Enhancer Journal Club: Functional and topological characteristics of mammalian regulatory domains

This month, I'm going to do a short journal club on a paper called "Functional and topological characteristics of mammalian regulatory domains", which comes from François Spitz's group at the EMBL. The original paper can be found at http://genome.cshlp.org/content/early/2014/01/07/gr.163519.113, and the abstract is as follows:

Long-range regulatory interactions have an important role in shaping gene expression programs. However, the genomic features that organize these activities are still poorly characterized. We conducted a large operational analysis to chart the distribution of gene regulatory activities along the mouse genome, using hundreds of insertions of a regulatory sensor. We found that enhancers distribute their activities along broad regions and not in a gene-centric manner, defining large regulatory domains. Remarkably, these domains correlate strongly with the recently described TADs, which partition the genome into distinct self-interacting blocks. Different features, including specific repeats and CTCF-binding sites, correlate with the transition zones separating regulatory domains, and may help to further organize promiscuously distributed regulatory influences within large domains. These findings support a model of genomic organisation where TADs confine regulatory activities to specific but large regulatory domains, contributing to the establishment of specific gene expression profiles.

In essence, the authors use the Sleeping Beauty transposon to generate mice with a single random insertion of a LacZ reporter gene. They then allow these single transposons to excise and re-insert themselves, generating a cohort of integrations which tend to fall into clustered domains due to the preference of Sleeping Beauty to transpose locally. This allows them to compare multiple identical insertions of the reporter gene over genomic distances up to around 2Mb. The paper reminded me of Bas van Steensel's TRIP paper from last year, but there are a couple of key differences. Since the TRIP paper was done in an ES cell line, their readout had to be gene expression level. In this paper, the use of mice allows the authors to score their insertions based on spatial distribution of expression. For me, this is a big plus. Most of the obvious clinical application for basic research on enhancers and other cis-regulatory elements is in developmental disease and this is where the contribution of enhancers to spatial or temporal patterns of gene expression is going to be most important.

They first look to see whether insertions showing expression in a given tissue are located close to EP300 binding sites (as a proxy for enhancers) in that tissue. They find a significant enrichment of limb EP300 sites around insertions which show expression in limb tissues. Furthermore, the closer an insertion was to an enhancer with a known spatial activity pattern, the more likely the insertion was to show a similar expression distribution. These effects were strongest within 50kb, but could be detected up to 200kb away, which is similar to the effect distances seen in the TRIP paper measured earlier. It would have been nice to see some analysis of TAD boundaries here. Are all of the enhancer/insertion pairs studied within the same TAD? If not, is it really that enhancers work better over 50kb range than 200kb or is it simply that there is more likely to be a TAD boundary between the pair the further apart they are? Similarly, since here they are only considering a subset of enhancers with known spatial activity, it would be a good idea to check the distribution of EP300 binding sites in the vicinity. If an insertion is 200kb away from the nearest well-characterised enhancer, perhaps there is more likely to be another, closer enhancer which has not been characterised.

A very interesting finding here is that insertions are more likely to show the expression pattern of a nearby enhancer if they occur between the enhancer and it's endogenous target gene. Insertions in the opposite direction from the enhancer than then endogenous gene show less concordant expression, and insertions that occur beyond a target gene are even less concordant. The most interesting interpretation of this result is that the gene and the insertion are competing for the enhancer, which makes sense given the evidence that this kind of enhancer competition can occur between endogenous genes. Here again though, it would have been useful to check that this is not influenced by the greater chance of enhancer/insertion pairs being separated by a TAD boundary if they are further apart. If it is the case that these pairs are generally within the same TAD, I wonder if one could use something like structured interaction matrix analysis (SIMA) to check whether enhancers make more contacts in the direction of their target gene in Hi-C interaction maps.

In the second half of the paper, they start to examine how Hi-C TADs might be affecting expression patterns of their reporters. They group local clusters of insertions together and classify those with similar expression patterns as "Regulatory Domains" and those with a change in expression pattern as "Transition Zones". They show that 80% of regulatory domains fall within a single TAD, whilst only 45% of transition zones do. They confirm the significance of this result by random permutation of the various regions. One big caveat of this analysis is that they use the TADs from the Dixon Hi-C paper, which are from mouse ES cells. The Dixon paper presents evidence that these TADs are largely cell type invariant, but of course there are some changes between the cell types examined. Changes in TAD boundary position are likely to be much more pronounced across the multiple tissues of a mouse embryo, and I think this makes the result they find all the more striking. Certainly it's some of the most persuasive evidence I've seen that topological domains might help to define regulatory modules with real developmental relevance.

The large distance scales separating insertions with very similar expression patterns suggests that enhancer activity (either alone or in concert) acts over broad domains. The confinement of these insertions to single topological domains suggests that either the local structure defines the range of this activity, or that some third party defines both the range of activity and the topological domain. Either way, the paradigm here would be that these domains are regulatory modules with all or most of the tissue specific genes in the same module being co-regulated. On the other hand, there is lots of evidence (even some from this paper) that genes have to compete with each other for activation by enhancers. If enhancers are largely captured by the closest gene, where is the evolutionary role for the domain? One might begin to answer this question by looking at the positions of genes that are expressed in a tissue specific manner. Can liver and brain specific genes be found in the same TAD for example, or are all tissue specific genes within a TAD generally specific for the same tissue? If anyone has ever done some analysis like this please let me know in the comments!