Enhancer Journal Club: Genome-scale functional characterization of Drosophila developmental enhancers in vivo

Following on from last month's journal club, there was a new paper this month from the Stark lab in Vienna also dealing with large-scale characterisation of enhancer activity patterns. The original paper can be found at www.nature.com/doifinder/10.1038/nature13395, and the abstract is as follows:

Transcriptional enhancers are crucial regulators of gene expression and animal development and the characterization of their genomic organization, spatiotemporal activities and sequence properties is a key goal in modern biology. Here we characterize the in vivo activity of 7,705 Drosophila melanogaster enhancer candidates covering 13.5% of the non-coding non-repetitive genome throughout embryogenesis. 3,557 (46%) candidates are active, suggesting a high density with 50,000 to 100,000 developmental enhancers genome-wide. The vast majority of enhancers display specific spatial patterns that are highly dynamic during development. Most appear to regulate their neighbouring genes, suggesting that the cis-regulatory genome is organized locally into domains, which are supported by chromosomal domains, insulator binding and genome evolution. However, 12 to 21 per cent of enhancers appear to skip non-expressed neighbours and regulate a more distal gene. Finally, we computationally identify cis-regulatory motifs that are predictive and required for enhancer activity, as we validate experimentally. This work provides global insights into the organization of an animal regulatory genome and the make-up of enhancer sequences and confirms and generalizes principles from previous studies. All enhancer patterns are annotated manually with a controlled vocabulary and all results are available through a web interface (http://enhancers.starklab.org), including the raw images of all microscopy slides for manual inspection at arbitrary zoom levels.

The idea here is to clone >7000 enhancer candidates upstream of a reporter gene and assay each candidate's activity by imaging throughout embryogenesis. The candidates are all cloned into the same genomic location, which should help minimize context dependent effects. They classified the candidates as being either constitutively active or active in early, middle or late embryogenesis. They then show that the candidates show DNase hypersensitivity, P300 binding and H3K27 acetylation at the relevant stage of embryogenesis (i.e. mid-embryogenesis enhancers show higher active marks during mid-embryogenesis than early or late). An interesting exception here is constitutive enhancers, which only show P300 binding in early embryogenesis - this perhaps indicates that P300 may be more important for establishing enhancer activity than maintaining it.

They assign putative target genes to the 874 strongest enhancers by matching the expression pattern of the enhancer with that of the closest five genes up or downstream. Like in the Symmons et. al. paper I discussed last month, they also find that Hi-C domain boundaries are highly depleted between enhancers and their target genes. However, out of 482 enhancer-gene assignments, only 28 enhancers appeared to regulate more than one gene. I think this is an interesting point - if Hi-C topological domains serve to constrain the genes affected by a certain enhancer then they clearly aren't the only story. Most domains contain multiple genes, yet it seems that enhancers may only affect a handful (or even just one) of those potential targets - what is it that adds the extra specificity? Of course the other explanation is that enhancer presence may have very subtle or partially redundant expression effects on multiple genes in the domain, of the type that are not easily assigned by matching expression patterns. On this point, of the 28 enhancers regulating multiple targets 23 target two paralogues with similar expression patterns, which might possibly indicate that this method of assigning enhancers to genes could be biased towards the most visually striking patterns.

Finally, they do some characterization of the identified enhancers. Interestingly, 36% occur inside of genes and of those 79% appear to regulate their "host" gene. I wonder whether those enhancers which do not regulate their host gene are very far from the promoter? It would be interesting to see whether being intragenic makes an enhancer more likely to regulate its host gene above and beyond simply being closer to the TSS. I also wonder if these enhancers are also acting as alternative promoters, as identified in Kowalczyk et al. They also identify motifs which are most characteristics of enhancers in different tissue types, which seem to be mostly transcription factor binding sites (e.g. Zelda consensus binding sequences, which were enriched in early embryonic enhancers.)  By mutating these transcription factor binding sites in their identified enhancers they are able to abrogate activity in 10 out of 11 cases, highlighting that these sequences are necessary for enhancer function. I think the more interesting question though is whether they are sufficient. It would have been interesting to take a set of motifs that together are highly predictive for enhancer function in a particular tissue and ask how many of the candidate sequences which did not show any enhancer activity have similar enrichment for these motifs.

All in all, a very thought provoking paper - also kudos to the authors for making all of their images available at enhancers.starklab.org, I think that's a very nice touch.