Functional classification of long non-coding RNAs by k-mer content

Abstract

The functions of most long non-coding RNAs (lncRNAs) are unknown. In contrast to proteins, lncRNAs with similar functions often lack linear sequence homology; thus, the identification of function in one lncRNA rarely informs the identification of function in others. We developed a sequence comparison method to deconstruct linear sequence relationships in lncRNAs and evaluate similarity based on the abundance of short motifs called k-mers. We found that lncRNAs of related function often had similar k-mer profiles despite lacking linear homology, and that k-mer profiles correlated with protein binding to lncRNAs and with their subcellular localization. Using a novel assay to quantify Xist-like regulatory potential, we directly demonstrated that evolutionarily unrelated lncRNAs can encode similar function through different spatial arrangements of related sequence motifs. K-mer-based classification is a powerful approach to detect recurrent relationships between sequence and function in lncRNAs.

Non-linear sequence similarity between the Xist and Rsx long noncoding RNAs suggests shared functions of tandem repeat domains

“…Thus, the enrichment of a particular protein-binding motif in an individual Xist or Rsx repeat domain does not provide direct evidence that the protein binds to the domain. Nevertheless, these results are consistent with the notions that lncRNA k-mer content encodes information about protein-binding potential (Kirk et al, 2018), and that the various repeats in Xist and Rsx encode function through the concerted recruitment of multiple RNAbinding proteins.…”

“…In our previous work, we found that SEEKR performed best when the length of the lncRNA or lncRNA fragment being studied was similar to 4^k, i.e. the total number of possible k-mers at kmer length k. In tests of Xist-like repressive activity, we found that comparisons of lncRNAs using k-mer lengths of k³7 underperformed relative to comparisons using smaller k-mer lengths, owing to the fact that most annotated lncRNAs are much less than 4^7 (16384) nucleotides long, and kmer profiles of individual lncRNAs at k³7 (³16384 possible k-mers) are dominated by "0" values (Kirk et al, 2018). Based on this observation, and because Repeats A and B, two essential repetitive regions within Xist (Almeida et al, 2017;Hoki et al, 2009;Pintacuda et al, 2017;Royce-Tolland et al, 2010;Wutz et al, 2002), are each about 4^4 (256) nucleotides in length, we reasoned that k-mer profiles at k=4 (4^4=256 possible k-mers) would provide a reasonable estimate of sequence complexity for the repeats without being dominated by "0" values.…”

“…We previously demonstrated that SEEKR can be used to quantify the similarity between any number of lncRNAs regardless of their evolutionary relationships or differences in their lengths, and that similarities in k-mer profiles correlated with lncRNA protein binding potential, subcellular localization, and Xistlike repressive activity. A major strength of SEEKR is that it ignores positional information in similarity calculations, allowing it to quantify non-linear sequence relationships (Kirk et al, 2018).…”

“…Despite their lack of linear sequence similarity, Xist and Rsx both harbor long, internal domains of tandemly repeated sequence (Grant et al, 2012;Johnson et al, 2018). We recently discovered that evolutionarily unrelated lncRNAs that encode similar functions often harbor nonlinear sequence similarity in the form of k-mer content, where a k-mer is defined as all possible combinations of a nucleotide substring of a given length k (Kirk et al, 2018). Below, we describe our use of k-mer based methods to investigate the possibility that the repeat domains in Xist and Rsx harbor non-linear sequence similarity that might be suggestive of shared function.…”

Combined Atomic Force Microscope and Volumetric Light Sheet System for Mechanobiology

“…Plasmid construction pEGFP-C1 F-tractin-EGFP was a gift from Dyche Mullins (Addgene plasmid # 58473; http://n2t.net/addgene:58473; RRID:Addgene_58473). PiggyBac plasmids PB-rtTA and PB-miRE-tre-Puro were kindly provided by Mauro Calabrese (The University of North Carolina at Chapel Hill); PB-rtTA encodes reverse tetracycline-controlled transactivator (rtTA) and G418 resistant gene under UbC promoter 44 , and PB-miRE-tre-Hygro encodes a protein of interest and a hygromycin resistant gene under tetracycline-dependent and EF1 promoters, respectively. pFtractin-Halo was first generated by cloning Halo-Tag cDNA (forward primer, aggggggctagcgctcgccaccatggcagaaatcggtactggctttc; reverse primer, cgaagcttgagctcgagatctagtcgactgaattcgcgttatcgc) between AgeI and BglII site of pEGFP-C1 F-tractin-EGFP using Gibson assembly (New England Biolabs, MA).…”