SRSF1 suppresses selection of intron-distal 5′ splice site of DOK7 intron 4 to generate functional full-length Dok-7 protein

Dok-7 is a non-catalytic adaptor protein that facilitates agrin-induced clustering of acetylcholine receptors (AChR) at the neuromuscular junction. Alternative selection of 5′ splice sites (SSs) of DOK7 intron 4 generates canonical and frame-shifted transcripts. We found that the canonical full-length Dok-7 enhanced AChR clustering, whereas the truncated Dok-7 did not. We identified a splicing cis-element close to the 3′ end of exon 4 by block-scanning mutagenesis. RNA affinity purification and mass spectrometry revealed that SRSF1 binds to the cis-element. Knocking down of SRSF1 enhanced selection of the intron-distal 5′ SS of DOK7 intron 4, whereas MS2-mediated artificial tethering of SRSF1 to the identified cis-element suppressed it. Isolation of an early spliceosomal complex revealed that SRSF1 inhibited association of U1 snRNP to the intron-distal 5′ SS, and rather enhanced association of U1 snRNP to the intron-proximal 5′ SS, which led to upregulation of the canonical DOK7 transcript. Integrated global analysis of CLIP-seq and RNA-seq also indicated that binding of SRSF1 immediately upstream to two competing 5′ SSs suppresses selection of the intron-distal 5′ SS in hundreds of human genes. We demonstrate that SRSF1 critically regulates alternative selection of adjacently placed 5′ SSs by modulating binding of U1 snRNP.


DOK7 intron 4 is alternatively spliced at 5′ SS. Since annotation databases indicate existence of two 5′
SSs in human DOK7 intron 4, we initially examined differential selections of these two sites by RT-PCR of total RNA extracted from various human tissues and cell lines ( Fig. 1b and c). We observed abundant expressions of DOK7 transcripts in the skeletal muscle, brain and, heart (Fig. 1b) in agreement with a previous report 23 , and found that the intron-distal 5′ SS was selected in nearly a quarter of DOK7 transcripts in these three tissues. Similar alternative selection of the intron-distal 5′ SS of DOK7 intron 4 was observed in immortalized human myogenic KD3 cells and HeLa cells (Fig. 1c). In contrast, DOK7 intron 4 was constitutively spliced in the smooth muscle, liver, and spleen (Fig. 1b).
Selection of the intron-distal 5′ SS of DOK7 intron 4 generates a non-functional Dok-7 lacking AChR clustering activity. Selection of the intron-proximal 5′ SS generates the canonical Dok-7 protein (isoform 1), which is essential for enhancing clustering of AChRs, whereas selection of the intron-distal 5′ SS causes a frame-shift in the middle of DOK7 and generates a truncated Dok-7 protein (isoform 2), which partially disrupts the PTB domain and lacks two tyrosine residues targeted by the SH2 domain (Fig. 1a). To examine whether isoform 2 retains a biological function to enhance AChR clustering, we made two constructs expressing isoform 1 and isoform 2 fused with EGFP, and overexpressed each of them in C2C12 myoblasts. After myotube differentiation of C2C12 cells, agrin was added to the culture medium, and cells were stained with α-bungarotoxin to observe AChR clusters. We found that overexpression of isoform 1 significantly enhanced AChR clustering, as previously reported 22 , while isoform 2 has a no effect on AChR clustering (Fig. 2a). Consistently, overexpression of isoform 1, but not of isoform 2, increased the number, length, and area of AChR clusters (Fig. 2b). Phosphorylation of MuSK was markedly induced by overexpression of isoform 1, as previously reported 23 , whereas induction of MuSK phosphorylation was insufficient with isoform 2 (Fig. 2c). These results suggest that selection of the intron-distal 5′ SS results in production of a non-functional Dok-7 retaining minimal MuSK phosphorylation activity with diminished AChR clustering activity.
Construction of minigenes to characterize two competitive 5′ SSs. To dissect the underlying mechanisms of alternative 5′ SS selection of DOK7 intron 4, we constructed a human DOK7 minigene spanning exons 3 to 5 in pcDNA3.1+ mammalian expression vector (Fig. 3a). As expected, RT-PCR analysis of HeLa cells transfected with this minigene demonstrated a similar splicing pattern to that we observed with endogenous transcripts (Fig. 3a, right panel).
Several algorithms efficiently predict strength of 5′ SS 10,27 . The strength of the intron-proximal 5′ SS (SD score −2.646 and MaxEntScan::score5ss 8.59) is higher than that of the intron-distal 5′ SS (SD score −3.084, MaxEntScan::score5ss 6.34). We made two mutant constructs, Mut-1 and Mut-2, in which splicing strengths of the intron-distal and intron-proximal 5′ SS, respectively, were drastically weakened, while a GT dinucleotide was retained. SD-scores of the mutated 5′ SS in Mut-1 and Mut-2 were reduced to −5.277 and −4.976, respectively. Similarly, MaxEntScan::score5ss of the mutated 5′ SS in Mut-1 and Mut-2 were reduced to −0.30 and −9.13, respectively. RT-PCR analysis of HeLa cells transfected with these constructs showed that the intact 5′ SSs were exclusively selected in both constructs (Fig. 3b). These results suggest that the two 5′ SSs in DOK7 intron 4 are competing for splicing each other, although the intron-proximal 5′ SS is preferentially selected.
Identification of a splicing regulatory cis-element that modulates alternative selection of 5′ SS of DOK7 intron 4. Splicing cis-elements around 5′ SSs play a crucial role in the selection of alternative 5′ SSs 28 .
To identify splicing regulatory cis-element(s) that regulate the selection of 5′ SS of DOK7 intron 4, we scanned the entire exon 4 by substituting 12 blocks excluding the first three nucleotides and the last fourteen nucleotides, which contained the intron-distal 5′ SS. We sequentially introduced a 15-nucleotide heterologous sequence block (5′-TCAGTATGACTCTCA-3′) into the first 11 blocks (Blocks-1 to -11) and a 19-nucleotide heterologous sequence block (5′-TCAGTATGACTCTCAGTAT-3′) for Block-12 of the minigene (Fig. 3c). The introduced sequence blocks were previously reported to have no effect on splicing 29,30 . These constructs were transfected into HeLa cells, and selection of 5′ SSs was examined by RT-PCR. Disruption of Block-12 prominently enhanced selection of the intron-distal 5′ SS (Fig. 3c, lower panel). We also introduced the 15-nucleotide splicing-neutral sequence block into the last 2 blocks (Blocks-13 and -14 in Supplementary Fig. S1). We found that disruption of Block-14, but not Block-13, enhanced selection of the intron-distal 5′ SS ( Supplementary Fig. S1). These results suggest presence of a splicing regulatory cis-element in Block-12. To characterize the minimal essential sequences in Block-12, we further mutated a segment of six or seven nucleotides in Block-12 (Mut-3, -4, and -5 in Fig. 3d). We confirmed that the mutated segments did not de novo gain binding of an RNA-binding protein according to SpliceAid2, which is a web service program to predict the binding sites of RNA-binding proteins for a given RNA sequence based on a database derived from experimentally proven RNA-binding protein-recognition sites 31 . We found that substitution of the 3′ third of Block-12 (Mut-5), but not the 5′ or the middle thirds (Mut-3 or Mut-4), altered the 5′ SS selection (Fig. 3d). Thus, the Mut-5 region in Block-12 close to the 3′ end of exon 4 harbors a critical splicing regulatory cis-element.

SRSF1 binds to the identified splicing regulatory cis-element. A lot of splicing factors including
hnRNP and SR proteins regulate selection of alternative 5′ SS through recognition of nearby cis-elements 12,18,19 .
To identify a trans-acting factor that binds to the identified cis-element, we performed an RNA affinity purification assay using biotinylated RNA probe including the wild-type Block-12 and HeLa nuclear extract (Fig. 4a). Our analysis identified one distinct band of ~30 kDa that was associated with an RNA probe for the wild-type Block-12 (Wt), but not for the nucleotide-substituted mutant (Mut-5) or the deletion mutant (ΔMut-5) (Fig. 4b). Mass spectrometry analysis of the excised band disclosed that the identified band was SRSF1, which was also confirmed by immunoblotting using antibody against SRSF1 (Fig. 4c). Indeed, the highest ESEfinder score 32, 33 of SRSF1 was 1.74 at GGGACCA in the Wt probe, whereas it was reduced to −3.77 at AGAATCA in the Mut-5 probe. Real-time RT-PCR of SRSF1 mRNA in various human tissues revealed that SRSF1 mRNA is abundantly expressed in smooth muscle, liver and spleen ( Supplementary Fig. S2), where the selection of intron-distal 5′ splice site of DOK7 intron 4 is suppressed (Fig. 1b). In contrast, the expression levels of SRSF1 mRNA are low in skeletal muscle, brain and heart ( Supplementary Fig. S2), where the selection of intron-distal 5′splice site of DOK7 intron 4 is facilitated (Fig. 1b), suggesting that SRSF1 expression is inversely correlated with the selection of intron-distal 5′ SSs of DOK7 intron 4. Coomassie blue staining of RNA affinity-purified products using HeLa nuclear extract with the indicated biotinylated RNA probes. A single protein band at ~30 kDa (black arrow) is associated with Wt probe, but not with Mut-5 or ΔMut-5 probe. Mass spectrometry analysis revealed that the identity of this protein is SRSF1. The other bands commonly observed in the Wt, Mut-5 and ΔMut-5 RNA probes are repeatedly identified in our RNA affinity purification analyses 43,45,55 . Mass spectrometry analysis of these bands revealed that none of the identified proteins carry an RNA-recognition motif, suggesting that these bands are likely due to non-specific binding of proteins to the streptavidin-sepharose beads. NuEx, HeLa nuclear extract. Binding of SRSF1 to the cis-element activates the intron-distal 5′ SS and suppresses the intron-proximal 5′ SS of DOK7 intron 4. We next asked whether binding of SRSF1 to the identified cis-element indeed regulates alternative 5′ SS selection of DOK7 intron 4 in cellulo. We knocked down endogenous SRSF1 mRNA in HeLa cells with two different siRNAs against human SRSF1, and found that down regulation of SRSF1 increased usage of the intron-distal 5′ SS in the minigene transcripts as well as in endogenous DOK7 transcripts (Fig. 4d).
To further demonstrate that direct association of SRSF1 with the identified cis-element modulates alternative 5′ SS selection, we artificially tethered SRSF1 to the cis-element using MS2-mediated artificial tethering system 34 . First, we made pcDNA-DOK7-MS2 minigene, in which the wild-type Block-12 was replaced with the MS2-binding hairpin sequence (Fig. 4e). We also made a cDNA construct expressing SRSF1 fused with the MS2 coat protein (SRSF1-MS2) to artificially tether SRSF1 to the MS2-binding hairpin sequence in pcDNA-DOK7-MS2. Western blotting showed efficient expression of SRSF1-MS2 protein in transfected HeLa cells (Fig. 4f). We observed that replacement of the cis-element with MS2 hairpin compromised usage of the intron-proximal 5′ SS (Fig. 4g, lane 2), and that tethering of SRSF1-MS2 specifically restored usage of the intron-proximal 5′ SS (Fig. 4g, lane 4). These results indicate that binding of SRSF1 to immediate upstream of the intron-distal 5′ SS shifts the splice site selection from the intron-distal 5′ SS to the intron-proximal 5′ SS.
Binding of SRSF1 to the cis-element suppresses assembly of U1 snRNP on the intron-distal 5′ SS. To dissect the underlying mechanisms of how binding of SRSF1 immediately upstream to the intron-distal 5′ SSs regulates alternative 5′ SS selection, we examined the effect of SRSF1 on the assembly of U1 snRNP to either of these 5′ SSs. We first made two 3 x MS2-attached RNA probes ( Fig. 5a and Supplementary Fig. S3); Probe-1 contained the intron-distal 5′ SS with the SRSF1-binding site, whereas probe-2 contained the intron-distal 5′ SS with a disrupted SRSF1-binding site. As a control, we used the 3 x MS2-attached human β-globin gene 35 (Fig. 5a and Supplementary Fig. S3). We added these 3 x MS2-attached RNA probes to the splicing-competent HeLa nuclear extract to make an early spliceosome assemble on the probe. Spliceosome on each probe was isolated using the MS2 coat protein-coated beads, and Western blot analysis was performed as previously described 35 . We found that association of U1 snRNP to probe-1, but not to probe-2, was markedly reduced compared to the control probe (Fig. 5b, lanes 2 and 3), suggesting that binding of SRSF1 to the cis-element has a suppressive effect on the assembly of U1 snRNP on the adjacent intron-distal 5′ SS. We next analyzed the effect of SRSF1 on the assembly of U1 snRNP on the intron-proximal 5′ SS. We made two additional 3 x MS2-attached RNA probes; Probe-3 contained the intron-proximal 5′ SS with the SRSF1-binding site, and probe-4 contained the intron-proximal 5′ SS with a disrupted SRSF1-binding site ( Fig. 5a and Supplementary Fig. S3). The intron-distal 5′ SS was mutated in both probes-3 and -4. Analysis of the associated complexes on these probes showed that the assembly of U1 snRNP on the intron-proximal 5′ SS was not affected by the disruption of the SRSF1-binding site (Fig. 5c), which was in contrast to the effect on the intron-distal 5′ SS (Fig. 5b).
We next made additional minigenes, where the intron-distal 5′ SSs was replaced with the intron-proximal 5′ SS (Mut-6 and Mut-7 in Fig. 5d). Substitution of the intron-proximal 5′ SS (Mut-6) for the intron-distal 5′ SS (Wt) compromised the suppressive effect of SRSF1 (Fig. 5d, lanes 1 and 2). This was likely due to a stronger splicing signal of the intron-proximal 5′ SS than that of the intron-distal 5′ SS. When the SRSF1-binding site was disrupted (Mut-5 and Mut-7), selection of the upstream 5′ SS was more enhanced (Fig. 5d, lanes 3 and 4). This indicates that SRSF1 is able to suppress the selection of an adjacent 5′ SS independent of its sequence context.

Binding of SRSF1 to a cis-element immediately upstream to tandem 5′ SSs suppresses selection of the intron-distal 5′ SS in other genes.
We have shown that binding of SRSF1 immediately upstream to tandem 5′ SSs suppresses selection of the adjacent intron-distal 5′ SS in the human DOK7 gene.
To examine whether what we observed with DOK7 intron 4 is applicable to other genes in the human genome, we analyzed RNA-seq of SRSF1-knocked down HeLa cells (GSE26463) and CLIP-seq of SRSF1 in HeLa cells (GSE71096), which were deposited in the GEO database 36,37 . Splicing analysis of RNA-seq with MISO 38 detected 1445 and 427 alternative 5′ splicing events, in which the intron-distal 5′ SS and the intron-proximal 5′ SS were selected by knockdown of SRSF1, respectively (Fig. 5e). SRSF1-regulated alternative splicing of DOK7 intron 4 is similar to the 1445 alternative 5′ splicing events, in which the intron-distal 5′ SS was selected by SRSF1 knockdown. We analyzed the distribution of SRSF1-CLIP tags around these alternative 5′ SSs. We found that SRSF1 clustered immediately upstream to the intron-distal 5′ SS in 1445 exons (double-headed arrow in Fig. 5e, upper left panel), which was selected by SRSF1-knockdown. In contrast, no noticeable SRSF1 cluster was observed around the intron-proximal 5′ SS in these 1445 exons (Fig. 5e, upper right panel). To further dissect SRSF1-CLIP tag coverage on the 1445 exons at 50-nt resolution, we divided 400 nt region spanning the 5′ SS into eight 50-nt sections ( Supplementary Fig. S4a, upper panel), and analyzed which 50-nt section has the highest SRSF1-CLIP tag coverage. We found that the highest SRSF1-CLIP tag coverage was most frequently observed (309 intron-distal 5′ SSs) immediately upstream to the 5′ SS (section 4 in Supplementary Fig. S4a and Supplementary Data S1). Three representative 5′ SSs are also indicated in Supplementary Fig. S4b. Although we observed less conspicuous binding of SRSF1 around both the intron-distal and intron-proximal 5′ SSs in the 427 genes, where knockdown of SRSF1 enhanced selection of the intron-proximal 5′ SS (Fig. 5e, lower panels), there was no noticeable difference between the intron-distal and intron-proximal 5′ SSs. Thus, the functional significance of SRSF1-binding on these 427 genes remains to be determined. To summarize, binding of SRSF1 immediately upstream to the intron-distal 5′ SS suppresses selection of the intron-distal 5′ SS in many human genes (Fig. 6).

Discussion
We analyzed the regulatory mechanisms of alternative selection of two 5′ SSs at intron 4 of DOK7, which encodes an indispensable adaptor protein for enhancing AChR clustering at the neuromuscular junction 23 . These 5′ SSs  Fig. S5a). Although alternative splicing in non-human species have not been reported or annotated, selection of the intron-distal 5′ SS causes a frame-shift in DOK7 mRNA with a premature termination codon (PTC) in all of these species (Supplementary Fig. S5b). This PTC-harboring DOK7 transcript constitutes around a quarter of DOK7 transcripts in the skeletal muscle, brain, and heart in human, where DOK7 is highly expressed (Fig. 1b). The truncated Dok-7 isoform 2 marginally enhanced MuSK phosphorylation (Fig. 2c), but significantly diminished the AChR clustering activity (Fig. 2a,b).
The canonical full-length Dok-7 isoform 1 carries pleckstrin-homology (PH) and phosphotyrosine-binding (PTB) domains in the N-terminal region, as well as two tyrosine residues, which are target motifs of the Src homology 2 (SH2) domain, in the C-terminal region 23 (Fig. 1a). The PH domain is essential for membrane association, and PTB domain is involved in Dok-7-induced activation of MuSK 39 . Activated MuSK phosphorylates the two tyrosine residues (Y396 and Y406) of Dok-7, which are essential for activating downstream signaling leading to AChR clustering 26 . Preservation of the PH domain and part of the PTB domain in Dok-7 isoform 2 ( Fig. 1a) may account for minimal induction of MuSK phosphorylation in C2C12 myotubes overexpressing the truncated Dok-7 isoform 2 (Fig. 2c). Lack of the tyrosine residues, however, is likely to have abolished AChR clustering activity of the truncated Dok-7 isoform (Fig. 2a,b).
We have identified SRSF1 as a regulator of selection of two 5′ SSs of DOK7 intron 4. Binding of SRSF1 immediately upstream to the two competing 5′ SSs suppresses selection of the intron-distal 5′ SS. SRSF1 is a member of serine and arginine-rich (SR) protein family, which regulates multiple steps of RNA processing, including pre-mRNA splicing 20 , transcription, mRNA stability, nuclear export, NMD, and protein translation 21 . In contrast to our finding, a previous study showed that artificial tethering of SRSF1 between two competing 5′ SSs facilitates use of the intron-proximal 5′ SS 17 . Thus, SRSF1 may exert the opposing effects on alternative 5′ SS selection in a position-specific manner. We showed by an integrated global analysis of CLIP-seq of SRSF1 and RNA-seq of SRSF1-knocked down cells that binding of SRSF1 immediately upstream to two competing 5′ SSs generally suppresses the intron-distal 5′ SS (double-headed arrow in Fig. 5e). In contrast, facilitation of the intron-proximal 5′ SS by binding of SRSF1 between two competing 5′ SSs reported by others 17 was not distinctly corroborated in our integrated analysis.
Isolation of an early spliceosome complex revealed that SRSF1 inhibits binding of U1 snRNP to the intron-distal 5′ SS to suppress selection of the intron-distal 5′ SS of DOK7 intron 4. In contrast to our study, a previous report shows that SRSF1 directly interacts with U1 snRNP to make an early spliceosome complex and facilitates splicing 40 . In this report 40 , SRSF1 bridges pre-mRNA to U1-70K, a specific component of U1 snRNP, through their RRMs. Contrarily, the SRSF1-bidning cis-element and the intron-distal 5′ SS are adjacently located in DOK7, which may sterically hinder simultaneous binding of SRSF1 and U1-70K snRNP to closely located RNA segments. In contrast to the intron-distal 5′ SS, we found that SRSF1 has no effect on binding of U1 snRNP to the intron-proximal 5′ SS (Fig. 5c). This can be accounted for by two possible mechanisms. First, as placement of the intron-proximal 5′ SS to the position of the intron-distal 5′ SS made the minigene construct sensitive to the suppressive effect of SRSF1 (Mut-6 and Mut-7 in Fig. 5d), the intron-proximal 5′ SS was immune to SRSF1 because of the distance from the SRSF1-binding cis-element. Second, the intron-proximal 5′ SS has a higher splicing strength than the intron-distal 5′ SS (Fig. 1a). Especially, intronic sequence of "gtaagt" of the intron-proximal 5′ SS is complementary to the 5′ end of U1 snRNA that binds to the pre m-RNA 41 . We conclude that alternative selection of the two 5′ SSs of DOK7 intron 4 is finely tuned by (i) binding of SRSF1 to their immediate upstream position, (ii) the distance between the SRSF1-binding site and 5′ SS, and (iii) the strength of splicing signals of the two 5′ SSs. Similar splicing regulations are likely to be operational in many other human genes.

Methods
Cell culture and transfection. HeLa and C2C12 cells were cultured in the Dulbecco's minimum essential medium (DMEM, Sigma-Aldrich) supplemented with 10% and 20% fetal bovine serum (Sigma-Aldrich), respectively. Immortalized KD3 human myoblasts were kindly provided by Dr. Naohiro Hashimoto at National Center for Geriatrics and Gerontology, Japan 42 . KD3 cells were grown in high-glucose (4.5 g/ml) DMEM (hDMEM) medium containing 20% FCS and 2% Ultroser G serum substitute (PALL), as previously described 43 . To induce myogenic differentiation of confluent C2C12 myoblasts, the culture medium was switched to DMEM supplemented with 2% horse serum and 1x Insulin-Transferrin-Selenium (Thermo Fisher Scientific). HeLa cells were transfected with FuGENE 6 (Roche), and KD3 and C2C12 cells were transfected with Lipofectamine 3000 (Thermo Fisher Scientific) according to the manufacturers' protocols. Exons 3,4, and 5, and their flanking intronic 150 nucleotides of human DOK7

Construction of minigenes.
were amplified by a proofreading DNA polymerase (PrimeSTAR, Takara) using DNA extracted from KD3 cells. The three PCR products were comprised of (i) exon 3 to intron 3 (IVS3+150); (ii) intron 3 (IVS3-150) to intron Sequences of the Wt and mutant constructs (Mut-5, -6 and -7) are shown in Supplementary Fig. S6. Mut-5 is identical to Mut-5 in Fig. 3d. (e) Distribution of SRSF1-CLIP tags centered around the intron-distal (left panels) and intron-proximal (right panels) 5′ SSs depicted by integrated genome-wide analysis of CLIP-seq of SRSF1 in native HeLa cells and RNA-seq in SRSF1-knocked down HeLa cells. Mean (green lines) and standard error (light green areas) of normalized CLIP-tag densities are shown. In 1445 (upper panels) and 427 (lower panels) genes, SRSF1-konockdown activates intron-distal and intron-proximal 5′ SSs, respectively. A doubleheaded arrow indicates a peak immediately upstream to the intron-distal 5′ SS in the 1445 genes, indicating the suppressive effect of SRSF1 on the intron-distal 5′ SS, as we observed in DOK7 intron 4. 4 (IVS4+150); and (iii) intron 4 (IVS4 −150) to the first 20 nucleotides of exon 5. As the sizes of introns 3 and 4 were 2705 and 8996 nucleotides, 2405 and 8696 nucleotides in the middle of introns 3 and 4 were excluded from amplification, respectively. PCR primers additionally carried a restriction site at their 5′ end that matched to the neighboring amplicon. Primer sequences and restriction sites are shown in Supplementary Table S1. The PCR products were digested by restriction enzymes, and ligated using DNA ligation kit (Takara). The ligated product was amplified again by PCR, digested by restriction enzymes, and cloned into pcDNA3.1+ vector (Fig. 3a).
Block scanning mutagenesis, site-directed mutagenesis to make Mut-1 to Mut-7 constructs, introduction of MS2-binding hairpin sequence, and placement of the intron-proximal 5′ SS were performed with the QuikChange site-directed mutagenesis kit (Agilent) with oligonucleotides indicated in Supplementary Table S1. All constructs were sequenced to ensure the presence of desired mutation(s) and the absence of unexpected artifacts.
Construction of expression vectors. cDNA for human DOK7 transcript variant 1 was amplified from total RNA of KD3 cells, and cloned into the pEGFP-N1 expression vector (Clontech) to make pEGFP-DOK7-T-var1. pEGFP-DOK7-T-var2 to express a fusion protein comprised of Dok-7 isoform 2 and EGFP was made from pEGFP-DOK7-T-var1 by deleting the 11 nucleotides at the 3′ end of exon 4, as well as a premature termination codon and its downstream nucleotides of DOK7. Construction of expression vectors for human FLAG-MuSK 44 , SRSF1 45 , SRSF1-MS2 45 and hnRNP H-MS2 46 were previously reported. The absence of artifacts was confirmed by sequencing the entire insert.

RT-PCR.
Total RNA was extracted 48 h after transfection using Trizol (Invitrogen), followed by DNase I treatment. cDNA was synthesized with oligo-dT primer (Invitrogen) for RT-PCR and random primer (Invitrogen) for Real-time RT-PCR using ReverTra Ace reverse transcriptase (Toyobo), and RT-PCR was performed with GoTaq (Promega) using primers shown in Supplementary Table S2. Real-time RT-PCR was performed using LightCycler 480 II (Roche) and the SYBR Premix Ex Taq II (Takara) to quantify endogenous human SRSF1 transcripts using primers shown in Supplementary Table S3. RNA affinity purification assay. Biotinylated RNA probes were synthesized with T7 RiboMAX large-scale RNA production system (Promega) using DNA templates generated by hybridizing two complementary oligonucleotides shown in Supplementary Table S4. RNA affinity purification was performed as previously described 35 with some modifications. Briefly, biotinylated RNA (0.75 nmol) and HeLa nuclear extract (30 μl) (CilBiotech) were mixed in a 500-μl binding buffer [20 mM HEPES, pH 7.8, 125 mM KCl, 0.1 mM EDTA, 1 mM DTT, 1 mM PMSF, 0.05% Triton X, 1× Protease Inhibitor Cocktail (Active Motif)], and were incubated at 30 °C for 3 h with gentle agitation. In parallel, 50 μl streptavidin-conjugated beads (Streptavidin-sepharose, GE Healthcare) were blocked with a 1:1 mixture of 1 ml binding buffer containing yeast tRNA (0.1 mg/100 μl of beads) and 1 ml PBS containing 4% BSA at 4 °C with rotation for 1 h. The beads were washed four times and incubated with the binding buffer at 30 °C for 1 h with gentle rotation. After washing the beads four times with 1 ml binding buffer, RNA-bound proteins were eluted in SDS loading buffer by boiling at 95 °C for 5 min. The isolated proteins were fractionated on a 12% SDS-polyacrylamide gel and stained with Coomassie blue or by immunoblotting.

Mass spectrometry.
Mass spectrometry analysis of affinity purified RNA binding proteins was performed as previously described 35 . Briefly, a Coomassie blue-stained band of interest was excised from the gel and was digested in-gel by Trypsin Gold (Promega) according to the manufacturer's protocols. For in-solution digestion, the RNA-bound proteins were eluted in an elution buffer (0.1 M glycine with 2 M urea, pH 2.9) and digested by Trypsin Gold according to the manufacturer's recommendations. Nanoelectrospray tandem mass analysis was performed using an LCQ Advantage Mass Spectrometry System (Thermo Finnigan). Multiple MS/MS spectra were analyzed by the Mascot program version 2.4.1 (Matrix Science) 35 . Figure 6. Schematic of alternative 5′ SS selection of human DOK7 intron 4. Binding of SRSF1 immediately upstream to the intron-distal 5′ SS suppresses binding of U1 snRNP to the intron-distal 5′ SS, but not to the intron-proximal 5′ SS. version 0.10.1, (ii) removal of adapters, polyA at the 3′ end, unreadable nucleotides at the 5′ end, and a short sequence <18 bp, (iii) mapping to UCSC hg19 by STAR version 2.5.3 52 , and (iv) conversion of the created BAM file to a BedGraph file by BEDTools version 2.26.0 53 . Finally, using the BedGraph file, we counted reads mapped on A5SS ± 200 nt by HTSlib version 1.4 54 . The normalized CLIP-tag density is calculated by dividing the CLIP-tag coverage at each position by total coverage of CLIP-tags in a 400-nt segment comprised of 200-nt upstream and 200-nt downstream regions of the 5′ SS.