Meta-analysis of tRNA derived RNA fragments reveals that they are evolutionarily conserved and associate with AGO proteins to recognize specific RNA targets

Abstract

Background: tRFs, 14 to 32 nt long single-stranded RNA derived from mature or precursor tRNAs, are a recently discovered class of small RNA that have been found to be present in diverse organisms at read counts comparable to miRNAs. Currently, there is a debate about their biogenesis and function.

Results: This is the first meta-analysis of tRFs. Analysis of more than 50 short RNA libraries has revealed that tRFs are precisely generated fragments present in all domains of life (bacteria to humans), and are not produced by the miRNA biogenesis pathway. Human PAR-CLIP data shows a striking preference for tRF-5s and tRF-3s to associate with AGO1, 3 and 4 rather than AGO2, and analysis of positional T to C mutational frequency indicates these tRFs associate with Argonautes in a manner similar to miRNAs. The reverse complements of canonical seed positions in these sequences match cross-link centered regions, suggesting these tRF-5s and tRF-3s interact with RNAs in the cell. Consistent with these results, human AGO1 CLASH data contains thousands of tRF-5 and tRF-3 reads chimeric with mRNAs.

Conclusions: tRFs are an abundant class of small RNA present in all domains of life whose biogenesis is distinct from miRNAs. In human HEK293 cells tRFs associate with Argonautes 1, 3 and 4 and not Argonaute 2 which is the main effector protein of miRNA function, but otherwise have very similar properties to miRNAs, indicating tRFs may play a major role in RNA silencing.

Figure 1

Non-random mapping of small RNAs (tRFs) on tRNA genes (HEK293 human cell line). (A) Numbers of unique tRFs that were present at a minimum of 20 reads per million are plotted against length of the tRF. (B) Length distribution of reads that mapped to a specific tRF-5 (GlyGCC), tRF-3 (ValCAC) and tRF-1 of (LeuTAG). (C) Illustration of a mature tRNA showing the cut sites that would generate the different subclasses of tRF-5s and tRF-3s.

Figure 2

Presence of tRFs in bacteria to human. (A) Frequency of the three types of tRF in different human cell lines. tRF alignments that start with the first or second base of tRNA were collated as tRF-5 and whose 3’ end mapped to the 3’ end of tRNA and have a CCA at their 3’ end were categorized as tRF-3. tRFs whose 5’ end matched with the first or second bases of the 3’ trailer sequence of a tRNA were categorized as tRF-1. The number of tRF-5, tRF-3 and tRF-1 mapped in each cell line was normalized with the total number of reads in the analyzed library. (B) Shows the frequency of tRF-5, tRF-3 and tRF-1 in mouse embryonic stem cells, mouse cell line NIH3T3, D. melanogaster, C. elegans, S. cerevisiae, S. pombe and R. sphaeroides. (C) tRF expression in different mouse tissues and embryonic stem cells (ESC).

Figure 3

A given tRNA does not yield tRF-5, −3 and −1 at equal abundance. (A) Number of reads per million of tRF-5s, tRF-3s and tRF-1s from selected tRNA genes. The tRNA genes were selected on the basis of tRF-1s that had >20 reads per million in HEK293 human cell line library. The tRF-1s were compared with tRF-5s and -3s from the same tRNA, regardless of whether the tRF-5 or −3 was derived from that specific tRNA gene or other members of the tRNA gene family. The duplicated tRNA genes (tRNAs with the same anticodon) are marked with special character “*”, “#”, “$”, “%” and &. (B) Scatter plots of tRF-3s versus tRF-5s, tRF-1s versus tRF-3s, and tRF-1s versus tRF-5s for the tRNA genes shown in A along with Pearson correlation coefficients.

Figure 4

Processing of tRFs is distinct from miRNAs and tRF-3 and tRF-1 are mostly cytoplasmic. (A) tRF read counts in wild type, dicer1 −/−, and dgcr8−/−mouse ES cells. (B) Same data sets as A, but read counts of various miRNAs are shown. (C) tRF read counts in Drosophila S2 cells either mock, dicer-1 dsRNA, or dicer-2 dsRNA treated. (D) Same data sets as C, but read counts of two miRNAs are shown. (E) tRF read counts from fly heads of either wild type, dicer-2 mutant, or r2d2 mutant flies. (F) tRF reads in wild type or dcr1 delta S. pombe. (G) tRF read counts in HeLa cell nuclear fractionation or whole cell.

Figure 5

PAR-CLIP analysis of miRNAs and tRFs. (A) Read counts for miRNAs, tRF-5s, tRF-3s and tRF-1s in AGO 1 to 4 PAR-CLIP data from Hafner et al [37]. Each microRNA and tRF is given an identifying number. The expression level of each microRNA and tRF is shown on the Y-axis and the assigned number is shown on the X-axis. (B) Normalized positional T to C mutation frequencies for miRNA, tRF-5 and tRF-3 reads found in the AGO 1 to 4 PAR-CLIP data. (C) Matches of canonical and noncanonical seeds of the 50 most abundant miRNAs, tRF-5s or tRF-3s seen in the AGO1 dataset to the 17,319 CCRs reported in Hafner et al.

Figure 6

tRF-mRNA chimeras are abundant in AGO1 CLASH data. (A) Numbers of CLASH reads that started with a perfect match to a miRNA, tRF-3, tRF-5 or tRF-1 and deemed to not be pre-miRNA, tRNA or a RNA in Ref-Seq. (B) Numbers of miRNA, tRF-5, tRF-3 or tRF-1 chimeras with mRNAs. (C) Alignments of the mRNA portion of the 45 most abundant reads for the tRF-3003a-HIST2H2AA4 interaction and the tRF-3034a-RPL35A interaction to the corresponding mRNA. The tRF-3 portion of the reads is depicted in dashed blue while the mRNA fragment is depicted in red. CLASH, cross-linking, ligation, and sequencing of hybrids.

Figure 7

Examples of tRF-3-mRNA CLASH chimeras. The most abundant read of 10 of the most prevalent tRF-3-mRNA interactions found in our analysis were analyzed with mfold’s RNA Folding Form using default settings. The output of mfold is represented with the tRF depicted 3’ to 5’ and the mRNA sequence 5’ to 3’. For each interaction the number of reads supporting the interaction is shown to the left, and the delta G from mfold is shown to the right. CLASH, cross-linking, ligation, and sequencing of hybrids.