una comparación internacional
1. Muestra de países y desempeño relativo
Here we sought to assess the potential of CTCF-dependent chromatin-looping on the intragenic scale. To do so, we first examined the distribution and orientation of CTCF motifs within genes. For this purpose we defined four regions of interest (detailed in section 2.4.1):
· promoter window: +/- 1kb from transcription start site (TSS) · transcription termination site (TTS) window: +/- 1kb of TTS
· intronic windows upstream of exons: 2kb intronic regions upstream of exon 5’ end
· intronic windows downstream of exons: 2kb intronic regions downstream of exon 3’ end
First exons, first introns and anything within 5kb from the TSS were filtered out.
We next scanned for CTCF binding sites within these windows (Methods: 2.4.1), and to minimise the number of non-functional CTCF binding sites, we only considered sites that overlap with a CTCF peak in Lymphoblastoid cell lines (LCLs) obtained from a previous study (Kasowski et al., 2013). Directionality was assigned relative to the gene strand defining sense CTCF motifs if both the motif and the gene had the same orientation and antisense motifs vice versa. Strikingly, we observe a strong bias for sense CTCF motifs at promoters (p-value = 7e-7, binomial test) with a very sharp increase within 500bp upstream of the TSS (p-value = 8.3e-11), while the window upstream and, to a much lesser extent, downstream of exons contained preferentially antisense motifs (p-value = 2e-5 and 1e-2, respectively; Figure 2.1A). Consistently, when analysing whole introns (scaled to the same length) we find the fraction of antisense motifs to increase with proximity to the 5’ exon boundary (Figure 2.1B). TTS windows showed very few CTCF sites overall and no
preference for any motif orientation. Interestingly, we did not find any evidence for such a pattern to occur in long noncoding (lnc)RNAs (Figure S2.1A), indicating that the three- dimensional chromatin structure for lncRNA genes is fundamentally different from protein coding genes. The following analyses are therefore based on protein coding genes only.
The strong enrichment observed for sense CTCF motifs at promoters and antisense motifs proximal to exons suggests pairs of convergent CTCF bound sites between promoter and exon proximal regions, which in turn would trigger the formation of exon- promoter chromatin loops. To evaluate this, we grouped motifs into the following pairs according to their genic region: promoter exon-upstream, promoter exon-downstream, promoter-TTS, exon-upstream TTS, and exon-downstream TTS (schematic in Figure 2.1B). And indeed, when we classified these CTCF-motif pairs as “convergent”, “divergent”, “both sense”, and “both antisense”, we found both classes of promoter-exon pairs strongly enriched for the convergent configurations (Odds Ratio (OR) =1.5 and 1.3, p-value = 3.4e-4, and 1.4e-2 for exon-upstream and downstream respectively, Fisher’s exact test, Figure 2.1C) and significantly depleted for divergent pairs. In line with exons containing preferentially antisense motifs, the exon-TTS motif-pairs were enriched for divergent pairs (OR=1.5 and 1.8, p-value = 8.3e-3 and 8.6e-5 for exon-upstream and - downstream) and depleted for convergent pairs, while very few pairs were observed for the promoter-TTS configuration. Together, these results provide evidence for CTCF-mediated DNA loop formation between promoters and exon proximal regions; overall we identified 181 promoter exon-upstream loops involving 136 promoters. By considering a less stringent threshold (i.e. lowering the motif score and considering a broader promoter region of +/- 1kb), the number of potential exon-promoter loops increases to 1,463 involving 956 promoters (LCL-specific set; Figures S2.1B-C, methods: 2.4.1).
Figure 2.1. CTCF motifs show preferential directionality along transcripts
Distribution of motifs in sense (blue) and antisense (red) orientation for (A) four genic regions (see schematic): promoter, upstream of exon, downstream of exon, and TTS, and (B) along introns [%]. (C) Orientation of CTCF interactions based on the motif pairs: promoter-exon:upstream, promoter-exon:downstream, promoter:TTS, exon- upstream:TTS, and exon-downstream:TTS. (D) Number of motif pairs with physical interaction based on ChIA-PET data of Rad21, H3K4me3, and CTCF (left), and CHiC (right). (E) Schematic of contingency table for testing the association of DUEs and promoter-exon:upstream (left) and promoter-exon:downstream (right) loops. Background color represents ratio of observed vs expected. (F) Enrichment of sQTLs (FDR=10%) overlapping with a particularly oriented CTCF motif (+/-20bp from midpoint) upstream (left) or downstream (right) of a middle exon (star = p-value < 0.05; Fisher’s exact test). This figure was produced by myself and is published in (Ruiz-Velasco et al. 2017).
p−val=7e−7 0 50 100 150 200 250 −1000 −500 0 500 1000 Number of motifs Antisense=1427 Sense=1706
Promoter -500bp Upstream (-2kb) Downstream (+2kb) TTS (+/-1kb)
prom-downstream prom-TTS downstream-TTS prom-upstream upstream-TTS C B D Figure 1 A 1e-2 0 50 100 150 200 250 0 500 1000 1500 2000 2264 2096 p−val=0.3 0 50 100 150 200 250 −1000 −500 0 500 1000 349 377
OR=1.5, pval=5e-2 OR=0.9, pval=0.7
Upstream Downstream E Middle exon * 0.0 0.5 1.0 1.5 2.0 2.5 1e−8 0 200 400 0 25 50 75 100 Intron length [%] Number of motifs Antisense=7220 Sense=6551 F 0.0 0.5 1.0 1.5 2.0 2.5 Enrichment of exon motifs overlapping sQTLs [Li 2016]
Observed/Expected 3,182 58 1,345 37 3,319 60 1,410 23 0.0 0.5 1.0 1.5 2.0 1.6e−5 0 50 100 150 200 250 −2000−1500−1000 −500 0 2254 1973 0.2 0 50 100 150 200 250 −1000 −500 0 500 1000 345 378 0.7 1.5 0.7 1.3 1.5 0.5 1.40.7 1.8 0.5 0 50 100 150 200 1.prom−up2.prom−d own
3.prom−end 4.up−end 5.do
wn−end
Number of motif pairs
ChIA−PET CHiC
1.prom−up 2.prom−d
own 3.prom−end4.up−end5.do
wn−end 1.prom−up
2.prom−d own 3.prom−end4.up−end5.do
wn−end 0 50 100 0 25 50 75 Number of inter actions CTCF−pair orientation Convergent Divergent Both antisense Both sense Antisense motifs Sense motifs A B C D E F
To validate the loops predicted from the motif analysis, we obtained ChIA-PET data for the cohesin subunit Rad21 and histone mark H3K4me3 (Grubert et al., 2015), ChIA-PET for CTCF (Tang et al., 2015), and promoter capture-HiC (CHiC) data (Mifsud et al., 2015). For all datasets, the physical contacts were measured in one of the cell lines that was used in the analysis above (GM12878). Using these contact maps, we captured 57% of our convergent predicted loops, a reasonable number given the high false negative rate in ChIA-PET and CHiC data (Phanstiel et al., 2015). In accordance with our predictions, we observed that most of the interactions occur between promoters and exon- proximal regions and that most of these interactions have a convergent orientation (OR=7.4 for ChIA-PET and 1.6 for CHiC, p-value = 2.5e-9 and 4e-2, Fisher’s exact test; Figure 2.1D).
Given this evidence of CTCF-mediated promoter-exon loops along with a previous report associating CTCF with alternative splicing at the CD45 locus (Shukla et al., 2011), we hypothesised that CTCF might regulate alternative exon usage through the formation of intragenic promoter-exon loops. In support of this hypothesis, we found that exons involved in a predicted promoter exon-upstream loop (convergent CTCF motifs) are more likely to be alternatively used across individuals (OR=1.5, p-value = 5e-2, Fisher’s exact test; Figure 2.1E). No such enrichment is observed for promoter exon-downstream loops. Interestingly, a recent study found that genetic variants affecting alternative splicing across individuals (splicing quantitative trait loci, sQTLs) often coincide with QTLs for CTCF (Li et al., 2016), suggesting a genome-wide role of CTCF in alternative splicing. Remarkably, when we overlapped the sQTLs with exon-upstream CTCF binding sites forming pairs with promoters, we observed a strong enrichment for sQTLs to participate in convergent interactions (OR=2.4, p-value = 2e-2, Fisher’s exact test), whereas all other configurations - including all of the downstream pairs - were not enriched for sQTLs (Figure 2.1F).
Together, these results suggest a relationship between CTCF-mediated promoter-exon looping and alternative exon usage.
One important prediction of such a mechanism is, that alternative exon usage will occur more often for exons that have the potential to form a loop with their promoter, i.e. convergent pair of CTCF motifs between exon and promoter, than for any other exon. In the following chapter we test this mechanism in the context of human genetic variation and explore the functional consequences on the protein isoforms in the human system.