Hyper conserved elements in vertebrate mRNA 3'-UTRs reveal a translational network of RNA-binding proteins controlled by HuR

Abstract

Little is known regarding the post-transcriptional networks that control gene expression in eukaryotes. Additionally, we still need to understand how these networks evolve, and the relative role played in them by their sequence-dependent regulatory factors, non-coding RNAs (ncRNAs) and RNA-binding proteins (RBPs). Here, we used an approach that relied on both phylogenetic sequence sharing and conservation in the whole mapped 3'-untranslated regions (3'-UTRs) of vertebrate species to gain knowledge on core post-transcriptional networks. The identified human hyper conserved elements (HCEs) were predicted to be preferred binding sites for RBPs and not for ncRNAs, namely microRNAs and long ncRNAs. We found that the HCE map identified a well-known network that post-transcriptionally regulates histone mRNAs. We were then able to discover and experimentally confirm a translational network composed of RNA Recognition Motif (RRM)-type RBP mRNAs that are positively controlled by HuR, another RRM-type RBP. HuR shows a preference for these RBP mRNAs bound in stem-loop motifs, confirming its role as a 'regulator of regulators'. Analysis of the transcriptome-wide HCE distribution revealed a profile of prevalently small clusters separated by unconserved intercluster RNA stretches, which predicts the formation of discrete small ribonucleoprotein complexes in the 3'-UTRs.

Figure 1.

HCEs are short, scattered and highly structured. The overall HCE identification pipeline is shown in (A), with the lower part detailing the algorithm used searching for seeds and extending them to lead to the final HCEs. (B–G) highlights the most relevant features of the HCEs. (B) shows the length distribution of HCEs and (C) their percent coverage of 3′-UTRs; (D) displays the predominance of AU base pairs content over CG base pairs in HCE bases composition and (E) the prevalence of highly structured HCEs, as indicated by the shown distribution of secondary structure density in HCEs. (F) displays the distribution of distances between HCEs on the same 3′-UTRs and (G) the HCE distance distribution from the 3′-UTR start, indicated in percent over the 3′-UTR length.

Figure 2.

HCEs can be classified according to their pattern of occurrence in 3′-UTRs and are organized in clusters. (A) shows the classification of 3′-UTRs in four classes, according to their HCE content (on the left). Numbers below each class box are the number of HCE-containing 3′-UTRs belonging to the class. On the right, a sample of six HCE-containing 3′-UTRs: HCEs are mapped onto their 3′-UTR and represented as yellow areas, a grey rectangle being the full-length 3′-UTR. Arrows from class boxes to UTRs indicate which UTR belongs to which class. (B) displays the increasing percentage of clustered HCEs when increasing the maximum intracluster distance allowed for an HCE to be considered part of a cluster. We span from 5 to 40 bases, and at 20 bases we can observe the beginning of a plateau. We therefore chose 20 bases as the maximum intracluster distance to consider. The graph is drawn excluding the 577 HCEs that are unique on their respective 3′-UTR. (C) Graphical representation of the proposed model of trans-factor binding to 3′-UTRs, assuming that HCEs are binding sites of one or more trans-factors. Clusters of closely occurring HCEs (composed by 3–4 HCEs on average) are separated by intercluster RNA stretches of variable length (from 20 to 1419 bases), suggesting a coordinated action on the 3′-UTR.

Figure 3.

HCEs in mRNAs encoding RRM-type RBPs share a sequence and secondary structure motif. HCEs contained in the group of RRM-type RBP genes 3′-UTRs were scanned for both sequence and secondary structure motifs. The first search returned two, almost identical, 12-bases motifs; the second one produced a 17-bases hairpin which, after multiple alignment, emerged to contain a 12-bases core markedly similar to previously identified sequence motifs. This core represents the loop part of the hairpin which, as the two searches are quite concordant on it, may indeed represent a binding motif for the trans-factor of the regulatory network we are trying to identify. (A) shows the alignment between sequence and secondary structure motifs. Light-yellow background highlights column sequence identity. (B) shows the secondary structure motif and its bidimensional structure. The green circle represents the biotinylated DNA polyC linker added to the RNA. (C) Motif instances (yellow areas) mapped on the full length 3′-UTR (grey rectangle) of the 19 RRM-type RBP mRNAs. HGNC gene names are on the left, UCSC UTR names are on the right in parenthesis.

Figure 4.

HuR is a trans-factor binding in vitro to the HCE motif shared by mRNAs encoding RRM-type RBPs. The different RNA probes used for the protein pull-down experiment are shown in (A). HuR pull-down probe: this probe was designed by using the secondary structure motif reported in Figure 3, slightly modifying the lowest part of the hairpin so as to make it fold correctly when not in context. The loop was designed by selecting the most probable bases of the sequence in the aligment and the most probable structure motifs. Positive controls probes are the known binding sites for the YB1 and PTB RBPs, experimentally obtained (11). Again, the lowest part of the stem was slightly modified so as to make it fold as desired. Negative controls HuR probes are Dbl-Mut1, Dbl-Mut2 and Degenerate. The Degenerate probe was synthesized by allowing all four nucleotides to be present at each loop position, so as to obtain a mixture of probes bearing all the possible eptameric loops. The Dbl-Mut1 and Dbl-Mut2 probes were obtained respectively by mutating two bases of the original probe loop in a way to preserve it and by mutating one base and deleting two others in a way to obtain a pentameric loop instead of a eptameric loop. In all probes, the 5′ circle represents the biotinylated DNA polyC linker. (B) shows the HuR pull-down western blots. From the leftmost band to the rightmost: Input, HuR probe, Dbl-Mut1, Dbl-Mut2, Degenerate probe and denaturated beads bands. As can be readily seen, the stem–loop probes bind to HuR with a marked specificity for the correct one. (C and D) YB1 and PTB RBPs pull-down. From the leftmost band to the rightmost: input, YB1/PTB probe and denaturated beads. As shown by western blotting, the stem–loop probes bind to PTB and YB1 respectively, thus confirming that the pull-down protocol works as expected.

Figure 5.

HuR has a preference for the binding of the 3′-UTR of RRM-type RBPs. (A) shows the enrichment of HuR 3′-UTR binding sites for several RNA-binding domains with respect to the most frequent human protein domains and to RBPs as a whole. Data are extracted by the PAR-CLIP experiment published in (49). (B) shows a Venn diagram indicating the overlap between our HuR RRM-type mRNA targets and the experimentally identified HuR PAR-CLIP RRM-type mRNA targets. (C) displays HuR 3′-UTR RRM-type mRNA targets, highlighted in different colours and shapes according to their belonging to our set of 23 mRNAs, to mRNAs we validated by RIP-qPCR and their intersection with the RRM-type mRNA targets from the PAR-CLIP data set.

Figure 6.

The network of HuR binding to mRNAs for RRM-type RBPs is a functional translational network. (A) shows the fold enrichment results (with respect to control) for four predicted RBP mRNAs (plus the CCNA2 mRNA as positive control and the RPL0 mRNA as negative control) subjected to ribonucleoprotein immunoprecipitation (RIP) from lysates of HuR overexpressing MCF-7 cells and quantitative RT-PCR, demonstrating interaction of HuR with these mRNAs. Enrichment significance P-values were computed for each mRNA with respect to the negative control RPL0 and are equal to 0.0163 for CCNA2, 0.0306 for MSI2, 0.01 for RBM15, 0.0003 for SRFS11 and 0.0054 for HNRNPA3. Level of significance is indicated by one, two or three stars (* ≤ 0.05, ** ≤ 0.01, *** ≤ 0.001). (B) reports the western blot confirming HuR silencing in MCF-7 cell line, β-tubulin is used as the housekeeping protein. (C) shows the statistically significant decrease of polysomal mRNA levels for the same four RRM-type RBP mRNAs when HuR is silenced, indicating a decrease of these mRNAs in the translatome and indirectly a translational-enhancing effect of HuR on these mRNAs. Increasing level of significance (* ≤ 0.05, ** ≤ 0.01) is indicated by one or two stars.