Human lncRNAs mirror adjacent protein-coding gene expression. (A) Violin plot of Pearson correlation coefficients between protein-coding gene and lncRNA expression profiles, binned by distance between transcripts’ transcriptional start sites. A generalized additive model (GAM) fit summarizes the relationship between distance and correlation of protein-coding/lncRNA pairs (e.d.f. = 8.197, P < 2−16). A simulation envelope, generated using a block-bootstrap approach and presented as a red shaded band (see Methods), demonstrates the expected trend under the null hypothesis that distance and correlation are unrelated. The trend in correlation against separation distance lies well outside the simulation envelope, indicating a relationship unlikely to be due to chance. Continuous GAM fit and simulation envelope values were overlaid by plotting the mean of each distance bin. (B) Similarity between expression profiles of coding/lncRNA distance-binned pairs, at time lags of −200–200 min. Solid lines represent the mean correlation coefficient calculated between distance-binned pairs at varying time-lags of lncRNA expression profiles relative to coding gene expression. Simulation envelopes generated using a block bootstrap approach show the expected cross-correlations versus time trends where there is no relationship with separation distance. (C) Produced as in B, with coding gene expression profiles replaced with mature mRNA expression, rather than pre-mRNA.
