Figure 3: Functional separation of distinct groups of promoters.

(a) Transcribed ENCODE HeLa DHSs were grouped into five major classes via k-medoids clustering based upon exosome sensitivity, expression levels and transcriptional strand bias (directionality). DHSs and their cluster memberships were visualized by principal component analysis (PCA). The first two principal components describe ~80% of the total variance in the data used for clustering. Colours and names given to each DHS class are utilized throughout the paper. (b) Biochemical properties of each DHS cluster are summarized by horizontal density plots. Illustrations on top of subpanels depict the typical arrangement of TSSs within each DHS cluster where the size and shade of the arrows indicate abundance and stability of emitted RNAs, respectively. The two unidirectional stable clusters are merged in subsequent analyses. (c) Enrichment odds ratios (vertical axis, log₁₀ scale) of DHS overlap with TSSs of GENCODEv17 transcripts (annotation classes are described in Methods). Stars (*) indicate Fisher’s exact test Benjamini–Hochberg FDR<0.001. (d) Enrichment odds ratios (log₂-transformed, vertical axis) of DHS overlap with ENCODE chromatin segmentation states. Although all clusters of transcribed DHSs are enriched for predicted promoters in HeLa cells (‘TSS’; OR ranging between 3.6 and 61.4), weak unstable DHSs are highly enriched for predicted strong enhancers (‘E’; OR=6.9). Note that untranscribed DHSs mainly fall into the classifications not associated with predicted active transcription initiation (‘repressed’, ‘CTCF’, ‘transcribed’ and ‘promoter-flanking’ states).