Identification of transcribed promoters and enhancers

To identify promoter and enhancer regions, TSSs that mapped close to each other on the same strand were grouped into clusters. This was performed using decomposition peak identification (Forrest et al., 2014)

(https://github.com/hkawaji/dpi1/blob/master/identify_tss_peaks.sh) with default parameters but without the decomposition composition parameter. TSS clusters with at least three supporting CAGE tags were retained and used as input to identify bidirectionally transcribed enhancers.

(https://github.com/anderssonrobin/enhancers/blob/master/scripts/bidir_enhancers).

Promoter TSS clusters that were defined as those that did not overlap enhancers and mapped to +/− 300bp of the 5′-end of GENCODE v 27 transcripts. For differential expression (DE) analysis between control and DUX4 expressing hESC, we first counted the TSSs mapping to promoters and enhancers. Next, coverage at single-base-pair resolution was calculated with BEDTools v 2.27.2 (http://bedtools.readthedocs.io/en/latest/) using only the 5′ ends of the reads. The resulting forward and reverse bedGraph files were then converted into bigWig files using the UCSC software bedGraphtobigWig. Counting was performed using in a strand-specific manner using UCSC software bigWigAverageOverBed. Normalization and DE was performed using egdeR v3.26.8 (McCarthy et al., 2012; Robinson et al., 2010).

(https://bioconductor.org/packages/release/bioc/html/edgeR.html). Promoter counts were normalized using calcNormFactors function with relative log expression, and counts were converted to log2 counts per million (CPM). A prior count of 0.25 was added to the raw counts. For enhancers forward and reverse counts were summed up. The counts were normalized using the same normalization factors as generated for promoters. Promoters (log2 CPM >−2.0) and enhancers (log2 CPM >−3.5) expressed in at least one library were retained. DE was performed between four controls (dox -) and four DUX4-expressing (dox +) expressing samples with Benjamini–Hochberg false discovery rate (FDR) correction.