We first carried out a comprehensive mapping study to determine the distribution of histone modifications across c-fos
in control, EGF-stimulated () or anisomycin-stimulated (Supplementary Figure S1
) mouse fibroblasts. A modified ChIP protocol, using MNase instead of sonication to generate predominantly mononucleosomal chromatin (MacDonald et al, 2005
) was used to produce high-resolution quantitative comparative maps of the distribution and dynamics of H3K4 trimethylation (K4me3), K36 trimethylation (K36me3), K79 dimethylation (K79me2) and K9 acetylation (K9ac) across these genes ( and Supplementary Figure S1
). Each antibody was first titrated in ChIP to arrive at concentrations that recovered virtually all available epitopes, leaving little or none in the unbound fraction as monitored by western blotting. Peaks of distribution of each modification across these genes are shown graphically ( and Supplementary Figure S1B
) and as bar charts with error bars ( and Supplementary Figure S1C
). This yielded highly reproducible maps with striking differences in the amounts of each modification recovered. Their interpretation is extremely complex because of the continuous dynamic turnover of acetylation at these genes, the fact that methylation may occur transiently with the traverse of RNA pol II, and the phenomenon of microheterogeneity of histone modifications at any single position. Evaluating these maps in overview, common trends applying to both c-fos
emerge. First, K9ac (, panel i) and K4me3 (panel ii) have remarkably similar overlapping distributions across the start site of both genes. For both modifications, there is a striking dip precisely at the start site of both genes, and the positions of peaks of modification largely coincide. The stability of this colocalisation at all points analysed is consistent with sequential ChIP assays showing these modifications coexist on the same nucleosomes at these positions (Hazzalin and Mahadevan, 2005
). The sharp dip in K4me3 and K9ac at the start site on both genes appears to be due to this position lacking a nucleosome and being more accessible, as indicated by MNase sensitivity maps across these genes ( and Supplementary Figure S2
). These show that the start sites of c-fos
are considerably more sensitive to MNase digestion with <5% (c-fos
−57) and <10% (c-jun
−396) of the corresponding genomic DNA signal remaining after digestion ( and Supplementary Figure S2
Figure 1 EGF-stimulated distribution of acetylated and methylated histone H3 across c-fos and c-jun. (A) Schematic representation of c-fos and c-jun showing positions analysed by real-time PCR. Primer positions indicate the 5′ position of the forward primer (more ...)
Figure 1 (C) Data shown for c-fos and c-jun are the same as in (B) but plotted in bar chart format with error bars representing the standard deviation (s.d.) of the mean of the two independent experiments. For each ChIP, primers spanning two regions of the inactive (more ...)
Figure 2 MNase sensitivity across c-fos and c-jun in quiescent and EGF-stimulated mouse cells. Quiescent cells were unstimulated (control) or stimulated with EGF (50 ng/ml) for 15, 30, 60 or 120 min and formaldehyde crosslinked mononucleosomes were prepared. Equivalent (more ...)
In contrast, K36me3 and K79me2 are found exclusively within the coding regions of c-fos and c-jun (, panels iii and iv). Although they do overlap, their precise distributions appear distinct. K36me3 peaks towards the 3′ end of the coding regions but drops off before the gene's terminus (, panel iii), whereas K79me2 peaks over the 5′ end of the coding region before declining towards the 3′ end (, panel iv). The presence of both these modifications in unstimulated control cells may be attributed either to their being deposited here in previous rounds of transcription prior to quiescence, or to their being part of a transcription-independent pattern of pre-existing, possibly epigenetic, mechanism; there is no evidence either way at present.
Finally, as a negative control, we monitored all four modifications at a site beyond the end of each gene (; c-fos
+5000) and found none of these modifications were present there. Stimulation with EGF leads to rapid increases in H3K9ac on c-fos
across promoter and coding region, peaking at 15 min over the promoter and 30 min over the coding region (, panel i left-side graph). These increases are transient and decrease substantially by 60 min (, panel i), correlating well with kinetics of c-fos
transcription (Thomson et al, 1999
). For c-jun
, basal level of K9ac is much higher than for c-fos
, and smaller increases are seen upon EGF stimulation, peaking at 60 min (, panel i, right-side graph). The elevated basal level of K9ac on c-jun
may be due to a low background level of c-jun
transcription in control cells (Hazzalin and Mahadevan, 2005
). K4me3 and K36me3 also increase within the coding regions on c-fos
upon stimulation (, panels ii and iii). Compared to K9ac, these increases are slower and more sustained, remaining after 2 h of stimulation (, panels ii and iii). Increases in K4me3 on c-jun
are much smaller than on c-fos
(, panel ii).
The entire mapping study was repeated using a different stimulus, anisomycin, which activates these genes predominantly via stress-activated protein kinases (SAPKs) as opposed to the ERK MAP kinase pathway, which EGF utilises (reviewed in Hazzalin and Mahadevan, 2002
). This yielded extremely similar MNase sensitivity maps and patterns of distribution and quantitative levels of these modifications across each gene (Supplementary Figures S1
), testifying to the high reproducibility of the methodology used here.
Finally, we address a concern that activation of these genes may produce marked changes of nucleosomes occupancy across the gene, thereby skewing the ChIP recovery data provided at various time points after induction. First, there is no indication of a great change in MNase sensitivity at various time points after stimulation ( and Supplementary Figure S2
), suggesting that average nucleosomal occupancy at any position did not vary greatly upon gene activation. Second, we carried out ChIP using anti-H3 antibodies directed to the unmodified C terminus across these genes (Supplementary Figure S3A and B
, panel i), and this too showed no major changes in recovery upon gene activation, confirming indications from MNase sensitivity maps. As expected, there was good correspondence between the relative H3 occupancy and the MNase sensitivity maps at all points compared across the genes (Supplementary Figure S3A and B
, panels ii and iii; see for example, the dip across the start site), indicating that MNase sensitivity seen here is a good approximation of average nucleosomal occupancy.
EGF-induced increases in H3K4me3 and H3K36me3 within c-fos and c-jun coding regions are dependent on transcription elongation
As shown in yeast (reviewed in Gerber and Shilatifard, 2003
; Hampsey and Reinberg, 2003
; Sims et al, 2004
), stimulation-dependent increases in K4me3 and K36me3 across coding regions suggest a link with elongating Pol II, a possibility we tested with the transcriptional inhibitor DRB. Quantitative reverse transcription PCR (qRT–PCR) shows that DRB ablates induction of c-fos
(). This correlates with loss of Pol II across the coding region of both genes under these conditions, shown by ChIPs using anti-Pol II antibodies raised against its N terminus (). Similar results were obtained at several different positions within the coding regions of c-fos
(Supplementary Figure S4A
), except for +444 of c-fos
, which showed increased Pol II loading in response to DRB at later time points. This may be explained by the transcriptional pause site reported at this region (Mechti et al, 1991
; Coulon et al, 1999
Figure 3 EGF-induced increases in H3K4me3 and H3K36me3 within c-fos and c-jun are transcription-dependent. (A) Quiescent cells were pretreated (10 min) with DRB (25 μg/ml) or untreated (con) prior to stimulation with EGF (50 ng/ml) for 15, 30 or 60 min (more ...)
EGF-induced increases in K4me3 and K36me3 within the coding regions of both genes were abolished by DRB pretreatment (, left and middle panels). Other coding region positions within c-fos
gave similar results (Supplementary Figure S4B and C
). As expected, in the negative control hbb
coding region, no significant change in either modification was observed in response to EGF or DRB (, right panel). Thus, EGF-induced increases in K4me3 and K36me3 within coding regions of these genes are dependent on transcription elongation.
Setd2 is non-redundantly responsible for H3K36 trimethylation in mouse fibroblasts
In Saccharomyces cerevisiae
, Schizosaccharomyces pombe
and Neurospora crassa
, Set2 is an H3K36 methyltransferase responsible for mono-, di- and trimethylation (Strahl et al, 2002
; Adhvaryu et al, 2005
; Morris et al, 2005
; Chu et al, 2006
). In mammals, several candidates have been proposed. The first discovered, NSD1, has in vitro
HMTase activity towards H3K36 and H4K20 (Rayasam et al, 2003
). Second, a SET domain protein that interacts with the Huntington's Disease gene (HD
) product, HYPB/SETD2, also methylates H3K36 in vitro
and, via its C-terminal Pol II interaction domain, can bind to phosphorylated Pol II (Sun et al, 2005
). Finally, Smyd2 has HMTase activity towards H3K36 in vitro
, and interacts with Sin3 HDAC complexes (Brown et al, 2006
). None of these studies addressed their ability to methylate H3K36 in vivo
, or the specific states (mono, di or tri) of methylation for which they might be responsible. Dimethyl-H3K36 antibodies were used in the NSD1 and Smyd2 studies, but capacity for mono- or trimethylation was not analysed.
Using NCBI protein BLAST search, we identified mouse homologues of yeast Set2 (data not shown). Setd2 (the mouse homologue of human HYPB/SETD2) and NSD1 (among others) had significant homology. In view of this homology, and the fact that HYPB/SETD2 has a Pol II interaction domain (Sun et al, 2005
), we investigated Setd2 and NSD1 as possible H3K36 methyltransferases in mouse fibroblasts.
Two different siRNAs targeting Setd2
transcripts knocked down Setd2
mRNA levels by >60% (Setd2 0) and >80% (Setd2 3) (). siRNA targeting NSD1
transcripts knocked down NSD1
mRNA levels by >80% (). These were specific, as mock-transfected cells, or non-targeting siRNA, did not show similar effects (). Additionally, Setd2- or NSD1-specific siRNAs did not knock down mRNAs of other genes tested ( and data not shown). We raised and characterised N- and C-terminal Setd2 antibodies to confirm that Setd2 protein levels were specifically diminished as a result of its mRNA knockdown (, panels i and ii; Supplementary Figure S5
). We were not able to assess NSD1 protein levels, as a good antibody is currently unavailable. This became less relevant since the Setd2 knockdown produced a clear H3K36 trimethylation defect.
Figure 4 Setd2 is a non-redundant H3K36-specific trimethyltransferase. (A) Cells were untransfected (−), mock transfected (no siRNA, mock) or transfected with Setd2 (0 and 3, two different siRNAs), NSD1 or non-targeting (non-t) siRNAs. Total mRNA was isolated (more ...)
Total cellular K36me3 was virtually absent in the Setd2 knockdowns (, panel i), whereas K36me2 and me1 levels remain unaffected (, panels ii and iii). In contrast, NSD1 knockdown had no effect on mono-, di- or trimethylation of K36 (, panels i–iii). Additionally, total levels of K4me3, K9me3 or K79me2 were unchanged in either Setd2 or NSD1 knockdowns (, panels iv–vi). Therefore, Setd2 appears to be a non-redundant methyltransferase specific for trimethylation of H3K36. This is the first demonstration in intact cultured cells of a H3K36-specific methyltransferase whose knockdown results in abrogation of virtually all trimethylation, but not mono- or dimethylation. It is not possible, however, to distinguish from these experiments if HYPB/Setd2 can catalyse all states of modification culminating in K36 trimethylation, or just the final conversion of dimethylated K36 to the trimethylated state. Since there are several recent reports of a link between H3K36 methylation and histone deacetylation in yeast (Carrozza et al, 2005
; Joshi and Struhl, 2005
; Keogh et al, 2005
), we also analysed global levels of various H3 and H4 acetyl modifications in Setd2 knockdown cells. No significant reproducible changes were observed for any acetyl modification (, panels ix–xv).
Setd2 is responsible for H3K36 trimethylation at IE and housekeeping genes
Next, we used ChIP to analyse all states of K36 methylation at c-fos- and c-jun coding regions (mid and 3′ end) after Setd2 knockdown. Four housekeeping genes were also analysed; glyceraldehyde-3-phosphate-dehydrogenase (gapdh), RNA-polymerase III subunit b (polr3b), glutaminyl-tRNA synthetase (glnrs) and cyclophilin b (cycb) (shown schematically in ).
Figure 5 Setd2 is responsible for H3K36 trimethylation at IE and housekeeping genes. (A) Schematic representation of the gapdh, polr3b, glnrs and cycb housekeeping genes showing regions amplified by primers used for real-time PCR. Primer positions shown indicate (more ...)
Setd2 knockdown produced almost complete loss of total cellular K36me3 in these cells () and a marked reduction of K36me3 at c-fos- and c-jun coding regions (, panel i). Stimulation with EGF for 60 min resulted in a 2.7- to 3.4-fold increase in the K36me3 observed at these regions (black bars panel i), whereas in the Setd2 knockdown, induced levels of K36me3 were much smaller—1.3- to 1.8-fold (middle bars panel i). K36me3 was also decreased by >50% within the coding regions of gapdh, cycb, polr3b and glnrs after Setd2 knockdown (, panel i). Setd2 therefore appears to be responsible for both basal and inducible K36me3 at IE gene coding regions, and K36me3 levels at constitutively active genes.
Figure 5 (C) Aliquots of formaldehyde crosslinked mononucleosomes prepared as in (B) were used in ChIP assays with H3K36me3- (i), H3K36me2- (ii) and H3K36me1- (iii) specific antibodies. Recovery of c-fos, c-jun and hbb coding region sequences were quantified by (more ...)
Mono- and dimethyl K36
For all points tested, the recovery of nucleosomes with H3 bearing mono- and dimethyl K36 was very much lower (>10-fold) than that of trimethyl K36 (). In untransfected cells, there is no increase in mono- and dimethylated K36 upon activation of c-fos and c-jun (, panels ii and iii), indicating no correlation between these modifications and transcription. In Setd2 knockdown cells, there is no decrease in these low levels of K36me2 or me1 within the coding regions of c-fos and c-jun, with the exception of the c-jun +1651 region where there is a small decrease in K36me2 (, panel ii). Thus, in general Setd2 does not appear to be responsible for the low levels of mono- or dimethyl K36 at these regions.
As described for c-fos and c-jun, recovery of K36me1 and me2 was also very much lower than that of K36me3 at coding regions of gapdh, cycb, polr3b and glnrs (). In fact, these levels are very similar to that seen at the inactive hbb coding region (, panels ii and iii bottom graphs). Knockdown of Setd2 resulted in slightly higher levels of K36me2 and K36me1 at these positions (, panels ii and iii) but recoveries remain relatively very low. Thus, K36me2 and K36me1 levels do not seem to correlate with active transcription as K36me3 does, and Setd2 does not appear to be responsible for these modifications. By their nature, knockdowns do not produce the complete loss of any enzyme and low residual activity of the remaining HYPB/Setd2 enzyme may explain the incomplete loss of H3K36 sometimes seen.
Loss of H3K36me3 does not affect the kinetics of IE gene transcription, intragenic transcription or expression of constitutively active genes
To investigate the potential role of K36me3 in transcriptional regulation, the effect of K36me3 loss on the transcription of IE and constitutively active genes was assessed by qRT–PCR (). As shown previously, knock down of Setd2 mRNA () led to virtually complete loss of all detectable K36me3 (). The gene induction kinetics of c-fos
and two other IE genes, core promoter element binding protein
and MAP kinase phosphatase 1
, was examined (). A recent report in yeast has shown that increased K36me leads to delayed induction of the HIS4
gene, whereas elimination of K36me accelerates HIS4
induction (Nelson et al, 2006
). However, loss of K36me3 did not affect the kinetics of induction of any gene tested here (). Second, the steady-state mRNA levels of four constitutively active genes (cycb
) were assessed in the absence of K36me3, and no significant changes were observed for any transcript (). Furthermore, we did not observe any intragenic transcription from c-fos
in K36me3-depleted cells (). This is in contrast to the observations in yeast where K36me appears to create a less permissive chromatin structure throughout yeast gene coding regions by recruiting HDAC complexes (Rpd3) via interaction with the chromodomain of Eaf3 (Carrozza et al, 2005
; Joshi and Struhl, 2005
; Keogh et al, 2005
). Loss of K36me leads to increased acetylation, K4me3 and Pol II loading within specific gene coding regions and intragenic transcript initiation from the FLO8
genes (Carrozza et al, 2005
; Joshi and Struhl, 2005
; Keogh et al, 2005
; Kizer et al, 2005
). Similar to the lack of any intragenic transcription in K36me3-depleted mouse fibroblasts (), we do not detect any increased H3 or H4 acetylation at 3′ regions of IE or housekeeping genes (Supplementary Figure S6
), nor any change in Pol II occupancy at these genes (Supplementary Figure S7
). These studies do not reveal a clear parallel at c-fos
in mammalian cells for the relationship between H3K36 methylation and histone deacetylation previously seen in yeast.
Figure 6 Effects of H3K36me3 knockdown on gene expression levels: lack of intragenic transcription from IE genes or gapdh. (A) C3H 10T½ cells were untransfected (−), mock transfected (no siRNA, mock) or transfected with Setd2, non-targeting (non-t) (more ...)
Figure 6 (D) Cells were transfected and stimulated as in (A) and cellular protein was isolated from the same TRIzol® reagent preparations used for RNA extraction. Protein was separated by 15% SDS–PAGE and transferred to PVDF membrane for (more ...)