The splicing regulators RBM5 and RBM10 are subunits of the U2 snRNP engaged with intron branch sites on chromatin

SUMMARY

Understanding the mechanisms of pre-mRNA splicing is limited by technical challenges to examining spliceosomes in vivo. Here we report the isolation of RNP complexes derived from precatalytic A or B-like spliceosomes solubilized from the chromatin pellet of mammalian cell nuclei. We found that these complexes contain U2 snRNP proteins and a portion of the U2 snRNA bound with protected RNA fragments that precisely map to intronic branch sites across the transcriptome. These U2 complexes also contained the splicing regulators RBM5 and RBM10. We found RBM5 and RBM10 bound to nearly all branch site complexes and not simply those at regulated exons. Deletion of a conserved RBM5/RBM10 peptide sequence including a zinc finger motif disrupted U2 interaction, and rendered the proteins inactive for repression of many alternative exons. We propose a model where RBM5 and RBM10 regulate splicing as components of the U2 snRNP complex following branch site base-pairing.

Graphical Abstract

eTOC blurb

Damianov et al. developed a method for isolation of complexes derived from spliceosomes formed in vivo, applicable both in examining functions of spliceosomal factors and mapping of spliceosome assembly sites. They demonstrated that the splicing regulators RBM5 and RBM10 are U2 snRNP-associated factors repressing exon inclusion following branch site selection.

INTRODUCTION

Pre-mRNA splicing is a key step in the expression of eukaryotic genes that is catalyzed by the spliceosome, a large multisubunit ribonucleoprotein (RNP) particle. Many essential cellular, developmental, and disease processes are driven by changes in splicing and the expression of alternatively spliced mRNA isoforms. These splicing choices are determined by a large number of regulatory RNA binding proteins that alter spliceosome assembly [reviewed in ^1–4].

Spliceosome assembly is highly dynamic [reviewed in ^5–7], and one key early step is the recruitment of the U2 snRNP by factors bound at the 3’ splice site, followed by base pairing of the U2 snRNA with the branchpoint sequence to form the pre-spliceosomal A complex. The U2/pre-mRNA helix contains a bulged branchpoint A residue whose 2’ hydroxyl will be the attacking group in the first catalytic step of splicing. The A complex then progresses through a series of mature spliceosomes with the sequential recruitment and ejection of many additional factors to ultimately catalyze exon ligation.

Spliceosomes must assemble on each intron and are subject to regulation that alters splice site choices according to cellular conditions. Despite much progress in resolving the structures of key spliceosome assembly intermediates and elucidating the mechanisms of catalysis, many aspects of splicing remain poorly understood. Most splicing in the nucleus occurs during transcription on chromatin, conditions that likely differ dramatically from standard in vitro splicing systems. Moreover, the assembly pathway and spliceosome structures have been examined only on a limited number of introns, chosen for their efficient splicing in HeLa nuclear extract. Of particular interest for biology are the many regulatory proteins that alter splicing choices. These factors’ interactions with pre-mRNA’s can be mapped using genomic sequencing approaches, but their interactions with the spliceosome have largely been studied in nuclear extracts and not a native context. New methods are needed for examining regulator/spliceosome interactions in different cellular contexts in vivo.

Two regulators studied both for their biology and the mechanism by which they alter splicing are the proteins RBM5 and RBM10, which are implicated in cancer development, where they alter apoptotic regulation and cell proliferation ^8–12. RBM10 is also seen to be mutated in the multisystem genetic disorder TARP syndrome ^13,14. RBM5 and RBM10 share domain structures ^11,15,16, and both proteins interact with several U2 snRNP-specific or U2-associated proteins ^15,17–20, and were detected in A- and B-complex spliceosomes ²¹. Despite high sequence similarity of their RNA binding domains, their RNA binding preferences differ ^8,16,22–26. In vivo crosslinking studies found RBM10 binding to be enriched upstream of branch sites, in proximity to the U2 snRNP ^16,25. RBM5 crosslinking was more prominent downstream of branch sites. Puzzlingly, these RBM5 and RBM10 crosslinking sites were not enriched adjacent to their target exons compared to other exons. Thus, the mechanism by which RBM5 and RBM10 can target particular exons for regulation is not fully clear.

In exploring the interactions of splicing factors within the chromatin compartment of cells, we previously isolated a large, multimeric complex called LASR that binds Rbfox family proteins and contains other splicing regulators ²⁷. We were interested in whether other regulatory interactions could be identified in this subcellular compartment, including between splicing regulators and spliceosomal proteins. Examining components of the branch site recognition machinery associated with chromatin, we identified an unusual U2 RNP particle that contains RBM5 and RBM10. We show that isolation of RNA from this particle allows the profiling of intronic branch sites assembled into endogenously formed precatalytic spliceosomes. We examine how U2-bound RBM5 and RBM10 might drive splice site choices through modulation of branchpoint interactions.

RESULTS

The splicing regulatory proteins RBM5 and RBM10 are bound within U2 snRNP complexes isolated from chromatin.

To assess protein interactions with U2 snRNPs assembling onto nascent RNA, we isolated nuclei, subjected them to gentle lysis, and pelleted the high-molecular weight (HMW) material containing chromatin and nascent RNA. To release spliceosomal components, this pellet was extracted with a cocktail of enzymes digesting RNA and DNA. Examining the sedimentation of spliceosomal proteins and splicing regulators in glycerol gradients, we identified factors present in higher order molecular complexes. We found that the U2 proteins SF3A3 and SF3B1 were present in a nuclease resistant complex that cosedimented with the splicing regulator RBM5 at approximately 15S. (Figure S1A). These complexes were observed in HMW extract of both mouse brain and human 293Flp-In cells, but not in the soluble nucleoplasmic fraction where the three proteins migrated near the top of the gradient. To characterize these complexes, we generated 293Flp-In cell lines inducibly expressing either SF3A3-Flag or RBM5-Flag under tetracycline control. Induced expression of Flag-tagged SF3A3 reduced the level of the endogenous SF3A3, but did not affect the levels of other SF3A and SF3B proteins (Figure S1B). We then immunopurified the complexes from the peak gradient fractions (Figures S1A and S1C, indicated in blue) on anti-Flag agarose and analyzed their components by gel electrophoresis and mass spectrometry. The protein profiles of the SF3A3-Flag and RBM5-Flag complexes were very similar (Figure 1A and Table S1). These profiles resemble the 17S U2 snRNP ²⁸ with a notable lack of the U2A’, U2B” and the Sm core proteins (Figure 1B). Unlike U2 snRNPs isolated from standard nuclear extracts, the U2 in the chromatin fraction of cells has additional components that include the regulatory protein RBM5.

Figure 1.

The SF3A3 and RBM5 RNP complexes isolated from HMW extract have similar protein components, and contain U2 snRNA and other RNA fragments.

(A and B) Protein profiles of immunopurified SF3A3-Flag or RBM5-Flag- containing complexes from gradient regions indicated in Fig. S1 AC. Proteins are detected either by staining with SyproRuby (A), or immunostaining (B). Lanes marked “HMW extract” in B correspond to 5% input.

(C) 5’-radiolabeled RNA, recovered from SF3A3-Flag or RBM5-Flag RNPs, separated by denaturing Urea-PAGE and detected by phosphorimaging. Samples were pretreated with full-length U2 snRNA-antisense DNA and RNase H, followed by TurboDNase digestion (lanes “U2-digested”), or with TurboDNase only (lanes “total”). Detected RNA fragments and short digested DNA oligonucleotides are indicated on the right.

(D) U2 snRNA read coverage per nucleotide from isolated RBM5-Flag RNP mapped according to position along the RNA. Only the nucleotides 1-122 are shown since no reads were mapped downstream. The read coverage relative to the overall structure of the 17S and A-complex U2 snRNP in is diagrammed below.

Interestingly, the SF3A3-Flag complex contained limited RBM5 but larger amounts of the related protein RBM10, reflecting the relative abundance of these proteins in the HEK293 cells. Since the RBM5 complex did not contain RBM10, these two paralogous proteins appear to bind U2 in a mutually exclusive manner (Figures 1B). To examine RBM10, we created a cell line expressing RBM10-Flag. In this case, we immunoprecipitated RBM10-Flag from total HMW extract without gradient fractionation (Figure S2A). The profile of proteins coprecipitated with RBM10 was almost identical to those of the RBM5 and SF3A3 complexes, but with excess RBM10, indicating a mix of free RBM10-Flag and protein bound to U2 snRNP. The RBM10 also coprecipitated some of the U2 proteins from the nucleoplasm, albeit less efficiently. Thus, the paralogous proteins RBM5 and RBM10 both bind to the U2 snRNP.

The third member of the RBM5 family is RBM6. In a cell line expressing Flag-tagged RBM6, we found the most abundant coprecipitating proteins were the scaffold attachment proteins SAFB and SAFB2 (Figure S3 and Table S1). Despite its sequence similarity and shared domain organization with RBM5 and RBM10, RBM6 did not interact with the U2 snRNP proteins, and appears to have a different function.

Despite extraction with nuclease, all of the isolated U2 complexes retain RNA components. As expected, several discrete bands were identified as U2 snRNA fragments by their sensitivity to RNase H degradation in the presence of U2 antisense oligonucleotide (Figures 1C, S2B). Sequencing RNA from the RBM5 containing U2 RNP, we found U2 snRNA reads were the most numerous among all snRNA reads (Figure S4). We also noted high levels of reads from the U11 and U12 snRNAs (see below). The U2 reads predominantly mapped to the 5’ half of the snRNA, with highest protection from nuclease digestion between U2 nucleotides 27 and 65, which includes the branch site interaction sequence (Figure 1D). The recovered reads drop at the Sm core binding site, and are absent after nucleotide 120. This is consistent with the observed absence of the Sm proteins and the U2A’ and B” proteins that bind in these regions (Figure 1AB).

The U2 Auxiliary Factor U2AF2/U2AF65 was found to interact with RBM5 ¹⁵, and could potentially recruit RBM5 to the U2 snRNP. However, U2AF2 was sub-stoichiometric in the isolated complexes, and bound to the RBM5-Flag complex at even lower levels than to SF3A3-Flag (Figures 1AB, Table S1). The presence of RBM5 in the U2 complex is thus unlikely to be mediated by U2AF2 binding. Similarly, RBM5 was found to bind the Sm B protein via its OCRE domain ²⁹. The Sm core proteins are absent from the U2 complex isolated here and thus are not a contact point for RBM5 (Figure 1B). These interactions may occur in the larger U2/pre-mRNA complexes present prior to nuclease digestion.

Another potential interaction for RBM5 and RBM10 within the isolated U2 RNPs is the RNA helicase DHX15. DHX15 enters the spliceosome as a component of the 17S U2 snRNP and remains associated until after catalysis, when it stimulates disassembly through interaction with the G-patch protein TFIP11 ^28,30. Several recent studies have also implicated DHX15 and the G-patch protein SUGP1 in a quality control step of early spliceosome assembly ^31–35. DHX15 was present in the RNPs isolated with Flag-tagged RBM5, RBM10, or SF3A3, but there was limited SUGP1 in these complexes (Figures 1A, S2A). Interestingly, RBM5 and RBM10 also contain G-patch motifs and the RBM5 motif was shown to interact with DHX15 and increase its helicase activity in vitro ¹⁸. With this in mind, we tested if the G-patch motifs in RBM5 and RBM10 mediated their recruitment to the U2 RNP. We found no differences between the complexes isolated with full length or G-patch deletion mutants of these proteins, indicating that the G-patch was not required for the U2 interaction (Figure S5AB).

To identify residues in RBM5 that mediate interaction with U2, we made a range of additional deletion mutations across the proteins. These mutations removed the known functional motifs within the proteins, including the two RRM domains, the first zinc finger, and the OCRE domain. We also deleted the N-terminal 90 amino acids, and the C-terminus that includes the G-patch. All of these deletion mutants maintained their interaction with the U2 complex, copurifying with the same complement of proteins as the full-length RBM5 (Figure 2AB). Finally, we deleted an internal segment of the protein upstream of the C-terminus that contains conserved peptide sequences and a second Zinc finger, but was not known to interact with other proteins (Figure S6). In contrast to the other mutations, deletion of RBM5 residues 544 to 702 eliminated its association with the U2 complexes (Figure 2AB). Deletion of the homologous region (residues 655-816) from RBM10 also completely disrupted RBM10 interaction with U2 (Figure 2C). These conserved residues of RBM5 and RBM10 thus include a likely U2 binding segment of the proteins.

Figure 2.

The region between the OCRE and the G-patch motifs in RBM5 and RBM10 is required for interaction with U2.

(A) Diagram of RBM5 domains. Significant motifs identified in the CDD/SPARCLE conserved domain database are colored ⁶⁹. A second Zinc finger motif (ZnF2) is also indicated ¹⁵. Amino acid residue positions are indicated on the top. Grey bars at the bottom indicate sequences deleted in RBM5. Each was deleted independently, or in combination with the G-patch motif.

(B) Proteins coprecipitating with the Flag-tagged full-length protein (lane RBM5wt) and with each deletion mutant as indicated on the top. Anti-Flag immunoprecipitations were performed from whole cell HMW extracts. The major coprecipitating proteins detected by protein staining are indicated on the right.

(C) Proteins coprecipitating with Flag-tagged full-length RBM10, and RBM10 with amino acids 655-816 deleted (equivalent to RBM5 residues 544-702). Anti-Flag immunoprecipitations were performed from nucleoplasm or HMW extract as indicated, the proteins indicated on the right were detected by immunostaining.

In some sedimentation experiments we noted an additional RBM5 peak, indicating the presence of a smaller-size complex. This peak can be seen in fraction 8 in Fig S1A, but not in Fig S1C. We immunopurified this smaller complex from the gradient fractions. Comparing the protein and RNA composition of this complex to the larger one from fractions 11-13, we found that it lacked the SF3A proteins (Fig. S7A, Table S2) and RNA (Fig. S7B). This smaller complex may be an inconsistent product of RNA degradation and SF3A dissociation from the larger one. The presence of RBM5 in this complex indicates that it likely makes a direct interaction one of the remaining proteins such as the SF3B subunits, but not with the SF3A heterotrimer.

RBM5 and RBM10 are components of branchpoint-engaged U2 snRNPs across many exons.

The U2 complexes isolated from chromatin contained RNA species that remained after U2 snRNA degradation with RNase H (Figures 1C, S2B). To characterize these RNA components, we generated sequencing libraries from the Flag isolated complexes after U2 removal by RNase H. The isolated RNA fragments (28 to 55 nt) were converted to cDNA and sequenced on the Illumina platform. The sample preparations were called RNP-seq if they included a gradient fractionation step, or IP-seq if the RNA was isolated with just a Flag-IP and peptide elution. Mapping the isolated RNA sequences identified them as pre-mRNA fragments. These fragments form tight clusters of mapped reads near the 3’ ends of introns (Figures 3A and S8, see tracks RNP-seq and IP-seq and Table S3 for details). The pre-mRNA sequences overlap branch sites, and thus likely derive from RNA base-paired with U2 snRNA in endogenously formed spliceosomes.

Figure 3.

The SF3A3, RBM5, and RBM10 RNPs are bound to pre-mRNA branch sites.

(A and B) UCSC Genome browser view of protected RNA sequencing library reads mapped to portions of the ZC3H4 gene. Total read coverage is shown in A below the gene diagram, with individual reads shown in B. Individual tracks show RNA recovered from the SF3A3-Flag and RBM5-Flag RNPs (isolated on gradients, see Figure 1), or RNA coprecipitated with SF3A3-Flag and RBM10-Flag directly from HMW extracts (see Figure S2) (lanes RNP-seq or IP-seq). Significant branch site clusters of overlapping reads are marked in tracks above the peaks. Above the gene diagram are tracks of branch points identified by others ^39,40. Precatalytic branch points predicted from reads in RNP-seq and IP-seq are shown as vertical bars with height proportional to the frequency of each predicted branchpoint in B (see Figure S10 for details).

These were protected from degradation during extraction from the HMW nuclear material. The peaks of protected branch sites are remarkably homogeneous, with consistent width distribution and few background reads away from branch sites. Virtually all introns in the expressed transcriptome generated a peak of branch site fragments, although the peaks vary in height from intron to intron. Surprisingly, this was seen for all complexes, isolated with SF3A3-Flag, RBM5-Flag, and RBM10-Flag. While some branch site fragments are more abundant in RBM5 or RBM10 complexes, and others more abundant in SF3A3 complexes, we found that virtually all branchpoints were bound by U2 complexes containing RBM5 and RBM10, in addition to SF3A3. These included branch sites for both constitutive and RBM5/10-regulated exons (see below). The reads from the SF3A3, RBM5, and RBM10 complexes together generated 385,806 clusters, 145,934 of which aligned with an annotated branchpoint.

The protection of the intronic 3’ ends was not limited to the major-class introns, but also included U12-type introns (shown for a ZC3H4 AT-AC intron in Figure 3A). This was unexpected since the SF3A heterotrimer was found to be absent from the 18S U11/U12 snRNP ^36,37 and is thought to be replaced with SCNM1 in the minor spliceosome ³⁸. Compared to U2 sites, the U12 branch site peaks were more efficiently recovered with RBM5 and RBM10 than SF3A3 (Fig. S9AB). Limited amounts of SF3A3 may be present in minor class spliceosomes at an as yet undefined stage of assembly. RBM10 and particularly RBM5 efficiently recovered branchpoints of minor class introns. It will be interesting to explore whether they play a role in their splicing.

Identified branched nucleotides from available datasets ^39,40 mapped near the middle of individual reads of the U2-protected RNA fragments (see Figure 3B for example). The reverse transcriptase used to generate the sequencing libraries of these RNAs is expected to terminate at a branched nucleotide. The efficient generation of reads spanning these sites indicates that the RNAs are unbranched, and the purified U2 complexes are likely derived from precatalytic spliceosomes. Considering the stability of the complexes, with high recovery of branch site fragments over the background, and the high protection of the branch site-interacting sequence of U2 snRNA, these RNPs resemble A- or B-like spliceosomes, rather than earlier complexes before the U2 snRNP interaction with the pre-mRNA is stabilized.

In A and B precatalytic spliceosomes the pre-mRNA is base-paired to the U2 snRNA with the future branched nucleotide unpaired and bulged from flanking helical stems ^41–45. To characterize the positions of the future branch nucleotides within our thousands of protected branchpoint regions, we aligned our branch site clusters to datasets of mapped and predicted branchpoints. The mapped branchpoints were identified in K562 and other human cell lines and tissues, but not HEK293 cells ^39,40. Thus, there were many protected fragments within our data where the branchpoint was not yet known. In a filtered set of clusters each containing a single previously mapped branchpoint (Figure S10A), the branch nucleotide was most frequently located downstream of the read midpoint, within a narrow 50% interquartile range (IQR) (Figure S10B). Focusing on this range and applying a tool for ranking branch site heptamers, we identified the best-fitting putative branch point for each read across all clusters (Figure S10C). We then ranked the most frequently called branchpoints within the reads of each cluster (Figure S10D). These are shown as branchpoint tracks in Figures 3B and S8. The ranks of these predicted branch points strongly correlated with the position of previously mapped sites (Figure S11). The nucleotide frequencies surrounding these predicted branch nucleotides closely matched the previously defined consensus sequences (see below). Thus, we can accurately predict the engaged branchpoints for each intron generating a read cluster in our data.

The sequencing of pre-mRNA fragments bound by U2 snRNPs provides a comprehensive map of branchpoint interactions across the transcriptome. The levels of unspliced introns across a gene are highly variable, with many present at low levels due to their rapid excision ^46,47. However, the low background of non-branchpoint reads and the precise alignment of the protected fragments allows identification of the U2 binding sites for nearly all the introns in the expressed RNA. The tracks presented here derive from one replicate of three for each condition. Together these replicates generated datasets of 24-84 million reads per experiment, over 65% of which mapped to branch site clusters and provided sensitive detection of the branchpoints (Table S3). The branchpoint peaks vary in height for each intron across a gene transcript. Some of this variation presumably results from different intron excision rates leading to branchpoints spending greater or lesser time within pre-catalytic spliceosomes. However, this does not appear to be the only factor affecting the peak heights, and we have not identified simple rules that determine the recovery of each branchpoint.

We also noted a large number of clusters located deep within introns distant from annotated 3’ splice sites (Table S3, see unidentified clusters). Approximately 19% of these undefined clusters were upstream of exonic sequences in the HEXEvent database⁴⁸, and thus likely arise from unannotated 3’ splice sites (See Table S4). Other undefined clusters were associated with recursive splice sites⁴⁹, or with pseudo or decoy exons⁵⁰ (Figures S12 and S13, see also Data S1). We found 38-40% of the recursive sites identified in a variety of human cells or tissues were associated with IP-seq or RNP-seq clusters in our data from HEK293 cells^51,52 (Table S4, see also Figure S12). In erythrocytes, SF3B1 contains an intron whose retention is mediated by the presence of several decoy exons that are bound by U2AF and other early splicing factors but are thought to not result in productive splicing⁵⁰. In our HEK293 data, we observe several prominent branch site clusters within this conserved SF3B1 intron, presumably corresponding to U2 assembly on the decoy exons (Figure S13). Thus, the IP-seq maps can identify splicing events that are not readily apparent in standard RNAseq splicing analyses.

Overall, this profiling approach allows comprehensive assessment of U2 assembly and accurate branchpoint prediction across the transcriptome.

Identification of RBM5 and RBM10 regulated splicing events.

To examine the effects of RBM5 and RBM10 on individual exons across the transcriptome, we established a system to identify RBM5 and/or RBM10-regulated splicing events. We engineered HEK293 cells with the endogenous genes disrupted and replaced with either with RBM5-Flag or RBM10-Flag, expressed from a tetracycline-inducible promoter. Using CRISPR-Cas9 to disrupt both alleles of the two genes, we could isolate cells lacking RBM5, but were unable to disrupt the reading frame of both RBM10 alleles in this RBM5^−/− background. We isolated a clone with both RBM5 alleles and one RBM10 allele inactivated, and with one RBM10 allele expressing limited amounts of a mutant RBM10 mRNA containing a cryptic exon (CE) that restores the reading frame after the targeted deletion (Figure S14). These RBM5 null, RBM10 mutant cells (RBM5^−/−; RBM10^−/CE) exhibited markedly slower growth than wildtype cells, and this growth was rescued by expression of either RBM5 or RBM10. We then compared gene expression and splicing in the mutant cells with cells rescued by either protein.

We isolated polyA+ RNA and sequenced it from four replicas each of the parental mutant cells and the RBM5-Flag and RBM10-Flag rescued cells induced to express the proteins at similar levels (Figure 4A, see also Figure S15). The RNAseq data were analyzed by rMATS to examine RBM5/10 dependent splicing ⁵³. These analyses identified many RBM5 and RBM10-dependent alternative splicing events, including cassette exons, alternative 5’ and 3’ splice sites and retained introns (Figure 4B, Tables S5, S6). Cassette exons were the most numerous. Previous studies identified smaller sets of RBM5 and RBM10 dependent exons in HeLa and mouse ES cells ^16,25. There was notable overlap between the different studies of RBM10 regulated exons. Of the exons affected by RBM10 depletion in HeLa, 23% were also RBM10 regulated in our HEK293 system, including a well characterized RBM10 target exon in Numb. We found that RBM5 and RBM10 predominantly downregulated their target exons and that a substantial fraction of these targets was coregulated by both proteins (Figure 4C).

Figure 4.

RBM5 and RBM10 control splicing of many cassette exons.

(A) RBM5 and RBM10 expression in 293Flp-In parental cells, in the RBM5^−/−; RBM10^−/CE cell line, grown in presence of 0.6 ng/mL Doxycycline for 72 hours, and in RBM5^−/−; RBM10^−/CE cells transfected with RBM5-Flag or RBM10-Flag transgenes induced for the same time with 0.6 ng/mL or 0.06 ng/mL Doxycycline, respectively. Whole cell RIPA lysates were subjected to immunoblotting. RBM5 and RBM10 were detected with antibodies recognizing endogenous epitopes and with Flag antibody as indicated. PRPF3 was also detected as an internal control. The immunoblot for one of four replicas is shown at the top, with the mean normalized expression of the Flag-tagged proteins from all replicas is graphed to the right (error-bars indicate standard deviation).

(B) RBM5 and RBM10 dependent splicing events were identified by rMATS⁵³ as altered by RBM5 and/or RBM10 expression compared to the RBM5^−/−; RBM10^−/CE cells. The number of upregulated and downregulated events from each type is shown on top or bottom, respectively. Note that there are approximately four-fold more exons repressed by each protein than activated.

(C) Comparison of RBM5 and RBM10 regulation of exon inclusion and skipping. Cassette exons altered by RBM5 and/or RBM10 expression compared to the RBM5^−/−; RBM10^−/CE cells are graphed by scatter plot. X axis reports ΔPSI values for RBM5 relative to the parental cells; Y axis reports the RBM10 ΔPSI values. Cassette exons with |ΔPSI| ≥ 10 and FDR ≤ 0.05 are color-coded to indicate regulation by RBM5 and/or RBM10. A control set of cassette exons in the central box with |RBM5 ΔPSI| < 5 and |RBM10 ΔPSI| < 5 is shown in brown.

Comparing branch site clusters isolated from RBM5-Flag or RBM10-Flag complexes with those isolated from SF3A3-Flag complexes, we found that the peak heights upstream of cassette exons did not correlate with the effects of RBM5 or RBM10 on splicing (Figure S16). We also did not find notable differences in the specific branch nucleotides selected by these complexes. The highest-ranking predicted branch point nucleotides matched at 60-65% of the RBM5, RBM10, or SF3A3 clusters. Neither this match frequency nor changes in predicted branch site position correlated with RBM5 and/or RBM10 downregulation (Figure S17).

We next looked for motifs enriched in sequences adjacent to the branch sites of regulated exons. In A- and B-like spliceosomes nucleotides adjacent to the U2-paired sequence are contacted by SF3B and SF3A proteins ^54–56. Regulatory motifs could also be present at sites more distal to the protected nucleotides of the branch site. The upstream and downstream regions differ in nucleotide composition due to the downstream polypyrimidine tract, the intron terminal AG dinucleotide, and exonic sequences. To separately examine these regions, we aligned the intronic sequences at the highest-ranking branchpoint predicted for RBM5-Flag or RBM10-Flag clusters, and then defined intervals upstream and downstream (see diagram in Figure S18A). Since the distance from the branchpoint to the exon can vary, we limited the downstream intronic region to range from 10 to 46nt. This excluded a small fraction of branchpoints at greater or lesser distance from the downstream exon. We separately analyzed the sequences at the intron-exon boundaries, including the last five intronic and the first twenty exonic nucleotides. For each set of branchpoint-adjacent sequences, we used the STREME motif analysis tool to compare the motif frequencies within the RBM5 and/or RBM10 regulated targets to those of a control set of cassette exons unaffected by RBM5 or RBM10 ⁵⁷. We found that both RBM5- and RBM10-downregulated targets are enriched for a variety of C-rich motifs upstream and downstream of the branchpoint. These enrichments were more significant for the RBM5 regulated exons. However, these C-rich sequences have only limited similarities to previously identified RBM5 binding motifs ^22–24. There was also enrichment for a G-rich motif in the region 16-36 nt upstream that is similar to an RBM5 binding motif identified by RNA Bind-N-Seq, although this enrichment did not have as strong a p-value (Figure S18A).

To test the contribution of individual sequence motifs in regulating RBM5- and RBM10-dependent exons we constructed several minigenes carrying target exons, and confirmed their regulation by RBM5 and/or RBM10 in the 293Flip-In cells. We then mutated individual sequence motifs in these minigenes to assess their role in splicing regulation. Overall, we mutated 8 sequence elements in 3 different RBM5/10 target minigenes. Mutations in the C- or G-rich motifs upstream of the branchpoint altered the overall inclusion of most target exons but did not alter their response to RBM5/10 (Figure S18B). Similarly, disrupting C-rich motifs downstream of the branch site of some target exons stimulated their baseline level of splicing, but did not affect RBM5/10 regulation (Figure S18B); One exception was an exon in DPP7 where mutation of the downstream C-rich motif (mt1) resulted in nearly 100% baseline inclusion that was minimally repressed by RBM5 (Figure S18B). Mutation of combinations of elements had no greater effect than single mutations. Moreover, many RBM5 and RBM10-responsive exons identified in the RNAseq analysis did not have these motifs. Such motifs may affect the efficiency of spliceosome assembly without providing specific contacts for RBM5 and/or RBM10.

Our finding that RBM5 and RBM10 are binding to U2 assembled at all exons may explain puzzling results on RNA binding by these proteins. Previous CLIP studies of RBM5 and RBM10 did not identify motifs that were strongly enriched in regulated exons^16,25. RBM10 crosslinking sites were enriched adjacent to 3’ splice sites but were equally present adjacent to constitutive exons and regulated ones. This analysis also identified high levels of RBM10 crosslinking to U2 snRNA. To make a parallel assessment of RBM5, we analyzed ENCODE RNAseq datasets to identify exons altered by CRISPR knockout of RBM5 in HEPG2 cells⁵⁸. These data were compared to ENCODE RBM5 eCLIP data from the same cells. Very similar to the published RBM10 data²⁵, we found that RBM5 crosslink sites clustered at 3’ splice sites but were found on all types of exons and were not enriched on exons showing regulation by RBM5 (Figure S19 and Data S2–S4). Again like RBM10, RBM5 crosslinked to the U2 snRNA²⁵. Given our findings by IPseq that RBM5 and RBM10 are bound with U2 across the transcriptome and not just on regulated exons, it is possible that the crosslinking²⁵ observed at 3’ splice sites is not due to sequence specific RNA recognition, but instead RBM5 and RBM10 are crosslinking to RNA in the proximity of their location within the U2 snRNP complex.

Consistent with this idea, previous analyses of an exon in FAS indicated that RBM5 was altering its splicing without a direct interaction with the pre-mRNA¹⁵. Instead of binding regulatory motifs in the pre-mRNA to mediate splicing regulation, RBM5 or RBM10 may alter spliceosome activity by direct interaction with the U2 snRNP.

Splicing regulation by RBM5 and RBM10 requires interaction with U2.

To test how the loss of U2 binding affects the splicing regulatory activities of RBM5 and RBM10, we attempted to express proteins defective for U2 interaction in RBM5^−/−; RBM10^−/CE cells. However, unlike what was seen for the wildtype proteins, the cells harboring these RBM5 or RBM10 mutant transgenes exhibited a severe growth defect. The limited number of recovered clones had very slow growth rate and were difficult to maintain. This indicated that the U2 binding mutants had a severe loss of function but precluded assaying their splicing activity in this system. As an alternative assay, we compared the activities of full-length and mutant proteins in the parental 293Flp-In cells, which continue to express wildtype RBM5 and RBM10 from the endogenous genes (Figure 5A).

Figure 5.

RBM5 and RBM10 require interaction with U2 snRNP to regulate splicing.

(A) Diagram indicating known motifs in RBM5 and RBM10, and the peptide segment required for interaction with U2 snRNP.

(B) 293Flp-In cells were induced to express Flag-tagged RBM5, RBM5 Δ544-702, RBM10, or RBM10 Δ655-816 as shown in A. Expression of each Flag-tagged protein was quantified after immunostaining, with the relative expression levels indicated below. SF3A3 expression served as a normalization control. Splicing of endogenous PCBP2 exon 12, TRPT1 exon 6, and NUMB exon 9, HOTAIR exon 3, and STAC3 exon 9 were assayed 48 hour post-induction by RT-PCR using radiolabeled flanking primers followed by denaturing PAGE and phosphorimaging. The quantified exon inclusion levels are indicated below, and graphed at the right. P-values indicate the significance of the splicing activity differences between the full-length and deletion mutant proteins on each target. These activities were expressed as the regression slope for the exon inclusion as a function of Flag-tagged protein expression. NS indicates not significant.

Assaying splicing of target exons by RT-PCR at multiple protein concentrations, the RBM5 deletion mutant had greatly reduced activity on most target exons compared to the full-length protein, and in many cases was completely inactive (Figures 5B and S20). The changes in activity in the RBM10 deletion mutant were less consistent, possibly due to higher endogenous RBM10 levels in these cells. For example, Exon 6 of the TRPT1 gene was repressed by RBM5Δ544-702, but not by RBM10Δ655-816 (Figure 5B). For NUMB exon 9, which was moderately repressed by wildtype RBM10 but not RBM5 in this system, both RBM5 and RBM10 deletion mutants had the opposite effect and promoted exon 9 splicing (Figure 5B). A limited number of targets, including ZDHHC16 exon 8, continued to be regulated by RBM5Δ544-702, perhaps through direct pre-mRNA binding (Figure S20). The variable responses of some exons indicate that additional factors are regulating these exons, and that the RBM5 and RBM10 proteins may act by multiple mechanisms.

To assess the contributions of shorter peptide segments adjacent to or within the potential U2-interacting region 544-702 of RBM5, we generated seven additional mutations. These deleted sequences were conserved in RBM10 and were designated RBM5 Δ1 through Δ7 (Figure S21A, top). These mutants were stably expressed as FLAG-tagged proteins, and tested for their interactions with the U2 snRNP, as done previously. Unlike the larger deletion of residues 544-702, all of these smaller deletion mutants exhibited some U2 binding, albeit at reduced levels compared to wildtype (Figure S21B). RBM5 lacking the ZnF2 motif (RBM5 Δ6) exhibited the weakest interaction. Deletion of other peptides, both upstream and downstream of ZnF2, also reduced U2 binding. However, deletion of a central peptide (RBM5 Δ4) had minimal effect on U2 binding. Thus, the interaction of U2 snRNP with the 544-702 segment of RBM5 appears to be bipartite, with none of the deleted sequences solely responsible for U2 binding (Figure S21B).

To assess the splicing-regulatory activities of RBM5 Δ4 and two mutants with reduced U2 snRNP binding activity, RBM5 Δ2 and RBM5 Δ6, we performed RT-PCR of endogenous RNA containing known RBM5 target exons. All three proteins exhibited reduced splicing repression activity compared to wild-type RBM5 (Figure S21C). Interestingly, the splicing repression by the different mutants varied among the tested cassette exons. Deletion of ZnF2 (RBM5 Δ6) almost completely inactivated RBM5 in repressing HOTAIR exon 3 and STAC3 exon 9, but this mutant retained some repression of TRPT1 exon 6. RBM5 Δ4, that still interacts with U2 snRNP, was inactive in silencing HOTAIR exon 3, but partially active in repressing two other cassette exons (Figure S21C).

Overall, the results indicate that segments of the RBM5 protein between the OCRE domain and the G-patch are required both for its interaction with branch site-bound U2 snRNPs and for splicing repression activity on most exon targets. Within this region, the ZnF2 domain plays an important role in U2 binding and splicing regulation, but other peptide segments also contribute to these activities.

DISCUSSION

In vivo isolation of early spliceosome complexes

We identified a U2 RNP particle that can be extracted from the chromatin fraction of cells using nuclease but without denaturants. This complex contains multiple U2 snRNP proteins, including SF3A and SF3B, and the 5’ portion of the U2 snRNA apparently base-paired to intron branchpoints. Unlike U2 snRNP’s and early spliceosome complexes isolated from standard nuclear extracts, the particle also contains the regulatory proteins RBM5 and RBM10. Within this U2 snRNP particle, both the branch site and the U2 snRNA are protected from RNase digestion, indicating it derives from an early spliceosome where the base-pairing of the branch site has been established. The presence of the SF3A and SF3B complexes, which are removed by DHX16 during formation of the B* complex, is consistent with the particle being an early assembly intermediate. The lack of branched nucleotides in the protected intron RNAs also indicate the particles were isolated prior to the first transesterification reaction ^59,60. From these considerations, these U2 complexes likely correspond to spliceosomes in the A, B, or B^act stages. They could also derive from an assembly intermediate not yet described in vitro.

Profiling U2/intron interactions in vivo

The RNA fragments protected within the isolated U2 complexes precisely map to intron branch sites across all of the expressed transcriptome. The cluster signal intensities largely correlate with gene expression levels, but vary among branch sites from the same transcripts. This variation did not limit the sensitivity of branch site identification. At the sequencing depth of our analysis, A/B^act-like spliceosomes were detectible from even weakly expressed transcripts showing exon RPKM of less than 0.1 in standard polyA+ RNAseq data. Using this approach, we unambiguously identified over 140,000 branch site clusters located upstream of annotated exons.

We find that the future branch nucleotide or nucleotides of each intron can be accurately predicted from the position of branch site motifs within the U2-protected pre-mRNA fragments. Since they cover nearly all the expressed introns in the cell, these predicted sites are expected to be more comprehensive than those where direct sequencing is used to detect intron lariats ^39,40,61. These predicted sites agree well with the physically mapped branched nucleotides in available datasets, but could likely be further improved using more sophisticated branchpoint-calling approaches that recognize alternative U2-branch site pairing modes ⁶¹.

The unannotated sites of U2 binding allow potential detection of isoforms, splicing events, and stalled precatalytic complexes that are not apparent in standard RNAseq analyses. In addition to read clusters at annotated branchpoints, we detect a large number of undefined clusters with lower signals that are located deeper within introns. Sequence databases that include spliced ESTs or other low frequency isoforms allow assignment of many of the undefined clusters to branch sites of poorly spliced 3’ splice sites⁴⁸. We found that other clusters align with recursive splice sites, and with pseudo or decoy exons (Figures S12 and S13). A recent study provided evidence that within long introns splicing can occur stochastically at unannotated, often recursive, sites prior to final splicing at an annotated 3’ splice site⁴⁹. Besides the clear clusters of U2 binding reads at recursive sites that we describe, there are many small clusters of reads that are less significantly enriched over background. Given the stochastic model, some of these small peaks may arise from low frequency splicing at many sites across introns. It will also be interesting to align human sequence variants with unassigned U2 binding sites as a means for identifying mutations with possible deleterious effects on splicing.

We also detect about 7,000 clusters overlapping with 5’ splice sites. These protected 5’ splice sites could derive from the same precatalytic spliceosomes as the branchpoints but which survive RNAse treatment with lower yield. In agreement with this, the isolated material contains substoichiometric amounts of other snRNAs as well as U5, U4, and U1-specific proteins. At the current sequencing depth, the numbers of 5’ splice site reads and clusters is limited, but future biochemical optimization may allow more comprehensive profiling of 5’ sites, using approaches similar to that described here. More efficient mapping of these 5’ splice sites may reveal if the pre-mRNA is base-paired with U1 or U6, and thus more precisely identify the stage of the isolated spliceosomes. In earlier work using this isolation strategy, we identified interactions between splicing regulatory proteins in the chromatin fraction of the nucleus that had not been observed in nuclear extracts ²⁷. In the future, we will be excited to test additional tags and extraction conditions in applying this method to studies of nuclear regulatory processes.

Splicing regulation by RBM5 and RBM10

We identify the splicing regulators RBM5 and RBM10 as components bound at high stoichiometry to the U2 snRNP/branchpoint complex. Splicing regulators have largely been studied through their interactions with the pre-mRNA. These pre-mRNA binding events are thought to direct changes in splice site choice through altering early steps in spliceosome assembly. In contrast, we find that RBM5 and RBM10 incorporate into spliceosome complexes on the nearly complete set of branch sites without an apparent dependence on specific interactions with pre-mRNA. Since RBM5 and RBM10 did not alter the selection of branchpoint or the recovery of the protected branch sites of regulated exons, they likely repress splicing at a later stage spliceosome than the isolated RNPs. Their interactions within the branch site complex could allow them to sense conformations that favor or disfavor further spliceosome assembly or to alter splicing kinetics on particular introns. Individual branch points on some unregulated exons were preferentially isolated with RBM5 or RBM10 compared to SF3A3. This is consistent with the idea that RBM5/10 binding is sensitive to the particular conformation of the bound U2 snRNP.

We identified a large set of cassette exons regulated by RBM5/10 in HEK293 cells. While a majority of these exons were repressed by RBM5/10, exons whose splicing was stimulated by the proteins were not uncommon. Some exons were sensitive to only one protein, while a large set responded to both RBM5 and RBM10, with splicing nearly always being altered in the same direction. A more limited overlap between RBM5 and RBM10 targets was observed in a previous study¹⁶. This is likely due to differences in the experimental systems: detection by microarray rather than RNAseq, and knockdown of proteins by RNAi rather than CRISPR-Cas9 gene deletion followed by transgene reexpression. RBM10 was absent from U2 snRNPs isolated with tagged RBM5 and vice versa, while both proteins are present but substoichiometric in the complexes isolated with tagged SF3A3. These observations are consistent with the two regulators being mutually exclusive in their interaction with U2. This, along with the overlap of their target exon sets, may indicate a shared mechanism on at least a subset of their targets.

We identify a highly conserved segment in RBM5 and RBM10 that interacts with U2, and is required both for cell growth and for their ability to regulate cassette exons. Multiple motifs within this segment affect the U2 interaction, and contribute to splicing silencing. Among these motifs, ZnF2 has the most substantial effect on both U2 binding and splicing regulation. Interestingly, ZnF2 is altered by several cancer-associated and TARP-syndrome missense mutations that were also shown to affect its splicing activity^11,62. ZnF2 is also conserved in RBM6, which we did not find interacting with U2 snRNP. RBM6 diverges from RBM5/10 in other segments of the U2 interaction region, and it is thus possible other RBM6 motifs block or weaken its U2 interaction. The motifs required for U2 snRNP binding also play a role in RBM10 sequestration in nuclear bodies, a process that may be U2 snRNP-mediated⁶³. It will be very interesting to identify the RBM5/10 binding partner(s) on U2, which may be one or more of the SF3B proteins.

We identified enriched RNA sequence motifs adjacent to the branchpoints of RBM5 and RBM10 repressed exons, but experiments to mutate these elements did not strongly support their function as binding motifs for the proteins or as regulatory elements for splicing. This is consistent with the regulation of FAS exon 6 by RBM5, where a direct pre-mRNA binding site was also seen to be lacking¹⁵. Interestingly, introns with C-rich polypyrimidine tracts were found to splice more slowly than those with U-rich tracts ⁴⁶. C-rich polypyrimidine tracts have also been identified as features in the splicing targets of DDX39B, another RNA helicase associated with U2 ⁶⁴. Thus, it is possible that RBM5 inhibits splicing of its targets by disrupting spliceosomes that are slow to transition from A to later complexes.

The G-patch motifs of RBM5 and RBM10 suggest possible mechanisms of repression. G-patch motifs act as essential coactivators of DEAH-box helicases ^65,66. DHX15 in the isolated U2 RNP is proposed along with the G-patch protein SUGP1 to mediate a quality control step of early spliceosome assembly ^31–35. This step is thought to be earlier than the complexes studied here that already have U2 base-paired at the branchpoint. The RBM5 G-patch could play a similar role in activating DHX15 to abort spliceosome assembly on certain exons before the first catalytic step. This hypothesis is compatible with studies indicating that RBM5 inhibits splicing by preventing the association of the tri-snRNP ¹⁵. The isolated U2 complexes also contain two other G-patch proteins, CHERP and RBM17/SPF45. CHERP and RBM17 have been shown interact with the DHX15, with each other, and to co-regulate alternative splicing ^17,67,68. All these proteins offer many possibilities for regulatory steps in both the quality control of early spliceosome assembly on constitutive introns, and in the selective rejection of particular complexes to control the choice of splice sites and exons.

Taking all the results together, we favor a model where RBM5 and RBM10 act primarily as U2-associated splicing repressors (Figure 6). These proteins are components of the branch site recognition machinery in A or B-like prespliceosomes assembled on nearly all exons. RBM5 or RBM10 could potentially disrupt this complex by activating the DHX15 helicase or otherwise preventing productive spliceosome assembly on particular exons. The specificity of their targeting could be influenced by branch site-proximal sequences that affect assembly kinetics rather than being directly recognized by the RNA binding domains, although direct pre-mRNA interactions are not ruled out. As integral parts of the assembling spliceosome, these factors differ from other well studied splicing regulators and it will be interesting to examine the interactions of RBM5 and RBM10 within the U2 branchpoint complex in more detail.

Figure 6.

Model for RBM5 and RBM10 regulation of exon skipping as component of the branch site selection machinery.

RBM5 and RBM10 enter early stage spliceosomes through interaction with U2AF2/U2AF65, as previously observed¹⁵. U2AF1/2 are eventually displaced from the spliceosome, while RBM5/10 remain associated through direct interactions with U2 snRNP proteins, allowing them to alter splicing at a later stages of spliceosome assembly.

Limitations of this study

We describe an experimental approach that allows biochemical characterization of endogenously formed spliceosomes. We demonstrate the versatility and sensitivity of this method both for understanding protein regulators of splicing and for the transcriptome-wide mapping of assembling spliceosomes on nascent RNA. Currently this approach is limited to cells engineered to express epitope-tagged proteins, and we are working to develop antibodies that allow the isolation of untagged RNP complexes. We are also optimizing the extraction method to broaden its application to additional spliceosomal components and to other informational processes in the cell nucleus. We show that two splicing regulators, RBM5 and RBM10 are integral spliceosome components that repress splicing in part through interactions with the branch site-recognition machinery. It is not yet clear which U2 snRNP components are direct binding partners of these two proteins. The presence of RBM5/10 within U2/branchpoint complexes helps explain earlier CLIP studies that found RBM5/10 crosslinking at 3’ splice sites but not correlation of crosslinks with regulated exons. It is not ruled out that RBM5/10 also make sequence specific contacts with some introns. Although we find that interaction with U2 is required for splicing regulation, the features that distinguish an RBM5/10 regulated exon are still obscure.

STAR METHODS

Lead contact:

Requests for information, resources, and reagents should be directed to the lead contact, Andrey Damianov (damianov@microbio.ucla.edu).

RESOURCE AVAILABILITY

Materials availability:

Plasmids and cell lines generated in this study are available upon request.

Data and Code Availability:

• Sequencing data generated in this study has been deposited at NCBI GEO under the accession number GSE240608.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Cell culture conditions:

Human Flp-In^™ T-REx^™ System (Thermo Fisher Scientific) were maintained with DMEM (Corning) supplemented with 10% (v/v) fetal bovine serum (Omega scientific). Cells were cultured at 37 °C with 5% CO2 and were monitored for mycoplasma contamination using PlasmoTest kit (InvivoGen).

METHOD DETAILS

Cell lines:

Stable HEK293 lines expressing RBM5, RBM6, RBM10, and SF3A3 proteins were prepared using the Flp-In^™ T-REx^™ System (Thermo Fisher Scientific). An RBM5/10-deficient clone (RBM5^−/−; RBM10^−/CE) derived from this cell line was obtained by pairs of CRISPR/Cas9-guided deletions of RBM5 exons 8 and 9 and RBM10 exon 5 using pX330 vector⁷⁰. Targeted sequences were as follows: RBM5 (TGCTCCCTGCCCCAGTTAGT_AGG; TAAGTTCATACACGATCTTT_TGG), RBM10 (TCGCTGGCAGCCACGAAGTA_AGG; GCTGAGGCTGGCCACTTGAA_GGG). Transfection of minigenes was performed in 40%-confluent monolayer cell cultures with Lipofectamine 2000 (Thermo Fisher Scientific), 10% minigene-encoding pcDNA3.1 vector, and 90% empty vector for 4 hours. Cells were then harvested 48 hours post-induction with doxycycline, samples for RNA and protein analysis were collected.

RT-PCR:

Total RNA was extracted with TRIzol (Thermo Fisher Scientific) from pelleted cells. DNA was removed with TURBO DNase (Thermo Fisher Scientific). Reverse transcription was carried with Superscript IV (Thermo Fisher Scientific) and d(T)₂₀ reverse primer for endogenous mRNA, or with the pcDNA3.1-specific reverse primer GCAACTAGAAGGCACAGTCG for RNA transcribed from minigenes. cDNAs were amplified for 15-20 PCR cycles with Phusion Hot Start II DNA Polymerase (Thermo Fisher Scientific) and 5’-radiolabeled primers, resolved by denaturing PAGE, and detected by phosphorimaging in Amersham Typhoon and quantified in ImageQuant (GE Healthcare Bio-Sciences). Primer and minigene sequences are listed below.

RNP sample preparation:

Nucleoplasm and HMW material from highly purified nuclei were obtained as described²⁷ in lysis buffer supplemented with 0.5 mM CaCl₂. Extraction of nuclease-resistant complexes was performed as before, in presence of approximately 5 U/μl of Benzonase Nuclease (Millipore-Sigma), 0.04 U/μl TURBO DNase (Thermo Fisher Scientific), 0.01 U/μl RNase A, and 0.4 U/μl RNase T1 (RNase Cocktail, Thermo Fisher Scientific). Whole cell soluble and HMW fractions were obtained similarly from pelleted cells lysed for 5 min in lysis buffer, and subjected to RNA and DNA degradation as above. Complexes from extracted material were separated in 10-30% glycerol gradients containing 20 mM HEPES-KOH pH 7.5, 150 mM NaCl, 1.5 mM MgCl₂, 0.5 mM DTT, and 1x Complete Protease Inhibitor Cocktail (Millipore-Sigma) by centrifugation in Optima XL centrifuge and SW41Ti rotor (Beckman Coulter) for 17h at 32,000 RPM, 4°C. Gradients were manually fractionated top to bottom into 0.5 mL aliquots. Fractions of interest were dialyzed against buffer lacking glycerol and protease inhibitors twice for 30 min at 4°C in Slide-A-Lyzer MINI Dialysis Devices, 10K MWCO Membrane (Thermo Fisher Scientific).

Extracts from nuclear or whole cell fractions, and dialyzed gradient fractions were incubated overnight at 4 °C with 5-7.5 μl packed-volume M2 FLAG agarose beads (Millipore-Sigma), washed four times with wash buffer (20 mM HEPES-KOH pH 7.5, 150 mM NaCl, 1.5 mM MgCl₂, and 0.05% Triton-X100), and eluted for two hours at 4°C in presence of 150 ng/μl of 3xFLAG peptide (Millipore-Sigma).

Protein analysis:

Immunopurified proteins were separated by SDS-PAGE and detected by immunoblotting with fluorescently-labeled secondary antibodies (GE Healthcare Bio-Sciences), or stained with SYPRO Ruby (Thermo Fisher Scientific). Images were visualized in Amersham Typhoon and quantified as above. Primary antibodies are listed in Extended Experimental Procedures. Proteins from these samples were also pelleted with four volumes of acetone at − 20°C overnight, washed with 85% Ethanol, and subjected to Liquid Chromatography Tandem Mass Spectrometry (LC-MS/MS). Spectral counts from proteomic analysis of the indicated purifications are shown in Tables S1 and S2.

RNA analysis:

RNA from deproteinized samples, each in a triplicate, was allowed to hybridize with U2- antisense DNA oligonucleotide (reverse-complimentary to GeneBank sequence NR_002716) and incubated with 0.375 U/μl RNase H (New England BioLabs) for 20 min at 37 °C. DNA was digested with TURBO DNase simultaneously with removal of 5’ and 3’ phosphates with FastAP alkaline phosphatase (Thermo Fisher Scientific) for 45 min at 37 °C. Analytical aliquots (5-10%) of RNA were 5’-radiolabeled by incubation in presence of γ-[³²P] ATP (PerkinElmer) and T4 Polynucleotide kinase (New England BioLabs), resolved by denaturing PAGE, and imaged in Amersham Typhoon. Sequencing libraries were obtained from preparative aliquots (90-95%) of RNA using iCLIP reagents⁷¹ and similar strategy with changes to allow for reactions in solution instead of on carrier surface. A detailed protocol is provided in Methods S1. These libraries were sequenced on a HiSeq4000 (Illumina).

Genome-wide splicing analysis:

Total TRIzol-extracted RNA was treated with TURBO DNase. cDNA libraries were prepared from polyA+ RNA with TruSeq Kits (Illumina).

Antibodies

Primary antibodies used for immunoblot assays: FLAG (F3165-1MG, Sigma), DHX15 (Novus Biologicals, NBP2-13919), hnRNP M (Novus Biologicals, NB200-314), RBM5 (Thermo Fisher Scientific, PA5-35963), RBM10 (Thermo Fisher Scientific, PA5-83253), U2AF2 (Thermo Fisher Scientific, PA5-30442), PRPF3 (Thermo Fisher Scientific, PA5-96337), U2-A’/SNRPA1 (Abcam, ab128937), SNRPB (Smith antigen antibody Y12, Novus Biologicals, NB600-546), RBM17/SPF45 (Bethyl Laboratories, A302-498A). SF3A3: Rabbit monoclonal antibody against the C-terminal sequence KKTYEDLKRQGLL, U1-70K, SF3A1 and SF3B1 are described⁷².

Gene expression vectors:

RBM5, RBM6, and RBM10 cDNAs were obtained by RT-PCR with 5’ and 3’ UTR primers from total HEK293 RNA. These were sequenced and confirmed to encode proteins identical to GeneBank sequences NM_005778 RBM6, NM_001167582, and NM_001204467, respectively. Human SF3A3 was obtained from plasmid DNA carrying an insertion identical with NM_006802. The protein coding regions were fused with C-terminal FLAG tag sequence and inserted in pcDNA5/FRT/TO vector (Thermo Fisher Scientific). Splicing reporter minigenes containing the full-length sequences of flanking exons and the intervening introns were obtained by PCR from HEK293 genomic DNA. These were inserted into the NheI and ApaI sites of pcDNA3.1 vector (Thermo Fisher Scientific). Sequences and descriptions of mutant derivatives are listed in Data S5.

Expression vectors

RBM5-Flag, 2481bp-long sequence, inserted between the BamHI and XhoI sites of the vector pCDNA5/FRT/TO:

gccaccATGGGTTCAGACAAAAGAGTGAGTAGAACAGAGCGTAGTGGAAGATACGGTTCCATCATAGACAGGGATGACCGTGATGAGCGTGAATCCCGAAGCAGGCGGAGGGACTCAGATTACAAAAGATCTAGTGATGATCGGAGGGGTGATAGATATGATGACTACCGAGACTATGACAGTCCAGAGAGAGAGCGTGAAAGAAGGAACAGTGACCGATCCGAAGATGGCTACCATTCAGATGGTGACTATGGTGAGCACGACTATAGGCATGACATCAGTGACGAGAGGGAGAGCAAGACCATCATGCTGCGCGGCCTTCCCATCACCATCACAGAGAGCGATATTCGAGAAATGATGGAGTCCTTCGAAGGCCCTCAGCCTGCGGATGTGAGGCTGATGAAGAGGAAAACAGGTGTAAGCCGTGGTTTCGCCTTCGTGGAGTTTTATCACTTGCAAGATGCTACCAGCTGGATGGAAGCCAATCAGAAAAAGTTGGTGATTCAAGGAAAGCACATTGCAATGCATTATAGCAATCCCAGACCTAAGTTTGAAGATTGGCTTTGTAACAAGTGCTGCCTTAACAATTTCAGGAAAAGACTAAAATGCTTCCGATGTGGAGCAGACAAGTTTGACTCTGAACAGGAAGTGCCTCCTGGAACCACAGAGTCGGTTCAGTCTGTGGATTACTACTGTGATACGATCATTCTTCGGAACATAGCTCCGCACACTGTGGTGGATTCCATCATGACAGCACTGTCTCCTTACGCGTCTTTAGCTGTCAATAACATCCGCCTCATAAAAGACAAACAGACCCAGCAGAACAGAGGCTTCGCATTTGTGCAGCTGTCCTCTGCAATGGATGCTTCTCAGCTGCTTCAGATATTACAGAGTCTCCATCCTCCTTTGAAAATTGATGGCAAAACTATTGGGGTTGATTTTGCAAAAAGTGCCAGAAAAGACTTGGTCCTCTCAGATGGTAACCGCGTCAGCGCTTTCTCTGTAGCTAGTACGGCTATTGCTGCTGCTCAGTGGTCATCCACCCAGTCTCAAAGTGGTGAAGGAGGCAGTGTTGACTACAGTTATCTGCAACCAGGTCAAGATGGCTATGCCCAATATGCTCAGTATTCACAGGATTATCAGCAGTTTTATCAACAACAAGCTGGAGGATTGGAATCTGATGCATCATCTGCATCAGGCACAGCAGTGACCACCACCTCAGCGGCTGTAGTGTCCCAGAGTCCTCAGCTGTATAATCAAACCTCCAATCCACCTGGCTCTCCGACTGAGGAAGCACAGCCTAGCACTAGCACAAGTACACAGGCCCCAGCCGCTTCCCCTACTGGTGTAGTTCCTGGTACCAAATATGCAGTACCTGACACGTCCACTTACCAGTATGATGAATCTTCAGGATATTACTATGATCCGACAACAGGGCTCTATTATGACCCCAACTCGCAATACTACTATAATTCCTTGACCCAGCAGTACCTTTACTGGGATGGGGAAAAAGAGACCTACGTGCCAGCTGCAGAGTCTAGCTCCCACCAGCAGTCGGGCCTGCCTCCTGCAAAAGAGGGGAAAGAGAAGAAGGAGAAACCCAAGAGCAAAACAGCCCAGCAGATTGCCAAAGACATGGAACGCTGGGCTAAGAGTTTGAATAAGCAGAAAGAAAACTTTAAAAATAGCTTTCAGCCTGTCAATTCCTTGAGGGAAGAAGAAAGGAGAGAATCTGCTGCAGCAGACGCTGGCTTTGCTCTCTTTGAGAAGAAGGGAGCCTTAGCTGAAAGGCAGCAGCTCATCCCAGAATTGGTGCGAAATGGAGATGAGGAGAATCCCCTCAAAAGGGGTCTGGTTGCTGCTTACAGTGGTGACAGTGACAATGAGGAGGAGCTGGTGGAGAGACTTGAGAGTGAGGAAGAGAAGCTAGCTGACTGGAAGAAGATGGCCTGTCTGCTCTGCCGGCGCCAGTTCCCGAACAAAGATGCCCTAGTCAGGCACCAGCAACTCTCAGACCTTCACAAGCAAAACATGGACATCTATCGACGATCCAGGCTGAGCGAGCAGGAGCTGGAAGCCTTGGAGCTAAGGGAGAGAGAGATGAAATACCGAGACCGAGCTGCAGAAAGACGGGAGAAGTACGGCATTCCAGAACCTCCAGAGCCCAAGCGCAAGAAGCAGTTTGATGCCGGCACTGTGAATTACGAGCAACCCACCAAAGATGGCATTGACCACAGTAACATTGGCAACAAGATGCTGCAGGCCATGGGCTGGCGGGAAGGCTCTGGCTTGGGACGAAAGTGTCAAGGCATTACGGCTCCCATTGAGGCTCAAGTTCGGCTAAAGGGAGCTGGCCTAGGAGCCAAAGGCAGCGCATATGGTTTGTCGGGCGCCGATTCCTACAAAGATGCTGTCCGGAAAGCCATGTTTGCCCGGTTCACTGAGATGGAGATGGACTACAAAGACGATGACGACAAGtga

RBM5A1 Δ1-90aa: nucleotides 10-276 deleted.

RBM5 ΔRRM1+ZF: nucleotides 10-663 deleted.

RBM5 ΔRRM2: nucleotides 664-1068 deleted.

RBM5 ΔOCRE: nucleotides 1360-1554 deleted.

RBM5 Δ544-702aa: nucleotides 1636-2112 deleted.

RBM5dG-patch: deletion of nucleotides 2239-2361.

RBM5 ΔCT: nucleotides 2239-2451 deleted.

RBM5 Δ1 / Δ533-543aa: nucleotides 1603-1635 deleted.

RBM5 Δ2 / Δ544-570aa: nucleotides 1636-1716 deleted.

RBM5 Δ3 / Δ571-596aa: nucleotides 1717-1794 deleted.

RBM5 Δ4 / Δ614-629aa: nucleotides 1846-1893 deleted.

RBM5 Δ5 / Δ631-646aa: nucleotides 1891-1938 deleted.

RBM5 Δ6 / ΔZnF2 / Δ647-677aa: nucleotides 1945-2037 deleted.

RBM5 Δ7 / Δ678-702aa: nucleotides 2038-2112 deleted.

RBM10-Flag, 2823bp-long sequence, inserted between the HindIII and NotI sites of the vector pCDNA5/FRT/TO:

gccaccATGGAGTATGAAAGACGTGGTGGTCGTGGTGACAGGACTGGCCGCTATGGAGCCACTGACCGCTCGCAGGATGATGGTGGGGAGAACCGCAGCCGAGACCACGACTACCGGGACATGGACTACCGTTCATATCCTCGCGAGTATGGCAGCCAGGAGGGCAAGCATGACTATGACGACTCATCTGAGGAGCAGAGTGCGGAGGATTCCTACGAGGCCTCCCCGGGCTCCGAGACTCAGCGTAGGCGGCGGCGGCGGCACAGGCACAGCCCCACCGGCCCGCCAGGCTTCCCCCGAGACGGCGACTATCGGGACCAGGACTATCGGACCGAGCAAGGGGAGGAGGAGGAGGAGGAGGAGGATGAGGAGGAGGAGGAGAAGGCCAGTAACATCGTCATGCTGAGGATGCTGCCACAGGCAGCCACTGAGGATGACATCCGTGGCCAGCTGCAGTCGCACGGCGTGCAAGCACGGGAGGTTCGGCTGATGCGGAACAAATCTTCAGGTCAGAGCCGGGGCTTCGCCTTCGTCGAGTTTAGTCACTTGCAGGACGCTACACGATGGATGGAAGCCAATCAGCACTCCCTCAACATCCTGGGCCAGAAGGTGTCGATGCACTACAGTGACCCCAAGCCCAAGATCAATGAGGACTGGCTGTGCAATAAGTGTGGCGTCCAGAACTTCAAACGCCGAGAGAAGTGCTTCAAATGTGGCGTGCCCAAGTCAGAGGCAGAGCAGAAGCTGCCCCTCGGCACGAGGCTGGATCAGCAGACACTGCCACTGGGTGGCCGGGAGCTGAGCCAGGGCCTGCTTCCCCTGCCGCAGCCCTACCAGGCCCAGGGAGTCCTGGCCTCCCAAGCCCTGTCACAGGGCTCGGAGCCAAGCTCAGAGAACGCCAATGACACCATCATTTTGCGCAACCTGAACCCACACAGCACCATGGATTCCATCCTGGGGGCCCTGGCACCCTACGCGGTGCTGTCCTCCTCCAACGTGCGCGTCATAAAGGACAAGCAGACCCAACTGAACCGCGGCTTTGCCTTCATCCAGCTCTCCACCATCGAGGCAGCCCAGCTGCTGCAGATCCTGCAGGCCCTGCACCCACCACTCACTATCGACGGCAAGACCATCAATGTTGAGTTTGCCAAGGGTTCTAAGAGGGACATGGCCTCCAATGAAGGCAGTCGCATCAGTGCTGCCTCTGTGGCCAGCACTGCCATTGCTGCGGCCCAGTGGGCCATCTCACAGGCCTCCCAAGGTGGGGAGGGTACCTGGGCCACCTCCGAGGAGCCGCCGGTCGACTACAGCTACTACCAACAGGATGAGGGCTATGGCAACAGCCAGGGCACAGAGTCTTCCCTCTATGCCCATGGCTACCTCAAGGGCACCAAGGGCCCTGGCATCACTGGAACCAAAGGGGATCCCACTGGAGCAGGTCCCGAGGCCTCCCTAGAGCCTGGGGCCGACTCTGTGTCGATGCAGGCTTTCTCTCGCGCCCAGCCTGGTGCTGCTCCTGGCATCTACCAACAATCAGCCGAGGCGAGCAGTAGCCAGGGCACTGCTGCCAACAGCCAGTCGTATACCATCATGTCACCCGCTGTGCTCAAATCTGAGCTCCAGAGCCCTACCCATCCTAGTTCTGCTCTCCCACCGGCTACCAGCCCCACTGCCCAGGAATCCTACAGCCAGTACCCTGTTCCCGACGTCTCTACCTACCAGTACGATGAGACCTCCGGCTACTACTATGACCCCCAGACCGGCCTCTACTATGACCCCAACTCCCAGTATTACTACAATGCTCAGAGCCAGCAGTACCTGTACTGGGATGGGGAGAGGCGGACCTATGTTCCCGCCCTGGAGCAGTCGGCCGACGGACATAAGGAGACAGGGGCACCCTCGAAGGAGGGCAAAGAGAAGAAGGAGAAGCACAAGACCAAGACAGCTCAACAGATTGCCAAGGACATGGAACGCTGGGCCCGCAGTCTCAACAAACAAAAAGAAAACTTCAAAAATAGCTTCCAGCCTATCAGCTCCCTGCGAGATGACGAGAGGCGGGAGTCAGCCACTGCAGATGCTGGCTATGCCATCCTCGAGAAGAAGGGAGCACTAGCCGAGAGACAGCACACCAGCATGGATCTCCCGAAATTGGCCAGTGACGACCGCCCAAGCCCTCCGCGAGGACTGGTGGCAGCCTACAGCGGGGAGAGTGACAGTGAGGAGGAGCAGGAGCGTGGGGGCCCTGAGCGGGAGGAGAAGCTCACCGACTGGCAGAAGCTGGCCTGTCTGCTCTGCCGACGCCAGTTCCCCAGCAAAGAGGCGCTCATCCGGCACCAGCAGCTCTCAGGGCTCCACAAGCAAAACCTTGAGATTCACCGGCGcGCCCACTTGTCAGAAAACGAGCTAGAAGCACTAGAGAAGAATGACATGGAGCAAATGAAGTACCGGGACCGTGCAGCTGAACGCAGAGAAAAGTATGGCATCCCCGAGCCGCCAGAGCCCAAGAGGAGGAAGTACGGCGGCATATCCACAGCCTCTGTAGACTTCGAGCAGCCTACTCGGGACGGGCTGGGCAGTGACAACATTGGCAGTCGGATGCTGCAGGCCATGGGCTGGAAAGAGGGCAGCGGCCTGGGCCGCAAGAAGCAGGGCATTGTAACGCCTATCGAGGCCCAAACACGGGTGCGGGGCTCCGGCCTGGGTGCACGGGGCAGCTCCTACGGGGTCACCTCAACCGAGTCCTACAAGGAGACACTGCACAAGACAATGGTGACCCGCTTCAACGAGGCCCAGATGGACTACAAAGACGATGACGACAAGtga

RBM10 Δ655-816aa: nucleotides 1966-2451 deleted.

RBM10 ΔG-patch: nucleotides 2584-2706 deleted.

SF3A3-Flag, 1535bp-long sequence, inserted between the BamHI and NotI sites of the vector pCDNA5/FRT/TO:

ccaccATGGAGACAATACTGGAGCAGCAGCGGCGCTATCATGAGGAGAAGGAACGGCTCATGGACGTCATGGCTAAAGAGATGCTCACCAAGAAGTCCACGCTCCGGGACCAGATCAATTCTGATCACCGCACTCGGGCCATGCAAGATAGGTATATGGAGGTCAGTGGGAACCTGAGGGATTTGTATGATGATAAGGATGGATTACGAAAGGAGGAGCTCAATGCCATTTCAGGACCCAATGAGTTTGCTGAATTCTATAATAGACTCAAGCAAATAAAGGAATTCCACCGGAAGCACCCAAATGAGATCTGTGTGCCAATGTCAGTGGAATTTGAGGAACTCCTGAAGGCTCGAGAGAATCCAAGTGAAGAGGCACAAAACTTGGTGGAGTTCACAGATGAGGAGGGATATGGTCGTTATCTCGATCTCCATGACTGTTACCTCAAGTACATTAACCTGAAGGCATCTGAGAAGCTGGATTATATCACATACCTGTCCATCTTTGACCAATTATTTGACATTCCTAAAGAAAGGAAGAATGCAGAGTATAAGAGATACCTAGAGATGCTGCTTGAGTACCTTCAGGATTACACAGATAGAGTGAAGCCTCTCCAAGATCAGAATGAACTTTTTGGGAAGATTCAGGCTGAGTTTGAGAAGAAATGGGAGAATGGGACCTTTCCTGGATGGCCGAAAGAGACAAGCAGTGCCCTGACCCATGCTGGAGCCCATCTTGACCTCTCTGCATTCTCCTCCTGGGAGGAGTTGGCTTCTCTGGGTTTGGACAGATTGAAATCTGCTCTCTTAGCTTTAGGCTTGAAATGTGGCGGGACCCTAGAAGAGCGAGCCCAGAGACTATTCAGTACCAAAGGAAAGTCCCTGGAGTCACTTGATACCTCTTTGTTTGCCAAAAATCCCAAGTCAAAGGGCACCAAGCGAGACACTGAAAGGAACAAAGACATTGCTTTTCTAGAAGCCCAGATCTATGAATATGTAGAGATTCTCGGGGAACAGCGACATCTCACTCATGAAAATGTACAGCGCAAGCAAGCCAGGACAGGAGAAGAGCGAGAAGAAGAGGAAGAAGAGCAGATCAGTGAGAGTGAGAGTGAAGATGAAGAGAACGAGATCATTTACAACCCCAAAAACCTGCCACTTGGCTGGGATGGCAAACCTATTCCCTACTGGCTGTATAAGCTTCATGGCCTAAATATCAACTACAACTGTGAGATTTGTGGAAACTACACCTACCGAGGGCCCAAAGCCTTCCAGCGACACTTTGCTGAATGGCGTCATGCTCATGGCATGAGGTGTTTGGGCATCCCAAACACTGCTCACTTTGCTAATGTGACACAGATTGAAGATGCTGTCTCCTTGTGGGCCAAACTGAAATTGCAGAAGGCTTCAGAACGATGGCAGCCTGACACTGAGGAAGAATATGAAGACTCAAGTGGGAATGTTGTGAATAAGAAGACATACGAGGATCTGAAAAGACAAGGACTGCTCGACTACAAAGACGATGACGACAAGtga

Primers

Primers for RBM5 deletion genotyping:

RBM5 intron 7 forward: aactgtgtcccctgtccagt

RBM5 exon 8 forward: CCTCCTGGAACCACAGAGTC

RBM5 intron 9 reverse: atacacatgcattcccacca

Primers for RBM10 deletion genotyping:

hRBM10 intron 4 forward: aagcaatccccaggactacat

hRBM10 exon 5 forward: GCTGATGCGGAACAAATCTT

hRBM10 intron 5 reverse: tcagagtcccagagaggaggt

RT-PCR primers for RBM10 cryptic exon detection and sequencing:

hRBM10 exon 4 forward: GGCCAGTAACATCGTCATGC

hRBM10 exon 7 reverse: TATTGCACAGCCAGTCCTCA

RT-PCR primers for detection of RBM5 and RBM10-dependent cassette exons expressed from endogenous genes and three-exon minigenes:

TRPT1 exon 5 forward: CATGCGGTCCCATTGTGAAA

TRPT1 exon 7 reverse: TTCTCTGGAGCTGTGCTTGG

STAC3 exon 8 forward: CCTGAAGGGGATAAGAAGGCT

STAC3 exon 10 reverse: CGCCACCATTCTTCATTGGAGT

ZDHHC16 exon 7 forward: AGCTATGGAAGTTGGGACCT

ZDHHC16 exon 9 reverse: GAAAGGAGAAGGTGGGTGGT

RNASET2 exon 2 forward: ACTAATTATGGTTCAGCACTGGC

RNASET2 exon 4 reverse: AAATTGAAGGGCCACGATCT

NUMB exon 8 forward: TCTGCTCCGATGACCAAACC

NUMB exon 10 reverse: GTACGTCTATGACCGGCCTG

HOTAIR exon 2 forward: CATTCTGCCCTGATTTCCGG

HOTAIR exon 4 reverse: AATCCGTTCCATTCCACTGC

PCBP2 exon 11 forward: TGGCAATGCAACAGTCTCAT

PCBP2 exon 13 reverse: TCGTTTGGAATGGTGAGTTCA

DPP7 exon 4 forward: CCATCGCCTTCGGTGGAA

DPP7 exon 7 130-111: GTGGGGTAGGGGTAGTCCAT

(for endogenous gene splicing analysis)

DPP7 exon 5 forward: GCGACTCCAACCAGTTCTTC

DPP7 exon 7 reverse: CAGGAAGTCAGTGGGGTAGG

(for three-exon minigene splicing analysis)

QUANTIFICATION AND STATISTICAL ANALYSIS

IP-seq data analysis:

PCR duplicates were removed using random barcodes. Unique reads from RBM5 library prepared without selective U2 snRNA degradation mapping to snRNA sequences were identified with BLAST ⁷³, allowing for one mismatch per 20 nt. Unique reads from all libraries were mapped to human genome hg19/NCBI37 using Bowtie2, excluding partially aligned reads and allowing for one mismatch per 20 nt⁷⁴. Regions of at least ten overlapping reads in each experiment were defined by YODEL⁷⁵, and trimmed to exclude terminal sequence with coverage lower than 50% of the maximum. Overlapping regions from one or more RNP-seq or IP-seq experiment were then combined to define clusters. These were further categorized as 5’ splice site or branch site-associated if being the closets to a 5’ or a 3’ end of an annotated intron, with maximum allowed distance of 10 and 100nt, respectively. Cluster RPKM was calculated with SeqMonk software (v1.45.4, Babraham Institute).

The distribution of known branched nucleotide positions on reads was determined from a subset of clusters, unambiguously identified as branch site-associated, with width of less than 53 nt. Clusters overlapping with less than one, or with multiple branch point identified in ^39,40 were removed. Also excluded were clusters where read coverage at the branch point nucleotide was less than 85% of the maximum coverage for the cluster. Median branch point positions and IQR intervals as fraction of read length were determined for each library. Using these distributions, a putative branch point was then assigned to each read from clusters other than categorized as 5’ splice site-associated. These were chosen as the position within the IQR for the read with the highest HSF branch site score^76,77. Finally, the putative reads within each cluster were ranked by the fraction of supporting reads out of all cluster reads. For branch site-associated clusters on U12-type introns⁷⁸, a single U12-type branch point with the highest similarity to the U12 branch site consensus sequence out of all predicted branch points was called separately.

RNAseq data splicing analysis:

Alternative splicing was analyzed by rMATS⁵³ and expressed as changes in percent-spliced-in values (ΔPSI). Exons showing splicing change (|ΔPSI|)>10 with FDR less than 0.05) between control RBM5^−/−; RBM10^−/CE cells, and cells expressing RBM5-Flag or RBM10-Flag were considered regulatory targets. RBM10-dependent splicing events in mouse ES and MEPA cells were as described²⁵. RBM5-dependent splicing events in HepG2 cells were derived from Encode⁵⁸ with rMATS⁵³ using cutoff criteria as in ²⁵.

Processing of immunoprecipitation-crosslinking data:

eCLIP and iCLIP data were processed following the standard Encode procedures. Mapping was done using STAR⁷⁹. Crosslink sites were detected using iCount⁸⁰. Significant crosslinked sites were defined with a FDR less than 0.05. Crosslink sites overlapped with control set were removed from the downstream analysis.

Motif enrichment analyses:

Motifs enriched upstream or within RBM5- and/or RBM10-downregulated cassette exons were determined using STREME⁵⁷. A set of cassette exons not regulated by RBM5 or 10 (|ΔPSI| < 5 and FDR > 0.05) served as control. Intronic regions were selected based on the position of rank1 branch point from the corresponding branch site cluster.

ADDITIONAL RESOURCES

A detailed protocol for IPseq and RNPseq is available as a supplementary file: Methods S1

Supplementary Material

1

Data S1. U2AF2 eCLIP ⁵⁸ significant crosslink sites in human HepG2 cells (Related to Figure 3).

2

Data S2. RBM5 eCLIP ⁵⁸ significant crosslink sites in human HepG2 cells (Related to Figure 3).

3

Data S3. RBM10 iCLIP ²⁵ CLIP significant crosslink sites in mouse MEPA cells (Related to Figure 3).

4

Data S4. RBM5-regulated cassette exons in human HepG2 cells ⁵⁸ (Related to Figure 4).

5

Data S5.

6

Table S1. Mass-spectrometric analyses of samples immunoprecipitated from HMW extract with the indicated Flag-tagged protein (Related to Figure 1).

Table S2. Mass-spectrometric analyses of samples immunoprecipitated with RBM5-Flag from the indicated glycerol gradient fractions of the HMW extract (Related to Figures 1 and 2).

Table S3. Summary of IP-seq and RNP-seq, and RNAseq data (Related to Figure 3).

Table S4. RNP-seq and IP-seq internal intronic clusters associated with recursive splice sites and exons annotated in the HEXEvent database⁴⁸ (Related to Figure 3).

Table S5. RBM5-dependent splicing changes (Related to Figure 4).

Table S6. RBM10-dependent splicing changes (Related to Figure 4).

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
FLAG	Millipore-Sigma	F3165-1MG
DHX15	Novus Biologicals	NBP2-13919
hnRNP M	Novus Biologicals	NB200-314
RBM5	Thermo Fisher Scientific	PA5-35963
RBM10	Thermo Fisher Scientific	PA5-83253
U2AF2	Thermo Fisher Scientific	PA5-30442
PRPF3	Thermo Fisher Scientific	PA5-96337)
U2-A’/SNRPA1	Abcam	ab128937
Smith antigen antibody Y12	Novus Biologicals	NB600-546
RBM17/SPF45	Bethyl Laboratories	A302-498A
SF3A3	This study, Rabbit monoclonal antibody against the C-terminal sequence KKTYEDLKRQGLL	N/A
U1-70K	Sharma et al. 2014. 72	N/A
SF3A1	Sharma et al. 2014. 72	N/A
SF3B1	Sharma et al. 2014. 72	N/A
Chemicals, Peptides, and Recombinant Proteins
TRIzol	Thermo Fisher Scientific	15596018
TURBO DNase	Thermo Fisher Scientific	AM2239
SuperScript IV	Thermo Fisher Scientific	18090050
Phusion Hot Start II DNA Polymerase	Thermo Fisher Scientific	F549S
Benzonase Nuclease	Millipore-Sigma	E1014-25KU
RNase Cocktail	Thermo Fisher Scientific	AM2286
Complete Protease Inhibitor Cocktail	Millipore-Sigma	4693132001
M2 FLAG agarose	Millipore-Sigma	A2220-5ML
3xFLAG peptide	Millipore-Sigma	F4799-4MG
SYPRO Ruby	Thermo Fisher Scientific	S12000
RNase H	New England BioLabs	M0297S
FastAP	Thermo Fisher Scientific	EF0654
γ-[32P] ATP	PerkinElmer	NEG035C005MC
Proteinase K	New England BioLabs	P8107S
acidic phenol:chloroform:isoamyl alcohol (125:24:1)	Thermo Fisher Scientific	AM9720
GlycoBlue	Thermo Fisher Scientific	AM9516
RNaseOUT	Thermo Fisher Scientific	10777019
T4 Polynucleotide Kinase	New England BioLabs	M0201S
T4 RNA ligase1	New England BioLabs	M0204S
phenol:chloroform:isoamyl alcohol (125:24:1), pH 7.7–8.3	Millipore-Sigma	77617-100ML
CircLigase II ssDNA ligase	Biosearch Technologies	CL9021K
FastDigest BamHI	Thermo Fisher Scientific	FD0054
dNTP Solution Mix	New England BioLabs	N0447S
ATP	New England BioLabs	P0756S
Deposited Data
ENCODE RBM5 eCLIP and RNA-seq	Luo et al. 2020. 58	N/A
RBM10 iCLIP and RNA-seq	Rodor et al. 2017. 25	N/A
RNA sequencing data	This study	GSE240608
Experimental Models: Cell Lines
Flp-In™ T-REx™	Thermo Fisher Scientific	R78007
RBM5−/−; RBM10−/CE	This study	N/A
Oligonucleotides
DNA and RNA oligonucelotide sequences listed in separate method sections
Recombinant DNA
pcDNA5/FRT/TO	Thermo Fisher Scientific	V652020
pOG44	Thermo Fisher Scientific	V600520
pX330	Ran et al. 2013. 70	N/A
Software and Algorithms
BLAST	Camacho et al. 2009. 73	N/A
Bowtie2	Langmead et al. 2009. 74	N/A
YODEL	Palmer et al. 2017. 75	N/A
SeqMonk v1.45.4	Babraham Institute https://www.bioinformatics.babraham.ac.uk/projects/seqmonk/	N/A
rMATS	Shen et al. 2014. 53	N/A
STAR	Dobin et al. 2013. 79	N/A

HIGHLIGHTS.

• Isolation of endogenous spliceosomes allows whole transcriptome mapping of branch sites.

• Regulators RBM5 and RBM10 are components of the U2 snRNP bound to all branch sites.

• RBM5 and RBM10 regulate splicing after branch site recognition.