Enhancer Journal Club: Functional roles of enhancer RNAs for oestrogen-dependent transcriptional activation

Update 12/08/2013: I've added a plain english summary of this paper at ScienceGist. Find it here: http://sciencegist.com/p/189

So, now for the second of the two papers about enhancer RNAs from the 27th June edition of Nature. This one is by Li et al. and comes from the Rosenfeld lab, who are also at UCSD. Like the Lam et al. paper, they examine the effect of a particular transcription factor (in this case ER-α) on human breast cancer cells. Their findings are also similar to the Lam et al. paper, but the two articles diverge when they start to explore the mechanism through which their eRNAs appear to be acting.

Here's the abstract:

The functional importance of gene enhancers in regulated gene expression is well established. In addition to widespread transcription of long non-coding RNAs (lncRNAs) in mammalian cells, bidirectional ncRNAs are transcribed on enhancers, and are thus referred to as enhancer RNAs (eRNAs). However, it has remained unclear whether these eRNAs are functional or merely a reflection of enhancer activation. Here we report that in human breast cancer cells 17β-oestradiol (E2)-bound oestrogen receptor α (ER-α) causes a global increase in eRNA transcription on enhancers adjacent to E2-upregulated coding genes. These induced eRNAs, as functional transcripts, seem to exert important roles for the observed ligand-dependent induction of target coding genes, increasing the strength of specific enhancer–promoter looping initiated by ER-α binding. Cohesin, present on many ER-α-regulated enhancers even before ligand treatment, apparently contributes to E2-dependent gene activation, at least in part by stabilizing E2/ER-α/eRNA-induced enhancer–promoter looping. Our data indicate that eRNAs are likely to have important functions in many regulated programs of gene transcription.

So they start by treating MCF-7 breast cancer cells with E2, which binds to their transcription factor of interest (ER-α) and allows it to bind to DNA. When they map the binding sites of activated ER-α by ChIP-seq, they find 31052. Of all of these sites, only around 900 are direct binding to promoters, whilst 7174 of the sites co-localise with H3K4me1 and H3K27Ac - histone marks known to be associated with active enhancers. They then map transcription of RNA by GRO-seq and identify 1309 genes which have up-regulated transcription on treatment with E2, finding that almost all of these genes (1145) have an ER-α bound site within 200Kb of their promoters. Again, the majority of these sites, even near to activated genes, are not themselves promoters (and so are likely to be enhancers). Finally, when they look at the GRO-seq data for these ER-α sites near activated genes, they find a large upregulation of eRNAs transcribed bidirectionally from the putative enhancer.

The next question is whether these putative enhancer regions are involved in actively regulating the expression of nearby genes. Since both the eRNA and the nearby gene are activated when they treat the cells with E2, they can address this question by E2 treatment whilst simultaneously reducing the level of the nearby eRNA. If the eRNA is not involved in activating the nearby gene, this treatment should have no effect on the activity of the gene itself. They choose three genes (FOXC1, TFF1 and CA12) and use two independent methods of knocking down the eRNA (siRNAs and LNAs), showing that when the level of RNA from the nearby enhancer is reduced, the genes are no longer activated in the presence of E2. One could argue that this is an effect of transcription, i.e. that the treatment with siRNAs or LNAs is, for example, reducing the occupancy of RNA Polymerase over the enhancer and therefore reducing delivery of RNA Polymerase to the promoter of the nearby gene. To answer these sorts of questions, they use GRO-seq to show that treatment with the LNA does not reduce the transcription of the enhancer RNA (only the level of the mature eRNA). I really like the fact that they also check that the LNA treatment does not affect methylation of the enhancer, nor does it increase the levels of H3K9me3 or H3K27me3 silencing marks. All this supports their contention that it is the RNA product which is important here, and not just the level of transcription of the enhancer.

Now for the really interesting parts. They start off with a standard luciferase assay, in which they show that the enhancer region they identified upstream of FOXC1 can activate luciferase driven from the FOXC1 promoter. The question is still whether it is the DNA sequence of the enhancer or the RNA it produces which has this effect. They delete the part of the enhancer that codes for the RNA and replace it with five UAS sites, and unsurprisingly the luciferase is no longer activated. Now they can cleverly use the UAS sites to artificially recruit a modified version of the enhancer RNA back to the enhancer - and the activity is restored. This is a clear demonstration that for this particular enhancer, the eRNA required for activation of the FOXC1 promoter. An interesting point that is not well highlighted in the paper is that when they tether the eRNA to a random piece of DNA (i.e. not the enhancer sequence) they do not get activation. This indicates that the eRNA alone is not sufficient to activate the gene, there is still a requirement for some part of the enhancer's DNA sequence. One experiment which would have been nice would be to express the eRNA from a separate plasmid without tethering it to the enhancer sequence. This would answer the question of whether the eRNA is a real enhancer (in the sense of needing to be close to the promoter it activates) or whether it is really a new class of non-coding RNA - essentially a separate gene which controls FOXC1 activation whether it is close to it or not.

The current thinking in the field is that many (if not most) enhancers physically contact the genes which they activate by forming large DNA loops. They tested two of their enhancer regions and found that when they treat with E2 (the small molecule that activates these genes) they see an increase in looping between the enhancer and the gene (as measured by 3D-DSL, one of the many chromosome conformation capture techniques). At a different pair of enhancers, they show that in the presence of E2 (where the genes and enhancers are active) removing the RNA product of the enhancer by treating with LNAs actually reduces not only the activity of the gene, but also the looping between enhancer and promoter. This suggests that the eRNAs produced by these enhancers are somehow actually involved in mediating the contacts between the enhancer and the gene, which if true would be a very interesting finding.

Finally they try to address how these eRNA might be causing looping between the enhancer and the gene. One protein thought to be involved in stabilising these loops is cohesin, which they show to increase its binding to the enhancers when the cells are treated with E2. They also show that some specific eRNAs can interact directly with cohesin and that knocking down the eRNAs decreased the amount of cohesin bound to enhancers after treating with E2. Last of all, they show that when the levels of some cohesin subunits are reduced, around two thirds of previously identified E2 target genes are no longer activated by E2. This leads them to their conclusion:

On the basis of these results, we speculate that many regulatory genomic regions, such as enhancers, harbour the cohesin complex, which ‘poises’ the enhancer for the stable eRNA-induced looping necessary for gene activation events. However, we cannot exclude the possibility that the role of cohesin could also reflect non-enhancer-based regulation.

All in all, a very interesting and powerful paper. My main criticism is that neither this paper or the Lam et al. paper adequately rules out the possibility of these eRNAs acting in trans - that is to say, does the RNA really need to be transcribed close to the promoter to have an activating effect? This paper lists the following as evidence that eRNAs work in this way:

1. Most eRNAs they studied were present at roughly 5-15 copies per cell
2. ChIRP mapping of the FOXC1 eRNA showed it was bound most strongly at its site of transcription (i.e. at the enhancer). It was only found at 15 other regions, and none of these regions were adjacent to a gene activated in the presence of E2
3. Gro-seq analysis showed that 95% of the E2 responsive genes were unaffected by knocking down FOXC1 eRNA

For the first point, I'm really not sure how this demonstrates that the eRNAs are working at close range. Lots of other non-coding RNAs have been discovered which have similar copy numbers but have clear location independent effects.

For the second two, I'm happy to take these as evidence that the FOXC1 eRNA acts primarily on the FOXC1 gene and does not have any major trans-effects. What they do not do is show a requirement for the eRNA to work in this way. In other words: if the eRNA was artificially added to a cell in which it was not transcribed from the endogenous enhancer, could it still cause activation of FOXC1 and/or looping between the enhancer and the gene? In my eyes, this possibility is not ruled out by the experiments in the paper.

What does the internet think? Does the eRNA cause the looping, the activation or both? What ought to be the next experiments to dissect these complicated causal relationships? Let me know in the comments!

Paper reference: http://www.nature.com/nature/journal/v498/n7455/full/nature12210.html