Diverse RNA-binding proteins interact with functionally related sets of RNAs, suggesting an extensive regulatory system

Abstract

RNA-binding proteins (RBPs) have roles in the regulation of many post-transcriptional steps in gene expression, but relatively few RBPs have been systematically studied. We searched for the RNA targets of 40 proteins in the yeast Saccharomyces cerevisiae: a selective sample of the approximately 600 annotated and predicted RBPs, as well as several proteins not annotated as RBPs. At least 33 of these 40 proteins, including three of the four proteins that were not previously known or predicted to be RBPs, were reproducibly associated with specific sets of a few to several hundred RNAs. Remarkably, many of the RBPs we studied bound mRNAs whose protein products share identifiable functional or cytotopic features. We identified specific sequences or predicted structures significantly enriched in target mRNAs of 16 RBPs. These potential RNA-recognition elements were diverse in sequence, structure, and location: some were found predominantly in 3'-untranslated regions, others in 5'-untranslated regions, some in coding sequences, and many in two or more of these features. Although this study only examined a small fraction of the universe of yeast RBPs, 70% of the mRNA transcriptome had significant associations with at least one of these RBPs, and on average, each distinct yeast mRNA interacted with three of the RBPs, suggesting the potential for a rich, multidimensional network of regulation. These results strongly suggest that combinatorial binding of RBPs to specific recognition elements in mRNAs is a pervasive mechanism for multi-dimensional regulation of their post-transcriptional fate.

Figure 1. Diverse Binding Specificity among RNA-Binding Proteins

(A) Estimated number of RNA species (y-axis) associated with each protein in this survey (x-axis) at a 1% FDR threshold. The proportions of bound RNAs in each of several classes are represented by colors: nuclear-encoded mRNAs (black), nuclear introns (red), mitochondrion-encoded mRNAs (green), C/D box snoRNAs (cyan), H/ACA box snoRNAs (magenta), and all other RNAs (grey), which includes ribosomal RNAs, LSR1, NME1, SCR1, SRG1, TLC1, mitochondrial introns, unannotated “intergenic” transcripts (named IGR* and IGX* in Datasets S1–S3), tRNAs, and splice junctions. An asterisk (*) denotes proteins whose targets were identified using DNA microarrays that did not contain probes designed to detect most non-nuclear-encoded mRNAs. Six proteins from our previously published work (Puf1–5 and She2) are also included (marked with a plus sign [+]). (B) Hierarchically clustered heat map representation of RBPs and their RNA targets. Rows correspond to specific RBPs and columns correspond to RNAs. The certainty that the RNA is a bona fide target of the specified RBP is represented by a continuous black (10% FDR or greater) to yellow (0% FDR) scale. (C) The distribution of the number of RBPs bound per mRNA at 1% FDR threshold is shown as a bar plot. The black bars represent the values for the 31 RBPs associated with at least ten mRNAs, and the grey bars represent the 22 RBPs associated with at least ten mRNAs, but fewer than 500 mRNAs.

Figure 2. RNA-Binding Proteins Bind mRNAs Encoding Functionally and Cytotopically Related Proteins

(A) Enrichment of “cellular component” GO terms (rows) in target sets (1% FDR) of RBPs (columns). The significance of enrichment of the GO term is represented as a heat map (scale is to the left of the figure) in which the color intensity corresponds to the negative log₁₀ p-value, calculated using the hypergeometric density distribution function and corrected for multiple hypothesis testing using the Bonferroni method. Only a subset of significantly enriched GO terms are shown. RBPs whose targets are significantly enriched (p ≤ 0.01) for at least one “cellular component” or “biological process” GO term are shown. (B) Same as in (A), except for “biological process” GO terms.

Figure 3. Differential Exon/Intron Association Suggests That Certain RNA-Binding Proteins Bind Their Targets during Specific Stages in Their Lives

The relative enrichment of exons and introns in association with RBPs (columns) is represented using a color scale. Results are shown for RBPs that associated substantially more or less strongly with exons or introns than with RNAs overall (mean enrichment of exons from intron-containing genes or introns 25% above or below the median IP enrichment of all RNAs, respectively).

Figure 4. Combinatorial Interactions among RNA-Binding Proteins and mRNAs

(A) The significance of the overlap between mRNA targets for each pair of RBPs (1% FDR threshold) is represented as a hierarchically clustered heat map in which the color intensity represents the negative log₁₀ p-value, which was calculated using the hypergeometric density distribution and corrected for multiple hypothesis testing using the Bonferroni method. (B) An example of a cluster of functionally and cytotopically related mRNAs defined by their pattern of binding to multiple RBPs. The heat map represents RBPs (rows) and mRNAs (columns) color coded to reflect the certainty with which we infer that a specific mRNA is a target of a specific RBP (10% FDR [black] to 0% FDR [yellow]). These 78 mRNAs were associated (at a 1% FDR threshold) with at least four of a set of six RBPs (Ssd1, Khd1, Pub1, Ypl184c, Scp160, and Nab6) whose targets are enriched for mRNAs encoding proteins localized to the cell wall.

Figure 5. Diverse Sequence Motifs Enriched in mRNAs Bound by Specific RNA-Binding Proteins

A pictogram (http://genes.mit.edu/pictogram.html) represents the regular expression patterns defined for FIRE motifs or the preferred base composition of the position-specific scoring matrices for REFINE motifs. For each motif, the negative log₁₀ p-value of the significance of genome-wide enrichment for motif sites in targets is represented (using a color scale) for segments of its mRNA targets (5′ = 200 bases upstream of start codon, CDS = protein coding sequence, 3′ = 200 bases downstream of stop codon). Arrows indicate motifs with a forward strand bias, i.e., the reverse complements of the motifs are not significantly enriched (p > 10⁻⁴ based on the hypergeometric distribution) in targets. “Cons” indicates the negative log₁₀ p-value measuring whether motif sites in targets are more likely to be conserved in orthologous sequence alignments in S. bayanus than are motif sites occurring in nontargets, based on the hypergeometric distribution. Asterisks (*) denote motifs matching previously described RNA-binding elements (details in text). Exact data values and full descriptions of all motifs are presented in Table S4.

Figure 6. Diverse Combinatorial Patterns of RNA-Binding Protein Interactions with a Choice Sample of mRNAs

(A–K) Putative binding sites of RBPs in target mRNAs. The relative lengths of the 5′-UTR, coding sequence, and 3′-UTR are drawn to scale. For mRNAs for which there are reliable measurements for untranslated sequence lengths (SUN4, DSE2, SCW4, CLN2, PUF2, PMA1, SUR7, and HHT1) [63], we added 50 bases onto the estimated 5′-UTR and 3′-UTR lengths, because the estimated UTR lengths are likely conservative. For mRNAs that do not have reliable untranslated region measurements (CTS1, EGT2, and MRP1), we used 250 bases upstream and downstream of the coding sequence as the 5′-UTR and 3′-UTR, respectively. The positions of the start and stop codons are indicated by stop signals. Putative binding sites for RBPs with strong evidence for association (1% FDR) are marked (Puf3-REFINE, Puf4-FIRE, Puf5-REFINE, Pub1-FIRE, Puf1/2-REFINE, Ssd1-REFINE, Nsr1-REFINE, Yll032c-REFINE, Pin4-REFINE, and Nrd1/Nab3-REFINE) (Figure 5 and Table S4). RBPs that we found to be associated with the mRNA, but for which the recognition elements are not yet known, are listed to the right of the mRNA. The number of Pab1 molecules shown bound to the poly(A) tail represents the degree of enrichment of the corresponding mRNA in the Pab1 IPs (log₂ immunopurification enrichment −6 = 0, −5 = 1, etc.) and not the number of Pab1 molecules bound per mRNA. The cap-binding proteins, Cbc1/2 and eiF4e, are shown by default at the cap site.