Exon-intron marking is laid down in the absence of transcription on silent genes, with some marking biases changing or becoming reversed for genes expressed at different levels.
Cell-type specific differences in marking biases were not due to inherent differences between the epigenetic states of cell lines and primary cells, since K562 showed as many concordant marking biases with CD14+, as it shared with U937 (Supplementary Table S1).
The complexity of this marking system was further elaborated when we examined histone modification exon-intron marking biases for each of the three cell types independently (Supplementary Figure S8).
For virtually all histone modifications, the marking biases we observed were evident both with and without nucleosome normalization (Figure 1; Supplementary Figures S6 and S7), demonstrating that underlying nucleosome distributions could not account for the differential marking levels we observed.
More specifically, (i) marking biases for both H3K36me1 and H3K27me3 switched from favoring exons, to favoring introns, and (ii) the differentials between canonical exons, alternative exons and introns for all four modifications changed.
Fourthly, at the very highest levels of expression (the top 5-10% of RMA values, decile 10), the marking biases and enrichment levels of all four modifications showed distinct shifts away from those seen in the previous deciles.
Furthermore, while other studies had observed exon marking across expressed genes by analyzing a single cell type, our data points to aspects of this marking as being cell-type specific, combinatorial, and related to levels of transcription – with even untranscribed genes having exon-intron marking biases.
These studies were performed in the K562 cell line for four modifications (H3K27me1, H3K27me3, H3K36me1 and H3K36me3) which typified the kind of exon or intron marking biases we had observed across the ENCODE regions, and which appeared to be distributed as combinatorial signatures in either expressed or non-expressed genes (see Results above and Supplementary Figures S9 and S10).
Furthermore, in several instances, marking biases were accentuated only through nucleosome normalization (Figure 1 and Supplementary Figure S6, eg. H3K36me1; Supplementary Figure S7a and c, eg. H3K27ac), providing evidence that nucleosome distributions could mask underlying differences in histone modification levels if not taken into account.
We used permutations of histone modifications which showed concordant marking biases in K562 cells in either expressed or non-expressed genes (H3K36me3 followed by H3K27me1 - both favoring exons of expressed genes; H3K27me3 followed by H3K36me1 - both favoring exons of non-expressed genes).
