Cell-type specific regulator RBPMS switches alternative splicing via higher-order oligomerization and heterotypic interactions with other splicing regulators

Abstract Alternative pre-mRNA splicing decisions are regulated by RNA binding proteins (RBPs) that can activate or repress regulated splice sites. Repressive RBPs typically harness multivalent interactions to bind stably to target RNAs. Multivalency can be achieved by homomeric oligomerization and heteromeric interactions with other RBPs, often mediated by intrinsically disordered regions (IDRs), and by possessing multiple RNA binding domains. Cell-specific splicing decisions often involve the action of widely expressed RBPs, which are able to bind multivalently around target exons, but without effect in the absence of a cell-specific regulator. To address how cell-specific regulators can collaborate with constitutive RBPs in alternative splicing regulation, we used the smooth-muscle specific regulator RBPMS. Recombinant RBPMS is sufficient to confer smooth muscle cell specific alternative splicing of Tpm1 exon 3 in cell-free assays by preventing assembly of ATP-dependent splicing complexes. This activity depends upon a C-terminal IDR that facilitates dynamic higher-order self-assembly, cooperative binding to multivalent RNA and interactions with widely expressed splicing co-regulators, including MBNL1 and RBFOX2, allowing cooperative assembly of stable cell-specific regulatory complexes.


INTRODUCTION
Alternati v e pre-mRNA splicing (AS) is a widespread phenomenon in eukaryotes that allows multiple transcripts to be generated from individual genes, often leading to the production of functionally distinct protein isoforms with profound effects on cell and organismal phenotype ( 1 , 2 ). Genome-wide studies demonstrate that most AS e v ents (ASEs) are mediated by the combinatorial and tissuespecific binding of multiple RNA binding proteins (RBPs) (3)(4)(5). The human genome encodes at least 1500 RBPs ( 6 , 7 ), many of which comprise one or more RNA binding domains (RBDs) along with a variety of intrinsically disorder ed r egions (IDRs) ( 8 ). While much focus has been placed on the role of structurally ordered RBDs in the recognition of specific RNA motifs ( 6 ), recent studies have also begun to unveil the biological significance of the IDRs ( 9 , 10 ).
Most splicing regulatory RBPs have pr eferr ed binding motifs, which act as splicing enhancers or silencers depending on RBP activity and motif position relative to the ASEs ( 11 ). RBPs work either synergistically or antagonistically to modulate spliceosome assembly at regulated splice sites, resulting in either e xon acti vation or r epr ession (12)(13)(14). Tissue-specific 'splicing codes' comprise different combinations of enhancer and silencer motifs along with other transcript features ( 11 , 15 ). Most features of cell-specific splicing codes are not unique to that cell type, reflecting the roles of widely expressed RBPs in cell-specific splicing decisions ( 15 ). Ne v ertheless, some splicing regulators show mor e r estricted expr ession and may act as master regulators of splicing programmes ( 13 , 16 ). The outcome of splicing decisions can be viewed as resulting from a competition between activating and r epr essive inputs, while switching between AS patterns can result from modulation of either or both sets of inputs. For example, many neuron-specific exons are included as a result of reduced repression by PTBP1 combined with increased activation by RBFOX or SRRM3 proteins (15)(16)(17)(18).
Splicing activators of the serine-arginine (SR) protein family can act by binding to exon splicing enhancers (ESEs) via their RN A reco gnition motif (RRM) domains, w hile their SR-rich IDRs either recruit core splicing factors to splice sites or stabilize interactions within splicing complexes [discussed in ( 19 )]. Increased numbers of ESEs additi v ely enhance splicing efficiency, but this arises from increased probability of initial weak binding to ESEs ( 19 ) or increased probability of interaction of ESE-bound SR proteins with core splicing factors ( 20 ) rather than cooperation between SR proteins bound to different ESEs. Splicing r epr essors broadly act in one of two ways. Their binding can directly occlude splice sites, ESEs or whole exons, blocking the binding of activating or core splicing factors ( 11 , 14 ). Alternati v ely, r epr essors can interact with RNAbound core splicing factors leading to dead-end splicing complex es (21)(22)(23). Ex emplified by the heterogeneous nuclear RNP (hnRNP) family, r epr essors have one or more RBDs and typically interact in a multivalent manner with target RNAs containing multiple cognate binding motifs. Multivalency can arise via multiple RBDs within a single protein or via oligomerization mediated by IDRs ( 14 ). The IDRs have a propensity for mediating both homomeric and heter omeric pr otein-pr otein interactions, including higher-order oligomerization and biological condensate formation, and have been shown to be functionally important in a range of splicing regulators such as RBFOX2 (24)(25)(26), hnRNPH1 ( 27 ), hnRNPA and hnRNPD ( 28 ). It has been proposed that some RBPs might act by promoting local 'binding region condensates' on target transcripts ( 29 ).
Detailed mechanistic understanding of the action of splicing regulatory RBPs can be gained from cell-free in vitr o investiga tions. For e xample, biochemical inv estigations of the SRC N1 exon have provided a detailed picture of how the archetypal repressor PTBP1 leads to exon skipping via cooperati v e binding to motifs flanking the exon ( 30 ), leading to hyperstabilized non-producti v e U1 snRNP binding at the N1 5 splice site (5 ss) ( 22 , 23 ). Here, PTBP1 acts widely as a splicing r epr essor and its reduced expression in neurons leads to N1 exon inclusion. In vitro analyses of the action of cell-specific r egulators ar e lacking, possibly due to challenges associated with expression and purification of acti v e full-length (FL) proteins with e xtensi v e IDRs. We recently found that the 22-kDa RNA binding protein RBPMS is sufficient to activate a splicing programme associated with differentiated contractile vascular smooth muscle cells (VSMCs) ( 31 ). Among the ASEs regulated by RBPMS was the switch between Tpm1 m utuall y e xclusi v e e xons 2 and 3, an e v ent that has been e xtensi v ely investigated using in vitro , in cellulo and in vivo approaches ( 32 ). Tpm1 exon 3 inclusion results from its dominant 5 ss and 3 ss elements, which outcompete the weaker exon 2 splice sites, except in dif ferentia ted SMCs where exon 3 is r epr essed ( 33 , 34 ). MBNL and PTBP1 proteins a ppl y a constituti v e r epr essi v e influence on e xon 3 by binding to flanking negati v e regulatory sequences (35)(36)(37)(38). Howe v er, both proteins are widely expressed ( 39 ), and despite the binding of up to six PTBP1 and three to eight MBNL1 proteins around Tpm1 exon 3, it is efficiently included in HeLa nuclear extract (NE) splicing reactions ( 35 , 37 , 38 ). Since RBPMS ov ere xpression is sufficient to switch Tpm1 splicing in cell lines such as HEK293T ( 31 ), we hypothesized that recombinant RBPMS might be able to confer tissuespecific splicing of Tpm1 in cell-free assays. RBPMS has a single RRM that mediates both homodimerization ( 40 , 41 ) and binding to closely spaced pairs of CAC motifs ( 42 , 43 ), a 14-amino acid N-terminal tail and an ∼80-amino acid proline-rich C-terminal IDR (Figure 1 A). The IDR is important for some functions (44)(45)(46) and can contribute to RNA binding ( 42 ), but the biophysical basis of its activity is unclear.
Here, we show that recombinant RBPMS confers cellspecific AS of Tpm1 exon 3 in vitro by remodelling the ribonucleoprotein (RNP) complexes that assemble around the exon, thereby pre v enting the formation of ATP-dependent splicing complexes. The IDR is essential for RBPMS splicing regulatory function and for its ability to bind to Tpm1 RNA in NE. It mediates higheror der oligomerization e xtending to liquid-liquid phase separ ation, cooper ati v e binding to the multivalent Tpm1 RNA, and interaction with other widely expressed splicing regulators such as MBNL1 and RBFOX2. In particular, the interaction with MBNL1 helps to recruit RBPMS to Tpm1 RNA in the competiti v e conte xt of NE, while RB-FOX2 and other proteins are recruited by RBPMS. Notably both MBNL1 and RBFOX2 co-regulate not only Tpm1 but also other VSMC regulated e v ents. Our results provide an important proof of principle for how a cell-specific splicing regulator can interact functionally and physically with more widely expressed regulators to direct their activity towards a co-regulated set of ASEs.

Cloning
The cloning of rat RBPMS-A cDNA, NCBI accession code: XM 006253240.2, into the pEGFP-C1 vector was described previously ( 31 ). To produce RBPMS-A with an N-terminal removable His 6 tag, PCR products of pET15b vector and Alternati v e e x on encodes 20-amino acid RBPMS-A isoform specific tail (yello w). C20, an experimental construct lacking the C-terminal tail. ( B ) Sequence alignment of C-terminal 20 amino acids of RBPMS-A vertebrate orthologues. Asterisk indicates fully conserved residues, colon indicates residues of strongly similar properties and period indicates residues of weakly similar properties. ( C ) Tpm1 exon 3 minigene reporter co-transfected with FLAG-tagged RBPMS in HEK293 cells. Schematic of the minigene is shown on the left. Re v erse tr anscriptase polymer ase chain reaction (RT-PCR) analysis of splicing patterns is shown above, and western blots for protein expr ession ar e sho wn belo w. Exon 3 percent spliced in (PSI) values are shown as mean ± standard deviation (SD), n = 3. Statistical significance from Student's t -test is shown as follows: ns, P > 0.05; ***, P < 0.001. ( D ) Sodium dodecyl sulphate-polyacrylamide gel electrophoresis (SDS-PAGE) analysis of purified recombinant RBPMS-A FL and C-terminal 20 amino acids truncated ( C20). ( E -H ) Sedimentation coefficient distribution plot, c ( s ), of analytical ultracentrifugation (AUC) analysis using FL or C20 RBPMS at 500 or 60 mM KCl, pH 7.9. Protein concentrations are colour coded. Data shown in panels (E) and (F) are normalized to the area under curve using GUSSI, while for panels (G) and (H) data are normalized to the maximum value of the dataset. ( I ) Cryo-electron microscopy (cryo-EM) image of glutaraldehyde cross-linked and purified RBPMS-A high-order oligomer of 660 kDa. Scale bar, 50 nm. ( J ) Fluorescence and differ ential interfer ence contr ast microgr aphs of His 6 -RBPMS-A droplets at 90 mM KCl. Fluorophore-conjugated RBPMS-Alexa 647 is added to 0.5 M to gi v e the final concentration shown. Images are representati v e of 5-10 acquired at each concentration. Scale bars are 25 m. Table S1) were treated with T4 DN A pol ymerase and joined  by ligation-independent cloning. Based on the resulting  pET15b-TEV-RBPMS-A plasmid, PCR product generated  by primers (Supplementary Table S1) was used to replace  the FL RBPMS open reading frame flanked by SalI and  XhoI restriction sites, producing pET15b-TEV-RBPMS-A-C20 plasmid. To produce RBPMS constructs for affinity purification, the segment of pET15b-TEV-RBPMS-A flanked by XbaI and SalI sites was replaced with DNA oligonucleotide (Supplementary Table S2) encoding ribosome binding, Strep tag II and His 6 tag sequences. Using the r esulting pET15b-Str epII-His 6 -RBPMS-A, r estriction enzyme cloning was conducted replacing the FL RBPMA sequence with either C20 or RRM (2-114 amino acid) sequences. As tabulated in Supplementary Table S2, Gibson assembly ( 47 ) was used to produce pET15b-StrepII-His 6 -RBPMS-A-K100E.

Expression and purification of recombinant protein
Expr ession vectors wer e transformed into Esc heric hia coli BL21(DE3) competent cells. An overnight primary culture was prepared by inoculating a single colony in 10 ml l yso geny broth (LB), 100 mg / ml ampicillin, at 37 • C with shaking. The primary culture was subsequently scaled up by using a 1:50 dilution with LB (100 mg / ml ampicillin), and protein expression was induced with 0.2 mM IPTG at OD 600nm of 0.8 for 2 h a t 37 • C . The post-induction culture was harvested by centrifuga tion a t 7000 × g for 10 min, and the pellet was resuspended in HisA buffer [50 mM Tris, 500 mM KCl, 40 mM imidazole, 10% (v / v) glycerol, 0.5 mM DTT, pH 8.5]. The resuspended cell pellet was lysed using a Fr ench pr ess. Nucleic acids wer e pr ecipitation on ice in HisA supplemented with 1 M LiCl and cOmplete protease inhibitor (Roche) for 10 min. Lysate was clarified by centrifugation (40 000 × g for 30 min), and the supernatant was filtered with 0.45 m filter, loaded on a 1-ml Histrap HP column (Cytiva) with an AKTA purifier (Cytiva) and eluted with a gradient of increasing imidazole concentration. The identified peak fractions were buffer exchanged into QA buffer (20 mM CAPS, 50 mM KCl, pH 10), and protein concentration was determined by absorbance at 280 nm. To 1 mg of recombinant protein, 25 g of TEV protease was added and incubated at 4 • C for 16 h. Tag-free protein was purified further via a Mono Q 5 / 50 GL column (Cytiva), with elution via an NaCl gr adient. RBPMS fr actions were pooled and polished with a Super de x 200 16 / 600 column (Cytiva).

Analytical ultracentrifigation
Sedimentation v elocity (SV) e xperiments were conducted using an Optima XL-1 analytical ultracentrifuge (Beckman Coulter). Samples were loaded into standard double-sector cells, 12 mm centrepiece thickness, and analysed at a speed of 40 000 rpm with a four-hole An60 Ti rotor, at 20 • C for 15 h, and 300 scans of interference optics were recorded in 90 s interval. All AUC experiments were performed in buffer containing 20 mM HEPES and 1 mM TCEP, at pH 7.9, but varied in KCl concentrations. Under 500 mM KCl con-dition, FL and C20 RBPMS of equal molar concentration were studied at 0.46 and 0.57 mg / ml, separately. At 60 mM KCl, a concentration series of FL RBPMS was analysed from 0.7, 0.48, 0.24 to 0.11 mg / ml. Analysis of C20 was conducted at either 0.5 or 0.1 mg / ml. SV data analysis was performed using SEDFIT (v14.1) program, assuming sedimentations of all species fit into a continuous c ( s ) model. The partial specific volume of the protein (FL = 0.73 ml / g, C20 = 0.74 ml / g) and the viscosity and density of the buffer ( ρ = 1.016 × 10 -2 ; ρ = 1.026) were calculated using the program SEDNTERP ( 48 ). Best c ( s ) fits were determined using over 60 scans, by fixing the meniscus, partial specific volume and solvent density, but floating the frictional ratio f / f 0 , until the overall root-mean-square deviation fall between 0.005 and 0.02. f / f 0 between 1 and 1.15 (Supplementary Figure S2A-D) was determined to be the compromised value that was used to describe both RBPMS dimer and oligomers in a single c ( s ) plot. f / f 0 values above 1.4 wer e r eached for fitting the FL RBPMS-A at 5 M (Supplementary Figure S2E) and all C20 sedimentation profiles (Supplementary Figure S2F-H).

Cryo-electr on micr oscopy
The sample was spotted on a Quantifoil 1.2 / 1.3 300 mesh Cu (10) grid (Agar Scientific), blotted and plunge frozen using a Vitrobots (Thermo Fisher Scientific). Image acquisition was carried out at the normal magnification of 92 000 × using a Falcon 3 counting detector in a Talos Arctica transmission electron microscope (Thermo Fisher Scientific).

Fluorophore labelling
Purified His 6 -TEV-RBPMS-A was exchanged to 500 mM KCl AUC buffer using a Zeba spin desalting column (7K MWCO, Thermo Fisher Scientific). Alexa Fluor 647 C2 maleimide (Thermo Fisher Scientific) was added to 10 × molar excess and incubated overnight at 4 • C in the dark. The reaction was quenched with excess ␤mercaptoethanol and buffer exchanged to QA buffer. Using an Amicon Ultra centrifugal filter (10K MWCO; Thermo Fisher Scientific), labelled protein was repetiti v ely concentrated until a 100 000 × dilution was achie v ed. The amount of free fluorophore in the mixture was estimated by SDS-PAGE (Supplementary Figure S4A).

Phase separation assays
Purified RBPMS-A was exchanged to ima ge b uffer (20 mM CAPS, KCl, 1 mM TCEP, pH 10) using a Zeba spin desalting column (7K MWCO, Thermo Fisher Scientific) and diluted as indicated. For fluor escence microscop y, the mixture included 0.5 M labelled His 6 -TEV-RBPMS-A in QA. Phase separation was induced by addition of 100 mM HEPES (pH 7.9) to final volume of 10 l. Additionally, polyvinyl alcohol (PVA) was added in experiments using tag-free RBPMS. The mixture was incubated at room temperature for 1 h in the dark. For microscopy, 5 l of the mixture was spotted onto a glass slide, covered and sealed.
Images wer e acquir ed using a Nikon ECLIPSE Ti microscope equipped with a 60 × oil-immersion differential interference contrast objecti v e. All images were acquired within 5 h of the time at which phase separation was induced.

Band shift
Recombinant proteins were buffer exchanged to buffer BS (20 mM HEPES, 100 mM KCl, 0.5 mM DTT, pH 7.9). RNA substrates (Supplementary Tables S5 and S11) were transcribed using T7 RN A pol ymerase (Thermo Fisher Scientific). A 10 l binding reaction contains 10 nM RNA, 25 mM HEPES, 100 mM KCl, 2 mM MgCl 2 , 0.625 mM DTT, 0.1 mg / ml bovine serum albumin (BSA), 5% glycerol and increasing concentrations of recombinant RBPMS, at pH 7.9. After 1 h incubation at 30 • C, 1 l heparin was added to a final concentration of 5 mg / ml and a 10 min additional incubation was performed at 30 • C. Before gel loading, the binding reactions were chilled on ice and 2 l of 50% (v / v) glycerol was added. Bound and free RNA was separated on a nati v e PAGE gel, 5%, 40:1 acrylamide:bisacrylamide ratio, using TBE running buffer at room temperature. Gels were dried and visualized by autor adiogr aphy on a Typhoon FLA 9000 (Cytiva). Binding curves were fitted with specific binding with Hill slope analysis, Y = B max × X h ( K h d + X h ) , using Prism 9 program.

Cell culture and NE preparation
HeLa S3 cells wer e cultur ed in suspension with a 5-l T-flask in SMEM (Thermo Fisher Scientific) supplemented with 10% foetal calf serum (Sigma-Aldrich), with constant agita tion a t 80 rpm a t 37 • C . Six to eight litres of HeLa S3 culture in log phase of the growth, at a cell density of 5 × 10 5 cells / ml, was harvested by centrifugation in a Megafuge (Heraeus) at 2000 rpm, at 4 • C for 10 min. The cell pellets were immediately washed twice with ice-cold phosphatebuffered saline, in total, 40 times the cell pellet volume. The downstr eam extract pr eparation was carried out strictly according to the S10 protocol detailed in ( 49 ).

In vitro transcription
Depending on the experiment, either [ ␣-32 P] CTP or UTP (Perkin-Elmer) labelled RNA transcript (specified in the figure legends) was transcribed from linearized pGEM vectors (Supplementary Table S7) with T7 polymerase. To make RNA for in vitro splicing, complex assembly and UV crosslinking assays, GTP to m7G(5 )ppp(5 )G dinucleotide cap analogue ratio was kept at 1:8 to ensure high capping efficiency. For electrophoretic mobility shift assays (EMSAs), the addition of cap analogue was omitted from the in vitro transcription mixture. The reaction mixtures are tabulated in Supplementary Table S11.

In vitro splicing
In vitro splicing was carried out as in ( 33 ( 50 , 51 ). Treated 20 l NE aliquots were used directly or stored at −80 • C.

Complex assembly
Pr e-spliceosomal complex es (10 l) were assembled on 2.5 nM of [ 32 P]-labelled pre-mRNA with 50% (v / v) HeLa NE, based on the standard in vitro splicing reaction conditions. De viations from standar d conditions are indicated in the figure legends. Reactions were incubated at 30 • C for 10 min or as indicated. After complex formation, an additional 10 min incubation was performed with heparin added to the final concentration of 0.5 mg / ml. Similar to the band shift assay, the r eactions wer e chilled on ice, to which 2 l of 50% (v / v) glycerol was added. Complexes were loaded on a prerun of nati v e PAGE gel, 4%, 80:1 acrylamide:bisacrylamide ratio, using 50 mM Tris-glycine (pH 8.8) running buffer, running at 160 V at room temperature for 5 h. Gels were dried on a filter paper, and autor adiogr aphy was performed as described above.

Protein-RNA UV cross-linking and immunoprecipitation
Twenty microlitres of pre-spliceosomal complexes subjected to UV cross-linking were assembled on 2 nM [ 32 P]-labelled RNA transcript without PVA, in otherwise identical fashion to those resolved on native gels. After complex assembly and incubation with heparin, reactions were radiated with 2 × 960 mJ 240 nm UV-C light. Non-cross-linked RNA was digested by 8 g RNase A and 0.024 U RNase T1 at 37 • C for 12 min. For immunopr ecipitation, RNase-tr eated sample was incubated with 90 l NETS buffer [10 mM Tris-HCl, 100 mM NaCl, 10 mM EDTA, 0.2% SDS (w / v), pH 7.4] and 5 l of antibody or pre-immune serum. After 1 h incuba tion a t 4 • C , pull-down was performed with 100 l pre-blocked [NETS buffer with 4 mg / ml BSA (NEB) and 2 mg / ml tRNA (Sigma-Aldrich)] 0.2% protein G slurry (Cytiva). Following 1 h incubation at 4 • C, protein enriched on the beads was washed (3 × NETS buffer via centrifugation at 1000 × g for 1 min) and released by 30 l of reducing Laemmli loading buffer. Pr otein-RNA cr oss-links were resolved on 15% SDS-PAGE gels and visualized by autoradio gra phy.

Psoralen RN A-RN A cross-linking and snRN A identification
Pr e-spliceosomal complex es (2.5 fmol, 10 l) were assembled on [ 32 P]-labelled RNA transcript as described above. Psoralen-AMT (1 l; Sigma-Aldrich) was added to the final concentration of 25 l / ml, heparin omitted. After a further 10 min incubation, complexes were radiated with UV-A light for 20 min; both steps were performed on ice. The total RNA content was harvested by standard PK digestion followed by ethanol precipitation with GlycoBlue co-precipitant (Thermo Fisher Scientific  Table S4). The digested RNA was purified via standard phenol extraction procedure, ethanol precipitated and analysed on a 4% denaturing urea-PAGE acrylamide gel. The crosslinking products and sensitivity to complementary DNA oligonucleotides were determined by autor adiogr aphy with a phospho-imaging screen and imaged with a Typhoon FLA9000 imager.

trans -Splicing
Transcription templates for AML E1 and TM4 40exU1 pr e-mRNA wer e generated by oligonucleotide synthesis and cloned into pGEM-4Z vector (Supplementary Table  S5). Pre-mRNA substrates used in the trans -splicing assay were in vitro transcribed with T7 RN A pol ymerase (Supplementary Table S11), 80% capped with m7G cap analogue (NEB) and treated with DNase turbo (37 • C, 30 min; Thermo Fisher Scientific). After column purification with RNA Monarch kit (NEB), the concentration of RNA transcripts was determined by UV absorbance. trans -Splicing reaction condition was similar to that of cissplicing condition described pre viously, e xcept splicing was performed concurrently with 5 nM of regulated (TM3 or TM23) and 50 nM constituti v e (AML E1 or TM4 40ex) RNA substra tes, a t 3.6 mM MgCl 2 and 2 mM ATP. After incubation, spliced products were phenol extracted from the PK digestion and ethanol co-precipitated with 20 g of Gly-coBlue (Thermo Fisher Scientific). Ten microlitres of RNA product dissolved in water was pre-incubated (65 • C, 5 min) with 20 pmol RT primer (Supplementary Table S6) and dNTP, cooled on ice and then re v erse transcribed with Su-perScript II re v erse transcriptase (Invitrogen) accor ding to the manufacturer's instructions. Ten percent of the RT reaction was used as the template in 25 l PCR reactions containing 1.25 U of JumpStart Taq Polymerase (Sigma, D9307), 1 × PCR buffer (Sigma, P2192), 400 nM of primers (Supplementary Table S6) and 0.2 mM dNTP. The reactions were hea ted (94 • C , 3 min) before 32 amplification cycles (94 • C for 30 s, 60 • C for 30 s and 72 • C for 60 s) and a final extension (72 • C, 60 s). PCR products were subsequently resolved on the QIAxcel Advanced system (QIAGEN) using a DNA screening capillary electrophoresis cartridge.

Affinity enrichment of pre-spliceosomal proteome
RNA-assisted pull-down was adapted from ( 52 )

Mass spectrometry
Liquid chromato gra phy-tandem mass spectrometry (LC-MS / MS) was used to identify and quantify proteins  Figure S13C). The serial gel slices were excised and digested in situ with trypsin. The extracted tryptic peptides were analysed using Q-Exacti v e mass spectrometer. Raw data were processed using Proteome Discoverer v2.3 (Thermo Fisher Scientific). Protein identification was conducted by searching human database downloaded in 2020, UniPort, using Mascot algorithm. This generated a list of 1081 entries containing common contaminant proteins (human kera tins, hea t shock proteins and BSA), which were identified and removed from downstream analysis. The data obtained from Proteome Discover er wer e abundance da ta a t the peptide le v el. Data were processed with R package and filtered to remove entries that only identified one out of three replicates of at least one condition. The resulting 978 entries (Supplementary File RNA MS A) were background corrected and normalized by variance stabilizing transformations. Inspection of the list re v ealed repetiti v e interpretation due to isoforms of the same protein and searching multiple databases. We collapsed the repetiti v e isoform entries of the same protein and shortlisted 178 unique identifications for further analysis (Supplementary File RNA MS B). Low-intensity missing values were biased to no RNA background samples and no RBPMS added conditions. To conduct the differential expression analysis, missing total precursor intensity was imputed using random draws from a Gaussian distribution centred around a minimal value, q th quantile = 0.01. We used R package Limma to test the significant changes between background subtracted groups as tabulated in Supplementary Figure S12B. The fold changes were estimated by the Bayes method, while the adjusted P -values were corrected by the Benjamini-Hochberg method.
For an af finity purifica tion-mass spectrometry (AP-MS) experiment, sample preparation, peptide identification and raw data processing were identical to those described above. Initial proteomic data processing was carried out in Scaffold ( 53 ) (Supplementary File AP MS TSC raw). Entries detected in the lack of NE condition (Supplementary File AP MS TSC raw, samples BL1-3) were excluded from analysis. To further enrich the list of significantly recovered proteins, total spectrum count of grouped technical r epeats was compar ed using unpair ed Student's t -test. FL v ersus negati v e contr ol pr oduced 131 significant interactors ( P < 0.05), while FL versus C20 generated 133 significant interactors (Supplementary File AP MS SL). C20 v ersus negati v e contr ol pr oduced 48 significant interactors. Control null-gene sets were generated based on protein expression levels ( 54 ). Gene Ontology (GO) analysis was performed on STRING with the following parameters adjusted: interaction sources set to experiments and databases only, and minimum r equir ed interaction scor e set to medium confidence (0.400).

PAC-1 cell culture and RNAi gene silencing
Rat pulmonary artery smooth muscle PAC-1 cells were cultured following standard procedures to maintain the differentia ted sta te ( 39 ). siRNA-media ted knockdown was performed using re v erse transfection. Briefly, 60-90 pmol siRN A and Lipofectamine RN AiMAX reagent (Thermo Fisher Scientific) were diluted by Opti-MEM without serum and then mixed and incubated at room temperature for 20 min. Diluted 2.5 × 10 5 dif ferentia ted PAC-1 cells per w ell w ere added to the RNAi duplex-Lipofectamine RN AiMAX complexes. An siRN A of scrambled sequence 'C2' was used as control in e v ery set of knockdown experiments. All siRNA sequences are tabulated in Supplementary Table S8. Each condition was repeated in ×6 format, to allow triplicates of RT-PCR analysis and sufficient material for verification of the knockdown at the protein le v el. In the two-hit experiment, the cells were treated again with the same reagent and procedure after 24 h. Cells were harvested 48 h after the terminal siRNA treatment.

RT-PCR and RT-qPCR
To verify the silencing of target gene, cDNA was pr epar ed using 1 g of total RNA, oligo(dT) (Merck) and Super-Script II (Invitrogen) based on the instructions gi v en by the manufacturers. RT-qPCR was performed as in ( 31 ), but using gene-specific primers (Supplementary Table S9). Tw o housek eeping genes were included in each analysis (CANX and Rpl32), and their geometric means were used to normalize the relati v e e xpr ession values. Expr ession values were acquired from three biological repeats.
To examine the usage of dif ferentia tion-specific exon usage, PCRs with 50 ng of cDNA were performed using the oligonucleotide primers listed in Supplementary Table S9. The PCR products wer e r esolved in a QIAxcel system as described previously. The visualization and quantification of PSI values were conducted using QIAxcel ScreenGel softwar e. PSI values ar e expr essed as mean (%) ± SD. Statistical significance was examined with unpaired Student's t -test.

RBPMS assembles into higher -or der oligomeric structur es via its C-terminus
Vertebrate RBPMS and RBPMS2 sequences show high conservation of the C-terminal 20 amino acids corresponding to the major RBPMS isoform (Figure 1 A), with complete conservation of aromatic and basic residues (Figure  1 B). Deletion of this region in transfected RBPMS ( C20) led to a significant loss in splicing r epr essor activity (Figure 1 C), demonstrating that the RRM alone is insufficient for RBPMS splicing regulatory function, despite being sufficient for dimerization and sequence-specific binding ( 40 , 41 ). For in vitro analyses, we pr epar ed r ecombinant untagged FL and C20 rat RBPMS (Figure 1 D), which shares 98% sequence identity with human RBPMS. Around 60% of FL RBPMS eluted as a dimer during size exclusion chromato gra phy (SEC), consistent with pr evious r eports (Supplementary Figure S1A) ( 40 , 41 ). The remaining 40% eluted as a broad peak before the 443-kDa marker approaching the void volume, indicati v e of heterogeneous higheror der assemb lies. In contrast, 96% of the C20 mutant eluted as a single peak corresponding to dimer, suggesting that the C-terminal region is r equir ed for RBPMS higherorder oligomerization. Both proteins migrated corresponding to their monomeric molecular weight on SDS-PAGE (Figure 1 D).
To examine RBPMS higher-order structures in more detail, we carried out SV AUC. At high ionic strength (500 mM KCl), FL RBPMS displayed polydispersity and existed as a series of dimers and higher-order oligomers (Figure 1 E). The largest oligomeric species sedimented at 18S, corresponding to a size of ∼460 kDa ( ∼21 monomers). In contrast, C20 sedimented homo geneousl y as a dimeric protein (Figure 1 F), consistent with its SEC profile (Supplementary Figure S1B). At low-salt conditions used for in vitro splicing assays (60 mM KCl), FL RBPMS oligomerized in a concentration-dependent manner with oligomers as large as 23S ( ∼600 kDa) observed at 32 M (Figure 1 G). In contrast, the C20 mutant was mainly dimeric, with a small amount of trimers forming at 25 M (Figure 1 H). Using glycerol gradients supplemented with a glutaraldehyde cross-linker (GraFix), we found that the C-terminal tail is r equir ed for the formation of higher-order oligomers above 250 kDa, but not trimeric and tetrameric species, which wer e mor e prominent with C20 (Supplementary Figure  S3). FL RBPMS and C20 also exhibit different friction ra tios tha t reflect potential shape differences; FL RBPMS and C20 dimers have friction ratios over 1.4 (Supplementary Figure S2E-H), whereas heterogeneous RBPMS oligomers have friction ratios of 1.0-1.15 (Supplementary Figure S2A-D). This indica tes tha t RBPMS dimers ar e mor e elongated in conformation, whereas RBPMS higher-order structur es ar e likely mor e spherical. We confirmed the spherical shape of RBPMS higher-order struc-tures by subjecting chemically cross-linked oligomers of ∼660 kDa to cryo-EM ( Figure 1 I; Supplementary Figure S3). Howe v er, two-dimensional projections of RBPMS oligomers were unamenable to further classification, indicating a high degree of structural heterogeneity. Consistent with this, His 6 -tagged and tag-free RBPMS were observed to undergo liquid-liquid phase separation in vitro ( Figure 1 J; Supplementary Figure S4B). The phase transition of tag-free RBPMS required the presence of the molecular cro w ding r eagent PVA (Supplementary Figur e  S4B). To examine the nature of phase-separated RBPMS droplets, His 6 -tagged RBPMS was fluorescently labelled with Alexa Fluor 647 (Supplementary Figure S4A), spiked into unlabelled RBPMS and monitored by fluorescence microscop y. Pr e-formed RBPMS droplets acquir ed fluorescence after mixing and showed protein concentrationdependent changes in volume, suggesting that they are liquid-lik e and dynamic. Tak en together, our biophysical analyses show that RBPMS exists as a heterogeneous mixture of concentration-dependent higher-order oligomers in addition to dimers, and that the C-terminal 20 amino acids that are important for activity in vivo are also essential for higher-order oligomerization.
The TM1-3-4 substrate, comprising Tpm1 exons 1, 3 and 4, showed e v en more emphatic changes in splicing outcome upon RBPMS titration (Figure 2 A). The default splicing pattern of TM1-3-4 in HeLa NE is exon 3 inclusion, with bands corresponding to fully spliced 1:3:4 and partially spliced 1:3-4 and 1-3:4 intermediates (Figure 2 A, empty circles). Only a very small amount of the exon skipping product (1)(2)(3)(4) and the corresponding lariat is seen (Figure  2 A, black filled circles). Strikingly, titration of FL RBPMS led to a complete switch from exon 3 inclusion to skipping, indica ting tha t RBPMS is suf ficient to confer SMC-specific splicing of Tpm1 in vitro (Figure 2 A, lanes 6-2). When tandem CAC clusters both upstream and downstream of exon 3 were mutated, the basal pattern of splicing was unaltered but RBPMS no longer mediated exon 3 skipping (Figur e 2 B). Furthermor e, the RBPMS C20 mutant failed to induce exon 3 skipping e v en with intact flanking CAC clusters (Figure 2 (Figure 1 C). The cell-free AS assay therefore faithfully reflects the specificity of cell culture assays.

RBPMS C-terminus is essential f or RBPMS cooper ative RNA binding
To test whether the inability of RBPMS C20 to mediate exon 3 r epr ession was associated with altered RNA binding, we performed UV cross-linking of RBPMS along with HeLa NE proteins to a radiolabelled RNA substrate TM3, which contains exon 3 and all flanking splice site and regulatory elements (Figure 2 D). While FL RBPMS efficiently cross-linked to TM3 (Figure 2 E, lanes 7-2), very little crosslinking was observed for the C20 mutant (Figure 2 E, lanes   15-10). Hence, the C-terminal 20 amino acids are critical for RBPMS binding to the regulated Tpm1 pre-RNA in NE and the RRM domain alone, which mediates both dimerization and RNA binding in vitro , is not sufficient for RNA binding in NE. We further examined the RNA binding properties of FL and C20 RBPMS using EMSA. Strikingly, with an RNA substrate containing a single pair of CAC motifs, (CAC) 2 , separated by 10 nt (indicated by parentheses in Figure 2 D), RBPMS C20 bound with an apparent K d of 2.5 M, whereas FL RBPMS did not bind within the concentration range assayed (Figure 3 A and C). The lack of FL binding to (CAC) 2 may be related to the lower effecti v e concentration of free dimeric RBPMS compared to C20. In contrast, both proteins bound to a longer substrate with three tandem CAC motifs [3 ×(CAC) 2 , indicated by square brackets in Figure 2 D], with an apparent K d of ∼150 nM, which is 20-fold higher in affinity than C20 binding to the shorter (CAC) 2 substrate. The observed binding was dependent upon CAC motifs, as shown with CCC mutants (Supplementary Figure S6C and D). Importantly, FL RBPMS showed an additional supershifted species with limited gel mobility, which we postulate to be higher-order RBPMS-bound substrates (Figure 3 B, Bound II). Indeed, a Hill factor of 1.7 (Figure 3 D), deri v ed by considering both Bound I and II complexes, suggests that FL RBPMS binds the 3 ×(CAC) 2 substrate in a cooperati v e manner. No cooperativity was apparent (Hill factor ∼1) if only Bound I was considered (Supplementary Figure S6A and B). C20 also showed no cooperativity of binding, but its affinity was similar to FL RBPMS (Figure 3 B and D). These data indicate that the C-terminal 20-amino acid tail mediates cooperati v e binding to multivalent RNA substrates, which could be a result of its propensity for driving RBPMS oligomerization ( Figure 1 ). Howe v er, this does not appear to be sufficient to explain the severe loss of RNA binding by C20 in the competiti v e environment of NE (Figure 2 E). Instead, we postula te tha t IDR-media ted interactions with other RBPs ar e a r equir ement for RBPMS to bind target RNAs effecti v ely under splicing conditions (see below).

RBPMS r emodels pr e-spliceosomal comple x es on target RNA substrates
RBPMS may mediate exon 3 r epr ession by dir ect modulation of pre-spliceosomal assembly pathways. To explore this possibility, we e xamined comple x assemb ly on two cissplicing incompetent substrates, TM3 and TM23, in HeLa NE (Figures 4 and 5 ). Both TM3 and TM23 contain Tpm1 exon 3 with the full complement of splice site and regulatory elements, but no flanking constituti v e splice sites, thereby allowing us to focus on comple x assemb l y onl y across the regulated exon 3 region. To ensure that the complexes forming on TM3 and TM23 are functionally relevant, we first tested their activity in trans -splicing assays ( 51 ).
To test the 5 ss functionality of TM3 and TM23, Tpm1 exon 4 with an 88-nt 5 intronic extension (TM4) was used as a 3 trans -splicing partner. TM4 (50 nM) was used at 10fold molar e xcess ov er either TM3 or TM23 (5 nM), in order to overcome splicing inefficiency caused by the physical separation of 5 and 3 substrates ( 57 ). The 89-nt TM34 spliced product was generated from paired TM3:TM4 and TM23:TM4 reactions in HeLa NE (band a, Figure 4 (Figure 4 B). This is an important observa tion tha t demonstra tes the specificity of RBPMS action. The 5 ss of exon 2 is only 41 nt upstream of the exon To determine the 'A-like' percentage, the ratio of the 'A-like' complex was first determined as described in panel (A). For e v ery titration, RBPMS spiked conditions were normalized against the condition where RBPMS-A was omitted [(' A-like' RBPMS / ' A-like' null ) × 100]. For panels (B) and (D), the data points and error bars depict the mean of three technical repeats and SD. Unpaired, one-tailed, Student's test was used to assess the differences in mean ratio or percentage of 'A-like' conversion across different conditions and shown as follows: ns, non-significant; **, P < 0.01; ***, P < 0.001. 3 branch point; the activation of exon 2 splicing demonstra tes tha t RBPMS is not 'smothering' the w hole RN A to make it splicing incompetent, but is acting in a precise and targeted manner. Again, for both TM3 and TM23, RBPMS tr ans -splicing regula tory activity was completely dependent on its C-terminal 20 amino acids (lane 4) and the presence of CAC motifs flanking exon 3 (lane 6). The identities of bands a, b and c were confirmed by sequencing, and control lanes showed that they only appeared upon incubation with both trans -spliced partner RNAs and in the presence of ATP (Supplementary Figure S7, lanes 19-24).
We tried to test the 3 ss functionality of TM3 and TM23 using Tpm1 exon 1 with its downstream intronic segment (TM1) as a 5 trans -splicing partner. We were unable to detect the TM1:3 splice product with either the TM3 or TM23 acceptor (data not shown), despite the fact that equivalent splice products were readily detected in cis -splicing experiments ( Figure 2 ). We suspected that the large size of the TM1 substrate (350 nt) might result in inefficient transsplicing kinetics. We ther efor e used the 34-nt AML exon 1 with 95 nt of 3 intronic sequence (AML E1) ( 51 ). We detected AML-TM3 splice products at 154 nt, confirming the 3 ss functionalities of TM3 and TM23 (band d, Figure  4 C and D). Again, RBPMS inhibited the trans -splicing of AML E1 to exon 3 for both substrates (Figure 4 C and D,  lanes 1 and 2). This effect was again dependent on an intact RBPMS C-terminal region (lanes 3 and 4) and the presence of CAC clusters on both sides of the regulated exon (lanes 5 and 6).
Having established that both TM3 and TM23 are competent for trans -splicing and are regulated specifically by RBPMS, we proceeded to investigate how RBPMS regulates the assembly of splicing-related complexes ( Figure 5 ). In the absence of RBPMS and with ATP present, both minimal substrates initially formed a heterogeneous (H) complex of fast mobility that developed into a lower-mobility complex within 5 min (Figure 5 A, ATP+). The lowermobility ATP-dependent complexes were sensiti v e to targeted partial digestion of U1 and U2 snRNAs (  Figure  S9). Gi v en the single exon configuration of TM3, we propose that the lower-mobility complex on TM3 corresponds to an exon definition A (EDA) complex ( 22 ). On the other hand, the advanced complex formed on TM23 could be a combination of an EDA complex across exon 3 and a sterically hindered 'A-like' complex between exons 2 and 3 ( 58 ). Notably, on TM23 but not TM3, a lower-mobility complex also formed in the absence of ATP (denoted by ATP −, Figure 5 A, right). This complex could be distinguished from the ATP-dependent 'A-like' complex by its slightly lower mobility ( ∼0.8-fold lower mobility).
Addition of FL RBPMS abolished the formation of ATP-dependent complexes on TM3 and TM23 in a concentration-dependent manner ( Figure 5 D; Supplementary Figure S10A). At the highest concentration (2 M) of RBPMS, ∼20% of lower-mobility complexes remained on TM23 (Figure 5 D). Howe v er, the residual low-mobility complex migrated more slowly than the ATP-dependent complex in the absence of RBPMS ( ∼0.8-fold lower mobility), similar to the ATP-independent low-mobility complex (Figure 5 A). RBPMS therefore appears to inhibit formation of all ATP-dependent complexes. Consistent with this, in the presence of RBPMS, U1 and U2 snRNA base pairing to TM23 was eliminated ( Figure S9), mirroring the r equir ements for RBPMS splicing regulation in cis -and trans -splicing assays (Figures 2 and 4 ). Taken together, our results establish a strong link between RBPMS splicing regulatory activity and its remodelling of spliceosomal complexes on model transcripts.

RBPMS remodels the RNA-bound proteome composition
Changes in gel mobility of complexes forming on TM3 and TM23 are expected to be caused not only by RBPMS binding and snRNP displacement, but also by the recruitment and displacement of other RBPs. In line with this hypothesis, FL RBPMS was observed to alter the crosslinking of other NE proteins to the TM3 substrate (Figure 2 E, lanes 2-7). To identify RBPMS binding partners in HeLa NE, we first generated recombinant RBPMS proteins with an N-terminal Strep tag II followed by a polyhistidine tag (StrepII-His 6 -RBPMS) for AP-MS experiments ( Supplementary Figures S12 and S14; Figure 6 B). StrepII-His 6 -RBPMS has similar oligomerization properties to untagged RBPMS, indicating that the tag has negligible effects on RBPMS biophysical and functional properties (Supplementary Figure S4A). A total of 118 FL RBPMS interactors were significantly enriched above background and 100 of these interactors were significantly depleted from the C20 pull-down (Supplementary Figure S14C and Supplementary File AP MS SL). Interactors were then classified using enriched GO terms on STRING ( 59 ). Analysis of both lists of interactors generated near-identical top fiv e enriched terms associated with mRN A processing, RN A splicing and RNA binding from each category (biological process, molecular function and cellular component) (Supplementary Figure S15). The enriched terms were not reproduced to the same significance in three independent gene expression matched sets of 118 genes (Supplementary Figure S16; gene sets in Supplementary File AP MS SL). Fifty RBPMS interactors were selected from the following terms: RN A splicing, RN A binding and RNP complex, and grouped (Figure 6 A) based on their annotated function as 3 -end processing factors, hnRNPs, splicing regulators , other RBPs , interaction network of RBFOX2 ( 24 ), RNA helicases, and components of pre-spliceosome complexes such as U4 / U6 ·U5 tri-snRNP and U2 snRNP. Among the splicing regulators was MBNL1, a known regulator of Tpm1 splicing ( 37 ). Remar kab ly, truncation of the C-terminal 20 amino acids led to a near-global loss of To validate some of the RBPMS-mediated interactions, we performed western blot analysis (Figure 6 B, right) and confirmed interactions between FL RBPMS and MATR3, RBM4, RBM14, RBM47, RBFOX2, ESRP2 and MBNL1. The lack of interaction with PTBP1, a known co-regulator of Tpm1 exon 3, serves as a control for the specificity of RBPMS interactions. With the exception of RBM14 and RBM47, all of the interactions were completely abolished by the C20 deletion. This included MBNL1, which did not show a statistically significant difference between FL and C20 in the AP-MS analysis (Figure 6 A), but clearly showed loss of binding by C20 in the western blot (Figure 6 B). Using Benzonase-treated NEs or the K100E RNAbinding mutant, most of the interactions were observed to be partially or completely dependent on RNA binding (Figure 6 B, lanes 6 and 7 compared to lane 3).
We next tested whether RBPMS detectably altered the glycerol gradient sedimentation profiles of a subset of its interactors (Figure 6 C). Upon addition to NE, RBPMS itself sedimented more ra pidl y than free RBPMS, indica ting tha t it forms heterogeneous high molecular weight complexes. Consistent with its loss of both homo-oligomerization (Figure 1 ) and heter otypic pr otein-pr otein interactions (Figure 6 A and B), the C20 RBPMS in NE sedimented in lighter fractions than FL RBPMS (Figure 6 C). Strikingly, RBFOX2 shifted into heavier fractions upon addition of FL RBPMS but not C20 to NE (Figure 6 C), suggesting that the two proteins are components of a common higher molecular weight complex. RBFOX2 is known to be present in the multicomponent Benzonase-resistant large assembly of splicing regulators (LASR) complex ( 24 ). The sedimentation profiles of other proteins, including the LASR components MATR3 and hnRNPM, were unaffected by RBPMS suggesting that the RBPMS-RBFOX2 complex is distinct from LASR. MBNL1 appeared to show a slight shift to heavier complexes (Figure 6 C), but the differences in MBNL1 were not significant between equivalent fractions in the presence or absence of RBPMS.
Having established that the RBPMS interactome includes numerous splicing factors and regulators, we proceeded to examine how RBPMS remodels the composition of splicing-related complexes on splicing substrates tagged with MS2 sites to facilitate affinity purification with MBP-MS2 (Figure 7 A). We initially attempted to use the TM23 substrate, but were unable to achieve purification of specific complexes, in part due to the large size of TM23 (820 nt). We ther efor e opted for the shorter TM3 substrate and omitted the molecular cro w ding agent PVA to facilitate comparab le recov ery of transcripts across different experimental conditions. Under these conditions, stable association of snRNPs with the RNA is expected to be very inefficient, so we would not expect to capture the displacement of snRNPs evident in Figure 5 D. How ever, w e hoped to capture remodelling of RBPs associated with the H-complex ( Figure  5 D, lanes 1-6) that might influence subsequent complex assembly under splicing conditions with PVA pr esent. Ur ea-PAGE analysis showed a slight increase in RNA recovery in RBPMS-spiked samples, but these differences were within the normalization range of downstream data pro-cessing (Supplementary Figure S13C). Howe v er, assemb ly of ATP-dependent low-mobility complexes was negligible under these conditions (Supplementary Figure S13A). Total proteins from each condition ( ±ATP, ±RBPMS) were submitted to LC-coupled quantitati v e label-free MS / MS. Consistent with the complex assembly conditions, very few snRNP proteins were detected even in the absence of RBPMS. Howe v er, differential pull-down analysis re v ealed that a large number of RBPs were either recruited to (e.g. ESRP2) or displaced from (e.g. SRSF3) the TM3 substrate by RBPMS (Figure 7 B and C). Proteins identified as RBPMS interactors in the AP-MS experiment were also found among both RBPMS-recruited proteins (e.g. RBM4, RBM14, RBFOX2) and RBPMS-displaced proteins (e.g. SRSF7, hnRNPC) ( Supplementary File RNA MS B). The differ ential r ecruitment of some proteins appear ed to be sensiti v e to ATP; for example, enrichment of bound ESRP2 and depletion of SRSF7 by RBPMS were only observed in the presence of ATP. We observed no significant changes in the transcript-bound le v els of PTBP1 and MBNL1, despite the fact that MBNL1 was identified as a direct interactor ( Figure 6 ) and both proteins are co-regula tors tha t bind to sites flanking Tpm1 exon 3 ( 37 , 38 ).
Differential RBPMS-sensiti v e binding of selected RBPs was confirmed by UV cross-linking of [ 32 P]-UTP-or [ 32 P]-CTP-labelled TM3 RNA to proteins in HeLa NE ( Figure  7 D-F). RBM4, RBM14, Rbfox2 and ESRP2 cross-linking only occurred in the presence of RBPMS, in agreement with results from the differential pull-down using MS2-tagged TM3. Notably, each of these proteins was seen to interact with RBPMS in an RNA-dependent manner (Figure  6 B). RBM47 cross-linking was not detected with either the [ 32 P]-CTP-or [ 32 P]-UTP-labelled transcript, possibly due to poor cross-linking efficiency. In line with the differential pull-down results, cross-linking of MBNL1 and PTBP1 was unchanged by RBPMS ( Figure 7 E and F, lanes 9 and 10), Ther efor e, PTBP1 and MBNL1 binding to TM3 does not r equir e acti v e recruitment, although their transcript-bound activities may still be regulated by RBPMS. Consistent with r epr ession of Tpm1 exon 3 splicing, cross-linking of the essential splicing factor U2AF2, w hich reco gnizes the polypyrimidine tract, was reduced by RBPMS (Figure 7 F).

RBFOX2 and MBNL1 cooperate with RBPMS in the VSMC AS programme
The preceding data indicate that RBPMS interacts with and acti v ely recruits a number of RBPs in HeLa NE to TM3 RNA. In contrast to other RBPMS interactors, MBNL1 binds stably to short YGCY clusters upstream and downstream of Tpm1 exon 3 independent of RBPMS (Figure 7 E) and promotes exon skipping ( 37 ). To test whether MBNL1 modulates RBPMS activity on Tpm1 splicing, we disrupted MBNL1 binding to the TM134 substrate either by deletion of both clusters or by mutation of all YGCYs to Y CG Y. These mutations led to reduced sensitivity to RBPMS regulation of splicing in vitro ; higher concentrations of RBPMS  Figure  S21D and E). Taken together, the preceding data suggest that pr otein-pr otein interactions with MBNL1 help to recruit RBPMS to TM3 RNA in NE, thereby explaining the inability of C20 RBPMS to bind to or regulate TM134 RNA ( Figure 2 ). To examine the wider functional relevance of the identified RBPMS interactions, we tested the effects on four VSMC-specific ASEs ( Tpm1 , Actn1 , Flnb , Hspg2 ) of siRNA-mediated knockdown in PAC-1 VSMCs of RBPMS, RBM4A and B, RBM14, RBM47, RBFOX2,  Figure S20C). Unit variance scaling was applied to PSI values to standardize each row, resulting in a mean of 0 and SD of 1 in either direction (colour coded). Rows are centred and clustered using correlation distance and average linkage. Columns are clustered using maximum distance and average linkage. The knockdown was verified via RT-qPCR or western blots (Supplementary Figure S20A and B). Statistical analysis was performed using unpaired, one-tailed, Student's t -test, indicated as follows: ns, P > 0.05; *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. For panels (D) and (E), signal-to-noise ratio between 25 and 75 nM is too low to conduct confidence comparison.  (14). ( A ) In the absence of RBPMS, co-r epr essors such as MBNL1 and PTBP1 can bind around Tpm1 exon 3 but are not sufficient to pre v ent binding of U1 and U2 snRNPs. ( B ) RBPMS forms dynamic oligomers that can interact with other RBPs, including MBNL1 and RB-FOX2, to block binding of U1 and U2 snRNPs. We propose that this assemb ly resemb les a 'binding region condensate' as described by Hallegger et al. ( 29 ). ( C ) C20 RBPMS fails to oligomerize, interact with other RBPs or bind to Tpm1 RNA.
Supplementary Figure S20C), suggesting that they might work widely as RBPMS co-regulators of VSMC-specific ASEs. The only non-significant change was the effect of RBPMS on Lsm14b ; howe v er, this target was pre viously shown to respond to RBPMS knockdown in PAC-1 cells that were more dif ferentia ted than those used here ( 31 ). We noted that there were some indications of cross-regulation between RBPMS, MBNL1 / 2 and RB-FOX2 (e.g. MBNL1 / 2 knockdown results in some depletion of RBPMS and RBFOX2; Supplementary Figure  S20A and B). Ne v ertheless, this is not sufficient to undermine the conclusion that each protein directly affects splicing (see the next section), and further supports the functional integration of their splicing networks.

DISCUSSION
The activity of recombinant RBPMS in vitro allowed detailed analysis of the relationship between its biophysical properties and splicing regulatory activity and insights into how cell-specific splicing regulators interact physically and functionally with more widely expressed RBPs (Figure 9 ). Tpm1 exon 3 is efficiently spliced in most cell types, and in HeLa NE in vitro , despite the binding of up to six PTBP1 and three to eight MBNL co-r epr essors around the exon (Figure 9 A). RBPMS exists as a het-erogeneous dynamic mixture of dimeric and oligomeric species (Figure 9 B, left), with the C-terminal IDR mediating both homomeric oligomerization and heterotypic interactions with other proteins. Oligomeric RBPMS can ther efor e make multivalent interactions with the multiple (CAC) 2 motifs flanking Tpm1 exon 3 as well as contacting other RBPs, which might further stabilize RNA binding. Notabl y, MBNL1 binds independentl y to YGCY motifs and by a direct pr otein-pr otein interaction helps to recruit RBPMS to the RNA. RBPMS in turn recruits further co-regulators, including RBFOX2, that do not have specific binding sites. As a result, a stable r epr essed complex forms that pre v ents splicing comple x assemb ly, including binding of U1 and U2 snRNPs (Figure 9 B, right). With deletion of the C-terminal 20 amino acids of the IDR, RBPMS exists only as a dimer, is unable to interact with other RBPs and consequently is inacti v e as a splicing regulator, being unable to promote regulatory complex assembly (Figure 9 C). We propose that the stable r epr essed complex, which encompasses a 500-nt region surrounding exon 3 ( Figure  2 D), resembles the 'binding region condensates' described by Hallegger et al. ( 29 ). Single-molecule analyses showed that the TM3 RNA binds ∼5-6 PTBP1 and 3-8 MBNL1 molecules ( 35 , 37 ). Equivalent analyses of RBPMS binding have not yet been carried out, but the size of RNA-free RBPMS oligomers (Figure 1 ) suggests that the r epr essed complex likely contains 5-10 RBPMS dimers, meaning that the size of the RBPMS:MBNL1:PTBP1 complex would be ∼1 MDa or larger, without taking into account RB-FOX2 and other RBPs that do not have specific binding sites around exon 3. Despite this size, the r epr essi v e mode of action must be very precisely targeted because the 5 ss of exon 2, only 41 nt upstream of the exon 3 branch point, is activated by the r epr essi v e mechanism operating on exon 3, e v en on a tr ans -splicing substra te (Figure 4 ; Supplementary Figure S5). It seems plausible that the 'zone of r epr ession' is limited by the two PTBP1 binding tracts, which flank the upstream and downstream MBNL and RBPMS binding sites (Figures 2 D and 9 ).
Recombinant RBPMS primarily exists as a heterogeneous dynamic mixture of dimeric and oligomeric species, char acteristic of phase-separ ating proteins below their critical sa tura tion concentra tion ( c sat ) ( 60 , 61 ) (Figure 1 D), although it can undergo phase separation forming liquid-like droplets in vitro ( Figure 1 J; Supplementary Figure S4B). While the reported association of RBPMS and RBPMS2 with cytoplasmic granules ma y in volve condensate behaviour ( 42 , 46 , 62 ), we envisage that as a splicing regulator RBPMS is in the form of dynamic co-regulator-containing hetero-oligomers smaller than the mesoscale assemblies that form visible cellular condensates. The size of the transcript-bound RBPMS oliogomers could be addressed in the future using single-molecule methods ( 35 , 63 ). The disordered 20-amino acid C-terminal tail, enriched in aromatic and basic residues, is essential for RBPMS oligomeriza tion (Figure 1 ), alterna ti v e splicing outcomes (Figures 1 ,  2 and 4 ), cooperati v e binding to multivalent RNA ( Figures  2 and 3 ), splicing complex regulation ( Figure 5 ) and most pr otein-pr otein interactions ( Figure 6 ). These effects of the C20 deletion are consistent with, and provide a physical basis for, previous reports that C-terminal truncation of RBPMS or RBPMS2 impaired localization to cytoplasmic granules ( 62 ), participation in RNP complexes ( 44 ), coimmunoprecipitation with its FL counterpart and mRNA binding ( 45 ).
The relati v e importance of the homomeric and heteromeric interactions mediated by the IDR remains an open question. Indeed, it is plausible that homotypic and heterotypic interactions share a common physical basis --for example,or cation-interactions mediated by aromatic and / or basic residues ( 64 ) --so mutants to distinguish their roles might be elusi v e. Ne v ertheless, gi v en the multiple CAC motifs around Tpm1 exon 3 the ability of RBPMS both to oligomerize and to mediate heterotypic interactions with other RBPs appears to be essential for its function. Indeed, the ability to interact with MBNL1 appears important for recruiting RBPMS to TM3 RNA in the face of competiti v e binding in NE (Figure 8 ). Modulation of RBPMS activity via deoligomerization also appears to be a physiological control mechanism. RBPMS is phosphorylated a t Thr113 / 118 immedia tely downstream of the RRM, and phosphomimetic mutants have reduced activity and RNA binding, which is related, in part, to deoligomerization, as well as direct occlusion of the RNA binding surface of the RRM in a phosphomimetic mutant ( 65 ).
While many new RBPMS interactors were identified by affinity pull-down ( Figure 6 ), 10 proteins in our dataset are known RBPMS interactors, including RBFOX2, MBNL1 and RBM14 (66)(67)(68)(69). That most interactions are direct but enhanced by the presence of RNA (Figure 6 B) is consistent with combinatorial models of splicing regulation, in which the low affinity of binary pr otein-pr otein interactions is tuned so as to enable specific cooperati v e assemb ly only upon regulated substrates with the correct combination of binding sites ( 70 ). Gi v en the heterogeneity of RNAs in NE, the captured RBPMS interactome is likely to contain RBPMS co-regulators involved in both splicing activation and r epr ession as well as other activities such as 3 -end processing ( Figure 6 ; Supplementary Figure S15). Indeed, the identification of numerous U2 snRNP components ( Figure  6 A) suggests possible mechanisms for RBPMS splicing activ ation b y U2 snRNP recruitment. Howe v er, it is less clear how this interaction could be involved in the observed displacement of U2 snRNP from Tpm1 transcripts ( Figure 5 ; Supplementary Figures S10 and S11), which is more likely explained by the earlier displacement of U2AF2 by RBPMS (Figure 7 F). In examining how RBPMS remodels the Tpm1 RNA-bound proteome, we would ideally have used similar conditions to those used to identify ATP-dependent comple xes on nati v e gels (Figure 5 D). Howe v er, we encountered insurmountab le technical prob lems in trying to purify comple xes assemb led in the presence of PVA and with the longer (820 nt) TM23 substrate. In the future, it would be useful to exploit single-molecule methods to assess how RBPMS affects binding of individual snRNPs to the TM RNAs as well as the number of RBPMS subunits associated with the r epr essed complex ( 35 , 63 ). Nevertheless, by analysing complexes formed on TM3 RNA in the absence of PVA, we identified potential splicing co-regulators of RBPMS (Figure 7 ), many of which were also pulled down directly by RBPMS from HeLa NE. In contrast, se v eral known splicing regulators detected in the AP-MS experiment were ei-ther depleted from TM3 by RBPMS (e.g. SRSF7 and hn-RNPC) or not significantly enriched (e.g. SRSF1). Some of these differences could be attributed to the presence or absence of specific cis -elements in the TM3 substrate, which may be a r equir ement for r ecruitment of some interactors, such as RBM4 whose interaction was completely RNA dependent (Figure 6 B). RBM4 was pr eviously r eported to promote Tpm1 exon 3 inclusion, antagonizing the activity of PTBP1 ( 71 ), but we saw no effects upon RBM4 knockdown (Supplementary Figure S17D).
Among RBPMS interactors, we identified many components of the 55S Benzonase-resistant LASR splicing regulatory complex ( 24 ) (Figure 6 A). LASR confers the RNA binding specificities of its other constituent proteins upon RBFOX, effecti v ely e xpanding the motif r ecognition pr eference of RBFOX, consistent with the lack of identifiable RBFOX motifs associated with many RBFOX CLIP tags ( 72 ). Ne v ertheless, the higher-or der Benzonase-resistant RBPMS interaction with RBFOX2 ( Figure 6 C) did not involve other LASR complex components (e.g. MATR3 or hnRNP M), so it may be a distinct complex. RBFOX2 has been shown to direct distinct splicing outcomes of opposing biological activities by partnering with different splicing regulators ( 73 ). RBPMS may have such a determining influence, r edir ecting RBFOX2 fr om pr omoting mesenchymal ( 74 ) to dif ferentia ted VSMC splicing programmes. One interesting example is the Flnb H1 exon, encoding a hinge region in filamin B, which is activated in PAC-1 cells by RBPMS, RBFOX2 and MBNL1 (Figures 8 ; Supplementary Figures S18-S20) ( 31 ). Howe v er, in human breast cancer cells RBFOX1 promoted skipping of the same exon, as part of epithelial-mesenchymal transition ( 75 ).
Se v eral lines of evidence converge to suggest that MBNL1 and RBFOX2 act as general co-regulators with RBPMS. Both proteins were found as direct RNAstimulated interactors with RBPMS ( Figure 6 ): RBFOX2 was recruited to the TM3 RNA by RBPMS (Figure 7 ), while MBNL1 bound TM3 independently via its own binding sites and helped to recruit RBPMS in NE ( Figure 8 ). Knockdown of both proteins affected 11 tested ASEs in the same direction as RBPMS (Figure 8 ; Supplementary Figure S20C). We noted that there was evidence of crossregulation between RBPMS, RBFOX2 and MBNL1, particularly at the protein le v el (Supplementary Figure S20). Despite this, the data support direct roles of all three proteins in the ASEs tested. First, RBPMS depletion did not affect RBFOX2 and actually led to small apparent increases in MBNL le v els, which would act to dampen RBPMS affects, so we can conclude tha t RBPMS ef fects are explained by its own knockdown. Second, RBFOX2 knockdown led to partial RBPMS and MBNL1 protein depletion (Supplementary Figures S18-S20). Howe v er, RBFOX2 knockdown had a larger effect than RBPMS knockdown on many e v ents, arguing that it acts directly as well as by reducing RBPMS le v els. Finally, although MBNL1 / 2 knockdown also partially reduced le v els of RBFOX2 and RBPMS, it also had the grea test ef fect on all but one ASE (Itga7; Supplementary Figure S20C), again arguing that it acts directly. We do not know the molecular basis of most of the cross-regula tory ef fects. Howe v er, RBPMS ( 31 ) causes an exon skipping in MBNL1 and 2, which has been shown to media te post-transla tional downregula tion of MBNL2 protein ( 76 ), which might explain the MBNL protein upregulation upon RBPMS knockdown.
Other lines of evidence also suggest widespread functional cooperation of RBPMS and RBFOX proteins, including enrichment around RBPMS-regulated exons not only of RBPMS dual CAC binding motifs but also of RB-FOX binding GCAUG motifs ( 77 ), the identification of a GCAUG-containing motif as the top intronic binding site for RBPMS in ES cells ( 78 ) and the association of both RBFOX and RBPMS binding motifs with ERG (E26r elated gene protein)-r epr essed exons in HeLa cells ( 79 ). Furthermore, the SMC transcription factor MYOCD was recently found to indirectly dri v e a set of SMC AS changes via changes in the expression levels of RBPMS, RBFOX2 and MBNL1 ( 80 ), and both Rbpms and Mbnl1 were found by single-cell RNA sequencing to be part of a contractile VSMC gene signature ( 81 ). Indeed, it has been suggested that gastrointestinal dysfunction in myotonic dystrophy is associated with dysregulation of an MBNL1-regulated splicing programme in visceral smooth muscle cells ( 82 ). Our data suggest that this dysregulated programme is likely dri v en by MBNL1-RBPMS co-regulation. The finding that recombinant RBPMS is sufficient in vitro to switch Tpm1 e xon 3 alternati v e splicing to the fully dif ferentia ted VSMC state ( Figure 2 ) is consistent with its proposed role as a master regulator of the AS splicing programme in dif ferentia ted VSMCs ( 31 ). Rbpms heterozygous knockout mice have no phenotype, while homozygous knockouts are inviable ( 83 ) but have phenotypes associated with dysfunction of both VSMCs and cardiomyocytes. Confirmation of the physiological roles of RBPMS in fully dif ferentia ted VSMCs in vivo will ther efor e r equir e conditional knockout models.
In conclusion, this study builds on previous work to suggest that the dynamic regulation of splice site choice is dependent on the existence of both constitutive and tissuespecific AS regulatory networks. The intricate connections and functional redundancy of the tw o netw orks may reflect the r equir ement of VSMCs to conduct phenotypic switching ra pidl y in response to environmental cues.

DA T A A V AILABILITY
The proteomes r ecover ed from RNA-assisted pull-down and Strep-tagged RBPMS affinity pull-down were analysed using label-free quantitati v e LC-MS / MS. The raw mass spectrometry data were submitted to the ProteomeXchange Consortium via the PRIDE partner repository ( 84 ) and are identified by dataset numbers PXD037617 and PXD037620, respecti v ely.

SUPPLEMENT ARY DA T A
Supplementary Data are available at NAR Online.

ACKNOWLEDGEMENTS
For the purpose of open access, the authors have applied a Creati v e Commons Attribution (CC BY) license to any Author Accepted Manuscript version arising from this submission. We thank Mike Deery and Yagnesh Urania (Cambridge Centre for Proteomics) for mass spectrometry and bioinformatic analyses, respecti v ely. We thank Ian Eperon for helpful and insightful comments on the manuscript and anonymous r efer ees for constructi v e comments.