To isolate a spliceosome assembly intermediate formed directly before the A complex that still contains Prp5 (Extended Data Fig. 1a), we carried out splicing in S. cerevisiae cell extracts with an actin (Act) pre-mRNA in which the BS-A is deleted (Extended Data Fig. 1b). With this ΔBS-A substrate, splicing is blocked before catalytic step 1 (Extended Data Fig. 1c), consistent with previous results6. Spliceosomal complexes formed on ΔBS-A Act pre-mRNA lack the U4/U6.U5 tri-snRNP, but contain stoichiometric amounts of U1 and U2 snRNPs (Extended Data Fig. 1d). The proteins Prp5, Msl5 and Mud2 are also abundant, whereas Cus2 is absent (Extended Data Fig. 1e and Supplementary Table 1), indicating that these complexes stall after Prp5 hydrolyses ATP, but before the tri-snRNP has docked. We next carried out single-particle cryo-electron microscopy (cryo-EM) and determined the structure of the ΔBS-A complex at an average resolution of 5.9 Å (ranging from roughly 4.5 Å for U1 to approximately 15 Å for U2) (Extended Data Table 1 and Extended Data Fig. 2). Further classification and multibody refinement improved the resolution of the stable U1 snRNP region and adjacent U2 5′ region to 4.1 Å and 8.3 Å, respectively (Extended Data Fig. 2). By fitting known X-ray structures of spliceosome components into the EM density map (Extended Data Table 2), together with protein crosslinking coupled with mass spectrometry (CXMS) (Supplementary Table 2), we generated a three-dimensional (3D) model of the ΔBS-A complex (Fig. 1). This complex consists of two major elongated domains—comprising the U1 and bipartite U2 snRNPs—that are connected by two main bridges (Fig. 1).

Fig. 1: Three dimensional cryo-EM model of the yeast pre-A complex.
figure 1

a, b, EM density map (a) and molecular architecture (b) of the S. cerevisiae spliceosomal pre-A complex. a, Purple, better-resolved U1 density; grey blue and green, better-resolved densities of the 3′- and 5′-regions of U2 snRNP; translucent grey, cryo-EM map of the pre-A complex.

The U2–BS helix is not clamped by Hsh155

Hsh155HEAT adopts a closed conformation after U2 has integrated stably into the spliceosome, clamping the U2–BS helix and binding the bulged BS-A in a pocket formed by Rds3 (PHF5A in humans; see Extended Data Fig. 1f for a summary of yeast and human protein names) and HEAT repeats 15–17 of Hsh155 (refs.7,8,9,10) (Extended Data Fig. 1g). It is unclear at present what triggers this functionally important structural change. In the ΔBS-A complex, Hsh155HEAT exhibits an open conformation, as in the isolated human 17S U2 (ref. 11) but in notable contrast to its conformation in yeast A complexes (Figs. 1, 2a and Extended Data Fig. 3a) and pre-B, B and Bact complexes7,8,12,13. Stem–loop (SL) IIa of U2 snRNA is bound by Hsh155, Cus1 and Prp9 in the ΔBS-A complex, in a similar manner to that seen in the human 17S U2 snRNP and subsequently formed spliceosomal complexes (Extended Data Fig. 3b). An extended helical density element is located directly upstream of SLIIa. This element is longer than the U2 branchpoint-interacting stem-loop (BSL), which is found in the isolated U2 snRNP and sequesters U2 nucleotides that base pair with the branch site11,14 (Extended Data Fig. 3c, d). A modelled, extended U2–BS helix—lacking a bulged BS-A and comprising 13 base pairs—fits well into this density element (Extended Data Fig. 3c), indicating that an extended U2–BS helix has formed. On the basis of CXMS data, the Prp11 zinc finger (Prp11ZnF) could be positioned at the top of the U2–BS helix (Extended Data Fig. 3e, f), akin to its position in A, pre-B, B and Bact complexes7,8,12,13,15. In the ΔBS-A complex, the U2–BS helix is located further away from the carboxy (C)-terminal HEAT repeats of Hsh155HEAT compared with its position in the A to Bact complexes, and it is not sequestered by Hsh155HEAT (Fig. 2a and Extended Data Fig. 3g). Thus, formation of the U2–BS helix alone does not appear to trigger closure of Hsh155HEAT. As the U2–BS helix has formed, but Hsh155HEAT still exhibits an open conformation and Prp5 is stably bound (see below), we conclude that ΔBS-A complexes are stalled at a pre-A-complex stage, after Prp5-mediated formation of the U2–BS helix, but during/before it carries out its proofreading function.

Fig. 2: The Hsh155 HEAT domain has an open conformation in the pre-A complex.
figure 2

a, Conformation of the SF3B1 and Hsh155 HEAT domains and position of the U2–BS helix and U2 snRNA SLI and SLIIa in human 17S U2 snRNP (Protein DataBank (PDB) ( accession number 6Y5Q) and in the S. cerevisiae pre-A and A complexes (PDB 6G90). These domains were aligned via Hsh155 heat repeats 19–20, Rse1BPA and U2 SLIIa. Olive green, SLIIa nucleotides; reddish orange, pre-mRNA branch-site nucleotides; purple, BSL nucleotides that later form the U2–BS helix; yellow, BSL nucleotides forming the extended part of the U2–BS helix; dark green, remaining BSL nucleotides; blue, SLI. b, Fit of Prp5RecA1 into the pre-A EM density. c, Location of the Prp5 RecA1 and RecA2 domains in the pre-A complex. d, Prp5RecA1 contacts U2 snRNA nucleotides that connect the U2–BS helix to U2 SLIIa. The positions of Prp5 amino acids (located outside of the SAT motif) that when mutated suppress branch-site mutations are indicated in black.

Location of Prp5 in the pre-A complex

Prp5 was initially proposed to ‘proofread’ formation of the U2–BS helix by coordinating the rate of U2–BS base pairing with its ATPase activity5. However, more recent studies uncovered a correlation between increased Prp5 retention in early spliceosomal complexes and decreased tri-snRNP recruitment, suggesting that the physical presence of Prp5, rather than its ATPase activity, has a key role in its proofreading function16. In the human 17S U2 snRNP, the Prp5 RecA (Prp5RecA) domains sequester the BSL together with the TAT–SF1 protein and the C-terminal HEAT repeats of SF3B1 (ref. 11), thereby preventing formation of the U2–BS helix. We find here that, in the yeast pre-A complex, Cus2 (human TAT–SF1) has dissociated and the Prp5RecA domains are located between the U2 3′-region and the U2–BS helix (Fig. 2b, c and Extended Data Fig. 4a–d). Compared with their position in 17S U2 snRNP, the Prp5RecA domains in the pre-A complex are located further away from Hsh155HEAT, with RecA1 fitting well into the niche formed by the amino (N)-terminal region of Prp9 (Prp9NTR), with which it interacts, and the U2–BS helix (Fig. 2b, c and Extended Data Fig. 4c–e). Moreover, Prp5RecA1 now contacts the U2 snRNA strand that connects SLIIa and the U2–BS helix (Fig. 2d). This is consistent with the crosslinks of Prp5 to this region (nucleotides 45–49) of yeast U2 snRNA observed previously with a Prp5-associated S. cerevisiae spliceosomal complex (designated the Prp5-associated intermediate complex, or FIC) formed on a pre-mRNA with a mutated branch site16. Finally, CXMS indicates that the NTR of Prp5 interacts extensively with Hsh155 HEAT repeats 1–7 in the pre-A complex (Extended Data Fig. 4a, b), consistent with previous biochemical studies17.

Pre-A formation involves U2 remodelling

Comparison of the structures of the yeast pre-A complex and human 17S U2 snRNP suggests that, in addition to Prp5RecA, there are other changes in the organization of U2 components during pre-A formation. Relative to its position in 17S U2, the U2 3′-region has moved (Extended Data Fig. 4d, e) and its new location in the pre-A complex is stabilized by newly formed contacts between the U2 3′- and 5′-regions (Extended Data Fig. 4d, e). This movement is a prerequisite for formation of the pre-A complex, as it generates the binding pocket for the repositioned Prp5RecA1 (Fig. 2b, c and Extended Data Fig. 4). In the pre-A complex, Prp5RecA1 establishes new contacts with the shifted 3′-region by interacting with Prp9NTR (Fig. 2b, c and Extended Data Fig. 4), preventing the further movement of the U2 3′-region that is ultimately required to allow the tri-snRNP to dock to the A complex (see below).

Prp40 bridges U1 and U2 snRNP in pre-A complex

The U1 snRNP structure in the pre-A complex (Extended Data Fig. 5a) is highly similar to that observed in the yeast E, A and pre-B complexes13,15,18, indicating that U1 does not undergo substantial remodelling during early spliceosome assembly. As in the aforementioned complexes, base pairing between the 5′-splice site and the U1 snRNA is also stabilized by Yhc1 and Luc7 in the pre-A complex (Extended Data Fig. 5b). Although only three FF domains of Prp40 (domains 4–6) could be located in the yeast E complex18, CXMS allowed us to map all six of the FF domains of Prp40 in the pre-A complex (Fig. 1 and Extended Data Fig. 5c, d). FF1 and FF6 bind Luc7 and Snp1, respectively, tethering Prp40 to U1, while FF2–FF5 form an extended binding platform that interacts with numerous proteins, including Snu71 and U2 Rse1. The interaction of FF4 with the WD40 β-propeller domain B (BPB) of Rse1 forms a bridge between the U1 and U2 snRNPs (denoted bridge 1) (Extended Data Figs. 5d, 6a, b) that is not observed in yeast A complexes15. Bridge 1 also contains the C-terminal region of Snu71, as numerous crosslinks between it and the FF2 and FF3 domains of Prp40, as well as with Rse1BPB, are detected (Extended Data Fig. 5d). In the pre-A complex, U1 and U2 are connected by a second bridge comprising intron nucleotides upstream of the branch site (Extended Data Fig. 6a, c). Although the Prp40 WW domain and Msl5–Mud2 could not be localized based solely on the EM density, CXMS indicates that Msl5–Mud2 is likely to be located near the U2–BS in the pre-A complex, and furthermore remains bound to the Prp40 WW domain (Extended Data Fig. 6d and Supplementary Table 2).

Pre-A complex with a U257A branch-site mutation

Mutations in the conserved yeast branch-site sequence upstream of the BS-A that weaken the U2–BS interaction—including a U-to-A mutation at position 257 of Act pre-mRNA (two nucleotides upstream of the BS-A) (Extended Data Fig. 1b)—do not completely block splicing but do lead to the accumulation of spliceosomes in which Prp5 is retained but the tri-snRNP has not yet joined5,16. To determine whether this mutation stalls spliceosome assembly at the pre-A stage, we purified the complexes that form on Act U257A pre-mRNA and determined their cryo-EM structure (Extended Data Fig. 7). The RNA and protein compositions of the purified U257A and ΔBS-A complexes were identical (Extended Data Fig. 1d, e and Supplementary Table 1), and an overlay of the U257A and ΔBS-A complexes revealed a highly similar, if not identical, structure at the present level of resolution (Extended Data Fig. 7). Indeed, the structural model of the ΔBS-A pre-A complex fits well without further adjustment into the EM density of the U257A complex (Extended Data Fig. 7). Thus, with the U257A mutant, an extended U2–BS helix has also formed, Hsh155HEAT is in an open conformation, and Prp5RecA1 is docked to Prp9NTR and situated close to the U2 snRNA, indicating that U257A complexes are also stalled at the same pre-A stage. The highly similar structure of both pre-A complexes indicates that they represent a physiologically relevant intermediate that—at least in the case of U257A—can also progress along the wild-type spliceosome assembly pathway.

Dynamics of the pre-A to A transition

Comparison of our pre-A complex with the previously published yeast A complex15 reveals that the transition from the pre-A to the A complex involves large-scale remodelling that requires displacement of Prp5 (Fig. 3, Extended Data Fig. 8 and Supplementary Video 1). First, the U2 3′-region rotates by roughly 55° relative to the U2 5′-region during A-complex formation. In the pre-A complex, this rotational movement is prohibited by the Prp5RecA domains, which bind in a mutually exclusive manner with the new position of Prp9NTR, the long α-helix of Prp21 and the Prp11 β-sandwich, in the subsequently formed A complex. Second, U1 snRNP rotates by roughly 45° during the transition from the pre-A to the A complex, such that Prp39 now interacts with Lea1. A prerequisite for U1 movement is the dissociation of Prp40 from Rse1, and thus the apparent dissolution of U1–U2 bridge 1 (Fig. 3). The repositioning of U1 and the U2 3′-region is essential to generate the binding platform needed to dock the U4/U6.U5 tri-snRNP during formation of the pre-B complex (Fig. 3). Our studies thus provide a structural explanation for why the docking of tri-snRNP is inhibited when Prp5 is retained in yeast prespliceosomes16.

Fig. 3: Prp5 blocks the repositioning of U1 and U2 snRNPs that is required to form the tri-snRNP-binding site.
figure 3

Molecular organization of U1 and U2 snRNPs in S. cerevisiae pre-A, A (PDB 6G90) and pre-B complexes (PDB 5ZWM and PDB 5ZWN). Movements of U1 and U2 snRNPs during the pre-A to A transition are indicated by curved arrows. All structures were aligned as in Fig. 2. For simplicity, the U1 snRNA stem-loops in the poorly resolved region of the U1 snRNP are not shown in the pre-A, A and pre-B complexes.

Mechanism of proofreading by Prp5

The cryo-EM structures presented here provide structural insights into the mechanism by which Prp5 proofreads the U2–BS helix (Fig. 4). The pre-A and 17S U2 structures are consistent with a model in which, after U2 interacts with the E complex, ATP hydrolysis by Prp5 leads to release of Cus2 and unwinding of the BSL. This allows formation of the U2–BS helix and repositioning of the U2 3′-region and Prp5RecA, generating the pre-A complex (Fig. 4). The new proofreading (or rather, ‘fidelity checkpoint’) position of Prp5RecA in the pre-A complex transiently prevents the further movement of the U2 3′-domain needed to form an A complex that can subsequently bind the tri-snRNP. As deletion of the BS-A hinders the closure of Hsh155HEAT, but does not affect the stability of the U2–BS helix per se, correct binding of the bulged BS-A by Hsh155HEAT and Rds3 is likely to be a major trigger for the conformational change in Hsh155HEAT. Furthermore, in the pre-A complex, the U2–BS helix is probably flexible, enabling it to intermittently move closer to Hsh155HEAT, which ‘probes’ for its presence. Thus, we propose that when a stable U2–BS with a bulged BS-A is formed, movement of the U2–BS into the open Hsh155HEAT leads to insertion of the BS-A into its binding pocket and closure of the HEAT domain (Fig. 4). Previous mutational analyses of Hsh155 indicated that alignment of the U2–BS duplex with conserved, positively charged amino acids in the C-terminal half of Hsh155 is crucial for closure19, and this alignment could thus help to properly position the bulged BS-A in its binding pocket.

Fig. 4: Model for Prp5-mediated proofreading of the U2–BS helix.
figure 4

For simplicity, in this depiction of our proposed mechanism of proofreading by Prp5, Rds3 of the BS-A-binding pocket is not shown, and the U1 snRNP has been omitted from the spliceosomal pre-A and A complexes. The closure of Hsh155HEAT and release of Prp5 may be highly coordinated events that occur simultaneously rather than sequentially. The dashed arrow indicates that some mutations in the branch-site sequence, including U257A, do not completely abolish conversion of the pre-A complex into the A complex.

Closure of Hsh155HEAT would destabilize not only the Prp5NTR that binds to it, but also the Prp5RecA domains (Fig. 4). Although the latter do not interact with Hsh155HEAT, the coordinated movement of Hsh155HEAT and Prp5NTR and the U2 snRNA nucleotides contacted by Prp5RecA1 could also lead to the displacement of Prp5RecA and subsequent release of the entire Prp5 protein. Release of Prp5 would then allow rotation of the U2 3′-region and formation of an A complex (Fig. 4). If the U2–BS helix lacks a bulged BS-A, binding of the latter by Hsh155HEAT and Rds3 would be blocked (Fig. 4). This would prevent the closure of Hsh155HEAT and release of Prp5, blocking the progression of spliceosome assembly and potentially targeting the stalled complex for discard (Fig. 4). As branch-site mutations that destabilize the U2–BS duplex also hinder spliceosome assembly and lead to Prp5 retention5,16, the stability of the U2–BS duplex per se may also affect the conformational state of Hsh155HEAT in pre-A complexes. A destabilized U2–BS helix could also potentially prevent proper bulging of the BS-A20, and in this way hinder closure of Hsh155HEAT.

Various mutations in Prp5 can suppress branch-site mutations, including those in the Prp5 DPLD motif (which is important for the interaction of Prp5 with the U2 snRNP21) and also mutations in its SAT motif and those in adjacent regions of RecA1 that have no effect on ATPase activity, such as K372E, N399D and G401E5,21. The Prp5N399D mutant, the SAT mutant TAG and the DPLD mutant AAAA have reduced affinity for yeast prespliceosomes and enhance the binding of tri-snRNPs to spliceosomes16. Our pre-A structures provide insight into how some of these mutations may destabilize Prp5 binding. In pre-A complexes, the Prp5NTR containing the DPLD motif still interacts with Hsh155HEAT (Extended Data Fig. 4b), and thus mutation of this motif is likely to destabilize the Prp5NTR–Hsh155HEAT interaction. In pre-A complexes, amino acids K372, N399 and G401 are located in the Prp5 region that interacts with the single-stranded U2 snRNA between SLIIa and the U2–BS helix (Fig. 2d). Consistent with less-stable Prp5 binding, these mutations introduce a negative charge that would destabilize the RecA1–U2 snRNA interaction. Although the SAT motif is not located at a Prp5–protein interface, some SAT mutations might alter the conformation of RecA1, thereby indirectly destabilizing its interaction with Prp9NTR. Indeed, several SAT mutations alter the equilibrium between the open and closed conformations of the Prp5RecA domains22. Together, our results indicate that Prp5 does not proofread the U2–BS helix directly, but instead proofreads the RNP conformation of pre-A complexes, and hinders progression of spliceosome assembly if mutations in the branch site alter the formation of a productive, closed conformation of Hsh155HEAT.


No statistical methods were used to predetermine sample size. The experiments were not randomized, and investigators were not blinded to allocation during experiments and outcome assessment.

Preparation of yeast whole-cell extracts

Yeast whole-cell extracts were prepared from the Saccharomyces cerevisiae 3.2.AID/CRL2101 strain (MATalpha, prp2-1, ade2, his3, lys2-801, ura3) (a gift from R.-J. Lin)23. Yeast were grown in a 100-litre fermenter and extracts were prepared as previously described7.

Affinity purification of pre-A complexes

Uncapped actin pre-mRNA lacking the branch-site adenosine (ΔBS-A Act pre-mRNA) was tagged at its 5′-end with three MS2 aptamers and transcribed in vitro using T7 RNA polymerase from a template prepared with the QuikChange II site-directed mutagenesis kit (Agilent). Yeast ΔBS-A pre-A spliceosomal complexes were assembled for 45 min at 23 °C in a 175 ml splicing reaction containing 40% yeast whole-cell extract and 1.8 nM ΔBS-A Act pre-mRNA with prebound MBP–MS2 fusion protein. The splicing reaction was subsequently chilled on ice and cleared by centrifugation for 10 min at 9,000 rpm at 4 °C in a Fibrelite F14-14 × 50 cy rotor (Thermo Fisher Scientific). It was then loaded onto two columns, each packed with 600 μl amylose resin (New England BioLabs) that were pre-equilibrated with GK75 buffer (20 mM HEPES-KOH pH 7.9, 1.5 mM MgCl2, 75 mM KCl, 5% glycerol, 0.01% NP40, 0.5 mM dithiothreitol (DTT) and 0.5 mM phenylmethylsulfonyl fluoride (PMSF)). The matrix with bound complexes was washed with 3 ml GK75 buffer and spliceosomes were eluted with 15 mM maltose in GK75 buffer. For electron microscopy, peak elution fractions containing approximately 40 pmol of spliceosomal complexes were pooled (1 ml total volume), crosslinked with 0.2 mM BS3 (Thermo Fisher) for 1 h on ice, and loaded onto a 17 ml linear 10–30% (v/v) glycerol/0–0.1% glutaraldehyde gradient containing GK75 buffer. Samples were centrifuged for 17 h at 24,400 r.p.m. in a SureSpin 630 rotor (Sorvall) and collected manually from the top in 28 fractions of 555 μl each. Crosslinking was stopped by adding 50 mM glycine, pH 7.7, and incubating for 2 h on ice. Fractions were analysed by Cherenkov counting in a Tri-Carb 2100TR scintillation counter (Packard). Two peak fractions containing 2.4 pmol of spliceosomal complexes were buffer-exchanged to GK75 with no glycerol, and concentrated to 250 μl in an Amicon Ultra-0.5 centrifugal filter unit Ultracel-50 (Merck), and then used for preparation of cryo-EM grids. Pre-A complexes assembled on the U257A Act pre-mRNA were purified as described above, with the following modifications. The template for in vitro transcription of the U257A Act pre-mRNA was purchased from Genscript. Pre-A complexes were assembled in a 252 ml splicing reaction containing 1.7 nM pre-mRNA; before preparation of cryo-EM grids, three peak gradient fractions were buffer exchanged to GK75 buffer containing 0.3% glycerol and samples were concentrated to 100 μl.

RNA and protein composition of pre-A complexes

To determine the RNA and protein composition of the pre-A complexes, we purified the complexes essentially as described above, except that we used a 12-ml splicing reaction, we washed spliceosomes bound to an amylose matrix with 10 ml GK150 and 10 ml GK75 buffer, and we did not incubate the eluted complexes with BS3. Furthermore, the complexes were fractionated on a 10–30% glycerol gradient lacking glutaraldehyde by centrifugation in a TH660 rotor (Thermo Fisher Scientific) for 16 h at 21,500 r.p.m. RNA and proteins were separated on NuPAGE 4–12% Bis-Tris gels (Invitrogen) and visualized by staining with SYBR Gold (Invitrogen) and Coomassie, respectively. The entire lanes were cut into 23 slices (680 fmol ΔBS-A pre-A) or 16 slices (170 fmol U257A pre-A) and proteins were in-gel digested with trypsin overnight. Resulting peptides were separated on a C18 column using an UltiMate3000 (Dionex) ultrahigh performance liquid chromatography system, and analysed by electrospray ionization mass spectrometry in a Thermo Scientific Q Exactive HF (ΔBS-A pre-A) or Orbitrap Exploris 480 mass spectrometer (U257A pre-A). Data were acquired using Thermo Exactive Series 2.8 SP1 and Orbitrap Exploris 480 3.0 software. The U257A pre-A complex was measured in duplicate, and a sum of both measurements is shown in Supplementary Table 1. Proteins were identified by searching fragment spectra against the S. cerevisiae Genomic Database (SGD; using Mascot v.2.3.02 as a search engine. For immunoblotting, proteins were separated on denaturing 4–12% NuPAGE gels, transferred to Amersham Protran 0.2-μm nitrocellulose membranes (Cytiva), immunostained with an Amersham ECL Western Blotting Detection Kit (Cytiva), and visualized with an Amersham Imager 680 (Cytiva). Antibodies against the yeast Prp5 and Lea1 proteins were provided by S.-C. Cheng.

Protein–protein crosslinking and identification

For CXMS experiments, spliceosomes were assembled in a 400-ml (experiment 1) or 300-ml (experiment 2) splicing reaction containing 40% yeast whole-cell extract from the 3.2.AID/CRL2101 strain. Following MS2 affinity selection, purified spliceosomal complexes were crosslinked with 250 μM BS3 for 1 h at 8 °C in a total volume of 3 ml. The reaction was split in half and loaded onto two 17-ml 10–30% (v/v) glycerol gradients and centrifuged in a Surespin 630 rotor (Thermo Fisher Scientific) for 16 h at 24,400 r.p.m. The gradients were fractionated by hand from the top into 28 fractions. Three peak fractions from each gradient, containing approximately 15 pmol of pre-A complexes, were pooled and the crosslinked complexes were pelleted by ultracentrifugation in a S100-AT4 rotor (Thermo Fisher Scientific) and analysed as previously described24. After tryptic digestion, peptides were reverse-phase extracted using Sep-Pak Vac tC18 1cc cartridges (Waters) and fractionated by gel filtration on a Superdex Peptide PC3.2/30 column (GE Healthcare). Next, 50 μl fractions corresponding to an elution volume of 1.2–1.8 ml were analysed in triplicate on a Thermo Scientific Q Exactive HF-X (Experiment 1) or Orbitrap Exploris 480 mass spectrometer (Experiment 2) using Thermo Exactive Series 2.9 and Orbitrap Exploris 480 1.1 software, respectively. Protein–protein crosslinks were identified using the pLink 2.3.9 search engine ( and filtered at a false discovery rate (FDR) of 1% or 5% according to the developer’s recommendations25,26.

EM sample preparation and image acquisition

Purified ΔBS-A or U257A pre-A complexes were absorbed for 15 min or 25 min, respectively, to a thin-layer carbon film that was subsequently attached to R3.5/1 Quantifoil grids. Next, 3.8 μl of sample buffer was applied to the grid and excess buffer was removed using an FEI Vitrobot loaded with pre-wetted filter paper, with a blotting force of 7, blotting time of 6.5 s, at 100% humidity and 4 °C. The sample was subsequently vitrified by plunging into liquid ethane. All cryo-EM data of the ΔBS-A pre-A complex and dataset 1 of the U257A pre-A complex were acquired at 300 kV on a FEI Titan Krios electron microscope (Thermo Fisher Scientific), equipped with a Cs corrector. Images were recorded in integration mode at ×120,700 magnification, corresponding to a calibrated pixel size of 1.16 Å at the specimen level, using a Falcon III direct electron detector. Micrographs were recorded via a Thermo Fischer EPU 2.1, using an exposure time of 1.02 s with 40 movie frames and a total dose of 44 e per Å2. In total, 87,604 and 9,170 micrographs were recorded for the ΔBS-A pre-A complex and dataset 1 of the U257A pre-A complex, respectively. Dataset 2 of the U257A pre-A complex (18,332 micrographs) was acquired at 300 kV on an FEI Titan Krios electron microscope (Thermo Fisher Scientific), in integration mode at a calibrated pixel size of 1.06 Å at the specimen level, using a Falcon III direct electron detector. Micrographs were recorded via a Thermo Fischer EPU 2.1, using an exposure time of 1.02 s with 40 movie frames and a total dose of 58 e per Å2.

Image processing

Frames were dose-weighted, aligned and summed with MotionCor 2.0 (ref. 27). The defocus values and equiphase averaging (EPA) of the micrographs were determined using Gctf28. Micrographs with a defocus range of 1 μm to 4 μm and a resolution of better than 4.3 Å based on EPA estimation were retained for further processing. For the ΔBS-A pre-A complex, 74,230 out of 87,604 summed micrographs were further processed. Initially, approximately 5.4 million particles were automatically picked using Gautomatch ( They were then extracted with a box size of 440 × 440 pixels, and binned to 110 × 110 pixels (pixel size of 4.64 Å) in RELION 3.0 ( Several iterations of reference-free two-dimensional (2D) classification were performed in RELION 3.0, and ‘bad classes’ showing fuzzy or uninterpretable features were removed, yielding 3,143,491 ‘good particles’. A subset of 308,419 particles was used to generate an initial 3D map using the ab initio reconstruction function in cryoSPARC29. Using the ab initio model, this subset of particles was further 3D classified into five classes in RELION 3.0. The best class showing clear features of U1 and U2 snRNP was selected, and the more flexible U2 snRNP region of this map was erased using UCSF Chimera v.1.13.1 (ref. 30). The remaining U1 snRNP was low-pass filtered to 35 Å and used as 3D reference for further 3D classification for the entire dataset. The retained 3,143,491 particles after 2D classification of the entire dataset were split into 3 subsets and each subset was 3D classified into 5 classes. The best classes of each subset were combined, yielding 986,393 particles, which were then centred and re-extracted to 200 × 200 pixels (pixel size of 2.32 Å) and further classified into four classes. After 20 iterations of consensus classification (7.5° sampling interval without local search), a mask was placed around the U1 snRNP and the local angular search range was limited to 20° with a finer sampling interval of 3.7°. The best class (with 504,547 particles) was selected, centred and re-extracted with an original pixel size of 1.16 Å with a box size of 400 × 400 pixels and refined with a mask around the U1 snRNP, resulting in a map at 4.3 Å resolution. Next, using the alignment parameters from the aforementioned masked 3D refinement, the 504,547 particles were focus classified with a mask around the high-resolution U1 core, into 4 classes. The best class (containing 226,656 particles) was selected and refined into a map of the entire pre-A complex with an average resolution of 5.9 Å (ranging from roughly 4.5 Å at the U1 region to roughly 15 Å at the U2 region). The U1 and U2 regions were further improved by multibody refinement to 4.1 Å and 10 Å respectively. To further improve the U2 region, we re-extracted the 504,547 particles into a smaller box size of 140 × 140 pixels (pixel size of 2.32 Å) with the U2 snRNP centred, and 3D classified into 5 classes with a mask around the U2 snRNP. The best class, with 160,894 particles, was selected and multibody refined with masks around the U2 5′- and 3′-regions, resulting in a map at 8.3 Å resolution for the 5′-region and one at 9.5 Å resolution for the 3′-region. All the aforementioned resolutions were estimated on the basis of the RELION gold-standard Fourier shell correlation (0.143 criterion).

For the U257A pre-A complex, initially 460,854 and 869,306 particles were extracted from dataset 1 and dataset 2, respectively, and rescaled to 110 × 110 pixels, to the same pixel size of 4.64 Å, in RELION 3.0 ( After several iterations of reference-free 2D classification, 697,892 ‘good particles’ from the two datasets were combined and classified into four classes by 3D classification with only the U1 part as the starting model, to avoid model bias. No class resembling the structure of the mature A complex was observed. Three classes had no discernible structural features of U1 or U2 snRNPs. One class (of 240,145 particles) clearly exhibited the structure of a pre-A complex, and was selected, centred and re-extracted with a pixel size of 2.32 Å with a box size of 220 × 220 pixels. Re-extracted particles were further 3D classified into four classes with a mask around the U1 part, yielding one class that showed clear secondary structures. This class (80,853 particles) was selected and refined into a map of the entire pre-A complex with an average resolution of 10.4 Å. Multibody refinement improved the U1 part to 7.5 Å. To improve the U2 part, we further classified the 80,853 particles with a mask around the U2 part into four classes, and two classes showing clear U2 density were combined, 3D refined and multibody refined, yielding the U2 part of the structure with 13 Å resolution. All of the aforementioned resolutions were estimated on the basis of the RELION gold-standard Fourier shell correlation (0.143 criterion).

ΔBS-A pre-A model building and refinement

Templates for the U1 and U2 proteins and RNA were obtained wherever possible from published structures (Extended Data Table 2). The U1 snRNP components, except Prp40, were initially docked as rigid bodies into the 4.1 Å EM map of the U1 region. In the central part of the U1 snRNP (resolution ranging from 3.7 Å to 4.3 Å), side chains were manually adjusted into the map using Coot v. (ref. 31). The entire model of the U1 snRNP, excluding Prp40, was combined and subjected to real-space refinement in PHENIX v.1.13-2998 (ref. 32), with secondary-structure restraints. The solution structure of the Prp40 FF1 domain and the homology model of domains FF2–FF6 predicted by the SWISS-MODEL suite33 were truncated to polyalanine, docked into the pre-A map as rigid bodies, and were not refined owing to the limited resolution. The model of Hsh155 (H1–H15) was based on human SF3B1 (H1–H15) but with the sequence changed to that of yeast Hsh155. The model of the ΔBS-A U2/BS helix (U2 nucleotides 32–46; Act1 pre-mRNA nucleotides 254–268) was generated by deleting the BS-A from the model of the wild-type U2–BS helix using Coot. All U2 snRNP components were docked into the U2 map as rigid bodies without further adjustments, except that Prp9 (amino acids 328–362) and Prp21 (amino acids 173–192) were slightly adjusted using Coot to better fit the EM density, and the linker between the U2–BS helix and SLIIa (U2 nucleotides 47–49) was de novo modelled using Coot. All modelled components in the U2 region were modelled as polyalanine and were not refined owing to the limited resolution. The structural model for the ΔBS-A pre-A complex was fit into the EM density obtained for complexes formed on the U257A mutant Act1 pre-mRNA. The video showing the structural dynamics seen during the transition from the pre-A to the A complex was generated using ChimeraX v1.1.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this paper.