Assembly of exon junction complexes on spliceosomes begins before spliceosome activation Authors: Elizabeth T Abshire, Lynne E Maquat Published: 2026 ABSTRACT AbstractDespite the importance of exon junction complexes (EJCs) to co- and post-transcriptional gene control, how and when the EJC core of EIF4A3, RBM8A, and MAGOH, and the peripheral factors of the alternative RNPS1-EJC and CASC3-EJC, assemble on spliceosomes remain unclear. Existing characterizations that order EJC assembly on spliceosomes offer different conclusions. Using immunoprecipitations (IPs) of EIF4A3, RNPS1, or CASC3 in the presence of RNase I followed by liquid chromatography followed by tandem mass spectrometry (LC-MS/MS) analyses of human HEK293T cells, we expand upon those splicing factors that associate with EJC constituents. Referencing these data against existing cryogenic electron microscopy (cryo-EM) spliceosome structures at different stages of pre-messenger RNA (pre-mRNA) splicing, coupled with data from a series of western blots of nuclear fractions before and after IP, we report that EIF4A3, RBM8A, MAGOH, and RNPS1 associate with assembled but inactive spliceosomes. This is followed by CASC3 joining to late activated spliceosomes. Our work bridges long-standing gaps in understanding EJC assembly on spliceosomes. Introduction Splicing i.e. the removal of introns from nuclear pre-messenger RNAs (pre-mRNAs) during mRNA biogenesis, involves carefully orchestrated interactions and structural rearrangements among spliceosome proteins and small nuclear RNAs (snRNAs) [1]. Most mammalian pre-mRNAs harbor multiple introns that must be removed to produce mRNAs that contain only exons and encode proteins with the correct sequence, subcellular location, and expression level. Intron removal is coupled to the deposition of an exon junction complex (EJC) at a position centered 27-nucleotides upstream of ∼80% of the resulting exon–exon junctions [2–4]. EJCs play crucial roles in both nuclear and cytoplasmic mRNA metabolism [5]. As examples, EJCs in the nucleus physically occlude m6A modifications [6] and promote mRNA export to the cytoplasm by recruiting ALYREF, a constituent of the TREX complex [7, 8]. EJCs in the cytoplasm augment mRNA translation by recruiting translation initiation factors, and they trigger nonsense-mediated mRNA decay (NMD) should one reside sufficiently downstream of a termination codon [5, 9]. All EJCs are thought to contain EIF4A3, which directly binds the RNA at the EJC deposition site, and a heterodimer containing RBM8A (also called Y14) and MAGOH [10]. Many EJCs also contain UPF3B (also called UPF3X), and the alternative RNPS1- or CASC3-EJCs contain, respectively, RNPS1, Acinus 1 (ACIN1 or, in its place, Pinin), SAP18, and UPF2, or CASC3 (also called BTZ or MLN51) and the cell-signaling kinase AKT [11–15]. Insulin, which augments AKT signaling [16], increases the number of AKT-containing EJCs and NMD efficiency [11]. However, there is much about EJCs that remains unclear. Here, we sought to determine how and when EJC core constituents and the alternative EJC constituents RNPS1 and CASC3 assemble on spliceosomes, and if any steps of EJC formation on spliceosomes are regulated by insulin. Using liquid chromatography followed by tandem mass spectrometry (LC-MS/MS) to identify proteins from HEK293T-cell lysates that co-immunoprecipitate with EIF4A3 in the presence of RNase I, we identified constituents shared by pre-B and B spliceosome complexes. We also identified the splicing factors CWC27 and CWC22, which are first detected in Bact-I [17]. Notably, the eight characterized Bact complexes describe multiple sub-stages having different structural conformations and splicing factor compositions [17–19]. Unexpectedly, we found that CWC27 and CWC22 form a complex with EIF4A3, even when pre-B complex formation, which precedes Bact-I formation, was inhibited using Pladienolide B (PB). Thus, we hypothesize that a second mechanism utilizes CWC27 and CWC22 as an “EIF4A3 recruitment complex”, consistent with previous reports that CWC27 and CWC22 are important for EIF4A3 recruitment to spliceosomes and deposition upstream of exon–exon junctions [20–22]. We find that not only EIF4A3 but also the entire EJC core join spliceosomes as early as the pre-B or B complex with the joining of RBM8A and MAGOH to EIF4A3. RNPS1 is also detected in pre-B and/or B complexes, while CASC3 is recruited to Bact complexes that lack detectable CWC27, i.e. likely after CWC27 departs the spliceosome. Together, our data show for the first time that EJC assembly in cells begins on spliceosomes prior to their activation and can proceed using multiple pathways. Materials and methods Plasmid constructions To generate pcDNA3.1-HA-EIF4A3, pcDNA3.1-HA-RBM8A, and pcDNA3.1-HA-MAGOH, complementary DNA (cDNA) was polymerase chain reaction (PCR)-amplified from sequence-verified plasmids encoding either human EIF4A3 (MHS6278-202755734, Horizon Discovery), RBM8A (MHS6278-202756105, Horizon Discovery) or MAGOH (MHS6278-202756286, Horizon Discovery), using Q5 Polymerase (M0491S, New England Biolabs) and gene-specific primers. The PCR products were gel-purified using the Zymoclean Gel DNA Recovery Kit (D4001, Zymogen) and assembled with EcoRI-HF (R3101S, New England Biolabs)-digested pcDNA3.1 HA (128 034, Addgene) using NEBuilder HiFi Assembly Master Mix (E2621S, New England Biolabs). To generate the sequence donor plasmid to knock-in the 2XHA tag at the C-terminus of the RNPS1 protein encoded by each copy of the HEK293T-cell RNPS1 gene, a DNA fragment (the “RNPS1-2XHA repair template”, Supplementary Fig. S2D) was synthesized (Twist Biosciences) to constitute 505-bps residing upstream of the stop codon in the RNPS1 gene, i.e. the upstream homology arm (UHA), followed by DNA encoding a 2XHA tag, a glycine- and serine-rich linker, the RNPS1 stop codon and then 508-bps residing downstream of the RNPS1 stop codon i.e. the downstream homology arm. The UHA harbors one synonymous mutation that renders the RNPS1-2XHA repair template insensitive to guide RNA (gRNA) #1-directed Cas9-mediated cleavage 4-bp upstream of the endogenous RNPS1 stop codon, where “GGG” on the reverse strand is mutated to “CGA”. For gRNA #2-directed Cas9-mediated cleavage, the RNPS1-2XHA repair template was rendered insensitive by the addition of the linker and 2XHA sequence. This 505-bp DNA fragment was amplified using Q5 Polymerase and sequence-specific primers to add nucleotides complementary to a BamHI-generated overhang at the 5′ end and complementary to a HindIII-generated overhang at the 3′ end. The targeting vector, which derived from pcDNA3 [23], was digested with BamHI-HF (R3136S, New England Biolabs) and HindIII-HF (R3104S, New England Biolabs) according to manufacturer instructions, gel-purified and assembled with the RNPS1-2XHA repair template using the HiFi Assembly Master Mix. RNPS1-targeting gRNAs were designed using CRISPOR [24] and using a 100-nt window centered around the endogenous stop codon in the RNPS1 gene as the input sequence. High-scoring gRNA sequences were used to design DNA oligonucleotides (IDT) containing the gRNA sequence and nucleotides complementary to Bbs I-generated overhangs (gRNA#1, 5′-CACCGCTTCAGTGGCCTGTTTATC-3′, sense, and 5′-AAACGATAAACAGGCCACTGAAGC-3′, antisense; and gRNA#2, 5′-CACCGTCCAACTCCTCCCGATAAAC-3′, sense, and 5′-AAACGTTTATCGGGAGGAGTTGGAC-3′, antisense). Each sense and antisense pair was annealed, phosphorylated at their 5′ end using T4 PNK (M0201S, New England Biolabs) and then ligated with BbsI-HF-digested (R3539S, New England Biolabs) pU6-(BbsI)-MYC-Cas9-T2A-mCherry (Addgene #64 324) using T4 DNA ligase (New England Biolabs). All plasmids were verified by direct plasmid sequencing (Plasmidsaurus) and prepared for mammalian-cell transfections using the NucleoBond Xtra Maxi EF kit (740424.50, Machery Nagel). HEK293T-cell culture and generation of RNPS1-2XHA HEK293T cells Human HEK293T cells (CRL-11268, ATCC) were cultivated in Dulbecco’s modified Eagle’s medium (DMEM; 11 995 073, Gibco) supplemented with 10% fetal bovine serum (FBS; 89 510 194, Avantor) and 1× penicillin/streptomycin (15140-122, Gibco). When specified, cells were transiently transfected with either small interfering (si)RNAs or plasmids using Transit-X2 (MIR6005, Mirus Bio). Where indicated, cells were deprived of serum by incubation in unsupplemented DMEM for 24 h and either harvested or incubated in 130 nM insulin (12 585 014, Gibco) for 30 min prior to harvesting. When specified, cells were incubated in 1 µM PB (445493-23-2, Santa Cruz) reconstituted in dimethyl sulfoxide (DMSO; D2650, Millipore Sigma) for the indicated amount of time prior to harvesting. Cells were harvested using 0.25% Trypsin–ethylenediaminetetraacetic acid (EDTA; 25 200 056, Invitrogen), pelleted and then stored in Gibco Cell Recovery Freezing Medium (12 648 010, Thermo Fisher Scientific) before lysis, as described below. RNPS1-2XHA HEK293T cells were generated by transfecting HEK293T cells at 80% confluency in a 10-cm dish with 0.6 µg of either RNPS1 gRNA #1 pU6-(BbsI)-MYC-Cas9-T2A-mCherry or RNPS1 gRNA #2 pU6-(BbsI)-MYC-Cas9-T2A-mCherry plasmid, 0.4 µg of the RNPS1-2XHA repair template plasmid, and 3 µl of Transit-X2. After 24 h, cells were cultured for 48 h in 10 µg/ml Blasticidin HCl (A1113903, Gibco) to select for transfected cells. After selection, cells were recovered in complete medium for 24 h and then individually sorted (Cytek Aurora CS) into 96-well plates. Edited cells were screened for the introduction of the 2XHA tag using anti-HA western blotting. Genomic DNA was isolated from selected clones (RNPS1-2XHA HEK293T #1 and RNPS1-2XHA HEK293T #2) using the Mag-Bind® Blood & Tissue DNA HDQ 96 Kit (M6399-00S, Omega Bio-Tek), subjected to PCR to amplify the RNPS1 gene around the edited region using the primer pair “RNPS1-2XHA S”, 5′-CCGCTCCAGCTCCAACTCCTCT-3′ and “RNPS1-2XHA AS” 5′-ACCAACAGCACTTCTGCAGCCG-3′ and then sequenced by Plasmidsaurus. RNPS1-2XHA HEK293T #1 and RNPS1-2XHA HEK293T #2 were detectably edited at two of the three copies of the RNPS1 gene in HEK293T cells. Immunoprecipitations For immunoprecipitations (IPs) of total-cell lysates or nuclear fractions, including IPs of RNPS1-2XHA, 5 µg of the indicated antibody (Supplementary Table S4) were pre-bound to 50 µl of Protein A Dynabeads (10002D, Invitrogen). For anti-HA IPs of HA-tagged EIF4A3, HA-RBM8A, and HA-MAGOH, 50 µl of anti-hemagglutinin (HA) magnetic beads (PI88837, Invitrogen) were used. For total-cell IPs, cells were resuspended in Gentle Hypotonic Lysis Buffer (10 mM Tris pH 7.5, 10 mM NaCl, and 10 mM EDTA) containing Pierce Halt Protease and Phosphatase EDTA-free Inhibitor Cocktail (78 442, Thermo Fisher Scientific), lysed for 10 min on ice and then sonicated to complete lysis. Lysates were clarified by centrifugation before use in IPs. For IPs using nuclear fractions, cells were fractionated using the Pierce NE-PER Nuclear and Cytoplasmic Extraction Kit (78 835, Thermo Fisher Scientific) according to manufacturer instructions, except that nuclei were suspended in Gentle Hypotonic Lysis Buffer containing 150 mM NaCl and Pierce Halt Protease and Phosphatase EDTA-free Inhibitor Cocktail prior to lysis by sonication. Antibody-bound beads were incubated with total-cell or nuclear lysates for 2 h at 4°C, washed three times with NET2 Buffer (50 mM Tris pH 7.4, 150 mM NaCl, and 0.05% NP-40), treated with RNase I (RNase ONE; M4265, Promega) for 15 min at room temperature followed by 15 min at 4°C where indicated, washed five times with NET2 Buffer and eluted in 2× Laemmli Sample Buffer (1610737EDU, Bio-Rad). Ratios of cell lysates analyzed before and after IP are reported in Supplementary Table S3. Liquid chromatography coupled to tandem mass spectrometry Sample preparation Triplicate anti-EIF4A3, anti-CASC3, and anti-HA IPs, in parallel with rIgG negative control IPs, were performed using total-cell lysates of HEK293T cells as described above. Lysates before IP were also analyzed in triplicate. Each sample was electrophoresed ∼1 cm into a 4%–20% sodium dodecyl sulphate (SDS) polyacrylamide gradient gel (4561 096, Bio-Rad), stained using Coomassie Brilliant Blue G-250 (1 610 406, Bio-Rad) and destained in deionized water. Bands corresponding to each sample were excised from the gel and diced into ∼1-mm pieces. Diced samples were reduced and alkylated using DL-dithiothreitol (D0632, Millipore Sigma) and Iodoacetamide (I1149, Millipore Sigma), respectively, dehydrated using acetonitrile (ACN; 325 730 025, Thermo Scientific) and covered with a solution of 10 ng/µl trypsin (V5280, Promega) in 50 mM ammonium bicarbonate for 30 min at room temperature. Additional ammonium bicarbonate was added to completely submerge the gel pieces, which were incubated overnight at 37°C. A solution of 50% ACN and 0.1% trifluoroacetic acid (TFA; 044630.AY, Thermo Scientific) was used to extract peptides, which were then dried down in a CentriVap concentrator (Labconco), desalted using a C18 spin column (89 873, Thermo Fisher), dried again and reconstituted in 0.1% TFA. Peptide concentrations were determined using a fluorometric peptide assay (23 290, Thermo Fisher). Mass spectrometry Peptides were injected onto a 75 µm x 2 cm trap column (164 946, Thermo Fisher) and then re-focused using an Aurora Elite 75 µm x 15 cm C18 column (AUR4-15075C18, IonOpticks) and Vanquish Neo UHPLC (Thermo Fisher) in-line with an Orbitrap Astral mass spectrometer (Thermo Fisher). Ions were introduced to the mass spectrometer using an Easy-Spray source operating at 2 kV. The gradient began at 1% solvent B (0.1% formic acid in 80% ACN) in 99% solvent A (0.1% formic acid in water), which was increased to 5% solvent B over 0.1 min, then to 30% solvent B over 12.1 min, then to 40% solvent B over 0.7 min, then to 99% solvent B over 0.1 min and finally washed in 99% solvent B for 2 min followed by re-equilibration in 1% solvent B for injection of the next sample. The Orbitrap Astral was operated in data-independent acquisition (DIA) mode. MS1 scans were acquired using the Orbitrap at a resolution of 240 000. The maximum injection time was 5 ms over a range of 380–980 m/z. DIA MS2 scans were acquired using the Astral mass analyzer with a 6-ms maximum injection time and a variable windowing scheme of 4-Da windows ranging from 380–750 m/z, and 6-Da windows ranging from 750–980 m/z. The higher-energy collisional dissociation (HCD) collision energy was 28%, and the normalized automatic gain control was 500%. Fragment ions were collected over a scan range of 150–2000 m/z, with a cycle time of 0.6 s. Analysis of IP-MS Data Raw data were processed using DIA-NN v1.8.1 (https://github.com/vdemichev/DIA-NN) [25] and library-free analysis mode. The library was annotated using the human UniProt ‘one protein sequence per gene’ database (UP000005640_9606, downloaded 4/7/2021) with ‘deep learning-based spectra and retention time prediction enabled, where retention time is the time it takes for a peptide to reach the detector after being injected onto the C18 column. The maximum number of missed cleavages was set to 1 for precursor ion generation and the maximum number of variable modifications was set to 1 for oxidized methionine. Also set were 7–30 for the peptide length range, 2–3 for the precursor charge range, 380–980 for the precursor m/z range, and 150–2000 for the fragment m/z range. The quantification was set to ‘Robust LC (high precision)’ mode with retention time-dependent normalization. Match between runs (MBR) was enabled, and protein inferences were set to ‘Genes’. ‘Heuristic protein inference’ was set to off. MS1 (precursor ion) and MS2 (fragment ion) mass tolerances and the scan-window size were set automatically by the software. Precursor ions were subsequently filtered at library precursor q-value (1%), library protein group q-value (1%) and posterior error probability (50%). Protein quantification was carried out using the MaxLFQ algorithm as implemented in the DIA-NN R package (https://github.com/vdemichev/diann-rpackage). The number of peptides quantified in each protein group was counted as implemented in the DiannReportGenerator Package (https://github.com/kswovick/DIANN-Report-Generator) [26]. Western blotting Cell lysates in 1× Laemmli Buffer were electrophoresed in a 4%–20% polyacrylamide gradient containing SDS and subjected to western blotting using antibodies and dilutions listed in Supplementary Table S4. Blots were imaged using a Chemidoc Imaging System (Bio-Rad). siRNA transfections The Silencer Negative Control No. 1 siRNA (AM4636, Ambion) served as control (CTL) siRNA. CWC22#1, CWC22#2, CWC27#1, CWC27#2, RNPS1#1–4, EFTUD2#1, and EFTUD2#2 siRNAs were sourced from Horizon Discovery (Supplementary Table S5). Where indicated, cells were transfected with 50 nM of the indicated siRNA using the Transit-X2 Dynamic Delivery System (MIR6005, Mirus Bio) according to manufacturer instructions. For CWC22, CWC27, and RNPS1 siRNAs, cells were transfected once and cultured for 48 h prior to lysis. For EFTUD2 siRNAs, cells were transfected with siRNA, transfected a second time after culturing for 24 h, and then harvested 24 h after the second transfection. For each siRNA experiment, CTL siRNA transfections were matched to the protocol used for the siRNAs with which it was paired. RNA purification For analyses of RNA before and after IP, RNA was isolated using Trizol Reagent (15596-018, Invitrogen), incubated with 2 µl of RQ1 DNase (M6101, Promega) for 45 min at room temperature, extracted using acidic phenol–chloroform–isoamyl alcohol and precipitated using 100% ethanol, 0.3 M Na acetate (pH 5.5), and 20 µg glycogen (R0561, Thermo Fisher Scientific) as previously described [11]. For analysis after PB treatment, RNA was isolated from total-cell lysates of ∼4–5 × 105 cells using Trizol Reagent and then further purified from the aqueous phase of Trizol extractions using the Mag-Bind® Total RNA 96 Kit (M6731-01, Omega Bio-Tek) according to manufacturer instructions, except the DNase I digest step included the addition of 2 µl of TURBO DNase (AM2238, Thermo Fisher Scientific). RNA quantity and quality were assessed by NanoDrop (Thermo Fisher Scientific). RT-qPCR and RT-semiquantitative PCR For IP analyses, the same number of cell equivalents was analyzed before IP and after IP, and for analysis of splicing inhibition by PB, cDNA was generated using 100 ng of RNA. In parallel, for quantitative analyses, two-fold dilutions of RNA ranging from 1000 to 0.122 ng were used to generate a 14-point standard curve. cDNA was synthesized using Maxima H Minus cDNA Synthesis Master Mix (M1662, Thermo Fisher Scientific) according to manufacturer instructions, in a total reaction volume of 20 µl. A “no reverse transcriptase, (RT)” control reaction that tested for the efficiency of DNase digestion was generated using 100 ng of RNA and the Maxima No RT Control mix. Reaction conditions for RT-quantitative (q)PCR were: 10 min at 25°C, 30 min at 50°C, and then 5 min at 85°C. qPCR was performed using 2 µl of 10-fold diluted cDNA, transcript-specific primers that were verified using NCBI Primer-BLAST (https://www.ncbi.nlm.nih.gov/tools/primer-blast/index.cgi), each at 0.5 µM (Supplementary Table S6), and PerfeCTa® SYBR® Green SuperMix, Low ROX™ (95056-500, QuantaBio) in a reaction volume of 10 µl. Cycling conditions using a Quantstudio 5 Real-Time PCR System (Thermo Fisher Scientific) were: one cycle for 1 min at 95°C and subsequently 40 cycles, where each cycle consisted of 15 s at 95°C followed by 15 s at 60°C, and then 1 min at 72°C. Data analysis was performed using the Quantstudio Design and Analysis Software and Microsoft Excel. Each transcript was quantitated using the standard curve generated for each primer pair. Primer specificity was verified using melt curves. To verify RNase I efficacy, the level of each RT-qPCR product after IP was normalized to its level before IP. For splicing inhibition assays performed in the presence of PB, the level of RT-qPCR products derived from pre-mRNA was normalized to the level of RT-qPCR products derived from 18S ribosomal RNA (rRNA), which was unaffected by PB treatment. Semiquantitative PCR (sqPCR) analyses of RNA after PB treatment were performed using 4 µl of 10-fold diluted cDNA, 0.5 µM of each verified transcript-specific primer pair that anneals to both the pre-mRNA and resulting spliced mRNA, 200 µM dNTPs, and Q5 DNA Polymerase (New England Biolabs). A “no RT” negative control was analyzed using each primer pair. Cycling conditions were optimized so as to assay each transcript during the linear phase of amplification, and were: incubation for 30 s at 98°C and subsequently either 20 cycles for 18S precursor rRNA, 25 cycles for BRD2 transcripts or 30 cycles for DNAJB1 transcripts, where each cycle consisted of 10 s at 98°C, followed by 30 s at 68°C and then 1.5 min at 72°C. Reactions were resolved using a nondenaturing 5% polyacrylamide Tris-borate-EDTA (TBE) gel and detected using SYBR Gold (S11494, Thermo Fisher Scientific) and a Bio-Rad Chemidoc Touch imager. Quantitation was performed using ImageLab (Bio-Rad). Primer sequences and amplicon sizes are listed in Supplementary Table S6. Northern blotting RNA was separated in a 10% polyacrylamide denaturing gel [27], transferred to Hybond Ny+ (45-000-927, Cytiva) and crosslinked at 120 000 µJ (254 nM) for 1 min using a Stratalinker ultraviolet (UV) 1800 Crosslinker (Stratagene). The membrane was blocked for 1 h at 42°C in 20 ml of DIG Easy Hyb Buffer (11 603 558 001, Roche) containing 1× DIG Blocking Buffer (11 585 762 001, Roche). After removing 10 ml, 1 µg of each DNA probe was added, and the incubation was continued overnight at 42°C. DNA probes were synthesized by IDT (sequences listed in Supplementary Table S7), 3′ end-labeled for 30 min at 37°C with digoxigenin-11-UTP (11 209 256 910, Millipore Sigma) using 3′ Terminal Deoxynucleotidyl Transferase (M0315S, New England Biolabs) and purified using a G-25 MicroSpin Column (27 532 501, Cytiva). Probe-labeling efficiencies were quantitated using dot blotting [28]. Membranes were washed at 42°C twice with 2× saline-sodium citrate (SSC; 300 mM NaCl and 30 mM Na Citrate) + 0.1% SDS, twice with 1× SSC + 0.1% SDS, twice with 0.1× SSC + 0.1% SDS and subsequently blocked for 30 min at 25°C in 1× DIG Blocking Buffer. Blots were probed for 30 min at 25°C in a 1:10 000 dilution of alkaline-phosphatase conjugated anti-Digoxigenin-alkaline phosphatase Fab fragments (11 093 274 910, Roche) in 1× DIG Blocking Buffer, followed by four washes in 1× DIG Wash Buffer (11 585 762 001, Roche) and one wash in 1× DIG Detection Buffer (11 585 762 001, Roche). Membranes were submerged in CDP-Star Chemiluminescence Detection Substrate (12 041 677 001, Roche) and imaged using a Chemidoc Imaging System. Confocal fluorescent microscopy RNPS1-2XHA #1 HEK293T cells were grown on 25-mm laminin- and poly-D-lysine-coated #1.5 glass coverslips (50-192-9634, Electron Microscopy Sciences). Cells were fixed in 4% paraformaldehyde (J61899.AK, Thermo Fisher Scientific) for 15 min at room temperature, permeabilized with 0.5% Triton X-100 (A16046.AE, Thermo Fisher Scientific) in 1× Phosphate-Buffered Saline (PBS pH 7.4; 137 mM NaCl, 2.7 mM KCl, 10 mM Na2HPO4, and 1.8 mM KH2HPO4) for 10 min at room temperature and then blocked with 0.025% Triton X-100 and 10% FBS in 1× PBS for 30 min at room temperature. Coverslips were incubated overnight at 4°C in the indicated primary antibodies, washed with Wash Buffer (0.025% Triton X-100 in 1× PBS) three times and then incubated for 1 h at room temperature in the corresponding secondary antibodies. Antibodies and dilutions are listed in the Supplementary Table S4. Cells were stained for 5 min at room temperature in 0.5 ng/µl 4′,6-diamidino-2-phenylindole (DAPI; 62 248, Thermo Fisher) in Wash Buffer and then washed three more times. Coverslips were mounted on slides using ProLong AntiFade Diamond Mountant (P36970, Invitrogen) and imaged using a Leica Stellaris 5 Laser-Scanning Confocal Microscope with a 63× oil objective. A 3.1× zoom factor was used for 1024 × 1024-pixel images in accordance with the Nyquist rule. Four fields of view containing 10 cells were imaged as Z-stacks. Image Analysis Software (Imaris, version 10.1.1) was used for post-imaging processing. For each Z-stack, the channels for RNPS1-2XHA and SC35 were deconvolved over 10 iterations. Quantification and statistical analyses Statistical analyses specified in the figures and figure legends were generated using Microsoft Excel. For all western blots, northern blots and RT-sqPCR gel images, standard dilution curves were used to ensure that quantitations using ImageLab (Bio-Rad) fell within the linear range of analysis. Quantitations are representative of the replicates indicated in each figure legend. For RT-qPCR, bar graphs and histograms represent the means measured from three or four technical qPCR replicates and error bars represent standard deviations. All RT-qPCR and RT-sqPCR experiments are representative of results from two to three experimental replicates. Significance cutoffs were defined as (*) P <.05, (**) P <.01, and (***) P <.001, which were calculated using a two-tailed Student’s t-test. Results Pre-B- and B-complex splicing factors co-immunoprecipitate with EIF4A3 Several models exist for the timing of EJC core assembly on spliceosomes during pre-mRNA splicing. In one model, based on data demonstrating that unspliced pre-mRNA co-immunoprecipitates with EIF4A3 in in vitro splicing reactions using HeLa-cell nuclear extracts, EIF4A3 was proposed to join the spliceosome in either the A or the B complex [29]. However, factors specific to either complex were not shown to associate with EIF4A3 [29]. In a different model, based on the findings that (i) EIF4A3 interacts directly, via CWC22, with a heterodimer of CWC27 and CWC22 in in vitro pulldowns of purified proteins and co-immunoprecipitates with FLAG-CWC27 in HeLa-cell lysates [21], and (ii) CWC27 is resolved in the cryogenic electron microscopy (cryo-EM) structure of the early Bact complex but not the mature Bact complex [18], EIF4A3 was proposed to join the Bact complex after CWC27 and CWC22 joining [21]. In yet another model, based on cryo-EM structures of spliceosomes demonstrating that EIF4A3 is not detected in any Bact complex but is detected in C and later complexes via a direct interaction with CWC22 and EFTUD2, EIF4A3 was proposed to join during the transition of the B* complex (for which there is no structure of the human B* spliceosome) to the C complex [17, 18, 30]. In any case, since in silico structural modeling indicates that RBM8A and CWC27 bind to EIF4A3 in a mutually exclusive manner [21], and yeast two-hybrid data indicate that RBM8A and MAGOH exist in cells as a heterodimer [10], RBM8A and MAGOH were predicted to join EIF4A3 on the C complex after CWC27 departs (Supplementary Fig. S1A) [17, 18, 21]. We sought to supplement these data, initially by assaying for spliceosome constituents that co-immunoprecipitate with EIF4A3. We performed anti-EIF4A3 IPs in triplicate and, as a negative control, a rabbit (r)IgG IP, each in the presence of RNase I, using HEK293T-cell lysates. In these and subsequent IPs, treatment with RNase I enabled the analysis of individual EJCs (Supplementary Fig. S1B and C and Supplementary Data) and individual spliceosomes (Supplementary Fig. S1D and E and Supplementary Data). This is critical since many newly made transcripts contain more than two exons and, thus, are typically bound by multiple EJCs and/or spliceosomes at different stages of formation. The resulting HEK293T-cell lysates, before and after anti-EIF4A3 IPs, were then analyzed by LC-MS/MS (IP-MS; Supplementary Table S1 and Supplementary Data). As expected, anti-EIF4A3 IPs, relative to the rIgG IP, were enriched in the known EJC constituents RBM8A, MAGOH, UPF3B, RNPS1, ACIN1, PNN, SAP18, and CASC3 (Supplementary Fig. S1F). The nuclear cap-binding protein heterodimer CBP80–CBP20 and the cytoplasmic poly(A)-binding protein PABPC1 were not detected in either IP (Supplementary Fig. S1F), consistent with efficient RNase I treatment. We next generated a high-confidence list of splicing factors that co-immunoprecipitate with EIF4A3 (Supplementary Table S1 and Supplementary Data) and are resolved in spliceosome cryo-EM structures of the pre-B complex to the C complex [17–19, 30–34]. Splicing factors known to constitute the pre-B complex, B complex, Bact complexes, and/or the C complex co-immunoprecipitated with EIF4A3 (Fig. 1A). Evidence that EIF4A3 joins pre-B and/or B complexes derives from the co-IP of PRPF3, PRPF31, PRPF4, PRPF6, and SART1, which cryo-EM structures indicate are confined to pre-B and B complexes (Fig. 1A and Supplementary Fig. S1A). Factors found in spliceosomes earlier than but absent from pre-B and subsequent complexes were absent from the high-confidence list of proteins associated with EIF4A3 (Supplementary Table S1). Together, these data indicate that the earliest spliceosome constituents that are detectably associated with EIF4A3 are found in the pre-B and/or B complexes. Defining splicing factors in the pre-B, B, pre-Bact, Bact, and C complexes of spliceosomes that co-immunoprecipitate with the EJC core constituent EIF4A3. (A) Heat map representation of the log2-fold enrichment of the denoted structurally resolved splicing factors in anti-(ɑ)-EIF4A3 IPs of HEK293T-cell lysates, performed in the presence of RNase I, relative to the level of each factor before IP. The vertical green box denotes pre-B and B complex factors that co-IP with EIF4A3; the horizontal blue box specifies spliceosome structures containing CWC22, CWC27, or both; *signifies spliceosome structures considered to be equivalent sub-steps of Bact. Results derive from triplicate samples, each analyzed before and after IP. (B, upper) Western blots of the denoted proteins in nuclear fractions of HEK293T cells that were cultured in complete medium or serum-deprived and subsequently harvested with no (−) insulin treatment or cultured for 30 min in the presence (+) of 130 nM insulin prior to lysis. Lysates were analyzed before (−) and after (+) IP in the presence of RNase I using anti-EIF4A3 or, as a negative control rabbit (r)IgG. *denotes rIgG heavy chain reactivity that does not interfere with analyses. Here and elsewhere, leftmost lanes under the wedge are three-fold serial dilutions of lysates. Results represent three experimental replicates. (B, lower) As in panel (B, upper) but using RNPS1-2XHA #1 HEK293T-cell nuclear fractions. (C) Schematic of the structurally resolved splicing factors detected only in either the pre-B/B, the pre-Bact, the Bact or the C complex(es) identified in anti-EIF4A3 IP-MS and/or anti-EIF4A3 IPs followed by western blotting. (D) As in panel (B), but for anti-SF3B1 and rIgG IPs in the presence of RNase I using nuclear fractions of HEK293T cells (upper) or RNPS1-2XHA #1 HEK293T cells (lower) cultured in complete medium. Results represent three experimental replicates. Insulin-mediated changes to alternative EJC composition in the cytoplasm are not evident in the nucleus To increase the sensitivity of assaying spliceosome-associated EIF4A3, and to independently corroborate our IP-MS data, we repeated the anti-EIF4A3 and rIgG IPs in the presence of RNase I using nuclear fractions of HEK293T cells and assayed for spliceosome constituents using western blotting (Fig. 1B, upper). Given our previous finding that EJC composition can be altered by insulin signaling [11], we also asked whether insulin stimulation alters EIF4A3 joining to the spliceosome. Therefore, prior to generating cell lysates for IPs, HEK293T cells were cultured either in complete medium, medium without serum for 24 h, which reduces AKT signaling and sensitizes cells to insulin (Supplementary Fig. S2A) [35], or medium without serum for 24 h followed by the addition of insulin to 130 nM for 30 min (Supplementary Fig. S2A and Supplementary Data). Notably, the cellular distribution of EJC constituents was not detectably altered by serum deprivation either without or with the subsequent addition of insulin (Supplementary Fig. S2B and Supplementary Data). IP specificity was confirmed by the detection of Beta-actin (BACT) before but not after either anti-EIF4A3 or rIgG IPs (Fig. 1B). As expected [5], RBM8A, MAGOH, UPF3B, ACIN1, SAP18, and CASC3 were detected in anti-EIF4A3 IPs. Each of these EJC constituents co-immunoprecipitated with EIF4A3 to the same extent in all three culture conditions, indicating that EJC core and CASC3-EJC formation is insulin-insensitive in the nuclear fraction at the level of sensitivity provided by western blotting of anti-EIF4A3 IPs (Fig. 1B, upper). Since we could not identify a commercial anti-RNPS1 antibody suitable for western blotting after anti-EIF4A3 IP (data not shown), we characterized two independently generated HEK293T-cell lines in which a 2XHA tag was appended using CRISPR-Cas9 to the C-terminus of two of the three copies of the RNPS1 gene (Supplementary Fig. S2D and data not shown). Anti-HA IPs revealed the co-IP of splicing factors with RNPS1-2XHA in both engineered cell lines (Supplementary Fig. S2E and Supplementary Data). Continuing with cell line #1 that more robustly immunoprecipitated splicing factors as previously observed [36], RNPS1-2XHA largely co-localized with SC35 in nuclear speckles (Supplementary Fig. S2F), as does endogenous RNPS1 [37, 38]. Anti-EIF4A3 IPs of nuclear fractions demonstrated that the co-IP of RNPS1-2XHA with EIF4A3 is unaltered by serum deprivation without or with insulin stimulation (Fig. 1B, lower). Thus, insulin does not alter the commitment to form either an RNPS1-EJC or a CASC3-EJC in nuclear fractions (Fig. 1B). Using total-cell lysates, the co-IPs of the RNPS1-EJC constituent UPF2 and the CASC3-EJC constituent AKT, both of which are largely cytoplasmic [39, 40] (Supplementary Fig. S2B), were, respectively, downregulated or upregulated by insulin [11]. Using nuclear fractions, the co-IP of either protein with EIF4A3 under all three culture conditions was either barely detectable or undetectable (Fig. 1B, upper). Given that serum deprivation without or with the subsequent addition of insulin did not alter the co-IP of any tested EJC constituent with EIF4A3 using nuclear fractions (Fig. 1B), all subsequent analyses utilized cells grown in complete medium. We next assayed the anti-EIF4A3 IPs of nuclear fractions from WT HEK293T cells and RNPS1-2XHA #1 cells (Fig. 1B upper and lower) for splicing factors detected in anti-EIF4A3 IP-MS analyses of total-cell lysates (Fig. 1A). PRPF3, PRPF6, PRPF31, SART1, and LSm7 (in pre-B and B complexes), WBP4 (in the B complex), and MFAP1, ZMAT2, KIN17, and CTBL1 (in pre-Bact complexes) were detected in anti-EIF4A3 IPs (Fig. 1B, upper and lower). Given the number of pre-B and B complex factors, this finding suggests that EIF4A3 joins pre-B and/or B spliceosomes before CWC27 and CWC22 join during formation of the Bact-I complex (Fig. 1B and Supplementary Fig. S1A). Corroborating anti-EIF4A3 IP-MS results (Fig. 1A), EFTUD2, CWC27, and CWC22 co-immunoprecipitated with EIF4A3 (Fig. 1B, upper), as did AQR, which is present in the pre-Bact-II complex and subsequent complexes [17–19, 30, 32] and is involved in EIF4A3 deposition on newly spliced transcripts [41]. SF3B1, found in all complexes between the A complex and the Bact complexes [1], and the Bact constituent DHX16 also co-immunoprecipitated with EIF4A3 (Fig. 1B, upper and Supplementary Fig. S1A), further verifying anti-EIF4A3 IP-MS results (Fig. 1A). Our detection of multiple splicing factors co-immunoprecipitating with EIF4A3 that each bind to the spliceosome at distinct sites from the EJC, including SF3B1, MFAP1, ZMAT2, AQR, DHX16, and others suggests that RNase I digestion largely did not disrupt spliceosome integrity (Fig. 1A and B). The inefficient co-IP of some splicing factors with anti-EIF4A3 may reflect their loose association with the spliceosome and/or their presence in short-lived splicing intermediates [42]. As for those EJC constituents analyzed, insulin did not detectably alter the co-IP of any tested splicing factors with EIF4A3 (Fig. 1B). Our finding EIF4A3 on spliceosomes that form prior to the C complex (Fig. 1A–C) was recapitulated in anti-SF3B1 IPs, also performed in the presence of RNase I, using nuclear fractions (Fig. 1D, upper). Based on cryo-EM structures of the C complex demonstrating a direct interaction between EIF4A3 and EFTUD2 [30, 32], we hypothesized that EFTUD2 mediates the recruitment of EIF4A3 to pre-B and/or B complexes. EFTUD2, a member of the U5 small nuclear ribonucleoprotein (snRNP), joins the A complex as part of the U4/U5/U6 tri-snRNP to form the pre-B complex [1, 43]. Since cryo-EM structures of pre-B and B complexes do not contain CWC27 and CWC22 [17, 33], and since CWC27 and CWC22 also promote recruitment of EIF4A3 to the spliceosome [20–22], EIF4A3 may be recruited to spliceosomes by multiple mechanisms (see below). To test our prediction that EIF4A3 can be recruited to pre-B and/or B complex spliceosomes in an EFTUD2-dependent manner, we transfected RNPS1-2XHA #1 HEK293T cells with either control (CTL) siRNA or one of two EFTUD2 siRNAs that each reduced the level of EFTUD2 protein to ∼30% of normal (Supplementary Fig. S2G). In a previous study that used HEK293T cells in which two of the three copies of the EFTUD2 gene had been knocked-out, RNA-seq revealed mis-splicing rather than a global inhibition of splicing [44]. Thus, we reasoned that EFTUD2 knockdown would not inhibit splicing prior to EIF4A3 joining to the pre-B and/or B complex. Relative to CTL siRNA, neither of the two EFTUD2 siRNAs reduced the co-IP of RBM8A or MAGOH, the pre-B and B complex factor PRPF6, the B complex factor WBP4, or the B and pre-Bact complex factor ZMAT2 in anti-EIF4A3 IPs of nuclear lysates (Supplementary Fig. S2G). These data suggest that a reduction in EFTUD2 protein levels to 30% of normal is insufficient to inhibit EJC core assembly and EIF4A3 joining to pre-B and/or B complexes, and/or that additional splicing factors mediate the interaction between EIF4A3 and the pre-B and/or B complex (Supplementary Fig. S2G and H). Inhibiting spliceosome assembly using PB implicates CWC27–CWC22 as an EIF4A3 recruitment complex In one model for EIF4A3 recruitment to the spliceosome, CWC27 and CWC22 serve as an EIF4A3 “landing pad” in the Bact complexes via direct contacts between EIF4A3 and CWC22 [21]. In support of this model, the introduction of amino-terminal FLAG-tagged EIF4A3 together with CWC22, each or both of which harbored a point mutation that disrupts the EIF4A3–CWC22 interface but does not inhibit splicing, reduced the deposition of EIF4A3 on mRNAs in in vitro splicing reactions [20]. In theory, however, EIF4A3 may join a CWC27–CWC22 heterodimer prior to joining the spliceosome. To test the possibility that CWC27 and CWC22 could constitute an EIF4A3 “recruitment complex” that delivers EIF4A3 to the spliceosome, splicing was inhibited using PB. PB binds to the branch-point adenosine-binding site of the U2 snRNP constituent SF3B1 to prevent tri-snRNP (i.e. U4/U5/U6) joining to the A complex and, therefore, pre-B complex formation (Fig. 2A) [45] before CWC27 and CWC22 are recruited to the spliceosome (Supplementary Data Fig. S1A). Confirming PB activity, lysates of HEK293T cells cultured in the presence of 1 µM PB for 120 min and immunoprecipitated with anti-SF3B1 showed that PB inhibits the co-IP of all tri-snRNP constituents assayed (Supplementary Data Fig. S3A and B, and Supplementary Data). CWC27 and CWC22 co-immunoprecipitate with EIF4A3 in the presence of PB. (A) Schematic depicting that PB inhibits splicing before recruitment of the U4/U5/U6 tri-snRNP to the spliceosome by binding to SF3B1 of U2 snRNP [45]. Created in BioRender. Abshire, E. (2026) https://BioRender.com/sx48pw2. The small red shape specifies SF3B1 bound to the branchpoint adenosine (A). (B, left) Histogram representation of RT-qPCR quantitations of the indicated pre-mRNA, normalized to the level of 18S rRNA, using RNA isolated from HEK293T cells treated with 1 µM PB and harvested at the indicated times (min). n = 3 technical replicates that are representative of two independent experiments. P-values, determined using a two-tailed Student’s t-test, were: ** ≤ 0.005; *** ≤ 0.001. (B, right) Schematics of DNAJB1 (orange exons) and BRD2 (green exons) pre-mRNAs, where AUG and Ter specify the translation initiation and termination codons, respectively, and blue arrows denote where PCR primers anneal to amplify cDNA of each pre-mRNA. (C) Western blots of the specified proteins from cell lysates analyzed in panel (B), but after IP of nuclear fractions using anti-EIF4A3 or rIgG, each in the presence of RNase I. Results represent two independent experiments. (D) Model of EIF4A3 recruitment to Bact complexes via an “EIF4A3 recruitment complex” that consists of CWC27 and CWC22. Created in BioRender. Abshire, E. (2026) https://BioRender.com/33ts2bj. HEK293T cells were next cultured with 1 µM PB for 0, 30, 60, 90, and 120 min, and the resulting cell lysates were split into two portions. Using one portion, RT-qPCR and RT-sqPCR quantitations of DNAJB1 and BRD2 transcripts, whose splicing has been shown to be inhibited by PB in HEK293T cells [46], demonstrated that splicing was inhibited by 30 min of PB treatment (Fig. 2B and Supplementary Fig. S3C). Western blots of anti-EIF4A3 IPs of nuclear fractions generated from the second portion revealed that the co-IPs of EFTUD2, SF3B1, KIN17, DHX16, and AQR with EIF4A3 were reduced by 30 min of PB treatment (Fig. 2C). Thus, PB inhibits splicing before EIF4A3 joins the spliceosome. Moreover, the PB-mediated reduction in the co-IP of EFTUD2 with both SF3B1 and EIF4A3 (Fig. 2C and Supplementary Data Fig. S3A) indicates that EIF4A3 does not interact with EFTUD2 when splicing is inhibited by PB. Surprisingly, PB did not decrease the co-IP of CWC27 and CWC22 with EIF4A3 (Fig. 2C), after which we confirmed anti-CWC27 and anti-CWC22 antibody specificity using CWC27 and CWC22 siRNAs, respectively (Supplementary Data Fig. S3D). These data indicate that EIF4A3 can interact with CWC27 and CWC22 even when PB inhibits splicing prior to when EIF4A3, CWC27, and CWC22 join spliceosomes (Fig. 1A and Supplementary Data Fig. S1A), suggesting that this interaction can occur independently of splicing. This result was recapitulated in anti-CWC22 IPs, which demonstrated that PB decreased the co-IP of EFTUD2, SF3B1, KIN17, AQR, DHX16, and MFAP1 but did not decrease the co-IP of either CWC27 or EIF4A3 (Supplementary Fig. S3E). We propose that CWC27 and CWC22 constitute an EIF4A3 recruitment complex that facilitates the joining of EIF4A3 to the spliceosome during formation of the Bact-I complex (Fig. 2D), consistent with CWC27 and CWC22 facilitating the interaction between the spliceosome and EIF4A3 [20–22]. RBM8A and MAGOH join the pre-B and/or B complex When an X-ray crystal structure of CWC22, EIF4A3, and a CWC27 peptide was used to dock the CWC27 peptide at its CWC22 interaction interface in the cryo-EM structure of the C complex, CWC27 was found to sterically clash with RBM8A in the C complex-associated EJC core [21]. Since CWC27 departs the spliceosome prior to C complex formation (Supplementary Fig. S1A) [17, 18, 32], RBM8A and MAGOH, which form a heterodimer [10], were proposed to join EIF4A3 on the C complex. However, our finding that EIF4A3 joins the spliceosome earlier than anticipated (Fig. 1A and B) encouraged us to reassess when RBM8A and MAGOH join EIF4A3 on the spliceosome. To this end, HEK293T cells were transfected with plasmids producing either HA-EIF4A3, HA-RBM8A, or HA-MAGOH, each of which was expressed at a level no more than two-fold of that of the corresponding endogenous protein (Supplementary Fig. S4) or, as a negative control, HA alone (Fig. 3). Western blots of nuclear lysates before or after anti-HA IP, the latter in the presence of RNase I, revealed the co-IP of SF3B1, WBP4, ZMAT2, CWC27, and CWC22 with HA-EIF4A3, HA-RBM8A, and HA-MAGOH (Fig. 3). The simplest explanation of these data is that RBM8A and MAGOH can join EIF4A3 on the spliceosome as early as the pre-B and/or B complex. The presence of RBM8A and MAGOH, as well as UPF3B, in anti-SF3B1 IPs (Fig. 1D, upper) indicates that the entire EJC core and UPF3B can join the spliceosome before SF3B1 departs during the transition of the Bact complexes to the C complex (Fig. 1D, upper and Supplementary Fig. S1A). Since RBM8A and MAGOH do not directly contact the spliceosome but do directly contact EIF4A3 in structures of the C complex [17, 30], either EIF4A3 joins the spliceosome first to facilitate their recruitment or the EJC core joins the spliceosome as a pre-assembled trimer. Despite structural modeling indicating that CWC27 and RBM8A cannot simultaneously bind EIF4A3 [21], our finding that CWC27 co-immunoprecipitates with both HA-RBM8A and HA-MAGOH suggests that CWC27 can exist with the EJC core in the same spliceosome, possibly in a conformation that differs from that indicated by in silico structural modeling. Splicing factors found in pre-B, B, pre-Bact, and Bact complexes co-immunoprecipitate with HA-EIF4A3, HA-RBM8A, and HA-MAGOH. Western blots of the denoted proteins before (–) and after (+) anti-HA IPs in the presence of RNase I from the nuclear fractions of HEK293T cells transiently transfected with pcDNA3.1 plasmids expressing either HA alone (as a negative control), HA-EIF4A3, HA-RBM8A, or HA-MAGOH. Results represent two experimental replicates. RNPS1 exists in pre-B and later splicing complexes Splicing factors from all stages of splicing, including the earliest spliceosome, which is the E complex, and late spliceosomes, such as the P complex, co-immunoprecipitate with FLAG-tagged RNPS1 expressed in HEK293T cells [1, 36], in agreement with our finding that RNPS1-2XHA co-immunoprecipitates with SF3B1 (Fig. 1D). The association of these splicing factors with RNPS1 likely reflects both EJC-dependent and EJC-independent functions of RNPS1 [5], such as promoting the alternative splicing of specific introns [47]. To determine when during splicing RNPS1 joins spliceosomes, anti-HA and rIgG IPs of RNPS1-2XHA #1 HEK293T-cell lysates were performed in the presence of RNase I in triplicate and analyzed by LC-MS/MS as described for anti-EIF4A3 IPs (Fig. 1A and Supplementary Data). As expected, EJC core constituents, RNPS1-2XHA, ACIN1, PNN, and SAP18 were detected in anti-HA but not rIgG IPs (Supplementary Fig. S5A). Our detecting CASC3, CBP80, and CBP20 in anti-HA IPs, but much less efficiently than EJC core and RNPS1-EJC constituents, may reflect short-lived or loosely associated interactions (Supplementary Fig. S5A) mediated by the spliceosome [30, 33]. As negative controls, PABPC1 and EIF4E were not detected in anti-HA IPs (Supplementary Fig. S5A). We then identified structurally resolved splicing factors in the IP-MS data for RNPS1-2XHA (Fig. 4A and Supplementary Data). In contrast to previous analyses [36], high-confidence splicing factors that co-immunoprecipitated with RNPS1-2XHA did not include splicing factors in complexes before the pre-B complex but did include splicing factors found in the pre-B complex (DDX23) [33] and B complex (PRPF6, PRPF38A, and MFAP1) [17] and later complexes [17, 18, 30, 32] (Fig. 4A and Supplementary Table S2). Whether EIF4A3 in pre-B or later complexes mediates the recruitment of RNPS1 has not been determined. Since in vitro pulldowns of purified proteins have shown that ACIN1 bridges the interaction of RNPS1 with EIF4A3 [47], we expect ACIN1 to mediate RNPS1 joining to pre-B and B complexes if joining is facilitated by EIF4A3. Whether ACIN1 joins the spliceosome at the same time as RNPS1 has not been determined. Our finding that ACIN1 and SAP18 co-immunoprecipitated with SF3B1 (Fig.1D) indicates that ACIN1 and SAP18 can be present on spliceosomes formed prior to the C complex. Our results for SAP18 are consistent with a previous study showing that, in the absence of RNase, the co-IP of pre-spliceosome constituents SNRNP70 [48] and U2AF2 [49, 50] with FLAG-MS2-SAP18 was reduced when FLAG-MS2-SAP18 recruitment to the ASAP complex was inhibited [51]. Evidence that RNPS1, relative to CASC3, joins earlier spliceosomes. (A) As in Fig. 1A but the heat map represents the log2-fold enrichment of the denoted structurally resolved splicing factors in anti-HA IPs relative to before IP using RNPS1-2XHA #1 HEK293T cells. Results represent three experimental replicates. (B) Western blots of the denoted proteins in nuclear fractions of RNPS1-2XHA #1 HEK293T cells before and after IP in the presence of RNase I with anti-HA or rIgG. Results represent 3 biological replicates. (C) As in panel (A) but for anti-CASC3 IPs from total-cell lysates of HEK293T cells. (D) As in panel (B), but for anti-CASC3 IPs from HEK293T-cell nuclear fractions. (E) Schematic of the structurally resolved splicing factors detected only in either the pre-B/B, the pre-Bact, the Bact or the C complexes identified in anti-CASC3 and/or anti-RNPS1-2XHA IP-MS and/or anti-CASC3 and/or anti-RNPS1-2XHA IPs followed by western blotting. Western blots of the anti-HA and rIgG IPs using nuclear fractions confirmed that RNPS1-2XHA associates with splicing factors found in both the pre-B and B complex (PRPF6), pre-Bact complexes (KIN17), and later complexes (Fig. 4B). In control experiments, neither SNRNP70 nor SF1 that, together with U2AF2, promotes E complex assembly by recognizing the 5′ splice site and branchpoint [49, 50], detectably co-immunoprecipitated with RNPS1-2XHA in the presence of RNase I (Fig. 4B). The association of RNPS1-2XHA and DDX23 identified in the IP-MS data (Fig. 4A) could not be recapitulated in western blots (Fig. 4B). We conclude that the earliest detectable association of RNPS1 with the spliceosome is in the pre-B complex or B complex (Fig. 4A and B), which can also contain EIF4A3, RBM8A, and MAGOH (Figs 1 and 3). CASC3 joins late Bact complexes without detectable CWC27 The X-ray crystal structure of an in vitro reconstituted EJC core in complex with a 149-amino acid CASC3 peptide [53] has been used to model this peptide into spliceosome-associated EJCs resolved in cryo-EM structures of the C and later complexes [17, 32, 53]. While the corresponding electron density in the cryo-EM structures does not unambiguously confirm the presence of CASC3 [17, 32], our finding that CASC3 immunoprecipitates with SF3B1 (Fig. 1D) indicates that CASC3 can join spliceosomes formed prior to the C complex. CASC3, like RNPS1, shuttles between the cytoplasm and nucleus, where it can associate with nuclear speckles, i.e. hubs for the splicing of some pre-mRNAs [54–56]. Moreover, IPs of lysates of Flp-In T-Rex 293 cells overexpressing either FLAG-tagged CASC3 or FLAG-tagged RNPS1 reported that splicing factors co-immunoprecipitate with both FLAG-RNPS1 and FLAG-CASC3 [12]. To determine when CASC3 is recruited to spliceosomes, we used anti-CASC3 IP-MS to identify which structurally resolved splicing factors co-immunoprecipitate with HEK293T-cell CASC3 in the presence of RNase I (Fig. 4C, upper, and Supplementary Table S1). As expected, the three EJC core proteins but not CBP80 or CBP20 co-immunoprecipitated with CASC3 (Supplementary Fig. S5B). Our finding that PABPC1 co-immunoprecipitates with CASC3 may reflect the detection of CASC3 in PABPC1 BioID experiments [57], possibly as a consequence of the direct interaction between CASC3 and EIF3 subunits in a mechanism that promotes translation initiation [9]. While ACIN1 and SAP18 were detected in anti-CASC3 IP-MS analyses of total-cell lysates (Supplementary Fig. S5B), neither detectably co-immunoprecipitated with CASC3 in the absence or presence of RNase I when nuclear lysates were analyzed (Fig. 4C, upper). However, we detect a largely RNase I-sensitive co-IP of CASC3 with RNPS1-2XHA (Fig. 4B) and, conversely, co-IP of RNPS1-2XHA with CASC3 using nuclear lysates (Fig. 4D, lower). These data suggest that CASC3 and RNPS1 largely do not exist in the same complexes in the nuclear fraction. The earliest splicing factor detected in anti-CASC3 IP-MS was PRPF38A, which is found in the B, pre-Bact I, and pre-Bact II complexes (Fig. 4A) [17, 31, 58]. However, no other factor found only in these complexes was detected (Fig. 4C), and using western blotting, neither PRPF38A nor KIN17, which constitute pre-Bact complexes [31, 34], was detected in anti-CASC3 IPs of nuclear lysates (Fig. 4D). Our finding that DHX16 and SYF2 [17, 18], which join Bact complexes, co-immunoprecipitated with CASC3 (Fig. 4C and D) indicates that the earliest spliceosomes that detectably associate with CASC3 are the Bact complexes. However, CWC27, which joins the earliest Bact complex [17], fails to co-immunoprecipitate with CASC3. The simplest explanation of these data is that CASC3 begins to associate with Bact complexes that form after the departure of CWC27 (Supplementary Fig. S1A). Discussion Our work clarifies when EJC constituents begin to assemble on spliceosomes, which is much earlier than previously thought. The EJC core constituent EIF4A3 is recruited as early as the pre-B and/or B complexes (Fig. 1), either just before or concomitantly with RBM8A and MAGOH (Fig. 3). The alternative RNPS1-EJC constituent RNPS1 is also recruited to the pre-B and/or B complexes, possibly independently of EIF4A3 (Fig. 4A, B, and E). This is followed by recruitment of the alternative CASC3-EJC constituent CASC3 to the Bact complexes after CWC27 departs (Fig. 4C–E). We hypothesize that EIF4A3 joins pre-B and/or B complex spliceosomes through a direct interaction with EFTUD2 considering that cryo-EM data show that EIF4A3 and EFTUD2 interact directly in the C complex [30] and with the C* complex, the latter of which forms between the first and second catalytic steps of splicing [32]. Our finding that the co-IP of pre-B and B complex constituents with EIF4A3 was not reduced by EFTUD2 siRNAs (Supplementary Fig. S2G) implicates additional constituents in EIF4A3 joining to pre-B and/or B complexes. Notably, EFTUD2 siRNAs may fail to inhibit the co-IP of pre-B and B complex factors with EIF4A3 if these factors are in excess of EJC constituents in HEK293T cells. This excess characterizes HeLa cells, for which there are an estimated 1 × 104 copies of EIF4A3 [59] and between 1 × 105 and 1 × 106 of each spliceosome snRNP [60, 61]. Our finding that EIF4A3 can form a complex with CWC27 and CWC22 when splicing is inhibited using PB suggests that CWC27 and CWC22, possibly with other factors, provide an additional mechanism for EIF4A3 joining to Bact complexes via an “EIF4A3 recruitment complex” (Fig. 2). In support of such a recruitment complex, anti-CWC22 IPs using lysates of HeLa cells detected EIF4A3 but failed to detect splicing factors from pre-B or B complexes [21], in agreement with cryo-EM structures indicating that CWC27 and CWC22 are absent from these complexes [31, 33]. Also, HeLa-cell fractionation studies detected EIF4A3 together with the pre-B and B complex factors PRPF6, PRPF4, and SART1, which we also detect in EIF4A3 IPs (Fig. 1A), but did not detect CWC27 or CWC22 [62]. Therefore, the simplest explanation of our data is that there are both CWC27- and CWC22-dependent and -independent mechanisms by which EIF4A3 can join spliceosomes (Figs 1 and 2). We previously found that insulin promotes the co-IP of AKT with EIF4A3 while decreasing the co-IP of UPF2 with EIF4A3 [11]. Here, we show using western blotting of nuclear lysates that serum deprivation, with or without insulin, did not alter the co-IP of ACIN1, SAP18, or RNPS1-2XHA, which are constituents of the RNPS1-EJC, or CASC3 of the alternative CASC3-EJC (Fig. 1B). Together, our data suggest a model whereby alternative EJCs can form in the nucleus but lack or largely lack, respectively, detectable AKT or UPF2, which are predominantly cytoplasmic (Supplementary Fig. S2B) [39, 40]. These “AKT-less” and “UPF2-less” EJCs do not contain the factors required for the activation of UPF1 until they are exported into the cytoplasm and can acquire either UPF2, in the case of RNPS1-EJCs, or CASC3, in the case of CASC3-EJCs, each of which is a step of EJC assembly regulated by insulin [5, 11, 40, 63]. Knowing when EJC constituents assemble on spliceosomes is key to defining how EJC assembly, alternative EJC composition, and EJC deposition at exon–exon junctions may be regulated in the nucleus. These findings supplement known EJC dynamics in the cytoplasm [11]. Three independent studies estimate that in human cells only 80% of exon–exon junctions are occupied by an EJC [3, 4, 64]. In one study, EJC footprints on RNA were sequenced after anti-FLAG IPs of lysates of cycloheximide-treated TRex-HEK293 cells stably expressing either FLAG-EIF4A3 or FLAG-MAGOH [3]. In the second study, EJC footprints on RNA were sequenced after UV-crosslinking HeLa cells followed by anti-EIF4A3 IPs of cell lysates [4]. A third study used UV-crosslinking and anti-HA IPs of lysates of HeLa cells stably expressing EIF4A3-3XHA [64]. The estimated 20% of exon–exon junctions that appear to lack an EJC may be conditionally regulated by the cell or due to deposition inefficiencies attributable to e.g. local RNA secondary structure [3, 65]. Understanding when EJC constituents assemble on the spliceosome provides new insight into when during splicing regulation of EJC deposition can occur and which splicing factors may be involved. Supplementary Material Notes Present address: Department of Molecular and Cellular Biochemistry, University of Kentucky, Lexington, KY 40508,United States Contributor Information Elizabeth T Abshire, Department of Biochemistry and Biophysics, School of Medicine and Dentistry, University of Rochester, Rochester, NY 14642, United States; Center for RNA Biology, University of Rochester, Rochester, NY 14642, United States. Lynne E Maquat, Department of Biochemistry and Biophysics, School of Medicine and Dentistry, University of Rochester, Rochester, NY 14642, United States; Center for RNA Biology, University of Rochester, Rochester, NY 14642, United States. Supplementary data Supplementary data are available at NAR online. Conflict of interest None declared. Funding This work was supported by the National Institutes of Health (NIH) National Institute of General Medical Sciences (NIGMS) R01 GM059614 to L.E.M., which is now NIH R35 GM149268, and NIH K99 GM152790 to E.T.A. The University of Rochester (UR) Mass Spectrometry Resource Laboratory is supported by an NIH Instrumentation Grant (S10OD025242). The URMC Center for Advanced Microscopy & Nanoscopy is supported by P50 HD103536 provided by the Department of Health and Human Services (DHHS)/Public Health Service (PHS)/NIH. The URMC Flow Cytometry Shared Resource is supported in part by the University of Rochester Wilmot Cancer Institute Support Grant P30CA272302. Funds for open access publication of this work were provided by NIH R35 GM149268. Data availability All data reported in this paper are included in the Supplementary Tables. Requests for further information should be directed to the lead contact. All IP-MS data are available on the MassIVE database; accession number: MSV000101000. References Associated Data Supplementary Materials Data Availability Statement All data reported in this paper are included in the Supplementary Tables. Requests for further information should be directed to the lead contact. All IP-MS data are available on the MassIVE database; accession number: MSV000101000.