|Home | About | Journals | Submit | Contact Us | Français|
Mutations in PHF8 are associated with X-linked mental retardation and cleft lip/cleft palate. PHF8 contains a plant homeodomain (PHD) in its N terminus and is a member of a family of JmjC domain-containing proteins. While PHDs can act as methyl lysine recognition motifs, JmjC domains can catalyze lysine demethylation. Here, we show that PHF8 is a histone demethylase that removes repressive histone H3 dimethyl lysine 9 marks. Our biochemical analysis revealed specific association of the PHF8 PHD with histone H3 trimethylated at lysine 4 (H3K4me3). Chromatin immunoprecipitation followed by high-throughput sequencing indicated that PHF8 is enriched at the transcription start sites of many active or poised genes, mirroring the presence of RNA polymerase II (RNAPII) and of H3K4me3-bearing nucleosomes. We show that PHF8 can act as a transcriptional coactivator and that its activation function largely depends on binding of the PHD to H3K4me3. Furthermore, we present evidence for direct interaction of PHF8 with the C-terminal domain of RNAPII. Importantly, a PHF8 disease mutant was defective in demethylation and in coactivation. This is the first demonstration of a chromatin-modifying enzyme that is globally recruited to promoters through its association with H3K4me3 and RNAPII.
Posttranslational modifications of histone tails play an important role in chromatin structure and function (27). While the presence of several modifications correlates with gene activation and transcription, others have opposing repressive functions and are enriched in transcriptionally inactive or heterochromatic regions. A well-studied type of histone modification is methylation of lysine residues, which is conferred by histone methyltransferases (KMTs). There are three possible states of lysine methylation, namely, mono-, di-, and trimethylation. As expected for reversible marks involved in dynamic gene regulation, the methyl groups can be removed by histone demethylases (KDMs) of either the LSD1 or the Jumonji C-terminal-containing (JmjC) family of proteins (43, 48).
The JmjC family of histone demethylases consists of approximately 30 members in humans (9, 25). Plant homeodomain (PHD) finger-containing proteins 2 and 8 (PHF2 and PHF8, respectively) and KIAA1718 constitute a subgroup containing a single N-terminal PHD followed by the catalytic JmjC domain. Several protein domains (including the PHD; the MBT, WD40, and Tudor domains; and chromodomains) have been shown to bind to peptides containing either an unmodified or a methylated lysine residue, and some PHDs specifically interact with histone H3 trimethylated at lysine 4 (H3K4me3) (42, 51, 53). H3K4me3 is considered to be an activating chromatin mark because it is enriched at active RNA polymerase II (RNAPII) transcription start sites (TSSs) (4, 17). In mammalian cells, H3K4me3 is mainly established by the SET1/KMT2 family, which includes the mixed-lineage leukemia 1 to 4 (MLL1 to -4) proteins. These chromatin modifiers can be recruited to promoters upon gene activation (16, 52), and certain MLL complexes also contain KDMs that concomitantly remove the repressive H3K27me3 mark and enhance expression (7, 21, 30).
On the other hand, H3K4me3 can be erased by KDMs of the JARID1/KDM5 family that are part of polycomb-repressive complexes (PRC1/2) (29, 35). These complexes also contain KMTs that are able to methylate histone H3 at lysine 9 or 27. Methylated H3K9 and H3K27 are repressive marks, which are read by PRC1 via chromodomain-containing proteins, a mechanism that is thought to be responsible for enforcement and inheritance of silenced chromatin (5, 18). This obvious coupling of writers (KMTs), readers (like proteins containing PHDs), and erasers (KDMs) enables cross talk between different chromatin modifications (45). Furthermore, many SET and JmjC domain proteins contain additional domains or putative DNA-binding modules, increasing their specificity and affinity for modified chromatin.
PHF8 is a ubiquitously expressed nuclear protein whose dysfunction is implicated in disease (28). Mutations in the PHF8 locus on the X chromosome have been linked to Siderius-Hamel syndrome, an X-linked mental retardation (XLMR) that is often accompanied by cleft lip and/or cleft palate (44). The described mutations result mostly in early truncations of the protein before or in the JmjC domain (1, 28). Moreover, a single point mutation (F279S) in the JmjC domain and a microdeletion of the whole locus were also reported to cause the disease phenotype (26, 36).
In the present study, we delineate crucial biochemical properties of PHF8. We report that PHF8 is a demethylase specific for H3K9me2 and a binder of H3K4me3-marked nucleosomes. Genome localizations show that PHF8, H3K4me3, and RNAPII cooccupy thousands of promoters. Association studies suggest that PHF8 interacts directly with the C-terminal domain (CTD) of the RPB1 subunit of RNAPII. The F279S mutant of PHF8 displays cellular mislocalization, defective histone demethylation, and aberrant coactivation function.
Full-length PHF8 (NCBI reference sequences NM_15107.1 for mRNA and NP_055922.1 for protein) was cloned into the pFlag-CMV2 vector for expression in mammalian cells and into pFastBacHTa-Flag for expression in insect cells. Truncated versions (containing amino acids 1 to 352 or 1 to 489 of PHF8) were cloned into pFlag-CMV2 with a short simian virus 40 (SV40) nuclear localization signal (NLS) (PKKKRKVG) added to the C terminus to ensure correct nuclear import. The mutations D28A/W29A, H31A/C34A, H247A/D249A, and F279S were generated by site-directed mutagenesis using the QuikChange protocol (Stratagene). The cloning primer and mutagenesis oligonucleotide sequences are described in the supplemental material. JARID1A- and JMJD3-HA-pCMV plasmids were kind gifts from Kristian Helin's laboratory. A PCR fragment encoding the PHD of human PHF8 (residues 2 to 66) (PHF8-PHD) was inserted into the pGEX-2T-derived pRPN265NB plasmid. Glutathione S-transferase (GST)-PHF8-PHD mutant plasmids were obtained by site-directed mutagenesis using the QuikChange protocol (Stratagene). The plasmids for GST-Taf3 (amino acids 857 to 924) and GST-UbcH5B were described previously (14, 51). The EB2 plasmid encoding GST fused to 27 repeats of the consensus CTD (YSPTSPS) was a kind gift of Marc Vigneron (ESBS, Strasbourg, France). All inserts were verified by DNA sequencing.
Full-length PHF8 containing N-terminal Flag and His6 tags was expressed in Sf9 cells using baculovirus (Invitrogen Bac-to-Bac system). The cells were lysed in whole-cell lysis buffer (20 mM Tris-HCl, pH 7.4, 137 mM NaCl, 1 mM EDTA, 1.5 mM MgCl2, 10% glycerol, 1% Triton X-100, 0.2 mM phenylmethylsulfonyl fluoride [PMSF], 1 μg/ml aprotinin, 1 μg/ml leupeptin, 1 μg/ml pepstatin) 2 days after infection and incubated with M2 Flag agarose beads (Sigma). The beads were washed four times with BC500 buffer (20 mM Tris-HCl, pH 7.4, 0.2 mM EDTA, 10 mM β-mercaptoethanol, 10% glycerol, 500 mM KCl, 0.1% NP-40, 0.2 mM PMSF, 1 μg/ml aprotinin, 1 μg/ml leupeptin, 1 μg/ml pepstatin), and recombinant protein was eluted with Flag peptide (400 μg/ml; Sigma) in BC500. After concentration of the eluates by centrifugation in Microcon YM-10 columns (Millipore), fractions were run on 4 to 12% Tris-glycine SDS-polyacrylamide gels and Coomassie stained (Invitrogen), and the protein concentration was estimated by comparison to bovine serum albumin (BSA) standards.
Four micrograms of bulk histones from calf thymus (Sigma) was incubated in 30 μl demethylation buffer [20 mM Tris-HCl, pH 7.4, 2 mM ascorbic acid, 1 mM α-ketoglutarate, approximately 100 mM KCl, 100 mM NaCl, 50 μM (NH4)2Fe(SO4)2] with or without up to 10 μg recombinant PHF8 for 4 h at 37°C. After denaturation in SDS loading buffer, the reaction mixtures were run on 4 to 20% Tris-glycine SDS-polyacrylamide gels and blotted to polyvinylidene difluoride (PVDF) membranes (Millipore). Western blots were blocked for >1 h at room temperature with 5% skim milk powder diluted in Tris-buffered saline containing 0.1% Tween 20 (TBS-T) before sequential incubation for 1 h at room temperature with antibodies. Primary mouse or rabbit antihistone antibodies were usually diluted 1:1,000, and secondary anti-IgG-alkaline phosphatase antibodies were diluted 1:10,000 (Promega) in TBS-T. Signals were developed by incubation with 5-bromo-4-chloro-3-indolylphosphate-nitroblue tetrazolium (BCIP-NBT) substrate, and the blots were dried and scanned. All antibodies used in this study are listed in the supplemental material.
293T, U2OS, Hs68, and HeLa-S3 cells were grown in Dulbecco's modified Eagle's medium (DMEM) containing 10% fetal bovine serum (FBS), l-glutamine, and antibiotics. Transient transfection of plasmid DNA was performed 6 h after seeding using FuGene6 transfection reagent according to the manufacturer's instructions (Roche). For immunofluorescence assays, 293T cells were seeded onto glass coverslips and transfected with 1 μg PHF8(1-489)NLS-pFlag-CMV2 wild-type or mutant constructs using 3 μl Metafectene Pro transfection reagent (Biontex) per coverslip. Twenty-four to 48 h later, the coverslips were washed with phosphate-buffered saline (PBS) and fixed in fresh 1% formaldehyde-PBS for 10 min. After being washed, the cells were permeabilized in 0.1% Triton X-100-PBS for >10 min and sequentially incubated for 1 h at room temperature with antibodies. Primary mouse or rabbit antihistone antibodies were diluted 1:200 to 1:500, and anti-Flag antibodies were diluted 1:1,000 in 0.1% Triton X-100-PBS. Secondary anti-mouse IgG-Alexa Fluor-488 and anti-rabbit IgG-Alexa Fluor-568 antibodies (Invitrogen) were diluted 1:1,000 in 0.1% Triton X-100-PBS. After DAPI (4′,6-diamidino-2-phenylindole) counterstaining, the coverslips were washed with PBS and mounted on slides using Mowiol 4-88 containing 2.5% 1,4-diazabicyclo[2.2.2]octane (DABCO) as an antibleach. Detailed information on image acquisition and statistical analysis is given in the supplemental material.
293T, HeLa-S3, or Hs68 cells were trypsinized from 2 (293T and HeLa) or 16 (Hs68; 4-day serum starved or 4-h serum restimulated) subconfluent 15-cm dishes, washed with PBS, and fixed in 10 ml of 1% formaldehyde-PBS at room temperature for 10 min. One milliliter of 1.25 M glycine was added and incubated for 5 min at room temperature. The cells were pelleted, washed with cold PBS, and lysed for 15 min on ice in 300 μl chromatin immunoprecipitation (ChIP) lysis buffer (1% SDS, 10 mM EDTA, 50 mM Tris-HCl, pH 8, and protease inhibitors). After the addition of 1.7 ml ChIP dilution buffer (0.01% SDS, 1% Triton X-100, 1.2 mM EDTA, 16.7 mM Tris-HCl, pH 8, 167 mM NaCl, and protease inhibitors), chromatin was sonicated on ice in a Bioruptor (Diagenode) for 10 min (30 s on, 30 s off; high power load). After dilution with another 1 ml of ChIP dilution buffer, solubilized chromatin was cleared and precleared by incubation with protein A-agarose beads (Millipore). Five hundred to 1,000 μl of this chromatin solution (25 to 50 μg DNA) was incubated overnight with 2 to 4 μg of antibody (listed in the supplemental material). Protein A-bead slurry (30 μl) was added, rotated for 2 h at 4°C, and washed 3 times with low-salt buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl, pH 8, 150 mM NaCl), one time with high-salt buffer (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl, pH 8, 500 mM NaCl), one time with LiCl buffer (250 mM LiCl, 1% NP-40, 1% deoxycholate, 1 mM EDTA, 10 mM Tris-HCl, pH 8), and one time with TE buffer (10 mM Tris-HCl, pH 7.6, 1 mM EDTA). Bound chromatin was eluted for >1 h at room temperature in 100 μl fresh elution buffer (0.1 M NaHCO3, 1% SDS). Cross-links were reversed overnight at 65°C, and DNA was purified using a QIAquick PCR purification kit (Qiagen).
For ChIP followed by high-throughput sequencing (ChIP-seq), approximately 10 ng immunoprecipitated DNA per sample was used for end cleanup and adaptor ligation. Size-selected preamplified inserts of 135 ± 25 bp were sequenced by the Illumina/Solexa technique. Total reads were mapped to the human genome version 18 using Paolo Ribeca's GEM mapper (http://www.paoloribeca.net/software/GEM/index.html) (see Table S2 in the supplemental material). Non-strand-specific cluster/peak analyses were performed with the NGS-Analyzer (Genomatix RegionMiner 2.02) using a 100-bp window and a Poisson distribution-derived cutoff value (depending on total reads; at least 8 to 10 reads per cluster). Regions overlapping with clusters of irrelevant IgG ChIP (for HeLa and 293T cells) or input DNA control (for Hs68 cells) were filtered out. Overlap analyses were done with Genomatix GenomeInspector (Eldorado 12-2008) using our own ChIP-seq data and published HeLa data for H3K4me3 (37) and for RNAPII (38). Strand-specific read analyses were performed with the ChIP-cor tool (http://ccg.vital-it.ch/chipseq/). Reads were correlated to oriented TSSs (49,392 ENSEMBL50 features). The examined distance from the reference feature was ±1,000 bp, the window width was 20 bp, the cutoff value was 10 counts, and the output format was normalized counts. Wiggle files for display in the UCSC genome browser were generated using the ChIP-center tool (hg18; tag shift, 60 bp; count cutoff, 10; resolution, 20 bp). For correlation studies, published HeLa expression data were used (10).
For quantitative ChIP (qChIP), purified immunoprecipitated DNA was used as a template for real-time PCR with 0.2 μM locus-specific primers (sequences are given in the supplemental material) and Bio-Rad IQ SYBR green mix. Samples and dilution series of input DNA were run in triplicate. The immunoprecipitated percentage of input was calculated using individual standard curves.
Expression of GST fusion proteins was induced, and lysates were prepared essentially as described previously (14). H3 peptides (~0.1 nmol) were coupled to magnetic streptavidin beads (M-280; Dynal) in H3-binding buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 0.1% NP-40, 10 μM ZnCl2, 1 mM dithiothreitol [DTT], and protease inhibitors), and excess peptide was removed by washing. Crude bacterial lysates were incubated with peptide-coated streptavidin beads in binding buffer (50 mM Tris-HCl, pH 8, 150 mM NaCl, 0.1% NP-40, 10 μM ZnCl2, 1 mM DTT, and protease inhibitors) for 2 to 3 h at 4°C. Following extensive washing, bound proteins were eluted in SDS sample buffer and analyzed by SDS-PAGE and Coomassie brilliant blue R-250 (Bio-Rad).
For GST pulldown experiments, glutathione-agarose beads were coated with bacterial lysates containing GST fusion proteins for 60 min at 4°C and subsequently washed three times with H3-binding buffer. Mononucleosomes from HeLa cells (prepared as described previously ) or transfected 293T cell lysates were mixed with 50 μl beads in a total volume of 400 μl and incubated overnight at 4°C in H3-binding buffer. The beads were washed five times with 500 μl of H3-binding buffer. Bound proteins were analyzed by immunoblotting.
For luciferase assays, U2OS cells were transfected in triplicate. The firefly reporter luciferase construct 5XGal4MLP-Luc was cotransfected with 50 ng TK promoter-driven Renilla luciferase plasmid (to normalize for differences in transfection efficiency) and PHF8-pFlag-CMV2 in combination with Gal4 activator constructs as indicated. Cell lysates were prepared 24 h after transfection, and the luciferase activity was determined using the Dual-Luciferase Reporter Assay System (Promega).
293T cells were lysed 24 to 40 h after transfection in IP buffer (50 mM Tris-HCl, pH 8, 150 mM KCl, 5 mM MgCl2, 0.5 mM EDTA, 0.1% NP-40, phosphatase, and protease inhibitor cocktails from Sigma Aldrich). Flag-tagged protein was bound to M2 beads (Sigma-Aldrich) for 3 h at 4°C, followed by 3 washes with IP buffer. Precipitated proteins were separated by SDS-PAGE and transferred onto PVDF membranes. The membranes were developed with the appropriate antibodies and enhanced chemiluminescence (ECL) (Pierce).
HeLa cells were transfected with pRetroSuperPuro plasmids containing short hairpin RNAs (shRNAs) targeting PHF8 (PHF8-shA or -shB) or nontargeting control (NT-sh) (sequences are given in the supplementary material). One day after transfection, selection with 2.5 μg/ml puromycin was performed for two more days. Cells were harvested with Trizol reagent (Invitrogen) for RNA isolation or with nondenaturing lysis buffer for protein isolation or were fixed and sonicated for qChIP analysis. PHF8 knockdown was confirmed by quantitative reverse transcription (qRT)-PCR and immunoblotting.
Three independent knockdown experiments were performed in HeLa cells with PHF8-shA and control NT-sh. Total RNA (500 ng) was reverse transcribed, labeled by in vitro transcription with Cy3 and Cy5, and hybridized to whole human genome arrays in dye swap (Agilent G4112F). After scanning of the arrays (Agilent G2565BA), features were extracted with GenePix Pro 6.0 and analyzed with AFM 4.0 (6). Probes with average changes higher than 2-fold were considered regulated. Nonannotated transcripts, PHF8, and three ambiguous transcripts (CCDC6, SCD, and CNOT6, which had different probes up- and downregulated) were excluded from further analysis. For 11 transcripts, regulation was confirmed by qRT-PCR using cDNAs obtained by reverse transcription of total RNA with Moloney murine leukemia virus (MMLV) enzyme and oligo(dT)18 primer. The sequences of qRT-PCR and qChIP primers are listed in the supplemental material.
Data from the ChIP-seq and microarray experiments were deposited as a SuperSeries under accession number GSE20753 in the GEO database (http://www.ncbi.nlm.nih.gov/geo/).
Several JmjC-containing proteins have been shown to specifically demethylate histone H3 methylated at lysine 4, 9, 27, or 36. These enzymes remove methyl moieties by an oxidative reaction that requires an Fe2+ ion and α-ketoglutarate as cofactors. Since the JmjC domain of PHF8 carries all the amino acids needed for cofactor binding (Fig. (Fig.1A),1A), PHF8 was suggested to be a histone demethylase (25) and was very recently demonstrated to be active in vitro (31). We expressed full-length PHF8 in Sf9 insect cells using the baculovirus system and isolated the recombinant protein by Flag affinity purification (Fig. (Fig.1B).1B). Bulk histones from calf thymus were used as PHF8 substrates and analyzed by immunoblotting using antibodies specific for different methyl lysine modifications of histones. We observed a profound reduction of H3K9me2 after incubation with PHF8. In contrast, we did not detect any change for the other histone methyl marks, including, H3K9me3, H3K27me3/2/1, H3K4me3/2, H3K36me3/2, H3K79me3/2, and H4K20me3/2/1 (Fig. (Fig.1C).1C). Demethylation was proportional to the amount of enzyme (see Fig. S1A in the supplemental material) and the reaction time (see Fig. S1B in the supplemental material) and was dependent on the cofactors Fe2+, α-ketoglutarate, and ascorbic acid (see Fig. S1C in the supplemental material). Like Loenarz and colleagues (31), we also observed weak activity on H3K9me1 (see Fig. S1B in the supplemental material), but in contrast, we could not detect significant reduction of H3K27me2 or H3K36me2 (Fig. (Fig.1C).1C). This difference may have been due to different recombinant proteins used in the two studies: ours was a full-length PHF8, while theirs was devoid of the PHD and the C-terminal half. While our work was under review, other studies were published that also showed H3K9me2-specific demethylase activity for PHF8 (15, 19, 55).
To verify its demethylase activity in vivo, we expressed a Flag-tagged truncated version of PHF8 (containing amino acids 1 to 489 and a C-terminal nuclear localization signal) in 293T cells. After 24 to 48 h, the cells were fixed and subjected to indirect immunofluorescence with antibodies against Flag and histone H3 methylated at different lysine residues. The mean fluorescence signals of 10 of each untransfected and transfected nucleus were quantified and compared using Student's t test (see Table S1 in the supplemental material). H3K9me2 was markedly reduced in cells that overexpressed wild-type Flag-PHF8(1-489) (Fig. (Fig.1D).1D). The JmjC domain conferred the H3K9me2 demethylase activity. This was indicated by the lost activity of an H247A/D249A mutant predicted to be defective in cofactor binding (Fig. (Fig.1E).1E). Interestingly, the XLMR point mutant F279S was devoid of activity (Fig. (Fig.1F).1F). This disease mutant displayed aberrant localization to the cytoplasm, formed aggregates in vivo (Fig. (Fig.1F),1F), and had reduced solubility at low salt concentrations in vitro (data not shown). Our observations with this XLMR mutant are consistent with recent work by Loenarz et al. (31). Intriguingly, we also observed dramatically impaired demethylase activity when we mutated the putative H3K4me3-binding residues D28 and W29 to alanine residues in the PHD (Fig. (Fig.1G).1G). However, we could observe a small demethylation activity with this mutant following prolonged expression (see Table S1 in the supplemental material).
While ectopic expression of PHF8(1-489) resulted in a decrease in H3K9me1 (Fig. (Fig.1H),1H), there were no changes for H3K9me3 (Fig. (Fig.1I),1I), H3K4me2 (Fig. (Fig.1J),1J), H3K27me2 (Fig. (Fig.1K),1K), or H3K36me2 (see Table S1 in the supplemental material). The demethylation of H3K9me1 was deemed to be less robust both in vivo (Fig. (Fig.1H)1H) and in vitro (Fig. S1B). The increased amounts of H3K9me1 following the demethylation of H3K9me2 may mask the extent of H3K9me1 activity of PHF8.
Further truncation of the PHF8 protein leading to a protein spanning amino acids 1 to 352 containing the PHD and JmjC domain was devoid of demethylation activity in vivo (see Fig. S2A in the supplemental material). Surprisingly, the full-length PHF8 was also deemed to be very low in demethylation activity when expressed in vivo (see Fig. S2B in the supplemental material). These results suggest that the approximately 100 amino acids following the consensus JmjC domain are important for demethylation activity in vivo. Moreover, it is likely that the C-terminal half of PHF8 exerts a negative effect on the demethylation activity in vivo, as full-length PHF8 displayed diminished activity. Finally, consistent with a published report (31), overexpression of PHF8(1-489) in HeLa cells did not yield a change in H3K9me2 (see Fig. S2C in the supplemental material), suggesting an important role for a cell type specificity factor for demethylation in vivo. Taken together, our results indicate that PHF8 displays demethylation activity toward mono- and dimethyl H3K9. However, we found evidence for cell-type-specific factors regulating the demethylation activity of PHF8 in vivo. Importantly, we showed the contribution of the PHD to the enzymatic activity of PHF8 in 293T cells in vivo, while the domain was shown not to be crucial for its activity in vitro (31).
Comparison of the PHF8 PHD with H3K4me3-specific PHDs of the bromodomain PHD finger transcription factor (BPTF) and TBP-associated factor 3 (TAF3) (39, 51) revealed conservation of crucial Zn2+- and H3K4me3-binding residues (indicated in red and green in Fig. Fig.1A).1A). In peptide pulldown assays, we found that bacterially expressed GST fusion with the PHD of PHF8 (amino acids 2 to 66) was specifically retained by biotinylated H3K4me3/2 peptides (amino acids 1 to 17), but not by H3K4me1, H3K9me3/2/1, and H3K36me3/2/1 (amino acids 22 to 39) or unmethylated H3 peptides (Fig. (Fig.2A).2A). The PHD of Taf3 was used as a positive control in these assays. Interaction of the PHF8-PHD with H3K4me3 was not compromised by modifications at neighboring residues, such as methylation of H3R2/H3K9 or phosphorylation of H3T6/H3S10. In contrast, concomitant phosphorylation of H3T3 blocked K4me3 binding (Fig. (Fig.2B).2B). H3T3 phosphorylation is mediated by the mitotic haspin kinase (11), and this correlates with the observation that PHF8 dissociates from mitotic chromosomes despite the presence of H3K4me3 (see Fig. S3A and B in the supplemental material). As expected, mutation of Zn2+-binding (H31A/C34A) (data not shown) or aromatic-cage (D28A/W29A) residues in the PHF8-PHD abolished H3K4me3 binding (Fig. (Fig.2B).2B). Similar results were obtained with a mammalian expression construct containing both the PHD and JmjC domain of PHF8 (amino acids 1 to 352). While wild-type PHF8, the JmjC H247A/D249A mutant, and the disease mutant (F279S) bound H3K4me3/2 peptides, the PHD D28A/W29A mutant did not (Fig. (Fig.2C2C).
We next assessed whether the PHD of PHF8 can also interact with nucleosomal H3K4me2/3. Using GST-PHF8-PHD, but not the D28A/W29A mutant, we could retain mononucleosomes derived from HeLa cells (confirmed by immunoblotting against H2A, H3, and H4), which were enriched in H3K4me3/2 but not H3K9me2 (Fig. (Fig.2D).2D). Again GST-Taf3-PHD served as a positive control. Analysis of PHF8 in vivo using antibodies against PHF8 revealed colocalization with H3K4me3 and RNAPII (see Fig. S3A and B in the supplemental material) in putative euchromatic regions of interphase 293T cells, while it was depleted at the nuclear periphery and in perinucleolar regions, where H3K9me2 was enriched (see Fig. S3C in the supplemental material).
Having established that the PHF8-PHD can bind the H3K4me3 mark, we examined the observation that PHD integrity was important for in vivo demethylase activity (Fig. (Fig.1G).1G). To test K4me3 involvement in this, we cotransfected 293T cells with wild-type Flag-PHF8(1-489)NLS and the H3K4me3-specific demethylase JARID1A/KDM5A (8). The ability of PHF8 to demethylate H3K9me2 was significantly reduced by concomitant H3K4me3 removal (see Fig. S4A in the supplemental material), while control cotransfections of PHF8 with JMJD3/KDM6B, an H3K27me3 demethylase (13), did not result in impaired PHF8 demethylase activity (see Fig. S4B in the supplemental material). The mean fluorescence signals of H3K9me2/H3K4me3 for PHF8/JARID1A and H3K9me2/H3K27me3 for PHF8/JMJD3 cotransfections were quantified and compared for transfected and untransfected cells (see Fig. S4C in the supplemental material). This is in agreement with the stimulation of H3K9me2 demethylation activity on peptides by the presence of H3K4me3 (19). Taken together, these experiments suggest that the PHD of PHF8 can bind specifically to H3K4me3-marked nucleosomes and that this binding is important for its histone demethylase activity in vivo.
In order to determine the genomic location of PHF8, we performed chromatin immunoprecipitation followed by high-throughput sequencing using the Illumina/Solexa technology (ChIP-seq). Sonicated chromatin from fixed 293T, HeLa-S3, and serum-starved and restimulated Hs68 cells was immunoprecipitated with rabbit anti-PHF8 antibody and normal rabbit IgG as a negative control. In Hs68 fibroblasts, we used input DNA as a control and performed ChIP with rabbit PHF8 or H3K4me3 antibody. On average, 14 million total reads were obtained per flow cell, and approximately 63% of these could be unambiguously mapped to the human genome. Cluster analysis of uniquely aligned reads revealed more than 10,000 significant peaks for PHF8 in all three cell lines. The ChIP-seq breakdown and cluster identities are given in Table S2 in the supplemental material. The peaks were almost exclusively located in close proximity to TSSs, and we observed strong overlap between PHF8, H3K4me3, and RNAPII clusters using our own and published data (37, 38) (Table (Table1),1), suggesting that PHF8 preferentially occupied thousands of active or poised promoters. ChIP-seq data were validated at several c-myc target loci by quantitative real-time PCR of chromatin-immunoprecipitated DNA obtained with rabbit PHF8, c-myc, or normal IgG. PHF8 exhibited high (for NCL [see Fig. S5A in the supplemental material]), medium (for CDK4, CCNB1, and ITGB1 [see Fig. S5B to D in the supplemental material]), or no (for TERT [see Fig. S5E in the supplemental material]) enrichment in these regions, which also corresponded to H3K4me3 and RNAPII occupancies.
A more detailed strand-specific analysis of sense (5′) or antisense (3′) reads in respect to oriented TSSs revealed a similar distribution and a phase shift of 120 bp for PHF8 in 293T (Fig. (Fig.3A)3A) and HeLa-S3 (Fig. (Fig.3B)3B) cells and for PHF8 (Fig. (Fig.3C)3C) and H3K4me3 (Fig. (Fig.3D)3D) in Hs68 cells. No enrichment of reads for controls, like IgG ChIP of HeLa cells (Fig. (Fig.3E),3E), or for input DNA of serum-stimulated Hs68 cells (Fig. (Fig.3F)3F) was observed at TSSs. Active and poised TSSs exhibit defined positioning of nucleosomes, which tend to be methylated at H3K4 (4, 41). Similar to previously reported experiments using native ChIP of mononucleosomes (40, 41) or sonicated cross-linked chromatin (37), we also observed several phased H3K4me3 peaks at the expected nucleosome positions in Hs68 (Fig. (Fig.3D).3D). The phase shift in our experiment was with 120 bp; however, it was smaller than the 150 bp obtained with mononucleosomes, possibly because of harsher fragmentation conditions during the sonication process. Analysis of PHF8 reads indicated one peak immediately before and one after the TSS (Fig. 3A to C). Therefore, we concluded that PHF8 bound preferentially to nucleosomes at the −2 and +1 positions. A short region at the TSS corresponding to nucleosome position −1 was depleted of H3K4me3 and PHF8 reads. In active or poised promoters, this nucleosome position is occupied either by RNAPII and the basal transcription machinery (41) or by unstable H2A.Z/H3.3-containing nucleosomes (23). Association of PHF8, H3K4me3, and PolII in vivo was further substantiated by their colocalization in immunofluorescence analysis (see Fig. S3A and B in the supplemental material).
Next, we determined the overlap of PHF8-occupied genes in the 3 examined cell lines. Out of 10,876 genes in HeLa, 9,210 in 293T, and 8,014 in Hs68 cells, 6,632 shared PHF8 enrichment at their promoters, while others were occupied in only one or two of the cell lines (Fig. (Fig.3G3G).
Subsequently, we correlated PHF8 occupancy with expression and H3K4me3 levels. Basically, all annotated genes occupied by PHF8 in Hs68 cells also carried H3K4me3 peaks, but we detected about twice as many clusters and occupied genes for H3K4me3 as for PHF8. We sorted the genes by decreasing counts of H3K4me3 reads at their promoters, grouped them into bins of 500, and determined the percentage of PHF8 cooccupancy per bin. While 70% of the promoters with high and medium H3K4me3 levels were cooccupied by PHF8, less than 10% of promoters with low H3K4me3 also carried PHF8 (Fig. (Fig.3H).3H). Furthermore, the average PHF8 read count at promoters with high H3K4me3 levels was about twice as high as at those with low levels (data not shown). In a similar analysis, we took advantage of Affymetrix data for HeLa from a previous study (10), sorted genes by decreasing expression levels, classified them into bins of 500, and for each bin determined the percentage of PHF8 occupancy at their promoters in HeLa cells. While 90% of genes with high transcript levels displayed PHF8 occupancy, only 10% of genes with low expression levels were occupied by PHF8 (Fig. (Fig.3I).3I). These results suggest a positive correlation between genes that are occupied by PHF8 and those that display transcriptional activity. However, about 30% of the genes with high H3K4me3 levels in Hs68 were not cooccupied by PHF8.
PHF8 may activate transcription, for example, by binding to an activating chromatin mark (H3K4me3) and/or by removing a repressive chromatin modification (H3K9me2). We used a reporter assay to test these possibilities (51). Cotransfection of human osteosarcoma U2OS cells with a firefly luciferase reporter construct and a construct encoding full-length Flag-tagged PHF8 indicated a 6-fold increase in promoter activity (Fig. (Fig.4A).4A). Moreover, ectopic expression of PHF8 revealed a coactivator function toward Gal4-tagged Ash2, p53, c-myc, and E2F (Fig. (Fig.4A).4A). Shorter versions of PHF8 (1-352 and 1-489) displayed impaired coactivation, although they were expressed at similar levels, as shown by Western blotting (Fig. (Fig.4A).4A). To extend this, we initially assessed the functional consequences of PHD and JmjC mutations in the context of truncated PHF8(1-489). While PHF8(1-489) displayed impaired coactivation capacity compared with full-length PHF8, it was still able to enhance E2F activity by about 2.5-fold (Fig. (Fig.4B).4B). Interestingly, while the PHD (D28A/W29A) and XLMR (F279S) mutants displayed diminished coactivation functions, the JmjC H247A/D249A mutant showed only a small decrease in coactivator function.
We next tested whether these observations extended to the full-length PHF8. Similar to PHF8(1-489), in the context of full-length PHF8, the PHD D28A/W29A mutant or the F279S mutant was impaired in coactivation of Gal4-p53 or Gal4-Ash2 (Fig. (Fig.4C).4C). Again, the catalytic JmjC H247A/D249A mutant displayed a minor reduction in coactivation. Taken together, these results indicate that a functional PHD is an important component of the coactivator function of PHF8 and that the activity of the JmjC domain may not be an essential determinant, at least in the context of luciferase assays.
Genomic localization analysis of PHF8 revealed a preference for transcription start sites that are H3K4 methylated and occupied by RNAPII. We hypothesized that PHF8 could interact with preinitiation complex components like the general transcription factor TFIID or RNAPII. To examine this, we performed CoIP experiments with full-length Flag-PHF8 in 293T cells. Immunoblot analysis revealed coprecipitation of the RNAPII subunits RPB1 (POLR2A) and RPB3 (POLR2C), but not of the TAF1 subunit of TFIID (Fig. (Fig.5A).5A). When we compared full-length Flag-PHF8 with truncated versions in CoIPs, the N-terminal half of PHF8 (1-489) displayed a weak association with RNAPII, and such an interaction could not be detected with the 1-352 version (Fig. (Fig.5B).5B). These results indicate that the C-terminal half of PHF8 plays an important role in RNAPII association. This contention is consistent with coimmunoprecipitation results of the full-length PHD D28A/W29A mutant and the JmjC H247A/D249A mutant, which interacted with RNAPII similarly to the wild type (Fig. (Fig.5C).5C). Even though larger DNA amounts were transfected, the XLMR F279S mutant showed lower levels of expression and consequently reduced amounts of precipitated RNAPII (Fig. (Fig.5C),5C), probably due to mislocalization and aggregation.
The CTD of the largest RNAPII subunit, RPB1, has been shown to be a binding platform for many proteins (33). To test PHF8 interaction with the CTD, we performed GST pulldown experiments using GST fused to the CTD (consisting of 27 heptad repeats) and lysates from Flag-PHF8-transfected cells. GST-CTD bound full-length PHF8 and truncated 1-489, but not 1-352, while GST-Taf3 was used as a negative control for these experiments (Fig. (Fig.5D).5D). Interestingly, none of the tested PHF8 mutations (D28A/W29A, H247A/D249A, or F279S) affected association of full-length PHF8 with GST-CTD (Fig. (Fig.5E5E).
Next, we tested whether phosphorylation of the CTD influences its interaction with PHF8. Phosphorylation of serine 5 in the heptad repeat is conferred by CDK7 and takes place in the initiation phase near the transcription start site, while phosphorylation of serine 2 by CDK9 is thought to occur in the transcribed region during processive elongation (33). Following preincubation of GST-CTD with the CDK7 or CDK9 kinase, we observed the expected mobility shift in CTD dependent on ATP addition (Fig. (Fig.5F).5F). However, phosphorylation of the CTD using either kinase did not alter the interaction with PHF8 (Fig. (Fig.5F).5F). These results suggest that PHF8 interacts mainly through its C-terminal half with the CTD of RNAPII and that this association is not influenced by the phosphorylation status of the CTD. Taken together, the interaction property of PHF8 with RNAPII is consistent with the genomic localization of these proteins and the transcription coactivator function of PHF8.
We also examined the possibility that PHF8 may physically interact with transcriptional activators and coactivators that were used for functional-coactivation assays (Fig. (Fig.4A).4A). To this end, we performed CoIPs with lysates from 293T cells that were cotransfected with Flag-PHF8 and Gal4-Ash2, Gal4-p53, or c-myc. We were unable to detect efficient association of PHF8 with any of these proteins (see Fig. S6A and B in the supplemental material). These results indicate that the coactivator function of PHF8 is not mediated by direct association with transcription factors, but more likely via its interaction with RNAPII.
Next, we investigated the effects of PHF8 knockdown on gene transcription. A specific shRNA (PHF8-shA) was able to significantly decrease the PHF8 level in HeLa cells 3 days posttransfection compared to a nontargeting control (NT-sh), as shown on the RNA level by quantitative RT-PCR (Fig. (Fig.6A)6A) and on the protein level by immunoblotting (Fig. (Fig.6B).6B). Total RNAs from three independent knockdown experiments were compared in two-color microarray dye swaps; 960 probes were more than 2-fold downregulated, while 384 had over 2-fold-increased levels after knockdown (Fig. (Fig.6C).6C). The down- and upregulated probes corresponded to 780 and 270 annotated transcripts, respectively (see Table S3 in the supplemental material). The observation that almost three times more transcripts were down- than upregulated is in line with a coactivator role of the knockdown target. However, knockdown did not result in a global defect compromising transcription of all PHF8-occupied genes. Furthermore, the up- and downregulated genes showed similar percentages of PHF8 occupancy (about 64%), similar levels of PHF8 reads at their promoters (Fig. (Fig.6D),6D), and comparable transcript levels in untransfected HeLa cells (Fig. (Fig.6E).6E). Functional-annotation analysis of downregulated transcripts using DAVID (12) revealed enrichment for pleckstrin homology domain-containing proteins and proteins that are involved in the focal-adhesion pathway (see Table S3 in the supplemental material). Taken together, these results indicate that PHF8 does not play a role in basal transcriptional regulation, as the number of genes displaying transcriptional responsiveness to PHF8 depletion is only a fraction (about 5%) of those occupied by PHF8. It is more likely that PHF8 is required for the fine tuning of the transcriptional output, which is governed by the signaling pathways that regulate the individual PHF8-responsive genes. This scenario is consistent with our coactivation function for PHF8, indicating that it could further enhance the activities of diverse transcriptional activators.
Next, we analyzed the genes that displayed downregulation following PHF8 knockdown and that have been reported to be linked to mental retardation. These comprise 11 candidate genes, i.e., MAOA, DPSYL2, CLN5, AP1S2, SMC1A, IDS, SLC9A6, CUL4B, OPHN1, OCRL, and AMMECR1. Using quantitative RT-PCR, we could validate the downregulation of the candidate genes (see Fig. S7A in the supplemental material). While the less effective short hairpin (PHF8-shB) also resulted in downregulation of these genes, the extent of this effect was smaller than that seen for the more effective PHF8-shA (see Fig. S7A in the supplemental material). Moreover, while we observed a reduction of PHF8 occupancy at the candidate gene promoters upon PHF8 knockdown, we could not see a consistent change in H3K9me2 or RNAPII levels (see Fig. S7B in the supplemental material). Since the family member KIAA1718 (also called JHDM1D or KDM7A), which is able to demethylate H3K9me2 and H3K27me2 (19, 20, 49), was induced 2.1-fold upon PHF8 knockdown, it might be compensating for the loss of PHF8.
The key findings of this study are as follows. First, PHF8 was delineated as a novel histone demethylase of the JmjC family specific for dimethyl H3K9 and, to a lesser extent, for the monomethyl state. Second, the PHD of PHF8 was identified as an important determinant in recognition of trimethyl H3K4. Third, a critical function for the PHD in histone demethylation and transcriptional coactivation was revealed. Fourth, it was shown that a single point mutant of PHF8 associated with X-linked mental retardation in humans is defective for histone demethylation and transcriptional coactivation. Fifth, by genome-wide location analysis of PHF8, a pattern of promoter occupancy similar to that of trimethyl H3K4 was revealed. Finally, a direct association between PHF8 and the CTD of the large subunit of RNA polymerase II was demonstrated.
We showed that PHF8 is an active H3K9me2 demethylase similar to the JHDM2/KDM3 family (54). Loenarz et al. recently additionally reported in vitro demethylation of H3K27me2 and H3K36me2 using a construct lacking the PHD and the C-terminal domain of PHF8 (31). However, in our study, full-length PHF8 was unable to demethylate H3K27me2 or H3K36me2 and was quite specific for demethylation of di- and monomethyl H3K9. Moreover, overexpression of PHF8(1-489) did not result in a decrease of any of the investigated histone modifications but H3K9me2 and H3K9me1 (see Table S1 in the supplemental material). Very recently, Feng et al. also reported that PHF8(1-690) has only marginal activity on H3K27- or H3K36-methylated peptides (15). H3K9me2 is an important repressive chromatin modification whose removal signals for coactivation of gene expression. Indeed the importance of the murine H3K9me2 demethylase Jhdm2a for proper activation of target genes by removing this repressive mark at the respective promoters has already been demonstrated (34, 47). This demethylation may result in the release of repressive factors, or it may facilitate the recruitment of transcription factors and coactivators.
We found that the PHD of PHF8 plays a critical role in its demethylation activity in vivo, as well as its transcriptional coactivation function. Our results indicate that the PHD of PHF8 serves as a specific reader of H3K4 tri- and dimethyl modifications and may act as an important determinant in anchoring PHF8 at transcription start sites of PHF8-occupied promoters. Moreover, PHF8 occupies a wide spectrum of H3K4-trimethylated promoters, suggesting that it might contribute to the activating effect of this histone modification on transcription. This contention is further supported by its ability to function as a coactivator for a number of transcriptional activators that we have examined, including Ash2, p53, c-myc, and E2F. These activators have been linked to recruitment of the MLL histone methyltransferase complex to mediate H3K4 methylation at promoters (32, 50). Therefore, we propose a model by which increased H3K4me3 levels, through the action of an MLL-related family of enzymes, would lead to further recruitment of PHF8 to the promoter of responsive genes. Once at the promoter, PHF8, through its association with RNAPII, may stabilize the preinitiation complex formation, leading to enhanced transcription (Fig. (Fig.77).
We found that the disease-causing mutant of PHF8 (F279S) displays aberrant cellular localization and is devoid of demethylation activity. Furthermore, this mutant exhibited reduced activity in our coactivation assays. While we did not observe a general defect in transcription or increased H3K9me2 levels upon knockdown of PHF8, a specific transcriptional defect in neuronal cells may underlie the disease phenotype of this mutant PHF8. Indeed, XLMR patients display specific defects in the development of neurons and the midline, arguing against a global role for PHF8 in transcription. Possible reasons for this lack of global effects on methylation levels or transcription following PHF8 mutations could be that in most tissues, the loss of PHF8 is compensated for by redundant proteins like JHDM2, PHF2, or KIAA1718, the last of which is induced by PHF8 knockdown, or that the activity of endogenous PHF8 is strictly regulated by signaling pathways and switched on in a tissue-, time-, and/or locus-specific manner.
Our data suggested that the interaction with H3K4me3 was crucial for PHF8 function, since mutation of aromatic-cage residues in the PHD, as well as removal of H3K4me3 by JARID1A, significantly reduced demethylation and coactivation. It has been shown that activation of neuron-specific genes occurs by recruitment of the H3K4 methyltransferase MLL1 during in vitro differentiation of P19 cells (52). On the other hand, deletions in the H3K9 methyltransferase EHMT1 cause the 9q34.3 subtelomeric deletion syndrome, one feature of which is mental retardation (24). Additionally, mutations in other PHD protein-coding genes, like ATRX and PHF6, are implicated in X-linked neurological disorders (3). Several H3K4me3 binders, like the inhibitors of growth protein family (ING1 to -5), the Taf3 subunit of TFIID, or the NURF subunit BPTF, have been demonstrated to be involved in gene transcription and chromatin remodeling (42, 51, 53). These proteins display different affinities for the H3K4me3 mark, and their recruitment to promoters could be stimulated via interaction with other transcription factors, DNA, or chromatin. Another JmjC protein, SMCX/KDM5C, is also implicated in XLMR, but in contrast to PHF8, one of its two PHDs binds H3K9me3. SMCX has been shown to act as a demethylase specific for H3K4me3 and a transcriptional repressor (22, 46). This argues for the importance of a balanced H3K4 and H3K9 methylation and readout in neuronal development and brain function (2). Writers, erasers, or readers exhibit synergistic or opposing roles by interaction at or competition for genomic binding sites in order to produce the desired transcriptional outcome. Disturbances in the fine tuning of this delicate equilibrium brought about by mutations in genes like PHF8 can result in diseases such as hereditary disorders or cancer.
We thank Kristian Helin, Min Gyu Lee, and Clément Carré for providing reagents and the Ultrasequencing and Microarrays Units at the CRG for performing Solexa and microarray analyses. We are particularly grateful to Marc Vigneron (ESBS, Strasbourg, France) for providing unpublished RNAPII antibodies and the GST-CTD construct. We thank Harm-Jan Vos for help in peptide synthesis, Hetty van Teeffelen for technical assistance, and Michiel Vermeulen for sharing unpublished results.
K.F. received an Erwin-Schrödinger fellowship from the Austrian Science Fund FWF (J2728-B12). P.D.G., N.S.O., and H.T.M.T. were supported by grants from the Netherlands Organization for Scientific Research (NWO-CW TOP 700.57.302), the European Union (EUTRACC LSHG-CT-2006-037445), and the Netherlands Proteomics Center. R.S. was supported by a grant from the NIH (CA090758).
We declare that we have no competing financial interest.
Published ahead of print on 26 April 2010.
†Supplemental material for this article can be found at http://mcb.asm.org/.