Transcription factor binding site landscapes of human lncrna and protein-coding promoters: a gc- and cpg-aware comparison

Long non-coding RNA (lncRNA) and protein-coding genes are thought to be governed by partly distinct transcriptional programs. A landmark study by Alam and colleagues (2014) reported that 140 transcription factor bind ing sites (TFBS) are over-represented in lncRNA promoters, including many forkhead (FOX), SOX, homeobox and nuclear receptor motifs, whereas 74 are over-represented in protein-coding promoters. lncRNA promoters, however, are markedly A/T-rich and rarely overlap CpG islands, raising the question of how much of this differential TFBS landscape reflects base composition rather than regulatory specificity. Here we revisit the question using GEN CODE v44 canonical transcripts (19,613 protein-coding and 10,988 lncRNA), 892 JASPAR position weight matrices and the full positional information of 11.0 million predicted TFBS in the [−2000,+49] region around the tran scription start site. Per-TF TFBS densities were almost identical between the two classes (Pearson r = 0.975), yet 543 TFs showed a significant class preference whose direction was largely predicted by the GC content of the motif (Spearman ρ = −0.655): A/T-rich motifs appeared lncRNA-enriched, GC-rich motifs coding-enriched. Modelling the expected lncRNA/coding ra tio as a smooth function of motif GC and CpG content explained 81% of the between-TF variance. After this correction, the lncRNA preference of 299 of 315 TFs, including essentially all FOX, SOX, HOXD and NKX motifs, was no longer significant, whereas a robust core of coding-promoter activators persisted (ETS/ELK, SP/KLF, CREB/ATF, YY1, NRF1, NF-Y, E2F6). The few TFs retaining a lncRNA preference were dominated by bHLH fac tors (NEUROD2, HES5, HEY1). Positional profiles further showed a sharp TSS-proximal accumulation of TFBS in coding promoters that is nearly absent in lncRNA promoters. Our results indicate that most of the previously reported lncRNA-specific TFBS signature is attributable to promoter base composition, and provide a composition-aware catalogue of TFs that gen uinely discriminate the two promoter classes.