Is H3K4me3 instructive for transcription activation?

Tri‐methylation of lysine 4 on histone H3 (H3K4me3) is a near‐universal chromatin modification at the transcription start site of active genes in eukaryotes from yeast to man and its levels reflect the amount of transcription. Because of this association, H3K4me3 is often described as an ‘activating’ histone modification and assumed to have an instructive role in the transcription of genes, but the field is lacking a conserved mechanism to support this view. The overwhelming finding from genome‐wide studies is that actually very little transcription changes upon removal of most H3K4me3 under steady‐state or dynamically changing conditions, including at mammalian CpG island promoters. Instead, rather than a major role in instructing transcription, time‐resolved experiments provide more evidence supporting the deposition of H3K4me3 into chromatin as a result of transcription, influencing processes such as memory of previous states, transcriptional consistency between cells in a population and transcription termination.


Introduction
The most highly cited paper, with >1,000 citations, under the search term 'K4' or 'Set1' is that from the Kouzarides group [1], with the title 'Active genes are tri-methylated at K4 of histone H3'. This was one of the first papers to propose a link between the level of H3K4me3 on a gene and the amount of transcript produced. Indeed, H3K4me3 is commonly referred to as an 'activating' histone modification. But whilst there is clearly a strong association between H3K4me3, active genes and RNA levels, for nearly 15 years many groups have searched for causality in this relationship and failed to find any overarching mechanism for how H3K4me3 is instructive for transcription activation. Here we aim to re-evaluate the evidence for whether H3K4me3 actually regulates transcription initiation and whether the failure to find an overarching mechanism reflects distinct functions in different organisms.

H3K4 is methylated by Set1 or Set1-like methyltransferase complexes
In eukaryotes, DNA is packaged into chromatin, with 147 bp DNA wrapped around a core of eight histone proteins to form a nucleosome [2]. The tails of these histone proteins protrude from the nucleosome core and are extensively posttranslationally modified. Histone H3 lysine 4 (H3K4) can be modified by the addition of one, two, or three methyl groups and it is likely that the levels and distribution of H3K4 methylation will reflect the combined action of methyltransferases and demethylases [3][4][5]. The enzymes that methylate H3K4 are highly conserved between yeast, worms, plants, flies, and humans [6]. They each contain a SET methyltransferase domain and are present in a complex with accessory proteins that are both required for lysine methyltransferase (KMT) activity and for influencing the state (mono-, di-, or tri-) of methylation deposited. In the yeast Saccharomyces cerevisiae, H3K4 is methylated solely by the lysine methyltransferase Set1/KMT2 in a complex containing seven accessory subunits: Swd1, Swd3, Swd2, Sdc1, Bre2, Spp1 and Shg1 (Fig. 1) [6]. Drosophila melanogaster has three Set1 homologues (dSet1, Trithorax and Trithorax-related) whereas mammals have six complexes that catalyse H3K4 methylation (SET1A/B and MLL1-4) (Fig. 1) [6]. These complexes are not redundant, as demonstrated by the embryonic lethality caused by deletion of individual MLL genes [7][8][9]. The Drosophila dSet1 complex and the two mammalian SET1 complexes are most similar to the yeast complex ( Fig. 1) [6] and are responsible for the bulk of H3K4me3 [10][11][12]. The Trithorax, Trithorax-related and MLL complexes are Set1 complex-like and consist of similar core components (Fig. 1), although the structural and functional organisation of these subunits differs from the yeast complex [13,14]. Additionally, these Set1-like complexes appear to have more specific functions: MLL1 and MLL2 methylate H3K4 at different subsets of homeotic genes [8,15] whereas MLL3/4 are linked to H3K4me1 and enhancer function [16]. Set1-related complexes have been implicated in cancer in mammals [17], ageing in C. elegans [18], embryonic development [19,20] and stem cell biology [21][22][23].

H3K4me3 peaks at transcription start sites
Using chromatin immunoprecipitation (ChIP), the distribution of H3K4 methylation has been mapped in a range of organisms [1,4,[24][25][26][27][28][29][30][31][32][33][34][35]. Whilst H3K4 methylation is predominantly associated with genes, leading to a strong correlation with transcription (Table 1), H3K4me3 also functions in DNA repair [36,37] and targeting of double strand breaks during meiotic recombination [38][39][40]. Despite the differences between yeast and metazoan gene length and structure, the most striking feature of the H3K4me3 gene distribution, conserved in yeast, plants, worms, flies and mammals, is the peak around the transcription start sites (TSSs) of genes [1,4,[24][25][26][27][28][29][30][31][32][33][34][35] (Fig. 2A). Moreover, closely spaced divergent Figure 1. The Set1-related complexes in yeast, Drosophila and humans. There is one Set1 complex in yeast but three in Drosophila and six in mammals. Homologous subunits are the same shapes and similarly positioned relative to the central methyltransferase protein. Subunits that are required for H3K4me3, H3K4me2/3 or all states of H3K4 methylation are shown in yellow, orange and red, respectively. D. melanogaster and mammals have complex-specific subunits and these are shown in blue. promoters, whether for coding or non-coding transcription, have an additional upstream peak of H3K4me3 separated by a nucleosome-depleted region (NDR) [41,42]. Interestingly, the promoters of antisense transcription at the 3 0 ends of yeast and human genes are also marked with a peak of H3K4me3, indicating that this modification is not specific to sense TSSs (Fig. 2B) [43,44]. As expected, the amount of H3K4me3 on both yeast and mammalian genes is related to the levels of nascent sense transcription, with highly expressed genes having more H3K4me3 (Fig. 2, Table 1). Unlike initially reported [4], both H3K4me3 peak height and peak breadth correlate positively with expression in a range of organisms [35,45]. Broad H3K4me3 peaks are also an epigenetic signature of tumour suppressor genes and are associated with increased transcription elongation and enhancer activity [35]. Furthermore, tumour cells have narrower H3K4me3 peaks and a lower level of tumour suppressor transcripts. The distributions of H3K4me2 and H3K4me1 are distinct from H3K4me3 ( Fig. 2A) and the differing correlations with transcription (Table 1) suggest that there are methylation state-specific functions.
The difficulty of studying the functions associated with different H3K4 methylation states The study of histone modifications in vivo is complicated by the fact that it is difficult to genetically affect one modification in isolation without being at risk of indirect effects caused by changing the levels of other marks. For example, deletion of yeast SET1 removes all three states of H3K4 methylation, giving potentially conflated results if each state has different functions. Further conflation comes from the fact that Set1, like many histone methyltransferases, has non-histone substrates, such as the kinetochore protein, Dam1 [46]. Similarly, in Arabidopsis thaliana, the H3K4 methyltransferase ATX1 has effects on transcription that are independent of its catalytic activity towards H3K4 [47]. A commonly used approach in yeast and Dictyostelium discoides is to substitute H3K4 with a non-methylatable residue such as alanine [48,49]. However, this again is not methylation state-specific and also removes H3K4 acetylation [33], which would result in effects that could not be attributed solely to H3K4 methylation. Furthermore, this approach is difficult in other organisms due to the high repeat number for histone genes. An alternative strategy to specifically alter H3K4me3 levels is to interfere with the integrity of the methyltransferase complexes.
Spp1 and CFP1 are required for the majority of H3K4me3 in yeast and mammals Both yeast Set1 and human SET1A/B complexes have subunits, Spp1 and CFP1 respectively, that when ablated specifically reduce H3K4me3 with little effect on global levels of H3K4me1 or me2 [50][51][52]. In yeast, reduced H3K4me3, by deletion of SPP1, is an important trigger for cell death Figure 2. Distribution of H3K4 methylation on genes. A: The distribution of H3K4 methylation on yeast genes (left) and mammalian (HeLa) genes (right) 3000 bp upstream or downstream of the sense transcription start site (TSS, dashed line) for highly expressed genes (dark blue) or poorly expressed genes (light blue). For yeast, the top (n ¼ 1155) or bottom (n ¼ 1163) quintiles of genes in terms of NET-seq reads [70] are used to map the distribution of histone H3normalised H3K4me1, 2 and 3 [29]. For HeLa cells, only genes with a strong 5 0 CAGE signal and thus a well-defined TSS were used for the analysis, and the top (n ¼ 1864) and bottom (n ¼ 1864) quintiles of these genes in terms of mNET-seq reads [139] are used to map the distribution of H3K4me1, 2 and 3 [34]. B: As (A) except H3K4me3 distribution on all genes with a defined antisense TSS, aligned to the antisense TSS (yeast n ¼ 1529, HeLa n ¼ 2468). (apoptosis) and is linked to poor chronological ageing [53] while absence of CFP1 causes peri-implantation embryonic lethality in mice [54], although murine embryonic stem cells (mESCs) are viable but unable to differentiate [55].
CFP1 is recruited to the $70% of mammalian promoters that contain a CpG island [52,56], regions of high CpG density that are not subject to DNA methylation, via a Zn finger CXXC domain that binds to unmethylated CpG dinucleotides [57]. CFP1 can also be recruited to chromatin via its PHD domain that can bind all three H3K4 methylation states [58]. CFP1 is not in the MLL complexes, but its binding functions may be recapitulated by the CXXC domains in MLL1/2 [59,60] and the third PHD domain in MLL1 [61]. Both SET1A/B and MLL2 complexes act on CpG island promoters, with SET1A/B methylating H3K4 on more highly expressed genes [62] whereas MLL2 preferentially methylates H3K4 at more lowly expressed bivalent gene promoters in embryonic stem cells (ESCs) [63,64].
Yeast Spp1, like CFP1, has a PHD finger that recognises H3K4me2 and H3K4me3 [65], suggesting that Spp1, through its PHD domain, may bind H3K4me2 and stabilise the Set1 complex on chromatin to bring about H3K4me3 [29]. Another possibility is that, by binding H3K4me2/3 on one histone tail, Spp1 can recruit Set1 to methylate the other histone H3 tail in the same nucleosome or neighbouring nucleosomes. It is common for components of histone modifying enzyme complexes to play a role in the deposition of a modification while also containing a domain that binds to that modification. Additionally, it was speculated that the binding of Spp1 to H3K4me2/3 may protect these modifications from demethylation [29].
Although Spp1 directly interacts with the Set1 N-SET catalytic domain [66,67], in vitro studies demonstrate that Spp1 is not required for tri-methylation of H3K4 on free histone H3, arguing against a strict enzymatic requirement of Spp1 for Set1-mediated H3K4me3 [66,67]. In agreement, in vivo, $20% global H3K4me3 remains after SPP1 deletion [3,40]. Since this residual amount is unevenly distributed between individual genes, with H3K4me3 on some genes such as the ribosomal protein genes being largely independent of Spp1 [3], it is possible to restrict analysis to the class of genes with H3K4me3, the most affected by SPP1 deletion. Similarly, whilst CFP1 only recruits SET1A/B complexes to CpG island promoters, analysis could be limited to these genes when ascertaining any effect of H3K4me3.

Loss of H3K4me3 has minor effects on global transcription
Deletion of CFP1 from mESCs causes loss of most H3K4me3 from expressed CpG island-associated genes but yet there are minimal changes in transcript levels at these genes, assessed by microarray, or nascent transcription as measured by levels of RNA polymerase II by ChIP and genomic run-on sequencing [62]. Furthermore, only a minority of the DNA damage-inducible genes that gain H3K4me3 in a CFP1dependent manner showed defects in transcript levels in cfp1 À/À mESCs [68]. However, it is plausible that the basal level of H3K4me3 remaining in cfp1 À/À cells is sufficient for transcription activation. Interestingly, small defects in repression upon DNA damage were also observed at some genes in cfp1 À/À mESCs, associating H3K4me3 with both genespecific activation and repression. Additionally, levels of both divergent upstream transcripts and antisense transcripts arising from the 3 0 regions of the DNA damage-inducible genes are significantly increased in the absence of CFP1 [68]. These data support a role for CFP1, and possibly H3K4me3, in repressing divergent and convergent antisense transcripts at the 755 genes induced by DNA damage.
Similar to mammalian systems, deletion of SPP1 from yeast, resulting in gene-specific H3K4me3 reduction [3], has virtually no effect on steady-state or dynamically changing mRNAs [43,69] or transcription (Table 1 and Fig. 3A, unpublished) assessed using native elongating transcript analysis, NET-seq [70]. We find only 57 (1.0%) genes are significantly (p < 0.05) upregulated and 23 (0.4%) genes are significantly downregulated at the level of transcription upon deletion of SPP1 at steady-state growth. To ask whether the genes that lose the most H3K4me3 are those whose transcription changes the most, we made use of a ChIP-seq dataset measuring H3K4me3 in a strain in which levels of H3K4me3 are affected similarly to those in the spp1D strain [3]. We found no correlation (r s ¼ 0.006) between the change in H3K4me3 and the change in transcription upon deletion of SPP1 at the þ1 nucleosome where H3K4me3 is most abundant. Moreover, when we studied a dynamic system during a change from glucose-containing to galactose-containing growth medium, a carbon source shift shown to alter the transcription of $20% of yeast genes [71], in the spp1D strain fewer than 15 genes are significantly (p < 0.05) differentially transcribed (unpublished data). Of note, the induction profiles of the GAL genes that are strongly upregulated upon shift to galactose-containing medium, are not altered in the spp1D strain. Yeast H3K4me3 levels can also be controlled by the JmJC family demethylase Jhd2 [72][73][74][75], which is capable of removing all three H3K4 methylation states in vivo [76]. Analysis of datasets from the Holstege [69] and Yan labs (accession number GSE67211; unpublished) shows that genes with increased H3K4me3 upon deletion of JHD2 do not have higher levels of mRNA ( Fig. 3B and C).

Context-dependent effects of H3K4me3 on gene expression
The results from the deletion of CFP1 in mESCs and SPP1 in yeast indicate that, whilst at the majority of genes no effect on transcription is seen, H3K4me3 can be associated with both activation and repression of transcription at a small subset of genes. For example, ectopic deposition of H3K4me3, either by mis-regulation of the SET1A/B complex in cfp1 À/À mESCs or by targeting of a meiotic H3K4 methyltransferase PRDM9 to sites normally depleted for H3K4me3 in human cells can lead to transcription of these regions [62,77], whereas H3K4me3 increases at yeast ribosomal protein genes during their repression in response to diamide stress [78,79]. This apparent disparity between activating and repressing functions can be resolved by the fact that H3K4me3 can recruit a number of factors, many of which have been linked to  (Table 2), via protein folds such as the PHD finger, chromodomain or Tudor domain [80]. The factor recruited may depend on the presence or absence of neighbouring modifications for example, BPTF in the NURF chromatin remodelling complex has both a PHD finger and a bromodomain and preferentially binds H3K4me3 when H4K16 on the same nucleosome is acetylated [81]. Conversely, Spp1 binding to H3K4me3 is inhibited by asymmetric di-methylation of neighbouring H3R2 [29]. Thus depending on which factor and other co-factors are recruited, the residency time and the affinity of those interactions, H3K4me3 has the potential to activate or repress transcription at different subsets of genes, illustrated by the capacity of H3K4me3 to recruit both histone acetyltransferases and deacetylases (Table 2). However, it should be noted that the conclusions of the studies that link these recruited factors to transcriptional activation or repression often rely on the consequential effect on other histone modifications that also correlate with transcription, for example, histone acetylation, without definitively demonstrating the requirement of the factor interaction with H3K4me3 for this transcriptional activation/repression. Furthermore, a number of the so-called 'activating' factors that specifically bind to H3K4me3 have subsequently been shown to also have repressive functions, for example, the chromatin remodeller yeast Isw1/human SNF2h [82][83][84].
A robust study in human colorectal carcinoma HCT116 cells has demonstrated a mechanism by which H3K4me3 promotes transcription activation at the subset of p53-dependent DNA damage-inducible genes whose transcription is affected by H3K4me3 loss [85]. This involves the TFIID subunit TAF3, which binds directly to H3K4me3 [86]. Using high quality biochemical analysis with engineered nucleosomes containing H3K4me3, TAF3 recruitment to H3K4me3 is shown to activate p53-dependent transcription in vitro and in vivo [85]. Additional support for the link between H3K4me3 and TFIID/TAF3 comes from the observation that TBP levels are reduced at the Il20ra gene promoter that loses most of its H3K4me3 in mll1 À/À mouse embryonic fibroblasts (MEFs) [15]. However, this mechanism may only be functional at a small subset of genes: depletion of TAF3 only altered the transcripts from 5.6% of genes [85]. Interestingly, this function of H3K4me3 may be specific to mammalian cells as the yeast TAF3 does not contain a PHD finger [87] and plant The average H3K4me3 ChIP profile for 5774 yeast genes in the wild type (light blue) and jhd2D (dark blue) strains (Yan lab, GSE67211). C: Despite the increase in H3K4me3 at 393 genes in the jhd2D strain (B), there are no significant changes in transcript levels of these genes (dark blue) compared to the change at all genes (light blue, n ¼ 5774) [69]. D: Histogram showing the timing of the H3K4me3 peak relative to the transcript peak for each gene whose transcript peaks during oxidative phase of the yeast metabolic cycle (n ¼ 702) [100]. Time labels refer to the difference between the 16 sampling time-points used by Kuang [88]. Whilst it is very conceivable that H3K4me3 may have different modes of action at distinct subsets of genes and in different cell types and organisms, the positive genome-wide correlation between H3K4me3 and transcription exists in all organisms studied and so there must be other mechanisms to explain this widespread association.

No instructive role for H3K4me3 in transcription?
The fact that at the majority of genes transcription is unaffected by loss of most H3K4me3 argues against a globally instructive role for H3K4me3 in transcription. This is supported by the observation that in vitro, H3K4me3 is not required for transcription of a chromatinised substrate [89]. Instead, studies suggest the Set1/Set1-like complexes may rely on pre-existing transcription-dependent events to deposit H3K4me3 into chromatin, ruling out a strictly instructive role but explaining the strong correlation with transcription. Transcription-dependent recruitment of the Set1/MLL proteins has been described for yeast, D. melanogaster, A. thaliana and mammals [90][91][92][93][94]. In yeast, the Paf1 elongation factor is required for the recruitment of the Set1 complex to RNA polymerase II post-initiation [90], although evidence disputes this requirement in mammals [93,95]. Additionally, in yeast, co-transcriptionally deposited histone H2B ubiquitination [96] is required for H3K4me3 [97,98], which acts in part by preventing Jhd2-dependent demethylation [99]. Further evidence for the deposition of H3K4me3 as a result of transcription comes from time-resolved experiments. The study of the oscillating expression of over half of all yeast genes in the yeast metabolic cycle reveals that at the majority of these genes, H3K4me3 peaks after the peak of the RNA transcript, especially for those genes whose transcripts peak during the oxidative phase (Fig. 3D) [100]. A delay of H3K4me3 relative to induced RNA is also observed during meiosis in yeast [38] and removal of H3K4 methylation does not have a strong effect on meiotic gene induction [101]. Similarly, for genes varying expression during diurnal cycles in the mouse liver, H3K4me3 on average peaked 1.3 h after RNA polymerase II on chromatin [102] and bulk H3K4me3 is not observed before the start of transcription during the zebrafish maternalzygotic transition [103].
Another possibility is that, although H3K4me3 and transcription correlate strongly, these are actually independent events. A simple explanation is that an NDR over DNA with a suitable base composition, such as a CpG island in mammals, recruits SET1A/B leading to H3K4me3. Independently, an NDR might also be sufficient to recruit RNA polymerase II, either supported by DNA-bound transcription factors or not, leading to transcription, hence the apparent link between H3K4me3 and transcription. However, as the non-perfect correlation between H3K4me3 and transcription suggests, it is possible to get H3K4me3 without transcription and vice versa. Evidence for this comes from mESCs: at engineered CpG islands with no nearby promoters, CFP1dependent H3K4me3 can be deposited when surrounding transcription levels, assessed using RNA polymerase II ChIP, are very low or absent and this H3K4me3 does not, in turn,  [52]. Furthermore, in Xenopus embryos, developmentally regulated hypomethylated but not hypermethylated CpG island promoters have H3K4me3 hours before their transcriptional activation [104]. In these contexts, H3K4me3 deposition appears to be solely a function of the unmethylated CpG content of the DNA, allowing specific recruitment of CFP1 and SET1A/B directly. The evidence presented so far has suggested that whilst H3K4me3 can influence the transcription of some genes positively and negatively, H3K4me3 can also be deposited after or independently of transcription. In this situation, the main function of H3K4me3 is unlikely to be regulation of transcription initiation, supported by the fact that the absence of the majority of H3K4me3 does not have pervasive effects on nascent transcription. So if H3K4me3 is not affecting transcription globally, what other functions might it have?
A role for H3K4me3 in splicing pre-mRNAs?
One of the H3K4me3-binding proteins in humans is CHD1 [105] ( Table 2). CHD1 is linked to splicing as it is able to stably associate with U2 snRNP in the spliceosome and depletion of CHD1 decreases splicing efficiencies eightfold at early timepoints of IRF1 gene induction in vivo [83]. However, the direct link between CHD1 association with H3K4me3 and splicing is challenged, as the splicing defect is only threefold upon ASH2 knock-down, a component of all Set1-like complexes in mammals that is required for H3K4me3 [14], and no splicing defect is observed in cfp1 À/À cells, specifically lacking the majority of H3K4me3 [62,68,83]. This might indicate an additional non-H3K4me3-dependent role of CHD1 in splicing. Other studies claim that H3K4me3 can regulate the alternative splicing of the FGFR2 pre-mRNA between different cell types, with a variety of histone post-translation modifications including H3K4me1 and H3K4me3 differentially enriched or depleted over alternatively spliced exons [106]. Splicing is extremely sensitive to transcription elongation speed [107,108] and many effects ascribed to H3K4me3 might simply reflect alterations to transcription elongation. For example, the dynamic acetylation of histones H3 and H4 in nucleosomes also containing H3K4me3 is proposed to affect the splicing of the MCL1 pre-mRNA [109]. Interestingly, splicing can also influence H3K4me3, perhaps indicating a feedback mechanism for this process [110]. Thus, direct links between H3K4me3 and splicing are currently tenuous.

H3K4me3 and transcription termination linked to transcript degradation
A role for H3K4me3 has been described in the promotion of efficient Nrd1-dependent termination of RNA polymerase II transcription in yeast [111]. Nrd1, in combination with its complex members Nab3 and Sen1, binds to specific sites in the nascent RNA and mediates the termination of non-coding RNAs and some short mRNAs that are then targeted to the exosome for degradation or trimming [112]. Whilst there is reduced Nrd1 recruitment in the absence of H3K4 methylation [111], the mechanism for H3K4me3-stimulated termination may also involve the ability of H3K4me3 to recruit both lysine acetyltransferases and deacetylases [113][114][115]. These control the levels of histone acetylation, which influence the kinetics of early elongation by RNA polymerase II and, therefore, may affect the efficiency of Nrd1mediated termination. Thus the putative role of H3K4me3 in Nrd1-dependent termination may be an indirect consequence of altered transcription speed.

H3K4me3 is linked to transcriptional responsiveness
Although H3K4me3 increases over genes during their transcription induction, the modification often persists for many hours after transcription is shut down. This was shown in yeast at GAL10, where H3K4me3 remained on the chromatin for 5 h after inactivation of transcription and dissociation of Set1 [116]. In larger-scale yeast studies, the slowness of H3K4me3 to change was also shown for genes that were downregulated during the induction of meiosis [38] or upon diamide or osmotic stress [79,117], especially compared to the rapid dynamic changes in levels of histone acetylation. Although this could just reflect the higher stability of H3K4me3 compared to histone acetylation, perhaps due to less efficient active removal by demethylation and/or reliance on a mechanism involving dilution of histones during replication [118], it was speculated that the maintenance of H3K4me3 might be advantageous in a reversal of conditions, allowing re-adjustment of the transcriptional response more rapidly [38]. In fact, high residual H3K4me3 has also been observed following transcription at trainable drought-resistance genes in A. thaliana and has been linked to faster transcriptional responses upon re-exposure to stress, a phenomenon known as transcriptional memory [119].

H3K4me3 promotes transcriptional consistency
Many genes can be expressed at very different levels in individual cells in a large population. This variation generates phenotypic heterogeneity and can be advantageous. For example, if not all cells in a population respond to a particular signal, the non-responders are primed for expression if conditions change again or return to pre-signalling, a phenomenon seen clearly in cell heterogeneity in cancer [120]. This variability is often referred to as gene expression noise. Methylation states at H3K4 are known to influence noise. In D. discoides, H3K4 methylation can stabilise transcriptional frequency between generations, perhaps transmitting a memory of the previous state to the progeny [49]. In agreement, relative to acetylation of H3K9, H3K4me3 levels are higher at essential, periodically expressed yeast genes than non-essential, responsive genes, leading to the hypothesis that H3K4me3 is associated with expression stability in yeast [121]. Further support to the idea that H3K4me3 reduces variability in gene expression between cells comes from correlations between the presence of H3K4 methylation and gene expression noise for over 2,000 GFP-tagged yeast proteins [122]. H3K4me3 in the region of the þ1 nucleosome ....Pr ospects & Overviews F. S. Howe et al.
(0-300 bp relative to the TSS) anti-correlates with noise (r s ¼ À0.410) while H3K4me1/2 correlate positively with noise (r s ¼ þ0.347; þ0.296, respectively). This reinforces the need to study each separate modification state in isolation, as loss of all methylation states in a set1D strain reduces noise at selected genes, probably reflecting the loss of H3K4me1/2, while loss of Spp1, and thus the majority of H3K4me3, increased noise (unpublished). These associations between H3K4me3 and transcriptional consistency are supported by a study of the breadth of the H3K4me3 peak in a range of cell types and organisms (Fig. 4A) [4,45]. The broadest peaks are found on the most essential genes for a particular cell type or key genes for basic cell functions across all cell types and correlate with transcriptional consistency. Variations in peak breadth are observed from yeast, flies, worms, plants to mammals but it remains to be seen whether breadth is an important determinant of consistency in many of these organisms. In mouse neural progenitor cells, however, knock-down of WDR5, a component of all the Set1/Set1-like complexes in mammals, decreases both H3K4me3 peak breadth and transcriptional consistency whereas depletion of one of the H3K4 demethylases JARID1B [123] increases both H3K4me3 peak breadth and transcriptional consistency at selected genes [4]. It should be noted that WDR5 is also present in complexes that contain the lysine acetyltransferases Gcn5 or MOF [124,125] and so the effects of depletion on consistency may be indirect. Further experiments more specifically affecting H3K4me3 are required, but the opposing effects of WDR5 and JARID1B knock-downs on consistency are promising. On a population level, H3K4me3 may, therefore, not affect the mean level of transcripts or transcription, but could be influencing the spread of transcriptional responses between individual cells. This highlights the importance of considering single cell experiments to address questions such as whether H3K4me3 and transcription co-occur in the same cell (Fig. 4B). Most data on H3K4me3 in vivo come from ChIP experiments, where levels of the modification correlate with average transcription at the population level. However, ChIP levels say nothing about how many cells contain the H3K4me3 signal or whether those cells are undergoing transcription. For example, PHO84 is a highly transcribed yeast gene with a broad peak of H3K4me3 [29], but RNA fluorescence in situ hybridisation reveals that PHO84 transcripts are present only in 20-25% of the population [126]. Understanding how H3K4me3 is distributed in this population of cells would certainly clarify the relationship between H3K4me3 and transcripts.

Concluding remarks
Reassessment of the evidence that H3K4me3 influences transcription has yielded examples of both activation and repression of small subsets of genes. However, the overwhelming finding from most genome-wide transcript and transcription studies upon removal of the majority of H3K4me3 is that actually very few transcription events/ mRNAs change at the population level, which is remarkable considering the strong correlations with transcripts and transcription. Rather than a major role in instructing transcription, H3K4me3 may instead be deposited as a result of transcription, influencing processes such as splicing, transcription termination, memory of previous states and transcriptional consistency. These events can alter the nature  [4]. At some genes, knocking down WDR5 reduces H3K4me3 peak breadth and increases transcriptional noise. At other genes, knocking down JARID1B increases peak breadth and increases transcriptional consistency/reduces noise. Not all genes show changes in peak breadth upon knockdown of WDR5 or JARID1B. B: Schematic showing the theoretical relationship between peaks of H3K4me3 and levels of transcription in three individual cells (blue circles) and, below, the sum of these representing a population average. For example, if H3K4me3 is instructive for transcription then the cells in column 1 should show the transcription profile in column 4. If there is no relationship at the single-cell level, then the profile in column 1 might lead to the transcription profile in column 3. Note at the population level these different states cannot be distinguished and H3K4me3 appears linked to transcription. F. S. Howe et al.
Prospects & Overviews .... and/or levels of protein produced in the population, without requiring a change to transcripts or transcription. One consideration is that because all studies are populationbased, the consequences of the action of H3K4me3 may only be understood at the level of an individual cell. Levels of H3K4me3 are usually assessed using ChIP and reported as peaks at various sites in the genome. As yet we do not know how this reflects events, particularly transcription, in individual cells in the population and whether transcription and H3K4me3 will ever occur contemporaneously at an individual gene. A relatively new concept in our understanding of functions associated with H3K4me3 is peak breadth, whose size distribution, determined by the combined actions of the methyltransferases and demethylases, is remarkably conserved across eukaryotes despite large differences in gene size. Peak breadth is associated with both transcript levels and transcriptional consistency. Key questions include how breadth is determined, how H3K4me3 functions to suppress noise, whether H3K4me3 is antagonised by another modification, limiting its spread, and what is happening at the individual cell level, particularly during dynamically changing conditions. So H3K4me3, the universal mark associated with 'active' genes that every undergraduate with an interest in genes knows about, remains enigmatic.