Cap analysis of gene expression (CAGE) sequencing

CAGE was performed in two independent experiments on normal and treated NCI-H1299 cells using the CAGE™ Preparation Kit from DNAFORM.jp according to the manufacturer’s instructions. Enrichment of capped RNAs versus uncapped ribosomal transcripts was used to assess sample quality. Samples with a minimum of 400-fold enrichment over ribosomal RNA were subjected to sequencing on the Illumina Hi-Seq 2000 system in 50 bp single-end (replicate 1) and 100 bp paired-end (replicate 2) mode by the DKFZ Genomics and Proteomics Core facility. Resulting raw sequencing data was processed as follows: Multiplexed samples were separated by barcode, trimmed at the first position to remove non-specific guanines⁵³ as well as to 50 bps in the case of the 100 bp paired-end reads, and aligned against the reference genome (hg19) using HISAT⁵⁴ version 0.1.6-beta. Only uniquely mapped reads were retained and in the case of SB939 and DAC+SB939, files were down-sampled to 25×10⁶ aligned reads. The resulting BAM files were loaded into CAGEr version 1.10.0⁵⁵ and CTSS were called using the following parameters (sequencingQualityThreshold = 20, mappingQualityThreshold = 20). After simple tpm normalization, clusterCTSS were generated using the paraclu method (threshold = 0.1, nrPassThreshold=2, thresholdIsTpm = TRUE, removeSingletons = TRUE, keepSingletonsAbove = 0.2, minStability=2, maxLength=100, reduceToNonoverlapping = TRUE). Finally, consensus TSSs across all conditions and replicates were created using the aggregateTagClusters function (tpmThreshold = 0.3, qLow = NULL, qUp = NULL, maxDist = 100, excludeSignalBelowThreshold=FALSE). Importantly, no confounding effects of the underlying sequencing protocol on TSS expression were observed (Supplementary Fig. 3a). Distance to the nearest Gencode GRCh37.p13 annotated TSS was calculated using HOMER⁵⁰ software and statistical analysis was performed in DESEQ version (1.18.0)⁴⁸. Size factors were calculated for the normalization of TSS expression and dispersion estimates for each gene were obtained using the estimateDispersions function with the following parameters (method=”per-condition”, sharingMode=”maximum”). Differential expression between control and DAC, SB939, SAHA, and DAC+SB treated cells was assessed by testing the differences between the base means of two conditions (nbinomTest). Benjamini-Hochberg adjusted q-values below 0.05 were considered as significantly differentially expressed.