Exon Junction Complexes Show a Distributional Bias toward Alternatively Spliced mRNAs and against mRNAs Coding for Ribosomal Proteins

Abstract

The exon junction complex (EJC) connects spliced mRNAs to posttranscriptional processes including RNA localization, transport, and regulated degradation. Here, we provide a comprehensive analysis of bona fide EJC binding sites across the transcriptome including all four RNA binding EJC components eIF4A3, BTZ, UPF3B, and RNPS1. Integration of these data sets permits definition of high-confidence EJC deposition sites as well as assessment of whether EJC heterogeneity drives alternative nonsense-mediated mRNA decay pathways. Notably, BTZ (MLN51 or CASC3) emerges as the EJC subunit that is almost exclusively bound to sites 20-24 nucleotides upstream of exon-exon junctions, hence defining EJC positions. By contrast, eIF4A3, UPF3B, and RNPS1 display additional RNA binding sites suggesting accompanying non-EJC functions. Finally, our data show that EJCs are largely distributed across spliced RNAs in an orthodox fashion, with two notable exceptions: an EJC deposition bias in favor of alternatively spliced transcripts and against the mRNAs that encode ribosomal proteins.

Graphical abstract

Figure 1

The Experimental HeLa Cell System Expressing Fully Functional EJC-GFP Fusion Proteins Is Suitable for iCLIP Experiments (A) Titration of doxycycline demonstrated a concentration-dependent increase of the eIF4A3-GFP fusion proteins and that a concentration of 1 ng/ml doxycycline was best suited to achieve an expression close to the endogenous level. This image shows a representative immunoblot of three biologically independent experiments. Endogenous and recombinant eIF4A3 were stained concurrently with an α-eIF4A3 antibody. (B) Representative immunoblot after siRNA treatment of three biologically independent experiments is shown. The endogenous and recombinant eIF4A3 were stained concurrently with an α-eIF4A3 antibody. (C) Schematic drawing of three SC35 mRNA isoforms adapted from Sureau et al. (2001). The grey boxes in SC35WT mRNA represent RNA regions that are spliced out in the other isoforms. The arrows show the position of the primers that were used for the amplification of the transcripts (Table S7). (D and E) Upregulation of SC35C (D) and SC35D (E) transcripts after depletion of endogenous eIF4A3 and rescue of efficient NMD upon induction of the fusion protein with doxycycline. (F) The expression of the NMD-insensitive SC35WT isoform did not change under the different conditions. The error bars represent SEM, and p values were calculated by one-way ANOVA with Dunnett's multiple comparison test (^∗∗∗p value < 0.001 with n = 3–5 independent biological experiments). (G) CoIPs show that the EJC core proteins eIF4A3 and Y14 were stably associated with UPF3B and BTZ under salt concentrations of 150 and 250 mM NaCl and disassembled at 500 mM NaCl. Therefore, 500 mM NaCl was used for the subsequent iCLIP experiments. GFP, BTZ-GFP, and UPF3B-GFP were stained concurrently with an α-GFP antibody. See also Figure S1 and Table S7.

Figure 2

The Distribution of iCLIP Peaks Validates the Predominantly Sequence-Independent Deposition of EJC Proteins at Exonic Regions (A) Relative distribution of peaks for all iCLIP data sets confirms that EJC components bind predominantly in the ORF. (B) Distribution of BTZ peaks indicates the high reproducibility of the biological replicates; horizontal lines highlight the mean of three biologically independent BTZ iCLIP replicates. (C) Enrichment of iCLIP peaks at mRNA regions calculated by dividing the number of iCLIP peaks by the cumulative length of indicated regions reveals that the ORF harbors by far the most EJC binding sites, followed by the 5′ UTR, and then the 3′ UTR regions. The error bars represent SEM of n = 3 independent biological experiments. (D–H) The higher enrichment of the splice site signal compared to motifs within the reads confirmed the predominant sequence-independent deposition of the EJC. These graphs display the distribution of 5mer motifs around the binding site of the protein (illustrated as the vertical line in the middle of the plots). (D) Average distribution of the 20 most enriched motifs (see also Table S2) relative to the binding site. (E) Average distribution of 5mers that do not contain GT in their motif. (F) Average distribution of 5mers that do contain GT in their motif. (G) The distribution of GT-containing 5mers in BTZ iCLIP experiments revealed three prominent motifs (AGGTA, GGTAA, and GTAAG). (H) These most enriched GT-containing 5mers concatenate to the canonical splice donor site AGGTAAG, which is enriched in all EJC, but not in the PTB, libraries. See also Figure S2 and Tables S1–S3.

Figure 3

Binding of All EJC Components Is Highly Enriched to Transcripts Coding for RNA Processing Proteins (A–F) The red dots show significant mRNA targets either up- or downregulated after differential analysis using edgeR controlled by the Benjamini-Hochberg procedure with an FDR <0.05. The targets below the red line provided a higher PTB signal, whereas targets above the red line exhibited a higher BTZ signal (A). The targets below the red line were used for GO enrichment analysis of PTB (B). All EJC iCLIP data sets were compared to PTB, and the targets with a log₂ fold change >0 (above red line) were analyzed for GO enrichment for BTZ (C), eIF4A3 (D), RNPS1 (E), and UPF3B (F). (G–I) Enriched EJC occupancy does not correlate with mRNA abundance. These plots display the correlation of iCLIP and RNA-seq data using the 2,194 common targets that were significantly enriched in BTZ iCLIP in both differential analyses (compared to PTB and GFP). (G and H) The RNA-seq (G) and BTZ iCLIP (H) replicates were highly reproducible. (I) Relationship between the mean count of all three RNA-seq and BTZ iCLIP libraries. The red dots highlight mRNA targets for BTZ with a log₂ fold change >3 compared to PTB in the differential analysis using edgeR and thus indicate mRNAs that are particularly strongly bound by BTZ. The specificity of these highly occupied mRNAs is demonstrated by the finding that these mRNAs were distributed across transcripts with high and low expression levels as measured by RNA-seq (CPM and R = Pearson's correlation coefficient). See also Figure S3.

Figure 4

High-Confidence EJC Binding Sites Reveal that BTZ Is an Essential EJC Component at the Canonical Deposition Site (A) Internal exons are predominantly bound by EJC components and constitutive exons display the strongest BTZ signal relative to their abundance. The average profile of reads covering exons are plotted for: (1) 5′ terminal exons; (2) 3′ terminal exons; (3) constitutive exons present in all isoforms; (4) variant exons not present in all isoforms; (5) exons with alternative acceptor sites (ALT acceptor); (6) exons with alternative donor sites (ALT donor); and (7) exons with both alternative donor and acceptor sites (ALT both) using ngsplot software (Shen et al., 2014). (B) Histogram on the left hand side shows the length of the constitutive exons and is aligned to the heatmaps showing RNA-seq and iCLIP coverage across individual exons ordered by their length. The length of the exons can be extracted from the histogram. This image confirms that most exons harbor an EJC at the 3′ end and demonstrate that the EJC signal strength is independent of exon length. The color key represents the signal strengths of the RNA-seq and iCLIP data. (C) Average coverage profiles across constitutive exons for BTZ iCLIP: (1) raw reads; (2) reads in peaks; and (3) reads in peaks that overlap with at least one of the other EJC proteins (eIF4A3, UPF3B, and RNPS1) are restricted to the 3′ end of exons. The peak detection and filtering approaches increased the BTZ signal. (D) By contrast, the average exon profiles for eIF4A3, UPF3B, and RNPS1 iCLIP binding sites that were not determined concurrently by BTZ binding sites were absent of an EJC signal close to the 3′ end of exons suggesting that non-canonical binding sites do not contain the fully assembled EJC. The profiles are plotted as read CPM mapped reads (RPM). (E) The integrated analyses over four distinct components of the EJC revealed that BTZ determines the position of fully assembled EJCs to the canonical deposition sites at 15–30 nt upstream of exon-exon junctions. Non-canonical depositions sites of EJC proteins (alone or in subcomplexes) located in other regions of the exon are less common and do not contain BTZ. See also Figure S4 and Tables S4–S6.

Figure 5

Binding of EJC Components Is Highly Enriched at Alternatively Spliced Exons in Transcripts with High EJC Occupancy and Enables Detection of Low-Abundance NMD-Sensitive mRNA Isoforms (A) We used the RNA-seq and BTZ maxima as shown in Figure 4 (see also Table S4) to calculate a log₂ fold difference for both all mRNAs and mRNAs that are highly occupied by EJCs. The iCLIP/RNA-seq ratio is enriched for alternatively spliced exons in RNAs that are highly occupied by EJCs (see Figures 3G–3I) compared to all mRNAs. (B) Genome browser view of SRSF2 (SC35) gene reveals EJC iCLIP peaks on exons corresponding to NMD-sensitive SC35C and SC35D mRNA isoforms. The NMD-insensitive SC35WT isoform is displayed in the Ensembl genes track as the upper isoform. The red box highlights the variant exon 3. (C) Candidate exon in the intronic region of the mRNA ADARB1. The track range displays CPM and was adjusted to the highest iCLIP signal obtained in the iCLIP libraries of this study in each genome browser view. The signals of the RNA-seq and literature data were not adjusted. The literature data were obtained from RIP (Singh et al., 2012) and HITS-CLIP (Saulière et al., 2012) of eIF4A3. See also Figures S4 and S6.

Figure 6

EJC Binding Is Underrepresented on mRNAs Coding for Ribosomal Proteins (A) Heatmaps of individual exons clustered by the differences between the RNA-seq and the iCLIP profile uncovered four different exon categories: (1) exons that are highly abundant and show a weak iCLIP signal; (2) exons that are expressed and harbor a corresponding EJC signal; (3) exons that are not expressed and therefore do not have an EJC signal; and (4) exons that are weakly expressed, but show enrichment of EJC binding. (B) The same clustering was performed only with reads in binding sites that were determined by at least two out of four EJC proteins. Each row of the heatmap represents one of the 123,585 constitutive exons of the human reference genome. The color key represents the signal strengths of the RNA-seq and iCLIP data. (C and D) EJC occupancy is highly diminished in mRNAs coding for ribosomal proteins. Average coverage profiles across canonical exons of ribosomal protein coding genes lack an increased EJC signal at the 3′ end of the exon for all EJC proteins (C) including published eIF4A3 data from RIP (Singh et al., 2012) and HITS-CLIP (Saulière et al., 2012) of eIF4A3 (D). The average profiles are plotted as read CPM mapped reads (RPM) calculated using peak data (C) or raw counts (D). (E) Exons of highly expressed mRNAs (>40 CPM) that do not belong to the TOP mRNA class show a higher BTZ signal analyzed by Welch two sample t test. See also Figures S4 and S7.