Characterization of the in vitro RNA-binding sequence specificity of CstF64 and CstF64τ by SELEX-seq

Previous studies have characterized CstF64 RNA-binding specificity in vitro using the conventional SELEX procedure (Takagaki and Manley 1997). However, the sequence specificity of CstF64τ remains unknown. To comprehensively characterize the RNA-binding specificity of both CstF64 and CstF64τ, we performed SELEX coupled with high-throughput sequencing (SELEX-seq) (Fig. 2A). Compared to the conventional SELEX, which only selects a small number of the most avid binders, SELEX-seq provides a more comprehensive characterization by identifying protein–RNA interactions with a wide range of affinities (Dittmar et al. 2012). To this end, we incubated the recombinant GST-CstF64-RRM or GST-CstF64τ-RRM fusion proteins with a pool of random 20-nucleotide (nt) RNAs. After GST fusion proteins were precipitated using glutathione beads and washed, bound RNAs were eluted and amplified by reverse transcription and PCR (RT-PCR) (Fig. 2A). After just one round of selection, CstF64 and CstF64τ cognate sequences were significantly enriched as shown by gel mobility shift assays (Fig. 2B). Additionally, CstF64-selected RNAs were bound by CstF64τ and vice versa (Fig. 2B), indicating that the two proteins have selected similar sequences. Next, we sequenced the CstF64- and CstF64τ-bound sequences along with the original RNA pool by using the Illumina HiSeq platform. Approximately 80 million reads were obtained for each library. Based on the sequencing data, we first compared the fold enrichment of all possible 10-mer sequences in CstF64- and CstF64τ-selected RNAs compared to the original RNA pool. Strikingly, the fold enrichment of all 10-mer sequences was almost identical for CstF64 and CstF64τ (Pearson's coefficient: 0.997) (Fig. 2C). We next compared the nucleotide composition of sequences that were enriched to different extents. For both CstF64 and CstF64τ, the most highly enriched sequences consisted almost entirely of Us and Gs (Fig. 2D). In sequences with intermediate enrichment levels, A was also present at a significant frequency. In contrast, C-containing sequences were depleted during selection (Fig. 2D). To verify our SELEX-seq results, we performed gel mobility shift assay with GST-CstF64-RRM or GST-CstF64τ-RRM and three selected 10-mer sequences. Consistent with our sequencing results, both CstF64 and CstF64τ bound to UG-repeats with the highest affinity (Fig. 2E; Supplemental Fig. S1). The affinities of both proteins were lower for the A-containing sequence and further decreased and became more divergent when C was present (Fig. 2E). Finally, we identified specific sequence motifs that are enriched at different levels (Supplemental Table S1). Long UG-repeats had the highest fold enrichment (Supplemental Table S1). Sequences containing shorter UG-rich regions were enriched to a lesser extent. CACGACG was among the most depleted motifs. These analyses revealed that CstF64 and CstF64τ have nearly identical RNA-binding sequence specificities in vitro, and both proteins preferentially bind to UG-rich motifs while avoiding C-rich sequences.

Open in a separate window

FIGURE 2.

Comparison of in vitro RNA binding specificity of CstF-64 and CstF-64τ using SELEX-seq. (A) A schematic representation of the SELEX-seq procedure (details described in the text and Materials and Methods). (B) Gel mobility shift assay using RNAs prepared from the random pool (library 1), CstF64-selected sequences (library 2), and CstF64τ-selected sequences (library 3), with GST-CstF64 (marked as CstF64) or GST-CstF64τ RRM (marked as CstF64τ). (C) Comparison of fold enrichment of all possible 10-mer RNA sequences in CstF64 and CstF64τ selected sequences: (x-axis) log10(frequency of each 10-mer sequence in library 2/its frequency in library 1); (y-axis) log10(frequency of each 10-mer sequence in library 3/its frequency in library 1). (D) Comparison of nucleotide composition of RNA sequences with fold enrichment in CstF64- and CstF64τ-selected sequences. Fold enrichment levels are marked on the bottom. The height of the boxes corresponds to the frequency of each nucleotide. (E) Quantification of the gel shift assays in Supplemental Figure S1. Three sequences were tested (different colored lines) with GST-CstF64-RRM (squares) or GST-CstF64τ-RRM (dots).

Mapping the in vivo RNA-binding sites of CstF64 and CstF64τ at the transcriptome levels by iCLIP-seq

We recently mapped CstF64-RNA interactions in vivo at the transcriptome level through iCLIP-seq analysis (Yao et al. 2012). To globally characterize CstF64τ-RNA interactions in vivo, we carried out iCLIP-seq of CstF64τ in HeLa cells. As shown in Figure 3A (lanes 1,2), CstF64τ was efficiently crosslinked to RNAs in vivo by UV irradiation and specifically purified by immunoprecipitation (IP) using our CstF64τ antibody described earlier. Following similar procedures that were used for CstF64 iCLIP-seq (Yao et al. 2012), we prepared CstF64τ iCLIP-seq libraries and subjected them to high-throughput sequencing using the Illumina HiSeq platform, and 6.1 million uniquely mapped reads were obtained for the CstF64τ iCLIP library. The distribution of CstF64τ iCLIP tags showed significant enrichment in the 3′ UTRs (Fig. 3B), consistent with its function in mRNA 3′ end processing. A global analysis of the distribution of CstF64τ-binding sites near actively used PASs (based on our RNA 3′ end mapping, see below) detected a peak at 30~35 nt downstream from the AWTAAA hexamer (Fig. 3C, red line), suggesting that CstF64τ binds to RNA together with CPSF. The distribution of CstF64τ iCLIP tags is very similar to that of CstF64 (Fig. 3C, blue line); but overall, CstF64τ iCLIP signals are lower than those of CstF64 (Fig. 3C, cf. red and blue lines). Motif analysis of CstF64τ crosslinking sites downstream from AWTAAA identified an UG-rich motif (Fig. 3C, inset), which resembles the UG-repeat motif identified by our SELEX-seq. As shown in two specific examples of the CstF64τ iCLIP analyses results (Fig. 3D,E), CstF64τ and CstF64 bind to the same regions downstream from the cleavage sites in Rps12 and Basp1 transcripts. Together, these results demonstrated that CstF64τ-RNA interaction landscape in vivo is highly similar to that of CstF64.

Open in a separate window

FIGURE 3.

Mapping CstF64τ-RNA interactions in vivo at the transcriptome level by iCLIP-seq analysis. (A) HeLa cells were UV irradiated or mock treated, and cell lysates were treated with RNase I as labeled. CstF64τ was IPed. After the 3′ linker ligation and 5′ end labeling, the purified RNP complexes were resolved by PAGE and visualized by phosphorimaging. CstF64τ and β-actin levels in all cell lysates were monitored by Western blotting. (B) Distribution of CstF64τ iCLIP tags in different regions compared to their frequency in the genome. (C) Distribution of the cleavage sites (green line), CstF64- (blue line), and CstF64τ binding sites (red line) relative to the closest upstream AWTAAA. Position 0 represents the 5′ end of AWTAAA. Weblogo of the top 20 over-represented motifs at CstF64τ crosslinking sites near poly(A) sites are shown in the inset. (D,E) CstF64 and CstF64τ iCLIP-seq results for Rps12 and Basp1 visualized on the UCSC genome browser.

The role of CstF64τ in regulating mRNA alternative polyadenylation

We next characterized the role of CstF64τ in regulating global APA. To this end, we established HeLa cell lines that stably express a CstF64τ-targeting shRNA (CstF64τ-RNAi cell lines). As shown in Figure 4A (right lane), CstF64τ was significantly depleted in CstF64τ-RNAi cells. In these cells, an increase in CstF64 protein levels was observed. Similar reciprocal changes were also observed in a cell line in which CstF64 is stably silenced by RNAi (CstF64-RNAi) (Fig. 4A, middle lane). These results suggest that CstF64 and CstF64τ negatively regulate one another's expression. Quantitative RT-PCR (RT-qPCR) analyses did not detect significant changes at the mRNA levels (Supplemental Fig. S2), indicating that CstF64 and CstF64τ modulate one another's expression primarily at the post-transcriptional level. Next, we characterized the global APA profile in CstF64τ-RNAi cells by direct RNA sequencing (DRS) analysis using the Helicos platform as described before (Yao et al. 2012). By comparing the APA profiles between control HeLa cells and CstF64τ-RNAi cells, we identified 96 genes (of the 3800 genes that have at least two polyadenylation sites, each with 10 DRS reads or more) that displayed significantly different APA patterns in CstF64τ-RNAi cells (false discovery rate <0.05, Fisher's exact test) (Fig. 4B). Among them, 57 showed a relative increase in the mRNA isoforms polyadenylated at the proximal PAS (distal-to-proximal or DtoP shift), whereas the remainder showed APA changes in the opposite direction (proximal-to-distal or PtoD shift) (Fig. 4B). For example, the APA profiles of CSTF3 in control HeLa cells and CstF64τ-RNAi cells are shown in Figure 4C. The APA profile of the same gene in CstF64-RNAi cells was also shown for comparison (Fig. 4C). A significant increase in the mRNAs polyadenylated at the distal PAS in CSTF3 was observed in both CstF64- and CstF64τ-RNAi cells (Fig. 4C). This and a number of other APA changes identified by our DRS analyses were validated by RT-qPCR (Fig. 4D), confirming the accuracy of our sequencing analysis. To further characterize the CstF64τ-mediated APA regulation, we next compared the APA changes induced by CstF64τ depletion and the previously reported APA changes caused by CstF64 depletion or CstF64&τ codepletion (Fig. 4E). It is apparent that depletion of CstF64 or CstF64τ had a relatively small effect on the global APA profile compared to the codepletion of both factors (Fig. 4E; Yao et al. 2012), strongly suggesting functional redundancy between the two proteins. Additionally, APA changes induced by CstF64 or CstF64τ did not seem to have strong biases between PtoD and DtoP changes (Fig. 4E; Supplemental Fig. S3). In contrast, codepletion of CstF64 and CstF64τ led to PtoD shifts in 85% of the affected genes (Fig. 4E; Supplemental Fig. S3). These observations are consistent with a model in which CstF64 and CstF64τ play largely redundant roles in APA regulation. Depletion of either proteins induces the up-regulation of the other, resulting in a relatively stable general mRNA 3′processing activity and relatively few changes in APA. However, codepletion of both proteins leads to a decrease in mRNA 3′ processing activity and thus lower efficiency of cleavage/polyadenylation at the proximal PASs. This, in turn, results in higher transcription read-through and usage of the distal PASs and thus PtoD shifts.

SELEX-seq

GST-CstF64/τ-RRM was expressed in E. coli and purified with glutathione-conjugated beads according to the manufacturer's instructions (GE Healthcare Life Sciences). SELEX was carried out as previously described with minor modifications (Takagaki and Manley 1997). Briefly, template DNA containing 20 nt random sequence was generated by annealing two oligos (T7 and N20), and the DNA ends were filled with Klenow. The RNA pool was prepared by in vitro transcription with T7 RNA polymerase and purified on a urea-polyacrylamide gel. For RNA pulldown, 7.4 µg GST fusion protein was conjugated with 20 µL glutathione sepharose 4B beads. After washing, the beads were mixed with 200 μL of binding buffer (8 mM HEPES, pH 7.9; 40 mM NaCl; 2 mM EDTA; 0.2 mM DTT; 0.2 mM PMSF; 8% glycerol) containing 16 µg of heparin and 32 µg random RNA pool (final concentration 7.5 µM), and incubated for 10 min at 30°C with occasional shaking. After washing, bound RNAs were released by proteinase K digestion and precipitated by ethanol. Recovered RNAs as well as the original RNA pool were amplified by RT-PCR to generate barcoded libraries for sequencing. Oligo sequences are provided in Supplemental Material.

iCLIP-seq

CstF64τ iCLIP-seq was carried out as described previously (Yao et al. 2012). For simplicity, one master iCLIP-seq library was generated and used for all analyses. For CstF64τ iCLIP motif analysis, we ranked all CstF64τ crosslinking sites according to their cDNA counts. We iteratively chose the crosslinking site with the highest cDNA count that did not overlap with the 21-nt region spanning a site that had been chosen previously. For all nonoverlapping crosslinking sites, we determined the enrichment score of 6-mer motifs in the 21-nt surrounding region. We further classified those CstF64τ crosslinking sites into two groups: those with A(A/U)UAAA within 40 nt upstream (AAUAAA+) and those without (AAUAAA−). We aligned the top 20 most enriched motifs in each group and generated sequence logos using WebLogo 3 (http://weblogo.threeplusone.com/) as previously described (Yao et al. 2012).

Gel shift assay

RNA oligos were prepared by in vitro transcription with T7 RNA polymerase and purified on denaturing polyacrylamide gels. For gel shift assays, 5′ radio-labeled RNAs (5 nM) were incubated with 0.5 µM to 60 µM GST-CstF64/t RRM protein in 10 µL binding buffer for 10 min at 30°C. The reaction mixtures were resolved on a 4% nondenaturing polyacrylamide gel and visualized by phosphorimaging.

Direct RNA sequencing analyses

Direct RNA sequencing (DRS) was performed by Helicos BioSciences, and DRS reads were aligned to hg19 using the index-DPgenomictool in Helisphere (Helicos BioSciences). Only uniquely mapped reads with a minimum mapped length of 25 and an alignment score of 4.0 were kept. We further filtered and clustered all individual PASs in the same way as described previously (Yao et al. 2012). To compare the alternative polyadenylation (APA) profiles in HeLa and CstF64τ-RNAi cells using DRS data, we first removed PASs that had zero or one read in both samples. For the remaining PASs to be mapped to RefSeq genes, we used the Fisher exact test to compare the ratio of the DRS read counts of one PAS to the sum of the read counts of all of the other PASs within the same gene. The P-values were adjusted by the Benjamini–Hochberg method for calculating the FDR. PASs with an FDR <0.05 were defined as significantly changed PASs. To create the scatter plot shown in Figure 4B, we selected two PASs with the smallest P-values for each gene with multiple PASs and calculated the corresponding proximal/distal ratio. In the figure, PAS pairs with an FDR <0.05 and log10(proximal/distal PAS read count-control/RNAi) >0.2 are highlighted in red for proximal-to-distal switches and in blue for distal-to-proximal switches.

We thank Dr. Paolo Casali's laboratory (University of California, Irvine) for the generous gifts of mouse tissues and Dr. Charles Limoli's lab (University of California, Irvine) for sharing equipment. This study was supported by grants from NIH (R01GM090056) and ACS (RSG-12-186) (Y.S); an NIH grant (R01GM088342) and a junior faculty grant from the Edward Mallinckrodt Jr. Foundation (Y.X.); grants from the National Center for Research Resources (5P41RR011823-17), the National Institute of General Medical Sciences (8 P41 GM103533-17), and the National Institute on Aging (R01AG027463-04) (J.R.Y.).