Chapter 2: Investigating chromatin features that determine promoter contacts and transcriptional regulation
2.5 Discussion
Chromosome conformation capture reveals interacting loci, which mediate transcriptional regulation. Chromatin features that confer CRE functionality can be used to find biologically relevant captured interactions. Through ANCOVA modeling of Capture-C signal with seven different genomic features, I characterized promoter contacts and how they vary with respect to gene expression. I found a high
proportion of promoter contacts within the viewpoint TAD when mESC TAD boundaries were used as a proxy for the TAD boundaries in E13.5 limb bud,
although this is potentially confounded by the proximity ligation effect. In line with the typically rapid signal decay produced by proximity ligation, distance was most
predictive of the variation in Capture signal. Intermediate promoters, however, had on average less variation explained by distance, and more unexplained variation. Intermediate promoters may be transitioning from an active to an inactive state, or the reverse. Particularly during development, this switch may need to occur quickly in order to establish precise control of gene expression at the right stages and sections of the developing limb bud. Therefore, there may be rapid turnover or dissociation of the bulky complexes associated with transcriptional activation or repression (Swift and Coruzzi, 2017), and promoters in transition may experience
less steric hindrance in forming contacts. They may exist in a poised state in which they are more promiscuous than active or silent promoters, contacting loci enriched for genomic features other than those in the ANCOVA model (Figure 2.10).
Interactions between poised promoters and enhancers have been observed at the Sonic hedgehog (Shh) promoter in anterior mouse limb bud cells, where Shh is not detectably transcribed despite contacting a known Shh limb enhancer (Amano et al., 2009).
Figure 2.10. Transition promoters are less constrained than silent or active promoters in the contacts they form. The ANCOVA results suggest that active and silent promoters
form fewer contacts than do transition promoters, possibly due to steric hindrance by transcriptional or repressive complexes. Curved lines, chromatin loops; blue rings, cohesin; colored teardrops, histone modification marks; black arrows, promoters; transparent
spheres, large chromatin-associated complexes.
This model is supported by the dynamics of chromatin loop formation. According to loop extrusion theory, which has now been visualized in real-time in yeast (Ganji et al., 2018), loops form when loop extrusion factors such as the
cohesin ring (in mammals) bind the DNA, bringing non-adjacent loci into contact with one another as they translocate along the DNA (Fudenberg et al., 2016). Depleting cohesin in human cells increased the average distance between interacting loci in the mouse, confirming its critical role in chromatin interactions (Wutz et al., 2017). When the extrusion factors encounter a boundary element like CTCF, often at the end of a TAD, translocation stalls and the loop cannot proceed beyond the TAD boundary. TAD boundaries are enriched not only for CTCF but also for active
promoters and their associated transcriptional machinery. The latter is hypothesized to be capable of itself acting as a boundary element by physically disrupting the translocation of cohesin along the DNA (Fudenberg et al., 2016). Active promoters
At inactive promoters, repressive complexes like the Polycomb Repressive Complex (PRC) may likewise contribute to steric hindrance if their dissociation rate from the DNA is sufficiently low. On a broader scale, long stretches of repressed chromatin interact with one another through the actions of PRC1 and PRC2 (Andrey et al., 2017). PRC2 both catalyzes and – in a positive feedback loop which enables it to maintain or foster the spread of repressive chromatin – recognizes the
trimethylation of histone H3, lysine 27 (Berry et al., 2017). Through the spread of repressive marks, the chromatin is compartmentalized into laminar-associated domains (LADs) and inter-LADs. LADs comprise gene-poor, low GC content, closed chromatin localized to the nuclear periphery, whereas inter-LADs comprise gene- rich, higher GC content, open and active chromatin closer to the center of the nucleus (van Steensel and Belmont, 2017). Any Capture-C viewpoints located in LADs may be restricted from forming as many contacts due to the higher level of chromatin compaction in these territories. In support, 4C-seq in mouse immune cells showed that inactive viewpoints contacted fewer loci per chromatin loop than active viewpoints (Jiang et al., 2016).
The ANCOVA models in this work suggest that when attempting to identify CREs regulating a gene, one should first consider activity level of the gene – not only because chromatin interactions are known to take place between regions with similar levels of transcriptional activity (Andrey et al., 2017), but because if a gene of
interest is expressed at an intermediate level or is poised, then it may be subject to fewer constraints than active or silent loci in the contacts it forms.
If relevant datasets other than those included here are available for the cell type of interest, they should be considered when attempting to predict regulatory function among promoter contacts. This is because the chromatin features I included predict less than half of the variation in Capture signal. If the lack of predictive power is because Capture signal is simply too noisy – due to exerimental conditions or to cellular heterogeneity within the limb bud (Andrey et al., 2017), including allele- specific differences (Davies et al., 2016) such as at the imprinted Igf2 locus on Chromosome 7, then adding additional chromatin features might only result in incremental increases in model fit. However, a study predicting enhancer activity in mouse erythroid progenitors found TF occupancy to be a better predictor than either chromatin accessibility or histone modifications (Dogan et al., 2015). Two studies using chromatin features to predict contacts of active promoters in human cell lines
found DNase hypersensitivity and histone modification marks to have some predictive power (Roy et al., 2015), in accordance with this work, but also found CTCF, cohesin subunit Rad21 (Yang et al., 2017), and TF occupancy to be predictive. The abundance of predicted TF binding sites in the genome and the tendency of TFs to be highly cell type-specific and follow a complex, sometimes sub- optimal motif grammar makes accurate computational prediction of their binding sites challenging (Spitz and Furlong, 2012; Farley et al., 2015; Khamis et al., 2018;
Keilwagen et al., 2019), but obtaining ChIP-seq data for each new cell type and TF would not be practical. In additon, determination of enhancer output is compounded by the cooperation between multiple TFs to out-compete nucleosomes for
occupancy of the DNA (Long et al., 2016). In the absence of relevant TF binding data, consideration of the presence or absence of transcriptional or repressive
complexes expected to localize at the viewpoint based on viewpoint gene expression level should guide the search for CREs.