adjacent normal tissue controls (Fig. 2 B and C). Consistent with the results obtained in breast cancer cell lines, KDM8 was widely overexpressed in breast tumors. IHC staining was quantified via computerized scanning analysis and manual scoring by an expert panel of pathologists who scored the IHC results on a scale from 0 through 3+, with 3+ indicating the most intense staining. The summarized IHC data indicated that 97.5% of the tumor samples were intensely stained for KDM8 expression (i.e., 3+ score) compared with 67.5% in normal patient-matched tissues (Fig. 2D). Both scoring systems (computerized and manual) concluded that KDM8 is overexpressed in tumor tissues compared to their nonmalignant counterparts. To extend these results and determine if KDM8 overexpression is potentially a general aberration occurring in cancer, a tumor array membrane consisting of spotted tumor and normal cell lysates was probed for KDM8 protein expression. Multiple types of tumors including thyroid, adrenal, bladder, uterine, and liver exhibited overexpressed KDM8 protein in comparison to their respective normal tissue controls (Fig. S2).

Fig. 2.

KDM8 is overexpressed in tumors and required for breast cancer cell proliferation. A quantitative examination of KDM8 expression in cancer cells was performed. (A) Western blot analysis showed significantly higher expression of endogenous KDM8 in a panel of breast cancer cell lines compared to primary HMECs. (B) Immunohistochemistry KDM8 staining of malignant breast cancer tumors (M) and patient-matched adjacent normal tissues (N). (C) Higher-magnification view of respective malignant and normal IHC staining of KDM8. (Scale bar, 10 μm.) (D) Intensity of staining was scored on a scale of lowest (0) to highest (3+). (E) (i) Loss of KDM8 contributes to a decrease in cell growth. Two KDM8 knockdown clones (shKDM8 no. 1, no. 5) or control cells (shControl) were maintained in culture for 5 d. Cell numbers were counted at indicated time points in triplicate. Growth data presented are means ± SD. from three independent experiments. Immunoblotted lysates showed decreased expression of KDM8 in knockdown lines (Upper). (ii) Enzymatic activity is required for MCF7 proliferation. KDM8 WT or catalytic inactive mutant (H321A) expression was induced (Dox) and cell growth was examined by counting cells for 6 d. Overexpression of KDM8 WT increased cell proliferation compared to noninduced control, whereas induction of mutant decreased cell proliferation. (F) Flow cytometry analysis of KDM8 shRNA MCF7 cells stained with PI identified 40% of total cells arrested in G2/M phase compared with 20% in control shRNA cells.

Because KDM8 was found to be overexpressed in cancer cells, we focused on its potential role in cell proliferation. To examine its role in cell growth, stable KDM8 knock-down cell lines were generated by lentiviral-mediated expression of KDM8-targeting shRNAs in MCF7 (sh-KDM8-MCF7-1 and -5). To control for potential off-target effects of the KDM8 shRNAs, two stable cell lines with independent target sequences were generated. KDM8 silencing was validated by immunoblot analysis and the effects on proliferation were monitored by standard cell counting with a hemocytometer. The shControl cells and shKDM8-MCF7s were plated in six-well dishes and the cell proliferation was examined for 5 d. Cell numbers were counted every day from three independent wells. The results demonstrated that stable KDM8 knockdown resulted in a significant inhibition of growth of MCF7 cells. At the end of the 5 d, shRNA-KDM8-MCF7 exhibited a fourfold decrease in growth compared to control shRNA cell lines (Fig. 2E, i). Further analyses indicated that the effects on MCF7 cell growth were dependent on enzymatic activity of KDM8, as induction of KDM8 WT but not catalytic mutant stimulated cell growth (Fig. 2E, ii). Flow cytometric analysis of propidium iodide (PI)–stained cells revealed that cells accumulated and arrested in G2/M as evidenced by an increase in the fraction of cells in G2/M from 20% to 40% in shRNA-Control-MCF7 and shRNA-KDM8-MCF7 cells, respectively (Fig. 2F). Taken together, these results indicated that KDM8 is required for G2/M phase cell cycle progression.

JMJD5/KDM8 Is Enriched in the Coding Region of Cyclin A1.

To define the mechanism by which KDM8 regulates the cell cycle, KDM8 target genes were identified by performing comprehensive gene expression profiling in combination with promoter tiling array analysis for the detection of genomic recruitment sites. MCF7 cells and MCF7 cells engineered for conditional expression of WT KDM8 or KDM8-H321A were used in these experiments. Gene ontology analysis was performed on the gene expression data to identify differentially regulated genes that favor cell cycle progression and were selectively up-regulated by induced expression of WT KDM8. Our initial gene expression screening revealed that several genes encoding cyclins and cyclin-dependent kinases were up-regulated in response to KDM8 overexpression and down-regulated by KDM8-H321A (Fig. S3). In parallel, binding sites for endogenous KDM8 were mapped using ChIP-on-chip analysis with promoter tiling arrays that provide 10 kb of coverage for each promoter region spanning 7.5 kb upstream through 2.45 kb downstream of the transcription start sites (GeneChip Human Promoter 1.0R Arrays; Affymetrix). The results indicated that KDM8 is recruited to the cyclin A1 locus. Importantly, cyclin A1 is required for progression through the G2/M phase of the cell cycle (9, 10). Of particular interest, KDM8 binding sites were found primarily within the coding region of this gene with the strongest ChIP-enrichment identified in exon 2, with surprisingly little binding to the promoter region (Fig. 3A). To confirm the ChIP-on-chip cyclin A1 binding results, quantitative PCR was performed using endogenous KDM8 ChIP DNA and primer sets designed to encompass the proximal promoter regions (primer sets A, B) and exon 2 (primer set C). This ChIP analysis demonstrated that exon 2 sequences were selectively enriched in anti-KDM8 chromatin immunoprecipitates (i.e., with respect to a control IgG ChIP) whereas sequences in the promoter were not. In conclusion, KDM8 was recruited specifically to the coding region of the cyclin A1 gene (Fig. 3B). In agreement with our initial microarray screening, corroborative quantitative RT-PCR data with induced KDM8 MCF7 cells showed 1.8-fold increase in cyclin A1 expression over the noninduced controls (Fig. 3C). To examine whether KDM8 also affects cyclin A1 at the protein level, MCF10A cells were transfected with siRNA targeting KDM8 and immunoblotted for expression of KDM8, cyclin A, and H3K36me2. Consistent with the gene expression data, immunoblot analysis revealed that transient knockdown of KDM8 in cells exhibited lowered levels of cyclin A protein, and increased levels of H3K36me2 (Fig. 3D). Taken together, these data strongly indicate that KDM8 is capable of up-regulating cyclin A1 transcription by demethylating H3K36me2 located in the cyclin A1 coding region.

Fig. 3.

KDM8 increases cyclin A1 gene and protein expression through binding in the coding region. (A) Endogenous KDM8 recruitment sites in MCF7 cells were mapped using ChIP-on-chip analysis. KDM8-associated DNA was enriched by ChIP with a KDM8-specific antibody and subsequently analyzed with Affymetrix Human Promoter 1.0R tiling arrays as described in SI Materials and Methods. CisGenome software was used for peak detection and annotation of genomic regions bound by KDM8. The figure demonstrates that endogenous KDM8 is recruited to the cyclin A1 coding region and most prominently to exon 2. The binding profile along the CCNA1 locus is depicted based on magnitude of enrichment at each location, which is signified by the height of each bar. Location of primer pairs are indicated in the panel by letters and black bars. (B) Quantitative real-time PCR confirming the binding of endogenous KDM8 in the cyclin A1 gene. (C) Quantitative RT-PCR analysis of cyclin A1 in Dox-inducible Flag-KDM8/H321A MCF7 cells. Induction of KDM8 increased cyclin A1 expression. (D) Immunoblotting analysis of MCF10A cells transfected with KDM8 siRNA and scramble siRNA shows that knockdown of KDM8 leads to a decrease in cyclin A expression and increases the amount of H3K36me2.

JMJD5/KDM8-Induced H3K36me2 Demethylation Leads to Increased H3/H4 Acetylation and Cyclin A1 Transcription.

In contrast to its function as a classic activating mark in the 3′ region of genes (11), the H3K36me2 chromatin mark acts as a transcriptional suppressor when located in the coding region (12), where it acts to suppress spurious transcription events. Having observed KDM8 demethylation of H3K36me2 in vivo, we speculated that KDM8 targeted this methyl mark at the coding region to enhance the transcription of cyclin A1. ChIP analysis performed with inducible KDM8 MCF7 cells demonstrated that KDM8 and its H321A inactive mutant form were enriched along exon 2 of cyclin A1 but not the promoter region (Fig. 4, bars 1–4, and Fig. S4). Additional ChIP analysis examining the presence of H3K36me2 at this coding region revealed that it was greatly diminished in the presence of KDM8 WT, yet enhanced with H321A expression (Fig. 4A, bars 5–8). ChIP analysis of HDAC1 enrichment, a transcriptionally repressive chromatin modifier (13) that is recruited by the H3K36me2 mark (14), showed that during KDM8 overexpression, HDAC1 decreased by 50% (Fig. 4A, bars 9–10), whereas H321A overexpression resulted in a threefold increase in HDAC1 recruitment (Fig. 4A, bars 11–12). Further ChIP analysis also revealed that the induction of KDM8 led to enhanced H3 and H4 acetylation (Fig. 4A, bars 13–14, 17–18), further supporting inhibition of HDAC1 recruitment. Finally, endogenous KDM8 knockdown led to increased levels of the H3K36me2 mark along with HDAC1 recruitment (Fig. 4B), which resulted in a significant reduction of cyclin A1 expression in MCF7 cells (Fig. 4C). These data establish a plausible relationship among KDM8-mediated H3K36me2 demethylation, reduced HDAC1 recruitment, and the transcriptional activation of cyclin A1.

Fig. 4.

KDM8 mediated demethylation of H3K36me2 leads to a decrease in HDAC1 recruitment in the cyclin A1 coding region. (A) ChIP analysis was performed to examine KDM8 effects on H3K36me2 and HDAC1 recruitment to the coding region of the cyclin A1 gene with inducible cells. ChIP assays were performed with indicated antibodies. The results were normalized to IgG controls and shown as an average fold change in enrichment. Data are presented as a mean ± SD from three independent experiments. (B). ChIP analysis was performed similar to A in the MCF7 cells that had stable knockdown of KDM8 with the indicated antibodies. (C). Immunoblot analysis of KDM8 knockdown MCF7 cells. The lysates from the KDM8 knockdown and control cells were probed with the antibodies as indicated.

Discussion

Identification of histone demethylases revealed the dynamic reversibility of epigenetic methylation marks (15–18). A complete understanding of the functional and spatial differences of histone lysine methylation has yet to be revealed. Among the various methylated lysine residues, H3K36me2 may be one of the least studied epigenetic marks. Although the tri- and monomethylated forms of H3K36 have been shown to be transcriptionally active marks (19), H3K36me2 in the coding region is typically defined as a transcriptional silencing histone modification and considered to influence transcript elongation (20, 21).

Although our in vivo studies conclusively demonstrated that full-length KDM8 can demethylate H3K36me2 in breast cancer cells, in vitro demethylation results were only successful when using an N-terminal–truncated (101–416 aa) KDM8 mutant. We speculate that the lack of full-length enzymatic activity in vitro may be caused by either an autoinhibitory domain of the protein or the requirement of a cofactor protein, whose interaction alters KDM8 protein conformation (Fig. S1B). Because the N-terminal mutants worked in vitro, we hypothesize that any potential regulatory modifications to KDM8 that direct its histone demethylase activity must occur within the first 101 amino acids. As previously seen in other histone modifiers (22, 23), these results further suggest that interactions with cofactor proteins or posttranslational modifications may regulate KDM8 demethylase activity. Our model is shown in Fig. S5. Preliminary analysis of the KDM8 amino acid sequence predicts a nuclear hormone receptor binding motif LXXL located in the N-terminal region. Additional preliminary studies have indicated that certain histone methyltransferases such as SUV39H1 and SETDB1 interact with KDM8 in the N-terminal region, indicating complex regulation of KDM8.

In this study, we demonstrated that KDM8 plays a critical role in the regulation of cell cycle by activating the cyclin A1 locus via demethylation of H3K36me2. Interestingly, KDM2B, another H3K36me2 demethylase, also has been shown to regulate cell cycle progression (24). Although the specific mechanism used by these two demethylases and the target genes affected may be different, both results point to the significance of the H3K36me2 epigenetic mark in cell cycle regulation.

Materials and Methods

Generation of Inducible KDM8/H321A Cell Lines.

MCF7 cell line expressing Tet-repressor, MCF7-pTR-7 (25), was a gift from Xin-Bin Chen (University of California, Davis). Lentivirus carrying an inducible expression cassette of KDM8 or KDM8 H321A was prepared by transfecting packaging plasmids (Invitrogen) with pLTRE-Flag-KDM8 or pLTRE-Flag-KDM8H321A plasmid using Lipofectamine 2000. The MCF7-pTR-7 cells were transduced with recombinant lentivirus and cells were selected with puromycin for 2 to 3 wk.

Immunohistochemistry.

IHC was performed and analyzed by Biomax by using laboratory-generated KDM8 antibody. Purified GST-KDM8 (residues 150–250) was used for the antigen. KDM8 antibody was validated by immunblotting bacculovirus-infected Sf9 cell lysates. Briefly, IHC staining on breast tumor tissue array (cat no. BR804), consisting of 40 patient cases (80 cores) matched with normal patient tissue, were analyzed by two methods: (i) software scanning analysis using ScanScope and (ii) staining intensity scoring by a panel of Biomax pathologists. Scores range from lowest (0) to highest (3+) intensity.

Histone Demethylation Assay.

Histone peptides were incubated with Flag-KDM8, Flag-KDM8 H321A, GST-KDM8 101-C, GST-KDM8 101-C H321A in histone demethylase buffer [20 mM Tris-HCL, pH 7.3, 150 mM NaCl, 50 μM Fe(NH)₄(SO₄)₂, 1 mM α-ketoglutarate, 2 mM ascorbic acid] at 37 °C for 2 to 4 h. Purified proteins used in this study are shown in Fig. S6. Peptides were commercially purchased from Millipore or custom synthesized by ThermoFisher Scientific. Reactions were analyzed by MALDI-TOF MS.

Immunofluorescence Microscopy and Quantification.

KDM8 Dox-inducible MCF7 cells were cultured for 72 h after induction. Dox-induced and uninduced cells were then mixed and spotted onto coverslips, fixed with 4% formaldehyde, permeabilized in 0.1% Triton X-100/PBS solution, and blocked with 1% BSA/PBS solution. Detailed staining and quantification methods are described in SI Materials and Methods.

ChIP Assays.

ChIP assays were performed following the University of California Davis Genome Center ChIP protocol (http://www.genomecenter.ucdavis.edu/farnham). The primary antibodies used in this study were as follows: KDM8, FLAG (Sigma), HDAC1, H3K36me2, acetylated histone H3, and acetylated histone H4. These antibodies were obtained from Millipore. The secondary rabbit anti-mouse IgG and goat anti-mouse IgG were purchased from MP Biomedicals. The nonspecific rabbit IgG and mouse IgG used as negative controls in the ChIP assays were purchased from Alpha Diagnostics.

siRNA Transfection.

siRNA negative control scramble sequence and sequences targeting KDM8 were designed (targeting 3′UTR) and purchased through Integrated DNA Technologies. The standard protocol for siRNA provided by IDT was followed. Briefly, cells were transfected with 10 nM of siRNA duplex sequences using Trifectin reagent, purchased from IDT. Cells were collected and prepared after 72 h. Duplex sequences were as follows: siKDM8, 3′UTR, 5′-CGCUGUCACUGAUCCCAAUUACUCT-3′; 3′-CGGCGACAGUGACUAGGGUUAAUGAGA-5′.