Matrin3 PRI motif is necessary and sufficient for PTB interaction

Matrin3 is a large nuclear protein with 847 amino acids that can bind both to DNA via two C2H2 zinc finger domains (ZF1 and ZF2) and to RNA by its tandem RNA recognition motifs (RRM1 and RRM2) (Hibino et al, 2006). A potential PRI motif, GILGPPP, is located between ZF1 and RRM1. This matches the PRI consensus (Fig​(Fig2A)2A) and is located in a disordered region, which is important for the function of short linear motifs (Dinkel et al, 2014). Moreover, the motif is absolutely conserved across 84 mammalian, avian, reptilian and amphibian species (UCSC browser, Vertebrate Multiz Alignment & Conservation, 100 Species). In order to test whether the GILGPPP motif is functional, we mutated it to GAAAPPA (mutated residues underlined) in a FLAG-tagged Matrin3 expression vector and tested the effect on PTB binding by anti-FLAG co-immunoprecipitation. As control, we used wild-type Raver1 and a mutant with all four PRI motifs mutated (Rideau et al, 2006). FLAG-tagged Matrin3 and Raver1 both co-immunoprecipitated PTB (Fig​(Fig2B).2B). Mutation of the single PRI in Matrin3 nearly eliminated PTB co-immunoprecipitation (lane 3), a more emphatic effect than mutation of the Raver1 PRI motifs (Fig​(Fig2B,2B, lane 5). We next tested whether the Matrin3 PRI is sufficient for binding to PTB. We in vitro transcribed and translated the Matrin3 and the Raver1_1 PRIs (Fig​(Fig2A)2A) fused to the bacteriophage MS2 coat protein. Both the Raver1 and Matrin3 peptides were pulled down by GST-PTB (Fig​(Fig2C,2C, lanes 1, 3). As negative controls, no binding was observed to an unrelated RNA-binding protein, GST-SXL, and MS2 alone was not pulled down by GST-PTB or GST-SXL (lane 2). The specificity of the interaction was demonstrated by mutation to alanine of the conserved leucine-3 of the PRI, which strongly impaired binding to PTB (Fig​(Fig2C,2C, lane 4). These data therefore demonstrate that the PRI motif of Matrin3 is both necessary and sufficient for interaction with PTB.

Matrin3 is a widespread regulator of alternative splicing

The interaction of Matrin3 with PTB led us to hypothesize that it may play a role in the co-regulation of some PTB-regulated alternative splicing events (ASEs). To test this hypothesis, we transfected HeLa cells with siRNAs targeting the Matrin3 mRNA and observed a > 90% decrease in the Matrin3 protein levels (Fig​(Fig3A).3A). Total RNA from knockdown and control samples was purified and analysed using Human Junction microarrays (HJAY), containing probe sets for all annotated human exons and exon–exon junctions (Llorian et al, 2010). The array data were analysed using the ASPIRE3 pipeline. Only 61 genes showed changes in RNA levels of greater than twofold, including the expected reduction of Matrin3 levels (3.7-fold; Supplementary Table S2). This suggests, in contrast to a previous report (Salton et al, 2011), that Matrin3 does not play a widespread role in stabilizing mRNAs. We did observe down-regulation of some of the previously reported mRNAs (Salton et al, 2011), but also observed alteration of alternative splicing events towards isoforms of these mRNAs with premature termination codons, which normally leads to nonsense-mediated decay (see Discussion).

Open in a separate window

Figure 3

Global splicing effects upon Matrin3 knockdown

• A Western blot probed for Matrin3 (top panel), PTB (middle panel) and actin (lower panel). Lanes 1–4 contain a twofold dilution of the control C2 sample (lane 1—12.5%, lane 2—25%, lane 3—50%, lane 4—100%). Lanes 4–7 contain equal amount of proteins, as can be confirmed by the anti-actin (lower panel), of control sample (lane 4), double-PTB and nPTB siRNA-treated sample (lane 5), Matrin3 siRNA-treated sample (lane 6) and triple knockdown of Matrin3, PTB and nPTB (lane 7). Lane 7 is from the same gel and exposure, but some lanes present in the original gel were cropped for clarity, and a black line to indicate cropping was placed.

• B Pie chart of the different categories of Matrin3-regulated alternative splicing events (ASEs), 330 cassette exons (50%), 116 alternative promoter usage (17%), 78 terminal exons (11%), 32 alternative 5′ (5%) and 34 3′ (5%) splice site and 18 intron retention (IR; 3%) events.

• C Gene ontology (GO) analysis of the Matrin3-regulated cassette exons. The x-axis represents the P-value in a logarithmic scale as shown.

• D Pie chart of the activated and repressed cassette exons by Matrin3.

• E Venn diagram of the overlap between the PTB- (blue) and Matrin3- (yellow) regulated cassette exons, showing the 813 events regulated only by PTB, 270 by Matrin3 only and the 61 that overlap.

• F RT–PCR validation of Matrin3-regulated alternative spicing events in the ST7, ACSL3, PLEKHA3, TCF12, VWA5A, PTBP2, PTBP3, C3orf17, ZMYND8, VEZT, PIGX and DMD genes. In each case, triplicates for each condition (C—control, M—Matrin3, PTB/nPTB and Matrin3/PTB/nPTB siRNA transfection samples) were analysed and exon inclusion (EI) percentage is shown beneath the corresponding lane, along with the standard deviation (s.d.).

Source data are available online for this figure

Next, we examined the potential role of Matrin3 in regulating alternative splicing. Significant changes in splicing were predicted by ASPIRE using a threshold of |dIrank| > 1 (Supplementary Table S3), which has previously been shown to produce a validation rate of > 80% (König et al, 2010; Wang et al, 2010). This identified 667 ASEs, half of which were cassette exons (n = 331; 50%; Fig​Fig3B).3B). Of the Matrin3-regulated cassette exons, 75% showed increased inclusion upon Matrin3 knockdown, indicating that Matrin3 represses inclusion of these exons (Fig​(Fig3D).3D). Notably, the degree of confidence in the changes observed in splicing of the 25% Matrin3-activated cassette exons was lower when compared to the Matrin3-repressed ones (Supplementary Fig S1).

We next examined the cassette exons that may be jointly regulated by Matrin3 and PTB, using the HJAY data set produced upon knockdown of PTB and PTBP2 (Llorian et al, 2010). The double knockdown is essential as upon PTB knockdown, its neuronal paralogue PTBP2 is upregulated and can partially compensate for loss of PTB (Boutz et al, 2007; Makeyev et al, 2007; Spellman et al, 2007). Only 61 (18.4%) of the 331 Matrin3-regulated cassette exons were also regulated by PTB (Fig​(Fig3E).3E). While the number of co-regulated cassette exons is 2.2-fold greater than expected by chance (expected 27.4, P = 5.5e⁻¹⁰, hypergeometric test), the majority of the Matrin3-regulated ASEs are PTB independent.

We validated a number of the cassette exon events predicted to be regulated by Matrin3 and PTB by knockdown of Matrin3 or PTBP1/PTBP2. We also tested the effects of combined knockdown of Matrin3/PTBP1/PTBP2 (Fig​(Fig3A).3A). RT–PCR was carried out using primers in flanking constitutive exons and the percentage exon inclusion determined (Fig​(Fig3F).3F). Four different classes of events were observed, depending on their response to Matrin3 and PTB knockdown: Matrin3 repressed, PTB independent (ST7 exon 11, ACSL3 exon 3 and PLEKHA3 exon 4); Matrin3 activated, PTB repressed (TCF12 exon 18, VWA5A exon 2, PTBP2 exon 10 and PTBP3 exon 2); Matrin3 repressed, PTB activated (C3orf17 exon 2, ZMYND8 exon 33 and VEZT exon 11); and repressed by both Matrin3 and PTB (PIGX exon 3 and DMD exon 78). In the cases where Matrin3 and PTB activities were opposed, knockdown of the repressor protein had a larger effect and tended to be dominant over the activator. In some cases, knockdown of the activator had no effect in the absence of the repressor (e.g. VEZT exon 11 and PTBP2 exon 10), suggesting that that the sole function of the activator is to antagonize the repressor.

Properties of Matrin3-regulated exons

In order to assess whether the exons regulated by Matrin3 possess any specific splicing features, we examined 5′ and 3′ splice sites, branch points, pyrimidine tracts, and nucleotide composition (Supplementary Fig S3) and flanking intron lengths (Fig​(Fig4A).4A). Few significant differences were observed compared to a control set of annotated cassette exons unaffected by knockdown of Matrin3 (or PTB) knockdown. One striking difference was that the introns flanking Matrin3 repressed exons are on average 1 kb longer than introns flanking Matrin3-activated, PTB-repressed, PTB-activated or control exons (Fig​(Fig4A4A).

Open in a separate window

Figure 4

Bioinformatic analysis of Matrin3-regulated splicing events

• A Intron lengths flanking exons regulated by Matrin3, PTB or control exons. Kruskal–Wallis rank-sum test was used to test for significant changes (Matr3 P-value = 3.947e⁻¹⁵, PTB P-value = 0.3896). ***P < 0.001.

• B Diagram of Matrin3-repressed cassette exons with enriched RBP motifs for human proteins (Ray et al, 2013) shown above and the number of enriched pentamers shown below in each of 7 locations. The RBP motifs are shown with their consensus binding site logo (Ray et al, 2013) and the respective motif enrichment score (odds ratio). The significant enriched k-mers are shown in Supplementary Table S4, and all enriched RBP motifs (multiple species) are shown in Supplementary Table S5.

We next looked for enrichment of pentamer sequence motifs associated with Matrin3-regulated exons, compared to control unregulated cassette exons, across seven transcript locations (cassette exons, flanking constitutive exons, 5′ and 3′ end of each flanking intron). Numerous motifs were enriched (FDR < 0.05) in the introns flanking Matrin3-repressed exons, but none within exons or in any location associated with Matrin3-activated exons (Fig​(Fig4B,4B, Supplementary Fig S4, Supplementary Table S4). Motifs associated with Matrin3-repressed exons were heterogeneous and included a number of pyrimidine motifs associated with PTB (e.g. TTCTT, TCTTT). The enrichment was also observed using a control set consisting of exons including PTB-regulated, Matrin3-independent exons. Most of the remaining motifs had high pyrimidine content, with one or two interrupting purines; more than half of the motifs immediately flanking Matrin3 repressed exons had a single purine. Individual analyses of RNA binding by Matrin3 have not revealed a clear consensus sequence (Hibino et al, 2006; Salton et al, 2011; Yamazaki et al, 2014). However, Matrin3 was one of 207 RBPs whose optimal sequence was determined by the RNA-compete array-based selection (Ray et al, 2013). We therefore used heptamer position frequency matrices to look for enrichment of RNA-compete motifs. Once again, enriched motifs were found only in the introns flanking Matrin3-repressed exons, and in no locations associated with Matrin3 activation (Fig​(Fig4B,4B, Supplementary Table S5). Strikingly, the Matrin3 motif was the only enriched RNA-compete motif upstream of repressed exons and was one of only four motifs on the immediate downstream side. Other enriched motifs included PTBP1, PCBP1, HuR and ZCRB1; this may be either due to a consequence of these proteins functioning as a co-regulator of Matrin3 or simply due to the similarity between the sequences of the binding sites. Taken together, the k-mer and RNA-compete motif enrichments suggest that Matrin3 might bind directly to the longer introns flanking repressed exons, but that activation by Matrin3 might be indirect.

Matrin3 binds widely around repressed exons

To directly address the relationship between Matrin3 splicing activity and RNA binding, we carried out crosslinking and immunoprecipitation (iCLIP) in HeLa cells (König et al, 2010). We obtained a total of 3,496,801 cDNA reads after collapsing PCR duplicates that mapped uniquely onto the genome. A Matrin3 splicing map was generated using the Matrin3 iCLIP binding microarray data sets (Fig​(Fig5).5). Matrin3 binding was elevated in long intronic regions immediately flanking repressed exons (Fig​(Fig5A,5A, blue). In contrast, the flanking constitutive exons, their immediate intron flanks and all regions associated with Matrin-activated exons (Fig​(Fig5A,5A, red) showed binding levels only slightly elevated above control cassette exons (Fig 5A, grey). This differential observed binding agrees well with the motif enrichments (Fig​(Fig4B,4B, Supplementary Tables S4 and S5). In contrast to many other splicing regulators (Licatalosi & Darnell, 2010; Witten & Ule, 2011), Matrin3 binding was uniformly elevated within 500 nt of repressed exons, with no discrete peaks (Fig​(Fig5A).5A). This uniform binding was not simply a result of higher steady-state levels of these RNA regions, because TIA1 (Wang et al, 2010) and U2AF65 (Zarnack et al, 2013) iCLIP tags showed a similar density on control and Matrin3-regulated exons (Supplementary Fig S2A). The uniform binding was also observed around individual exons (e.g. ZMYND8 and ADAR1B, Supplementary Fig S2D) and so did not result from averaging across large numbers of introns with discreet peaks at different locations. We also noted that elevated Matrin3 binding extended into the repressed exons (Fig​(Fig5A),5A), even though no enriched motifs had been observed in this location (Fig​(Fig4B).4B). A possible explanation for the extended region of elevated binding including the exon is that binding initiates at high affinity specific motifs in the flanking introns and then extends by Matrin3 oligomerization with less specific RNA binding (see Discussion). We did not observe a clear correlation between tag density and degree of splicing change upon Matrin3 knockdown, which may be related to lack of saturation of the iCLIP library. Nevertheless, the density of Matrin3 tags around repressed compared to control exons was highly significant (P < 0.0001, χ² test).

Open in a separate window

Figure 5

Matrin3 and PTB splicing maps

• A Matrin3 crosslinking in Matrin3-regulated pre-mRNAs where position of crosslinked nucleotides was mapped onto the regulated exon and the 500 nucleotides upstream and downstream of its 3′ss and 5′ss, respectively, the upstream flanking exon and 500 nucleotides downstream of its 5′ss and the downstream flanking exon with 500 nucleotides upstream of its 5′ss. The iCLIP tags were mapped onto silenced ASE (blue, n = 421), enhanced ASE (red, n = 187) and to control ASE (grey, n = 16,242), and percentage of occupancy is plotted.

• B Same as in (A) but using iCLIP tags obtained from PTB iCLIP and mapped onto PTB/nPTB-regulated ASE, silenced ASE (blue, n = 729), enhanced (red, n = 820) and control ASE (grey, n = 14,599).

• C Matrin3 crosslinked nucleotides mapped onto PTB-regulated ASE.

• D PTB crosslinked nucleotides mapped onto Matrin3-regulated ASE.

In view of our initial identification of Matrin3 as a PTB-interacting protein (Figs​(Figs11 and ​and2),2), the small overlap of target splicing events was somewhat surprising (Fig​(Fig3).3). We therefore explored the relationship of Matrin3 and PTB binding to each other's target RNAs. We carried out iCLIP for PTB in HeLa cells and used the previous HJAY data set for PTB/nPTB knockdown (Llorian et al, 2010). We obtained a total of 5,981,600 cDNA reads that mapped uniquely onto the genome. The resultant PTB splicing map showed the characteristic peak within 100 nt upstream of PTB-repressed exons, while PTB-activated exons showed highest levels of binding on the downstream side of activated exons and within the upstream constitutive exon (Fig​(Fig5B)5B) (Xue et al, 2009; Llorian et al, 2010). Mapping Matrin3 iCLIP tags onto PTB-regulated exons showed binding only slightly above control cassette exons (Fig​(Fig5C).5C). PTB had a higher level of occupancy upstream of Matrin3-activated exons, even though Matrin3 itself does not appear to bind in this location (Fig​(Fig5A5A and D). It is possible that for PTB-repressed, Matrin3-activated exons, such as PTBP2 exon 10 and PTBP3 exon 2, Matrin3 could antagonize PTB activity without binding RNA. How Matrin3 would activate PTB-independent exons without binding RNA remains unclear. PTB iCLIP tags showed a marked enrichment across the ± 500 nt intron flanks of Matrin-repressed exons, very similar to the pattern of Matrin3 binding (Fig​(Fig5D),5D), despite the fact that 80% of these exons are PTB independent. Indeed, PTB iCLIP tags could be observed in individual introns flanking Matrin3-repressed, PTB-independent exons (e.g. ADAR1B, Fig​Fig2D).2D). The lack of positional binding in vicinity to the 3′ splice site of Matrin3-repressed exons clearly distinguishes this non-functional mode of PTB binding from authentic PTB-regulated target sites (Fig​(Fig5B5B and D). We thus exclude that this binding reflects co-regulated events. Instead, this strongly suggests that PTB is recruited by Matrin3 to the introns flanking Matrin3-repressed exons even if it does not contribute to exon repression.

Mechanism of Matrin regulation of splicing

Matrin3 function as a direct splicing repressor is supported by its observed binding around repressed exons (Fig​(Fig5)5) along with the enrichment in flanking intron segments of optimal binding motifs for its RRM motifs (Fig​(Fig4B).4B). Consistent with this, it requires intact RNA-binding domains, but not DNA-binding domains, for its splicing repressor activity (Fig​(Fig6).6). Nevertheless, further evidence of its direct mode of action could be provided by in vitro analyses of its binding to regulated RNAs, demonstrating that specific binding sites are required for its splicing repressor activity, and more detailed analysis of the roles of the individual RRM domains. Unexpectedly, deletion of the ZF1 DNA-binding domain actually enhanced the activity of transfected Matrin3 (Fig​(Fig6).6). One possible explanation for this observation is that deletion of ZF1 affects the distribution of Matrin3 between pools that are active for splicing regulation or that are localized elsewhere.

The majority of Matrin3 target exons are unaffected by PTB (Fig​(Fig3),3), even though we initially identified Matrin3 via its interaction with PTB RRM2, part of the minimal repressor domain of PTB (Figs​(Figs11 and ​and2)2) (Robinson & Smith, 2006). Indeed, we observed enhanced recruitment of PTB around Matrin3-repressed exons and, importantly, the distribution of PTB resembled the broad distribution of Matrin3 around its own targets (Fig​(Fig5A5A and D) and lacked the distinctive peak of PTB binding observed upstream of PTB-repressed exons (Fig​(Fig5B).5B). This suggests that PTB can be recruited to Matrin3-regulated exons, even when it is not functionally required, as in the ST7 exon 11 (Figs​(Figs3F3F and ​and6).6). However, we observed that repression of ST7 exon 11 depended absolutely upon the Matrin3 PRI motif (Fig​(Fig6),6), which suggests that Matrin3 repressor activity at this exon might require interaction with proteins other than PTB via its PRI. The binding of Matrin3 across extensive regions flanking and within repressed exons (Fig​(Fig5A)5A) is suggestive of a mechanism of initial binding to specific sites followed by spreading. This is consistent with the observed enrichment of k-mers and the RNA-compete motif exclusively in the immediate 250 nucleotides of the flanking introns (Fig​(Fig4),4), and with its reported propensity for self-association (Zeitz et al, 2009). Similar models were originally suggested for repression by PTB (Wagner & Garcia-Blanco, 2001) and hnRNPA1 (Zhu et al, 2001), but analysis of numerous model systems has shown that this is not a common mechanism for these proteins (e.g. Cherny et al, 2010), and their splicing maps show distinct peaks of enriched binding (Xue et al, 2009; Llorian et al, 2010; Huelga et al, 2012). In contrast, the Matrin3 splicing map (Fig​(Fig5A)5A) is consistent with a general mechanism of action in which initial binding of Matrin3 at specific sites is followed by propagative binding across a wide region of RNA, leading to repression of the targeted exon. This model is amenable to testing by various methods including single molecule analysis (Cherny et al, 2010).

How Matrin3 might promote exon inclusion is less clear. Splicing maps indicate that binding around activated exons is not much above background (Fig​(Fig5A)5A) and no motifs were enriched (Fig​(Fig4).4). Nevertheless, in the ABI2 minigene assay, the RRMs were required, along with the PRI motif (Fig​(Fig6).6). One explanation could be that activated exons are mainly indirect targets, and changes in their splicing result from the primary actions of Matrin3, which require RNA binding. Another possibility, in cases where Matrin3 activation directly opposes PTB repression, is that interaction with Matrin3 might antagonize PTB activity. Although Matrin3 binding is low around Matrin3-activated exons (Fig​(Fig5A),5A), PTB binding shows an upstream peak (Fig​(Fig5B)5B) consistent with PTB repression. A notable example of this is the Matrin3-activated nPTB exon 10 (Fig​(Fig3F),3F), a well-known target of PTB repression (Boutz et al, 2007; Makeyev et al, 2007; Spellman et al, 2007).

DNA constructs

GST-PTB1 and GST-SXL used for pull-down of PRI motifs have been described previously (Rideau et al, 2006). GST-RRM2 and GST-RRM2 Y247Q were PCR-amplified and cloned into the EcoRI site in pGEX3 as described (Joshi et al, 2011). To generate the FLAG-Matrin3 PRI-MS2, we cloned into the AvrII and MluI sites pre-annealed oligos encoding the indicated protein sequence. FLAG-MS2, FLAG-Raver1 full-length wild-type and mutant, FLAG-Raver1 PRI have all been described previously (Rideau et al, 2006). Full-length Matrin3 was PCR-amplified from human cDNA and cloned into the MluI sites of a pCI-FLAG vector (Promega). Deletion mutants were generated by divergent PCR and mPRI by site-directed mutagenesis (Promega). The splicing reporters were constructed by PCR amplifying the regulated exon and the flanking region and cloning them into the Asp718I and EcoRV sites of a GFP expression cassette where an intron has been inserted into the second codon. The cloning sites are localized in the middle of this intron so will generate three exon splicing reporters upon cloning (Wollerton et al, 2004). All DNA constructs were confirmed by sequencing.

GST expression, purification and pull-down

Pull-down of nuclear proteins using GST-RRM2 and GST-RRM2 Y247Q was carried out as follows: 2 μg of purified recombinant protein was pre-bound to glutathione sepharose 4B beads for 1 h. To each GST protein, 300 μl of HeLa nuclear extract was added, together with RNase A to a final concentration of 5 μg/ml. The mixture was incubated for 2 h at 4°C rotating, followed by extensive washes and elution using 30 μl of protein loading buffer. Five microlitre was used for silverstain analysis, and the remainder of the pull-down was run on a SDS–PAGE followed by mass spectrometry of the entire section of the lane above the recombinant protein (see Fig​Fig1A).1A). Expression and purification of GST-RRM proteins has been described elsewhere (Joshi et al, 2011). Pull-downs of in vitro translated protein by GST-PTB were carried out as described (Rideau et al, 2006) using GST-SXL as negative controls. GST-SXL protein was produced from the plasmid pGEX CS NR SXL XW II, which was a kind gift from J. Valcárcel, Centre de Regulació Genòmica. RNase treatment of pull-downs was carried out in two steps: the in vitro translations were terminated with a 15-min incubation at 30°C with 25 μg/ml RNase A, and RNase A was added at 0.5 μg/ml to the pull-down pre-incubation for 3 h at 4°C.

Human Junction microarray experiments and validation

RNA from four biological replicates of control or Matrin3 siRNA-transfected HeLa cells was isolated using TRI reagent (Life Technologies). Total RNA was hybridized to Human Affymetrix Exon-Junction Array (HJAY) by Genecore (EMBL). The microarray data were analysed using ASPIRE (Analysis of SPlicing Isoform REciprocity) 3.0 (König et al, 2010; Wang et al, 2010). It analyses signal in reciprocal probe sets to monitor changes in alternative splicing events, from which we applied a threshold of |dIrank| ≥ 1 and obtained 667 alternative splicing events (Supplementary Table S3). Matrin3, PTB/nPTB and triple Matrin3/PTB/nPTB were carried out and cDNA was produced from extracted RNA using oligo dT. The splicing pattern of each ASE was analysed by RT–PCR (for primer sequence, see Supplementary Materials and Methods).

Bioinformatic analysis of Matrin3 regulated ASE

Analysis of enriched functional gene categories was carried out using PANTHER (Mi et al, 2013). Analysis of Matr3 intron length was done using ASPIRE [CS1] annotation in R version 3.0.1 and the ggplot2 and pgirmess packages. Since some alternative splice events have 0-length introns, intron length was analysed for alternative cassette exons only. For statistic analysis, data were tested with Kruskal–Wallis rank-sum test before doing multiple comparison testing. Motif enrichments were calculated using 100 bp of the flanking exons and the complete sequence of the cassette exons. For introns, we used maximum intronic flanks of 250 nt, removing SS context to avoid BP, SS and PPT signals, 9 nt at donor side and 30 nt at the acceptor side (Bland et al, 2010), retrieving introns with a minimum length of 60 nt. We assess the enrichment of 5-mers and RNA-compete motif matches (Ray et al, 2013) with the following procedure implemented in a custom PHP script: Given a sample set S of N sequences and a control set S⁽⁰⁾ of N⁽⁰⁾ sequences, the number of times n_a,i that each motif a appeared in each sequence i was calculated. Likewise, for the control set, the number of occurrences of each motif a per sequence i was also calculated. The expected density of each motif was also calculated as the ratio between the total number of occurrences in the control set over the total sequence length of the control set:

where l_i is the length of each sequence in the control set. For each sequence i in the sample set each motif a, it was recorded whether the observed motif count (n_a,i) is greater than the calculated expected count ():

Similarly, for the counts in the control set:

The sum of the d_i,a values over the sequences i represents the number of sequences for which the motif a has an observed count greater than expected. Thus, for each motif, the odds ratio (motif score) and corresponding P-value were the motif a has an obtained by performing a Fisher's test (one-tailed) with these sums counts for the sample set and the control set:

5-mer a	More than expected	Less than expected
S
S (0)

Open in a separate window

Statistically over-represented motifs were selected based on the Benjamini's and Hochberg's false discovery rate multiple test-corrected P-value (BH-FDR < 0.05). Statistical tests were performed and graphics were generated with R Development Core Team (2011). Additional scripts were written in PHP and Awk, and sequence logos were generated with seqlogo (Crooks et al, 2004).

CLIP and splicing maps

iCLIP experiments were performed for Matr3 and PTB using antibodies targeting the endogenous protein, GTX47279 (GeneTex) and polyclonal anti-PTB serum (Spellman et al, 2007). Exponentially growing HeLa cells were washed once with PBS and cross-linked at 0.15 mJ/cm² with a Stratalinker 2400 equipped with 254 nm light bulbs. Retrieval of protein-bound RNAs and preparation of Illumina-compatible DNA libraries were done as described in Huppertz et al (2014). To compute RNAmaps of Matrin3 and PTB binding on exon–intron boundaries, we assessed the positioning of cross-link sites. Cross-linked nucleotides are defined as the nucleotide upstream of mapped iCLIP cDNA tags as described before (König et al, 2010). For each position within the RNAmap, the number of cross-link nucleotides was counted as 1 if one or more cDNA tags matched the position, and then summed across all splice events. The summed cross-link count was divided by the number of splice events and plotted in 10 nucleotide bins. Thus, the resulting occurrence value reflects the number of exons with cross-linked nucleotides within the 10-nt window.

We thank Nejc Haberman (UCL) for assisting in the generation of the splicing maps and preparation of the iCLIP data. This work was supported by Wellcome Trust programme grants to CWJS (077877 and 092900) and by grants to EE and NB BIO2011-23920 and RNAREG (CSD2009-00080) from the Spanish Government and by the Sandra Ibarra Foundation for Cancer (FSI2013). JA was supported by a Boehringer Ingelheim Fonds studentship.