Mechanism for the initiation of spliceosome disassembly

Abstract

Pre-mRNA splicing requires the assembly, remodeling, and disassembly of the multi-megadalton ribonucleoprotein complex called the spliceosome¹. Recent studies have shed light on spliceosome assembly and remodeling for catalysis^2–6, but the mechanism of disassembly remains unclear. Here, we report 2.6 to 3.2 Å resolution cryo-electron microscopy structures of nematode and human terminal intron-lariat spliceosomes along with biochemical and genetic data. Our results uncover how four disassembly factors and the conserved RNA helicase DHX15 initiate spliceosome disassembly. The disassembly factors probe large inner and outer spliceosome surfaces to detect the release of ligated mRNA. Two of these factors, TFIP11 and C19L1, and three general spliceosome subunits, SYF1, SYF2 and SDE2, then dock and activate DHX15 on the catalytic U6 snRNA to initiate disassembly. U6 thus controls both the start⁵ and end of pre-mRNA splicing. Taken together, our results explain the molecular basis of canonical spliceosome disassembly and provide a framework to understand general spliceosomal RNA helicase control and the discard of aberrant spliceosomes.

The spliceosome is a dynamic and multi-megadalton ribonucleoprotein complex that excises introns from messenger RNA precursors (pre-mRNAs) to generate mRNA¹. This macromolecular machine forms anew on each intron via the orchestrated assembly of five small nuclear ribonucleoprotein particles (snRNPs; U1, U2, U4, U5, U6), the nineteen complex (NTC), nineteen related complex (NTR), and other non-snRNP proteins^7–9. After several dramatic, ATP-driven changes the U6 snRNA forms the spliceosome’s RNA-based catalytic center. Splicing then proceeds, resulting in a post-catalytic (P) spliceosome, in which the ligated mRNA and the excised intron-lariat are bound to the active site RNA network. The ligated mRNA is then released, yielding an intron-lariat spliceosome (ILS) that represents the terminal spliceosome state^7–9. The ILS and its catalytic center must then be disassembled to trigger the recycling of U2 and U5 snRNPs, U6 snRNA, NTC, NTR, and non-snRNP proteins for the next round of splicing and to degrade or process the intron-lariat RNA (Fig. 1a). Aberrant spliceosomes resulting from splicing errors are thought to be discarded by a similar process¹⁰.

Figure 1. Structures of a metazoan intron-lariat spliceosome (ILS) in two states.

a. Cartoon schematic of specific ILS disassembly. The ILS ‘prime’ (ILS’) and ‘double-prime’ (ILS”) states were identified in this study.

b. Composite ILS’ and ILS” cryo-EM densities from C. elegans (Ce) are shown from a front view⁵⁰. The maps range from 2.6 Å to 8.0 Å resolution (U2 3’ domain) and were generated from 15 (ILS’) or 18 (ILS”) local three-dimensional refinements. Subunits are colored according to snRNP identity (U2, green; U5, blue; U6, red; disassembly factors, shades of purple). A protein color code for each ILS subunit is shown underneath and is used throughout.

Spliceosome disassembly requires the essential RNA helicase DHX15 (yeast Prp43), a member of the DExH-box ‘translocases’. Although DHX15 has been extensively studied^11–15, the mechanisms governing its regulation are poorly understood. Notably, DHX15 lacks intrinsic RNA sequence specificity and its RNA target within the spliceosome remains unclear, with recent studies suggesting either the intron-lariat RNA, U2 snRNA, or U6 snRNA^10,13. Four conserved proteins have additionally been implicated in metazoan ILS disassembly: TFIP11 (yeast Ntr1), PAXBP1, C19L1 (yeast Drn1), and C19L2 (referred to by their human names; Extended Data Table 1)^16–22. However, their functional roles in disassembly are unclear. Owing to the incompleteness of available structural studies of yeast^19,23 and human²² ILS complexes, it further remains unknown how DHX15 and these additional factors act together to orchestrate the disassembly of terminal spliceosomes but not of earlier spliceosome intermediates.

Here we addressed these long-standing questions enabled by new high-resolution cryo-electron microscopy (cryo-EM) structures of nematode and human ILS spliceosomes. We combined our structural analysis with biochemical and genetic data, and a revised structure of the human P complex spliceosome, to reveal the functions of all four disassembly factors and the regulation of the disassembly helicase DHX15. Based on our results, we derive a model for disassembly of the ILS via U6, the catalytic center snRNA, thus providing critical insights into the stepwise dismantling of the terminal spliceosome and the discard of aberrant spliceosomes.

ILS structures at high resolution

To understand how the four disassembly factors and DHX15 specifically trigger spliceosome disassembly in metazoans, we determined the cryo-EM structure of a complete ILS complex. To overcome past challenges in ILS structure determination in yeast and humans, we purified spliceosomes from another metazoan, the nematode C. elegans (Ce). We chose Ce spliceosomes as a model system because Ce introns are remarkably short, with a median length of 65 nucleotides, which could help to increase spliceosome complex stability and the steady-state abundance of specific states²⁴. Moreover, Ce spliceosomes contain an identical protein composition to their human counterparts, unlike budding yeast spliceosomes, another extensively used model system (Extended Data Fig. 1g). For spliceosome purification, we endogenously tagged the NTC subunit PRP19 with an N-terminal 3xFLAG-tag using CRISPR-Cas9 and obtained spliceosomes from the extract of ~12 million adult worms. Ce spliceosomes sedimented around 35S in sucrose gradients and contained known spliceosome components, including the disassembly factors, TFIP11, PAXBP1, C19L1, C19L2, and DHX15 (Extended Data Fig. 1b,e). We collected cryo-EM data from this sample and obtained four million cryo-EM single particle images. Analysis of these particles revealed two ILS complexes (Supplementary Data Fig. 1). The general enrichment of the Ce ILS from worm extract mirrors the enrichment of the fission yeast ILS from the extract of logarithmically growing cells²⁵, suggesting that ILS disassembly could be a rate-limiting splicing step in certain conditions. Extensive image classification and local refinements of the Ce ILS data yielded twenty-seven cryo-EM maps, from which we assembled two composite maps that capture the ILS in two major states. We arranged these states based on their compositional complexity and named them ‘primed’ (ILS’, maps 1-7 and 8-15) and ‘double-primed’ (ILS”, maps 2-7 and 16-27), with the two structures differing in the bound disassembly factors (Fig. 1b, Extended Data Fig. 1f, Extended Data Table 2, Video S1, S2, Supplementary Data Fig. 1, 2, Supplementary Table 1). Local map resolutions reached 2.6 Å in the spliceosome’s core, with most densities resolved at a nominal resolution better than 3.5 Å (Fig. 1b, Supplementary Data Fig. 2). Together, these maps reveal a metazoan ILS in unprecedented detail (Fig. 1b, Supplementary Data Fig. 1, 2). AlphaFold2 Multimer predictions combined with manual building allowed us to prepare near-complete atomic models of Ce ILS’ and ILS” complexes (Extended Data Fig. 2c, d). In those ILS regions, where high-resolution human densities are available Ce and human maps are nearly identical (Extended Data Fig. 2a, b), indicating that structural insights from Ce reveal features common to metazoan spliceosomes.

The Ce ILS’ and ILS” models are substantially improved over human catalytic stage spliceosome structures that share the 8-subunit NTC and 9-subunit NTR complexes (Extended Data Fig. 2c, d). These improved models revealed the location of the essential NTR protein CCDC12 and previously unresolved parts of SYF1, SYF2, SYF3, ISY1, CDC5L, SDE2, PRP19, SPF27, SNW1, AQR, PPIE, PLRG1 (Extended Data Fig. 2d, Supplementary Data Figs 1, 2), which collectively contribute to NTC and NTR formation. In the ILS’, we also resolved the disassembly factors TFIP11 and PAXBP1. In the ILS”, we additionally resolved the disassembly factors C19L1, C19L2, and DHX15.

The ILS’ and ILS” states may thus be sequential on-pathway intermediates for disassembly, consistent with evidence for the step-wise formation of the ILS” (see Supplementary Text 1). Alternatively, the ILS’ could form through the partial breakdown of the ILS” during sample preparation, which would however not affect the conclusions drawn from the structural comparisons of ILS’ with ILS” (see Supplementary Text 1). Taken together, the high-resolution cryo-EM structures of the Ce spliceosome suggest that recognition of the ILS for disassembly occurs in two steps.

Recognition of outer ILS surfaces

The two disassembly factors TFIP11 and PAXBP1 are essential for viability in human cells (https://depmap.org/portal/) and in C. elegans^26–29. While they are necessary for ILS disassembly, their specific roles had been unclear. In the ILS’ structure, TFIP11 and PAXBP1 bind each other to form a ~250 Å long rod-like structure that spans the length of the ILS exterior surface (Fig. 2a, b). In TFIP11 we observe five structured regions, the ‘G-patch’, ‘hairpin’, ‘coiled-coil’, a helical repeat ‘GCFC’ domain, and a ‘C-terminal domain’ (CTD) (Fig. 2a, Extended Data Fig. 3b). In the ILS’ complex, all these regions engage in specific contacts (Fig. 2b, c), except for the TFIP11 G-patch, which is mobile. G-patch domains bind DHX15 with high affinity^30,31, suggesting that DHX15 is already tethered to the ILS’ but remains mobile, which is also consistent with similar abundances of TFIP11 and DHX15 in our mass spectrometry data of Ce spliceosomes (Extended Data Fig. 1e). The TFIP11 G-patch is followed by a β-hairpin, which binds the PRP8 RNaseH domain. The TFIP11 hairpin connects to a long helix (α3), which leans against the PRP8 RNaseH domain, and forms an anti-parallel coiled-coil with PAXBP1, consistent with a reported TFIP11–PAXBP1 interaction in Ce and humans^20,26. TFIP11 then continues with its α-helical repeat domain (residues 391-723), which binds the U5 snRNP subunit SNU114 domains III (residues 696-829), IV (residues 830-910), and V (residues 911-945) and connects via a linker to the TFIP11 CTD (residues 759-830). This CTD binds between SNU114 domains II (residues 587-665) and III and CWC15 (residues 58-108). PAXBP1 contains an additional α-helical ‘GCFC’ domain (residues 382-809), which contacts the PRP8 JAB1/MPN domains (residues 2065-2329) as well as the BRR2 plug element (residues 97-173). Metazoan BRR2 could thus aid disassembly by acting as a protein scaffold for TFIP11–PAXBP1, without using its RNA translocase activity³². In contrast, budding yeast Brr2 was reported to aid ILS disassembly via its ATPase activity³³. However, yeast Brr2 was also reported to be repressed by Ntr2 (yeast PAXBP1) in vitro³⁴. The role of yeast Brr2 in spliceosome disassembly thus warrants further investigation.

Figure 2. Disassembly factors recognize inner and outer ILS surfaces.

a. Domain organization of the disassembly factors TFIP11, PAXBP1, C19L1, C19L2, and DHX15. Solid lines indicate regions included in the atomic model. CTD, C-terminal domain; MMP, Metallophosphatase.

b. TFIP11–PAXBP1 recognize the ILS’ (left) and ILS” (right) exterior, whereas C19L1, C19L2, and DHX15 recognize the ILS” interior (right). Spliceosome regions not in contact with TFIP11–PABP1 are shown as transparent surfaces, except for the RNA active site, which is shown for reference. The black outline indicates the regions shown in panel c.

c. Interfaces between TFIP11–PAXBP1 and ILS’ (left) and ILS” (right) subunits. On the ILS” (right), the numbers 1, 2, and 3 mark regions of change during the ILS’ and ILS” transition: 1, movements at the TFIP11 ‘Hinge’; 2, movements of the PRP8 RNaseH (RH) domain and TFIP11 ‘Hairpin’; and 3, the newly liberated site in 2 is bound by C19L1–C19L2. See main text for details.

Taken together, the TFIP11–PAXBP1 dimer binds multiple exterior surfaces of the ILS’. Through these interactions, TFIP11–PAXBP1 would deliver the disassembly helicase DHX15 to the spliceosome. However, even though TFIP11 tethers DHX15 to the ILS’, this tethering is apparently not sufficient for DHX15 to dock onto the complex, suggesting that this requires additional structural changes.

Recognition of inner ILS surfaces

In the ILS”, we could resolve the remaining disassembly factors C19L1, C19L2, and a docked DHX15. Binding of these three proteins follows the dissociation of BRR2 from PAXBP1 and a substantial movement of TFIP11–PAXBP1 and PRP8 regions. We note that the location of BRR2 in the ILS’ and its dissociation in the ILS” may prime BRR2 for binding of the recycling chaperone TSSC4 and thus aid the recycling of the 20S U5 snRNP for the next splicing round^35,36. During the ILS’ to ILS” transition, TFIP11 bends at a hinge located between the TFIP11–PAXBP1 coiled-coil and the TFIP11 helical domain (Fig. 2b, c), swinging the TFIP11–PAXBP1 coiled-coil outwards from the ILS by ~20° degrees (Extended Data Fig. 4a-c). As a result, the TFIP11–PAXBP1 coiled-coil helices swings around the TFIP11 hinge, displacing the PRP8 JAB1/MPN–BRR2 complex and re-positioning the connected TFIP11-hairpin–PRP8 RNaseH complex by ~25 Å from its ILS’ position. The new space generated by movement of the PRP8 RNaseH domain is now occupied by C19L1 and C19L2 (Extended Data Fig. 4a-e). Based on the mobility of the TFIP11–PAXBP1 in the ILS’ (Fig. 1b, Supplementary Fig. 2), we speculate that TFIP11–PAXBP1 may stochastically sample its ILS” conformation, allowing for C19L2-binding to selectively stabilize the ILS” conformation.

C19L2 makes extensive contacts to the newly oriented PRP8 RNaseH domain, the PRP8 Large domain as well as the RNA-based active site, similar to the C19L2–spliceosome interaction network observed in the partial fission yeast¹⁹ and human ILS structures²² (Extended Data Figs 3d; 4b, j). Compared to the Ce ILS’, the RNA densities near the RNA active site and the nearby U2/U6 snRNA helix II are much better defined in the ILS” (Extended Data Fig. 4g,h). ILS recognition by C19L2 may thus conformationally lock the ILS RNA active site, which is expected to be mobile immediately after the release of mRNA and branching-specific splicing factors from the P complex. This conformational lock is thus likely a structural consequence of C19L2-binding, which may also play a role in ILS disassembly.

In the ILS” structure, we could also assign C19L1, which comprises an N-terminal metallophosphatase domain (MMP) and two C-terminal CwfJ domains (CWFJ) (Fig. 2b, c, Extended Data Fig 5, Supplementary Data Fig. 2). The C19L1 CWFJ domain binds C19L2 helices α1 and α2 (Extended Data Fig. 5a-d) and connects through a linker to its N-terminal MMP domain, which binds the now docked DHX15 through conserved interfaces (Extended Data Fig. 5e-j). Notably, a conserved peptide of SYF1, which we name the ‘tether’, binds in between C19L1 and DHX15 (Extended Data Fig. 5e), consistent with their in vitro interaction using recombinant proteins (Extended Data Fig. 5h) and the yeast-two-hybrid interaction of their yeast orthologs¹⁶. The SYF1 tether may thereby enhance the C19L1–DHX15 affinity and facilitate DHX15 docking in the ILS”.

Taken together, C19L1–C19L2 are likely to be part of a two-factor authentication mechanism to ensure that the RNA helicase DHX15 docks at the ILS only when both inner and outer ILS surfaces have been recognized by all disassembly factors.

Specificity of ILS recognition

To prevent premature disassembly in the P complex spliceosome and earlier spliceosome states, ILS recognition must be specific. However, our Ce spliceosome preparation did not contain the P complex, precluding a meaningful comparison with an ILS from the same species.

We therefore set out to determine the architecture of a complete human ILS and to prepare a revised model of available, but incomplete human P complex structures^22,37 (Fig. 3a, b, Supplementary Data Fig. 3). Since TFIP11 and PAXBP1 were missing from available human ILS structures²², we overexpressed GFP-tagged TFIP11 in human K562 cells and purified TFIP11-bound spliceosomes from nuclear extract (Extended Data Fig. 6). We collected 40,043 micrographs of human TFIP11-containing spliceosomes from which we obtained an overall density of a human ILS at 3.5 Å resolution (Extended Data Fig. 6, Supplementary Data Fig. 4). Surprisingly, TFIP11–PAXBP1 had largely dissociated from these complexes. However, further particle classification could reveal a subset of ~10,000 ILS particles that did contain TFIP11–PAXBP1 density with nominal resolution of 8 Å. This human ILS architecture is in excellent agreement with its Ce ILS” counterpart, including the locations of TFIP11–PAXBP1, C19L2, and weak densities for DHX15 and C19L1 (Extended Data Fig. 6). The high-resolution regions of the human ILS structure, including C19L2, closely resembled cryo-EM densities of previously reported partial human ILS structures²² (Extended Data Fig. 2b), which also showed weak densities for DHX15, but lacked the disassembly factors TFIP11–PAXBP1 and C19L1. By combining our new cryo-EM data with AlphaFold2 Multimer predictions and the Ce ILS” model, we generated an integrative architectural model of the human ILS” (Fig. 3b, Supplementary Data Figs 5, 6, Video S3). The quality of the human ILS densities was noticeably lower compared to Ce ILS densities in peripheral regions, where disassembly factors bind. This could reflect differences in complex stability between species or in complex heterogeneity due to the variable intron length in humans (~7,000 nucleotides) compared to Ce (~65 nucleotides). Despite these differences in quality, the locations of the disassembly factors are conserved between human and Ce ILS complexes (Figs 2b, 3b).

Figure 3. Human P complex and ILS” structures reveal determinants of state-specific disassembly.

a. The revised P complex coordinate model shown from the front. Subunits are colored according to snRNP identity (U2, green; U5, blue; U6, red; stage-specific proteins, shades of purple). The ‘#’ indicates that the P complex structure was generated by combining previous cryo-EM densities and models of human B^act, C*, and P complex spliceosomes. Below, regions of the human P complex that clash with the disassembly factors are shown as cartoons, the remainder is rendered as a transparent surface. The numbers 1, 2, and 3 highlight regions of the P complex that are used to discriminate P from ILS” complexes by the ILS disassembly factors.

b. The integrative human ILS” structure is shown from a front view. Below, human disassembly factors are highlighted, revealing that they bind the ILS” similar to their Ce counterparts (compare Fig. 2b). Colors as for the Ce ILS” in Fig. 1.

c. Structural comparisons of the human P and ILS” structures elucidate specific recognition of the human ILS. P-ILS clashes 1 (left): Shows the P complex-bound mRNP (mRNA 5’-exon, Exon Junction Complex), and the subunits SRRM2, NOSIP, CWC22 and SLU7 that clash with the ILS subunits TFIP11–PAXBP1. Structures were aligned on SNU114 (transparent surface). P-ILS clashes 2 (middle): Shows clashes between the P complex subunits SLU7, PPWD1, and the PRP8 JAB1/MPN domain with the ILS subunits TFIP11–PAXBP1. Structures were aligned on the PRP8 L domain (transparent surface). P-ILS clashes 3 (right): Shows clashes between the P complex-bound mRNA 3’-exon, the subunits FAM50A, CACTIN, and the PRP8 RH domain with the ILS subunit C19L2. Structures were aligned on the PRP8 L domain (transparent surface); the U5 snRNA Loop 1 is shown for reference to panels a and b.

To prepare a revised model of the human P complex, we next combined available cryo-EM densities and models of human B^act, C, C*, and P complexes^22,37–39, and the Ce ILS” (solved here), with AlphaFold2 Multimer and manual building (Fig. 3a, Supplementary Data Fig 3, Video S4). Compared to available P complex models, our revised structure additionally contains CCDC12, PPWD1, TLS1, NOSIP, STEEP1, FAM50A, ESS2, ISY1 and extensions in BRR2, SYF1, SYF2, SYF3, CDC5L, SDE2, PRP19, SPF27, and SNW1, which together result in the most complete P complex model available to date.

Structural comparisons of human P complex and ILS revealed the basis for specific ILS disassembly (Fig. 3, Extended Data Fig. 7). In the P complex, the mRNA ribonucleoprotein complex (mRNP) 5’-end, which comprises mRNA 5’-exon and the mRNA-bound exon junction complex, and the spliceosome subunits SRRM2, CWC22 would clash with the TFIP11 α-helical domain. The re-oriented PRP8 JAB1/MPN–BRR2 complex and BRR2-bound PPWD1 would further clash with PAXBP1 (Fig. 3c, Extended Data Fig. 7). These clashing P complex proteins are also present in all earlier catalytic spliceosomes (B^act to C* complex), explaining how binding of TFIP11–PAXBP1 discriminates against spliceosome states before the ILS. In addition, the C* and P complex-specific proteins CACTIN, FAM50A, SLU7, and TLS1 and the mRNP 3’-end containing mRNA 3’-exon would clash with C19L2 (Fig. 3c, Extended Data Fig 7). Taken together, the disassembly factors distinguish the ILS from earlier spliceosomal states by sensing the release of the spliced mRNA, mRNA-associated proteins, and catalysis-specific spliceosome proteins.

To gain insights into the conservation of ILS recognition, we compared our Ce ILS’ and ILS” structures with the budding yeast ILS structure²³. This revealed surprising differences in how metazoan and budding yeast ILS complexes are recognized by their disassembly factors, particularly among TFIP11–PAXBP1 and their yeast homologs, Ntr1–Ntr2 (Extended Data Fig. 3). The yeast Ntr1 (ref.²³) and Ce TFIP11 CTDs bind similarly to either yeast Snu114–Cwc23, where Cwc23 is yeast-specific, or the metazoan SNU114–CWC15. However, the remaining interfaces between TFIP11 and the Ce ILSs are entirely different from its yeast counterpart (Extended Data Fig. 3e). Further, although we identify yeast Ntr2 as the homolog of Ce and human PAXBP1, Ntr2 binds the budding yeast ILS through non-overlapping surfaces compared to Ce and human PAXBP1 (Extended Data. Fig. 3). Nevertheless, the alternate yeast Ntr1–Ntr2 interfaces also allow for specific ILS recognition, compared to yeast catalytic spliceosome states²³. We speculate that owing to the increased protein complexity of metazoan mRNPs (e.g. the presence of the exon junction complex) and of the spliceosome, TFIP11–PAXBP1 may have evolved to recognize metazoan mRNP and ILS features that discriminate metazoan catalytic-stage from terminal spliceosomes through a larger surface area. Notably, while the C19L2–spliceosome interactions are highly similar between fission yeast, Ce, and human complexes (Extended Data Fig. 3d), C19L2 is absent from budding yeast, which instead contains Cwc23 that binds at a different site of the budding yeast ILS. Drn1, the yeast homolog of C19L1, was not observed in any yeast ILS structure, precluding a structural comparison to the Ce and human ILS. Despite the apparent differences in ILS recognition between species, all employ multiple independent factors to verify the release of mRNA 3’-exons from the spliceosome, while Ce and human complexes in addition verify the release of the mRNA 5’-exon. Once release is verified, the highly conserved disassembly helicase DHX15 (yeast Prp43) locates to the same region in the ILS in various species, indicating conservation of the disassembly mechanism (Extended Data Fig. 3, ref.^19,22,23). However, owing to the low resolution of these regions in previous ILS structures^19,22,23, the mechanism of DHX15-mediated disassembly, and its RNA target, remained unclear.

DHX15 acts on U6 for disassembly

In the Ce ILS” structure, we resolved DHX15 to 3.9 Å resolution, revealing how DHX15 is positioned and activated to disassemble the spliceosome (Fig. 4a, b). DHX15 is an RNA helicase that translocates in a 3’-5’ direction along an RNA substrate, and comprises RecA1 and RecA2 lobes, and a C-terminal domain (CTD). In the ILS”, DHX15 and the bound RNA are in a relaxed, ATP-unbound conformation (Extended Data Fig. 8f). At 3.9 Å resolution, we could unambiguously trace U6 snRNA from U2/U6 Helix II into the DHX15 active site (Extended Data Fig. 4i), revealing that the spliceosome’s catalytic center snRNA U6 (ref.⁴⁰) is the target for metazoan spliceosome disassembly. This is consistent with biochemical data in budding yeast¹⁰, and in contrast to reports that proposed that U2 snRNA or the intron-lariat RNA play this role¹³. The TFIP11 G-patch bridges the DHX15 RecA2 and CTD domains, similar to crystal structures of human RNA-unbound DHX15 in complex with the ribosomal biogenesis factor NKRF1 (ref.³⁰) and the human splicing quality control factor SUGP1 (ref.⁴¹) (Extended Data Fig. 8a, b). Consistent with our structure, the recombinant Ce TFIP11 G-patch stimulates the Ce DHX15 helicase activity approximately 30-fold in vitro (Extended Data Fig. 8c, d), akin to human NKRF1 and yeast Ntr1 (ref.^12,30). The DHX15 RecA2 lobe is additionally bound by the C19L1 MMP domain and the SYF1 ‘tether’. The DHX15 CTD domain is bound to the NTR subunits SYF2 and SDE2, which are both essential for viability in humans (https://depmap.org/portal/) (Fig. 4b, c). These combined interactions, possibly together with a peptide of SYF3 (Extended Data Fig. 5k-o), guide DHX15 onto the U6 snRNA 3’-end, from where DHX15 could translocate along U6 snRNA to disassemble the catalytic center and the spliceosome.

Figure 4. DHX15 is primed for spliceosome disassembly via U6 snRNA.

a. Interactions of DHX15 (surface) with U6 snRNA and proteins of the Ce ILS” and disassembly factors (cartoons) are shown from the front. Non-interacting ILS regions are shown as a transparent surface. DHX15 is rendered as a sliced-through surface to highlight the U6 snRNA segment bound in its active site.

b. DHX15 is positioned on the ILS” by the NTR subunits SYF1, SYF2 and SDE2, and the disassembly factor C19L1 MMP domain. SYF2 and SDE2 act as a wall to protect U2/U6 helix II and guide the U-rich U6 snRNA 3’-end into the DHX15 active site. The TFIP11 G-patch activates DHX15 (Extended Data Fig. 8 and ref.⁴⁹).

c. SYF2 and SDE2 use a network of positively charged amino acids to guide the path of the U6 snRNA 3’-end towards DHX15 and possibly assist the separation of U2/U6 helix II upon the ATP-dependent translocation of DHX15 on U6 snRNA, from 3’- to 5’- ends. On the right, a cartoon schematic visualizes the key interactions of SDE2 and SYF2 with U2 and U6 snRNAs.

These observations suggest that the Ce ILS” is poised for disassembly but lacks ATP to initiate translocation (Extended Data Fig. 8a-f). To test this, we performed an in vitro ILS disassembly assay. We purified spliceosomes on beads via a 3xFLAG-PRP19 pulldown and then added ATP to bead-bound spliceosomes (Extended Data Fig. 8g). We would expect that ATP addition dissociates the ILS” into a complex comprising PRP19, the remainder of the NTC, NTR, and the U5 snRNP, but lacking the disassembly factors, U2 snRNP, and U6 snRNA. Consistent with this expectation, we observed the depletion of DHX15, C19L1, C19L2, TFIP11, PAXBP1, and U2 snRNP proteins, by quantitative mass spectrometry (Extended Data Fig. 8f). We thus conclude that the ILS” is competent for in vitro disassembly, but we do not exclude the possibility that additional factors, such as DBR1 (ref.^16,42), might contribute to ILS disassembly in vivo. Since C19L1 is located near the branch site in the ILS” and binds DBR1 in vitro¹⁸, C19L1 could for example target DBR1 to initiate intron-lariat RNA decay.

The choice of U6 snRNA as the target for DHX15 has implications for ILS disassembly and spliceosome discard during quality control. After transcription by RNA polymerase III, the U6 snRNA 3’-end is extended by oligo-uridylation^43,44. In the ILS”, the U6 snRNA 3’-end is bound to DHX15, indicating that post-transcriptional oligo-uridylation of U6 is not only required during spliceosome assembly⁴³, but might also be important for spliceosome disassembly by providing a single-stranded RNA 3’-end that acts as a landing pad for DHX15 (Fig. 4a, Extended Data Fig. 4f).

In our structure, the U6 snRNA 3’-end emerges from U2/U6 snRNA helix II (U2/U6 helix II; Extended Data Fig. 4h). U2/U6 helix II is embraced by the NTR subunits SYF2 and SDE2 (Fig. 4c), which together may use three mechanisms to promote disassembly. First, SYF2 and SDE2 spatially fix the position of U2/U6 helix II on the NTR complex, near DHX15. Second, the positively charged SYF2 helix α2 (‘wedge’) and SDE2 helix α2 (‘lid’) act as a wall that may protect the U2 snRNA 5’-end from the environment and guide the U6 snRNA 3’-end towards DHX15 (Fig. 4b, Extended Data Fig. 9a-b). The SYF2 residue W66 stacks onto the first nucleotide of the U2 snRNA, which together with the ‘wedge’ and ‘lid’ domains may further facilitate separation of U2/U6 helix II upon the translocation of DHX15 (Fig. 4c). Third, SYF2 helix α1 and SDE2 helix α3, jointly named the ‘anchor’, and the SYF1 ‘tether’ may aid in docking DHX15 at the NTR, near the U6 snRNA 3’-end. We propose that the NTR subunits SYF1, SYF2 and SDE2 contribute to priming the spliceosome for eventual disassembly.

To probe the importance of SYF2, we mutated Ce syf-2 in vivo. Consistent with the ILS” structure, deletion of the combined syf-2 helices α1 and α2 was lethal (Extended Data Fig. 9d, e). Deletion of only syf-2 helix α1 showed a partial loss of function. We could raise homozygous animals carrying the syf-2 helix α1 deletion at 20ºC but these animals were cold-sensitive, as has been reported for other splicing defective strains^45–47 (Extended Data Fig. 9f). Moreover, the partial loss of viability of this allele was enhanced by knock-down of sde-2 by RNAi, supporting a joint role for these proteins (Extended Data Fig. 9g).

Since SYF1, SYF2, and SDE2 join the spliceosome during its catalytic activation, these proteins could also be used for splicing quality control, to assist in the discard of aberrantly formed spliceosomes. Thus, we speculate that spliceosome disassembly and discard pathways may not only share DHX15 as the disassembly helicase^11,14,41,48, but also a common DHX15-binding site and RNA target¹⁰. Disassembly and discard pathways would nevertheless differ depending on the respective G-patch protein⁴⁹ and putative accessory factors for the specific multi-factor authentication of aberrant spliceosomes, before committing to disassembly via the U6 snRNA.

Conclusions

Here we presented structures of the Ce ILS’, Ce ILS”, human P complex, and human ILS”, which together reveal the conserved architecture of the metazoan spliceosome and a mechanism to initiate disassembly of the terminal spliceosome (Fig. 5, Video S5). We establish Ce as a novel model organism for the structural study of pre-mRNA splicing, yielding substantially improved cryo-EM densities over their human counterparts. Our results reveal an elaborate ‘multi-factor authentication’ system that involves four disassembly factors and three general spliceosome subunits, which collectively probe spatially distant surfaces to identify the terminal spliceosome. These surfaces are inaccessible in earlier splicing steps, revealing how specific spliceosome disassembly is achieved and pre-mature disassembly is prevented. After successful authentication of the terminal state, these proteins guide DHX15 onto U6 snRNA and activate DHX15 for disassembly. DHX15 would then translocate along U6 snRNA to unfold the spliceosome’s catalytic RNA center. Thus, both the beginning⁵ and the end of pre-mRNA splicing are orchestrated by the three-dimensional organization of U6 snRNA.

Figure 5. Model for terminal spliceosome disassembly.

The disassembly factors act together with the NTR subunits SYF1, SYF2 and SDE2 to initiate the specific dismantling of the ILS. After ligated mRNP release, the disassembly factors TFIP11–PAXBP1–DHX15 may recognize the ILS first, yielding the ILS’. This may license the binding of C19L1–C19L2 to form the ILS”. This multi-factor authentication would prime DHX15 to unwind the U6 snRNA-based active site, (iv) initiating ILS disassembly for spliceosome recycling and intron-lariat degradation (see Movie S5).

Our work also reveals how structural cues are read out and integrated in the terminal spliceosome prior to irreversible remodeling by DHX15. We speculate that similar principles govern the regulation of other DExH-box RNA translocases that act during splicing progression, discard of aberrant spliceosomes, and other aspects of cellular RNA metabolism.

Methods

C. elegans strain maintenance

All C. elegans strains were maintained on nematode growth media (NGM) plates seeded with OP50 bacteria at 20 °C as described previously⁵¹. Three strains were generated for this study, all of them were made using CRISPR/Cas9 starting with the reference background strain, N2 (Supplementary Data Table 3). Detailed information on their construction is provided below.

C. elegans genome engineering for endogenous tagging

For endogenous tagging of prp-19 with 3xFLAG tag, animals were subjected to Cas9-mediated genome engineering via RNP-microinjection following the approach described in ref.⁵². Briefly, the Alt-R CRISPR/Cas9 system (Integrated DNA Technologies) was employed and young adult animals were injected with a mixture containing 300 mM KCl, 20 mM HEPES, 4 μg/μl recombinant Cas9 (from S. aureus, purified in-house), 500 ng/μl TracerRNA, 100 ng/μl crRNAs targeting the endogenous prp-19 locus (Supplementary Data Table 4), and 150 ng/μl repair template encoding the 3xFLAG tag with around 150 bp flanking homology arms. Identification of candidates with the correct insertions was done by genotyping with the primers (Supplementary Data Table 4) and verified by Sanger sequencing. This yielded allele luc205 (strain MLC2610).

Large-scale cultivation and preparation of whole-worm extract

For large-scale cultivation of worms, Peptone + Streptomycin plates were seeded with 2 mL concentrated HB101 bacteria and left to dry overnight at 37 ºC. The following day, 100 000 synchronized L1 larvae of strain MLC2610 (prp-19::3xFLAG) were seeded onto each plate and grown at 25ºC until they had reached adulthood (between 2-2.5 days). Young adult animals were harvested by washing the 15 cm plates repeatedly with ice-cold M9 buffer (22 mM KH2PO₄, 42 mM Na₂HPO₄, 86 mM NaCl, 1mM MgSO₄), collected in 50 mL tubes, and washed three times in ice-cold M9 to remove residual bacteria. For extract preparation, the collected worm pellet was resuspended in one volume of Lysis buffer (50 mM HEPES pH 7.9, 1 mM MgCl₂, 100 mM KCl, 10% (w/v) glycerol, 0.05% NP-40, 0.5 mM DTT, cOmplete EDTA-free protease inhibitor cocktail (Roche)), frozen drop-by-drop in liquid Nitrogen, and subjected to cryo-milling using the Spex SamplePrep™ Freezer/Mill™ Dual Chamber Cryogenic Grinder (2 cycles of 2 minutes, each at 15 CPS). The resulting extract was cleared of debris by two consecutive rounds of centrifugation (10 min 15 000 g, 4 °C), frozen in liquid Nitrogen, and stored at -70 ºC until use. For the large-scale purification that yielded the sample used for cryoEM, we harvested 50-60 full plates of worms that yielded ~18 mL of cleared extract that was used for immunoprecipitation.

C. elegans genome engineering to generate syf-2 alleles

Truncations of syf-2 in C. elegans were generated by CRISPR/Cas9 mediated genome engineering. Cas9 ribonucleoprotein complexes (RNPs) were injected into synchronized young adult hermaphrodites. Cas9 RNPs were assembled by incubating a mix containing 50-100 ng/μL Alt-R crRNAs targeting the syf-2 gene (two guides were simultaneously injected in equal concentration, Supplementary Data Table 5), 500 ng/μl TracerRNA and 4 μg/μl recombinant Cas9 (from S. aureus, purified in-house) in buffer containing 300 mM KCl, 20 mM HEPES. As repair templates we used 100-200 ng/μL dsDNA repair template encoding the delta helix 1 truncation with ~150 bp flanking homology arms for homology directed repair (Supplementary Data Table 5); or 75 ng/μL ssDNA repair template with ~20 bp homology arms for the deletion of helices 1 and 2 (Supplementary Data Table 5) plus 75 ng/μL pBSK dsDNA. In all cases we added 2.5 ng/μL dsDNA encoding myo-2^prom:mCherry as a co-injection marker. The complete mix was incubated at 37 ºC for 15 minutes, followed by centrifugation for 10 minutes at maximum speed in a tabletop centrifuge at room temperature to pellet any precipitates in the injection mix. Following microinjection, P0 worms were moved to individual plates and grown at 15 ºC. At 48 hours post injection, F1 progeny were screened for myo-2^prom:mCherry expression. Plates containing at least three myo-2^prom:mCherry positive larval animals were separated. The injected P0 animals from these plates were transferred to fresh plates and were grown at 15 ºC. All F1 progeny laid after 48 hours were moved to single plates, allowed to lay progeny, and were genotyped by PCR of the syf-2 locus. The syf-2:Δ^anchor (delta helix 1) mutant was isolated and is maintained in homozygosity (strain MLC2722). The syf-2:Δ^anchor+wedge (delta helices 1-2) was isolated in heterozygosity; it is sterile in homozygosity. It is maintained in heterozygosity with the qC1(qls26) chromosome III balancer (strain MLC2727).

Viability, temperature Sensitivity, and RNAi

Animals were maintained at 20 ºC on standard NGM plates seeded with E. coli OP50, unless otherwise noted. RNAi plates used were NGM plates supplemented with 50 μg/ml Carbenicillin and 1 mM IPTG. RNAi plates were seeded with E. coli HT115 expressing the corresponding dsRNA which elicits an RNAi response against target gene mRNAs. Genes targeted were as follows: sde-2 (ORF-ID F53F4.14), syf-2 (ORF-ID K04G7.11), mog-7 (homolog of human PAXBP1, ORF-ID F43G9.12), cwf-19L1 (ORF-ID F17A9.2), and empty vector control (pL4440-dest-RNAi Destination vector). RNAi strains were streaked onto Carbenicillin plates from glycerol stocks of the RNAi libraries (ORFeome-RNAiv1.1). Single colonies were grown in Carbenicilin + Tetracycline containing LB, and the identity of the target gene was confirmed by Sanger sequencing. RNAi plates were freshly made for each experiment by growing 5 mL of RNAi culture O.N., concentrating the bacteria to 1 mL, and pipetting 150 μL of concentrated culture into the center of an RNAi plate. Worms plated for RNAi experiments were synchronized as follows: L4 larvae were picked and allowed to mature into young adults at 20 ºC overnight. Young adults were sliced open with a razor blade in a drop of M9. Early-stage embryos were collected by mouth pipetting and washed in M9. Washed embryos were allowed to hatch into L1 in 500 μL of M9 buffer O.N. Individual L1 worms were transferred to RNAi plates by mouth pipetting. Plates were grown at their respective temperatures (15 ºC or 20 ºC) until the empty vector control plates had laid a sufficient number of embryos (~150-250) to allow reliable counting of progeny, at which point all mother worms were removed from plates at that given temperature (note that 15 ºC and 20 ºC RNAi experiments were conducted at different time intervals because of the slower developmental time of C. elegans at lower temperatures – thus absolute viable worm counts are only comparable within a single temperature, not across different temperatures). F1 worms were grown at their respective temperatures until they reached L4 stage. L4 worms were removed by hand and counted manually to assess total number of viable worms from a given plate. Plates were screened for multiple days in a row to ensure that all L4s were counted and removed to avoid allowing animals to reach adulthood and start laying eggs again.

Generation of the human GFP-TFIP11 K562 cell line

For endogenous purification of the human ILS, lentiviral particles carrying the GFP-3C-TFIP11 constructs were generated in Lenti-X 293T cells (Takara) via polyethylenimine transfection (Polysciences) of the viral carrier plasmid and helper plasmids pCMVR8.74 (Addgene #22036) and pCMV-VSV-G (Addgene #8454), according to standard procedures. K562 (DSMZ) cells were infected at limiting dilutions and GFP-positive cells were isolated using a BD FACSAria III cell sorter (BD Biosciences). Viral integration was confirmed by immunoblotting for TFIP11 and GFP. Lenti-X and K562 cells tested negative for mycoplasma.

Preparation of human nuclear extract

To prepare nuclear extract (NE), 30 L of human K562 cells overexpressing GFP-3C-TFIP11 were grown to a density of 1.5x10⁶ cells mL^-1 at 37 °C, 5% CO2, stirred at 70 rpm. The NE was prepared as previously described⁵³ and dialysed against buffer F (20 mM HEPES pH 7.9, 100 mM KCl, 20% (w/v) glycerol, 0.2 mM EDTA, 2 mM DTT).

Cryo-EM sample preparation

C. elegans spliceosome anti-FLAG affinity purification and grid preparation

Whole-worm extract prepared from prp-19::3xFLAG animals and incubated for 2 h at 4 °C with anti-FLAG M2 resin previously equilibrated with Equilibration buffer C (20 mM HEPES pH 7.9, 50 mM KCl, 2 mM MgCl₂, 0.05% NP-40, 8% (w/v) glycerol, 1 mM TCEP). After washing, samples were eluted by incubation with 300 ng/μl FLAG peptide dissolved in TBS (Tris-buffered saline) for 2 hours at 4 °C. The eluate was loaded onto a GraFix⁵⁴ 15–30% (w/v) sucrose density gradient containing 0.05% glutaraldehyde in buffer D (25 mM HEPES pH 7.9, 50 mM KCl, 15% sucrose, 1 mM TCEP) and spun at 22000 rpm for 16 h in a SW60 Ti rotor (Beckman coulter). The sedimentation coefficients were simulated using the CowSuite software (https://www.cow-em.de). Peak spliceosome fractions were quenched for 15 minutes using a final concentration of 50 mM lysine, pooled, concentrated in a 0.5 mL 30 kDa MWCO Amicon Ultra concentrator (Sigma) and the buffer exchanged to buffer E (20 mM HEPES pH 7.9, 100 mM KCl, 2 mM MgCl₂, 1 mM TCEP) and immediately used for EM grid preparation. Briefly, 4 μL of concentrated and crosslinked samples was applied to glow discharged R2/1 200 holey carbon grids (Quantifoil) coated with a home-made 2 nm continuous carbon layer. Grids were blotted at 4 °C and 80% humidity and plunged into liquid ethane using a Leica EM GP2.

Purification and grid preparation of endogenous human ILS complexes

30 ml of GFP-3C-TFIP11 K562 NE were incubated with GFP-Trap Agarose resin (Chromotek) pre-equilibrated with binding buffer G (20 mM HEPES pH 7.9, 100 mM KCl, 2 mM MgCl₂, 0.05% (w/v) NP-40, 8% (v/v) Glycerol, 1 mM TCEP, cOmplete EDTA-free protease inhibitor cocktail (Roche)) for two hours at 4 °C under constant rotation. After five washes with six times the bead volume of binding buffer G, the human ILS complexes were eluted by cleavage using 3C PreScission Protease diluted in elution buffer H (20 mM HEPES pH 7.9, 100 mM KCl, 2 mM MgCl₂, 8% (v/v) glycerol, 0.04 μg/μl protease, 1 mM TCEP) for 1.5 hours.

The eluate was loaded onto a GraFix⁵⁴ 15–30% w/v sucrose density gradient containing 0.05% glutaraldehyde in buffer I (20 mM HEPES pH 7.9, 50 mM KCl, 1 mM TCEP) and centrifuged at 21000 rpm for 16 h in 4 °C in a SW60 Ti rotor (Beckman coulter). The sedimentation coefficients were simulated using the CowSuite software (https://www.cow-em.de). Fractions containing the human ILS were quenched for 15 minutes using a final concentration of 50 mM lysine, pooled, concentrated in a 0.5 mL 100kDa MWCO Amicon Ultra concentrator (Sigma) and exchanged to buffer J (20 mM HEPES pH 7.9, 100 mM KCl, 2 mM MgCl₂, 1 mM TCEP) and immediately used for EM grid preparation. Briefly, concentrated, and crosslinked human ILS complex were incubated with a home-made 2 nm continuous carbon layer for 20 minutes and picked up with glow discharged R2/1 200 holey carbon grids (Quantifoil). Grids were blotted at 4 °C and 68% humidity and plunged into liquid ethane using a Leica EM GP2.

Cryo-EM data acquisition

C. elegans spliceosomes

We collected two datasets of the C. elegans spliceosome sample, encompassing 17,600 and 25,358 micrographs, respectively. Both datasets were imaged on the same microscope (Titan Krios G3 at IST Austria, equipped with a Gatan K3 direct detector). For both datasets, we collected movies with a total dose of 60 e^-/Ų fractionated over 40 frames at a pixel size of 1.06 Å using ThermoFisher’s EPU software. The target defocus range was set to -0.9 μm to - 2.1 μm for dataset 1 and -0.75 to -1.9 μm for dataset 2. The electron filter was set to a filter width of 20 eV, and we used a 50 μm C2 aperture and no objective aperture.

Human TFIP11-containing spliceosomes

We collected two datasets of the human TFIP11-spliceosome sample, encompassing 12,344 and 27,699 micrographs, respectively. Both datasets were imaged on the same microscope (Titan Krios G4 at Vienna BioCenter, equipped with a Falcon 4i direct detector). The datasets had a defocus range of -0.75 μm to 2 μm. For both datasets, we collected movies with a total dose of 50 e^-/Ų in EER format at a pixel size of 0.945 Å using ThermoFisher’s EPU software. The electron filter was set to a filter width of 10 eV, and we used a 50 μm C2 aperture and no objective aperture.

Cryo-EM data processing

C. elegans spliceosomes

Pre-processing

Data was pre-processed with cryoSPARC live⁵⁵. Movies were gain- and motion corrected using the 'Patch Motion' program and the defocus was estimated using ‘Patch CTF’. Particles were picked using the ‘Blob picker’, with the minimal and maximal particle diameters set to 320 Å and 370 Å, respectively. This yielded 1,782,017 and 2,358,078 particle coordinates from dataset 1 and 2, respectively. Particles were initially extracted with a box size of 583 Å and binned to a pixel size of 2.45 Å for dataset 1 (box size of 256 px). For dataset 2, particles were binned to a pixel size of 1.3 Å/px and a box size of 448 pixels. Datasets 1 and 2 were subjected to initial cleaning and classification independently, and merged only after high-quality particle sets were identified.

Initial particle cleaning

To generate an initial model, we selected 173,220 particles from dataset 1 and cleaned the set using 2D classification, yielding 48,465 particles. From these, we calculated three initial volumes using the cryoSPARC ‘Ab-initio reconstruction’ algorithm, yielding one high-quality spliceosome class containing 76% of the input particles, and two ‘junk’ classes. These volumes were then used as reference volumes to classify all extracted particles from dataset 1 using ‘Heterogenous refinement’, yielding a dominant class of 924,617 dataset 1 ILS particles. Attempts to identify other spliceosome states in the data using reference volumes of human Bact, C, C*, or P complex maps as additional references were unsuccessful. Dataset 1 ILS particles were re-extracted at a pixel size of 1.3 Å/px and refined using ‘Homogenous Refinement’, enabling particle scale optimization. This yielded a map at a global resolution of 3.1Å. CTF refinement using global or local CTF refinement strategies did not improve resolution and was not further pursued.

For dataset 2, we split the 2,358,078 auto-picked particles into four batches and subjected each batch to ‘Heterogenous refinement’ using the same reference volumes as for dataset 1. ILS particles from each batch were combined and subjected to a second round of ‘Heterogenous refinement’. The best class contained 1,339,014 particles and was refined using ‘Homogenous Refinement’, enabling particle scale optimization. This yielded a map at a global resolution of 2.84 Å (dataset 2 ILS particles).

We next classified particles of each dataset into ILS’ and ILS” states, using global 3D classification without image alignment. We set the target resolution to 15 Å and the initial low-pass filter to 40 Å and used 50 classes (dataset 1) or 40 classes (dataset 2). We chose a slightly lower number of classes for dataset 2 because we were facing run-time issues when attempting to use fifty classes using the larger particle number in dataset 2.

Classes from 3D classification runs were inspected and grouped into either ILS’ (87% and 86% for datasets 1 and 2, respectively) or ILS” (13% / 14%), based on the presence or absence of DHX15 density. From ILS’ classes we excluded a large number of particles (360,899 / 433,898 for dataset 1 and 2, respectively) that showed no or poor density for the U2 snRNP, presumably due to mobility. These particles were otherwise indistinguishable from ILS’ and yielded high-resolution reconstructions (~3 Å).

We also separately pooled all particles that showed BRR2 density (155,690 / 299,666 particles). We exclusively observed BRR2 density in ILS’ classes. To further improve homogeneity within the ILS’ and ILS” populations, we subjected ILS’ particles with strong U2 density and ILS” classes to another round of global 3D classification, discarding classes with residual DHX15 density from the ILS’ particle set and discarding classes with weaker DHX15 density from the ILS” particle set. These stringently classified particle sets contained 321,346 and 558,116 particles for the ILS’ state, 155,690 and 299,66 for the ILS’ subset with BRR2 density, and 105,996 and 151,874 particles for the ILS” state. At this stage, particles from both datasets were combined, and we calculated consensus refinement using ‘Non uniform refinement’, yielding an ILS’ consensus map at a global resolution of 2.9 Å (Map 1) and an ILS” consensus map at a global resolution of 3.06 Å (Map 16).

Local refinements for peripheral regions

While the core of the consensus refinements reached a resolution of 2.6 Å, peripheral regions remained less well-defined due to molecular motion. To overcome this, we optimized local masks and local refinement parameters for different regions of the ILS that appeared to behave approximately as rigid bodies, guided by ‘cryoFlex’ analysis as implemented in CryoSPARC. While we initially subjected ILS’ and ILS” particle sets to independent local refinements, we noticed six regions that appeared indistinguishable at high resolution in the ILS’ and ILS”. For these regions, we combined ILS’ and ILS” particles for local refinements. Please refer to Supplementary Table 1 for a detailed list of map boundaries. Regions common between ILS’ and ILS” included i) the U5 snRNA 5’ end and the associated sm ring (map 2), ii) the intron binding complex (AQR and associated proteins) (map 3), the NTC core (map 4) and the PRP19 WD-40 domain (map 5), and the most peripheral region of the NTR (containing the SYF1 and SYF3 C-termini and associated proteins) (map 6), and the U2 snRNP 5’ end and associated sm ring (map 7).

To generate map 2, we performed an initial local refinement with a wider mask focused on the U5 snRNA sm ring using gaussian priors and setting the search ranges for translations and rotation to 21 pixels and 45°, followed by a second local refinement with a tighter mask and limiting the search ranges to 1 pixels /45°, yielding a map at a nominal resolution of 3.29 Å.

To generate map 3, we performed an initial local refinement with a mask around AQR and associated proteins using gaussian priors and setting the search ranges for translations and rotation to 21 pixels and 45°, followed by a second local refinement with a tighter mask and limiting the search ranges to 1 pixels /45°, yielding a map at a nominal resolution of 2.99 Å.

To generate map 4, we performed an initial local refinement with a mask around the NTC core, limiting the search ranges to 5 pixels /12°, yielding a map at 3.14 Å.

To generate map 5 and better resolve the PRP19 WD 40 (which was visible in only one of the four PRP19 subunits), we subjected Map 4 particles to 3D classification using four classes and a 20Å low-pass filtered reference volume, yielding two classes with 599,066 particles with improved WD40 density. Local refinement with gaussian priors and search ranges for translations and rotation of 21 pixels and 45° yielded a 5.57 Å nominal resolution density.

To generate map 6 and better resolve the NTR periphery, we first used local refinement with a mask around the SYF1 and SYF3 C-termini and associated proteins, limiting the search ranges to 5 pixels /5°, followed by 3D classification using four classes and a 20 Å low-pass filtered reference volume. Class 1 contained 345,902 particles and yielded a 3.31 Å nominal resolution map after reconstruction.

To generate map 7, we performed masked 3D classification of combined ILS particles using 20 classes and a 20 Å low-pass filtered reference volume. This revealed substantial mobility, and we selected class 5 containing 60,582 particles that yielded a 6.35 Å nominal resolution map upon reconstruction.

Local refinements in the ILS’

Next, we performed local refinements on ILS’ particles to generate focused refinement in regions that differ in the ILS’ and ILS”.

We first generated map 8 and 9, that together with map 6 completed the NTR lobe. To generate map 8, we subjected ILS’ particles to local refinement using a mask around the SYF1 central region, limiting the search ranges to 3 pixels /3°, yielding a 3.09 Å nominal resolution density.

To generate map 9, we subjected map 8 particles to local refinement using a mask focused on a central region of SYF1 adjacent to the region resolved in map 8 and limiting the search ranges to 3 pixels /3°, yielding a 3.69 Å nominal resolution density. While a map calculated with the same mask from ILS” particles showed clear and high-resolution density for SDE2 bound to SYF1, SDE2 was entirely absent, indicating that SDE2 which is recruited to the spliceosome during catalytic activation, is prone to dissociation in the absence of catalytic-stage splicing proteins or disassembly factors.

We next calculated local refinement maps to better resolve disassembly factors and their binding sites, starting at the base of the U5 snRNP where the TFIP11 CTD binds and moving towards the periphery, where the highly mobile BRR2 binds, yielding maps 10 to 14. To generate map 10, we subjected ILS’ particles to local refinement using a mask around the TFIP11-CTD and U6S1 domains III+IV, limiting the search ranges to 3 pixels /3°, yielding a 2.73 Å nominal resolution density.

To generate map 11, we subjected ILS’ particles to local refinement using a mask around the TFIP11 helical region, limiting the search ranges to 3 pixels /3°, yielding a 3.03 Å nominal resolution density.

To generate maps 12 to 14, we used the ILS’ subset containing BRR2 density, as BRR2 binding appeared to conformationally lock the mobile PRP8 RNaseH and as and PAXBP1. We initially performed focused 3D classification using a mask on BRR2 and a 20 Å low-pass filtered reference map and 10 classes, from which we selected a subset of 148,075 particles with improved density. We next performed focused refinement with a search range of 5 px/ 5° to pre-aligned particles, followed by a second refinement using gaussian priors and a search range of 21 px and 30°, yielding a 4.24 Å density of BRR2 and the PRP8 JAB domain (map 12). Starting from these particles, we performed local refinement using a wider mask encompassing the PRP8-RNaseH and JAB domains, PAXBP1 and BRR2, using a search range of 10 px /20° to yield map 12.

To generate map 13, map 14 particles were pre-aligned using local refinement with a similar mask as used for map 12, but larger, and a search range of 10 px/20°. A second local refinement with a mask around the PAXBP1-CTD yielded map 13 at a nominal resolution of 6.4 Å.

Finally, we noticed that density around the U2/U6 helix was poor in the ILS’, precluding conclusions on whether the SYF2 ‘wedge helix’ is bound on top of the U2-U6 helix as a consequence of DHX15 binding, or prior to it. To address this, we further classified ILS’ particles, using masked 3D classification around U2/U6 Helix 10 with 10 classes and a 20 Å low-pass filtered reference volume. We selected class 7, which showed clearly improved U2/U6 Helix II density and contained 127,957 particles. We also noticed very weak and blurry density in the region where DHX15 binds the ILS”. To exclude that stabilized Helix II density was only observed because of a small proportion of ILS” particle in the set, we further classified the 127,957 particle set using masked classification with a DHX15 density. From this we removed all classes that showed any residual DHX15 density, yielding a set of 78,633 ILS” particles with improved U2/U6 Helix II density at a nominal resolution of 3.61 Å (map 15).

Local refinements in the ILS”

We followed a similar refinement strategy for the ILS”, starting from a ‘Non-uniform refinement’ of ILS” particles that yielded a density at a nominal resolution of 3.06 Å (map 16). First, we pre-aligned particles on the NTR-IBC-DHX15 region using a local refinement with a wide mask encompassing these regions, using gaussian priors and a search range of 21 pixels 45° (NTR-IBC pre-aligned particles). To improve density for DHX15 and its interactors, performed a local refinement with a mask around DHX (search range 3 px/ 3°) to generate pre-aligned particles. These particles were imported into RELION 5 (ref. ^56,57) and subjected to 3D classification using 10 classes without image alignment and using ‘blush regularization’. While all classes showed clear density for DHX15, a major class of 31% (84,004 particles) showed improved density for the DHX15 RecA2 lobe and the C19L1 N-terminus. These particles were re-imported into cryoSPARC and subjected to local refinement using a mask on DHX15 (search range 3 px/ 3°) to generate a map 25 at a nominal resolution at 3.94 Å/px or a local refinement on the DHX15-RecA2 lobe and the C19L1 NTD (search range 1 px/ 1°) to generate map 27 at a nominal resolution of 3.88 Å.

To improve density for the DHX15-NTR interaction site, we used DHX15 pre-aligned particles to perform a local refinement with a mask focused on SYF1 central region and the SYF2 and SDE2 ‘anchor helices’ (search range 1 px/1°) to generate map 23 at a nominal resolution of 3.62Å.

In the ILS”, we noticed additional weak density near the IBC. To better resolve this, we used masked 3D classification of NTR-IBC pre-aligned particles using 10 classes and a 6 Å low-pass filtered reference volume and a mask focused on the additional density, and selected the class with the strongest density (57,951 particles), which we refined using a local refinement (search range 3px/ 3°) to generate map 17, showing the ISY1 N-terminus at a nominal resolution of 5.7 Å. An AlphaFold2 Multimer^58,59 prediction supported the density assignment. We also performed the same processing strategy with ILS’ particles, however ISY N-terminus could not be resolved, indicating it might be stabilized in the ILS”. We next sought to better resolve TFIP11 and PAXBP1 and its interactors. Starting from Mapp 16 particles, we performed local refinement (search range 3 px/3°) with wide mask focused on the TFIP11 helical domain, followed by a second local refinement with a tighter mask (search range 1px/1°) to generate map 19 at a nominal resolution od 3.14 Å. Map 19 particles were further refined using a local refinement with a mask focused on the TFIP11-CTD and U5S1 domains III and IV (search range 3px/3°) to generate map 18 at a nominal resolution od 2.82 Å.

To generate map 20 and focus on the PRP8 RNaseH domain and PAXBP1, we performed focused refinement on map 19 particles with a mask focused on the PRP8-RNase H and (search range 3px/3°) to generate map 20 at a nominal resolution of 3.13 Å. To generate map 21 and focus on PAXBP1-CTD, we performed focused refinement on map 20 particles with a mask focused on the PAXBP1-CTD and (search range 3px/3°) to generate map 21 at a nominal resolution of 6.33 Å.

To better resolve U2-U6 Helix II and its interactors, we first aligned map 16 particles using local refinement (search range 3px/3°) with a mask focused on U2-U6 Helix II to generate map 22.

To further improve density for U2/U6 Helix II and the SYF2 ‘wedge helix’ and SDE2 ‘anchor helix’, we classified map 22 particles using the same mask as for map 22 and selected a subset of 69,968 particles that improved connectivity for the U6 snRNA between Helix II and DHX15 and SDE2 density. The selected particles were refined with a tight mask around U2-U6 (search range 1px/1°), yielding map 24 at a nominal resolution of 3.92 Å and clearly revealing the U6 snRNA as the target for DHX15.

Finally, to improve the mobile C19L1 CWFJ density and generate map 26, we pre-aligned map 16 particles using a local refinement with a mask focused on C19L2 and the C19L1 CWFJ (search range 1/1). Focused 3D classification using 10 classes and a 15 Å low-pass filtered reference volume revealed a class of 26,170 particles with improved C19L1 CWFJ density, which fere reconstructed to generate a nominal resolution map of 3.19 Å. Note that this resolution value reflects overall resolution of the reconstruction, and that local resolution of the C19L1 CWFJ is lower.

Model building

To generate an initial model, we downloaded AlphaFold2 models for all subunits present in the human ILS2, except for subunits CDC5L, SNW1, whose highly extended conformation was better captured by Phyre2 homology models (ref.⁶⁰). We further generated an AlphaFold2 Multimer model for a complex containing four copies of PRP19, SPF27, and CDC5L, which fitted Map 4 almost perfectly, and AlphaFold2 Multimer models for the TFIP11–PAXBP1, TFIP11-G-patch–DHX15, and DHX15C–C19L1-CTD, and C19L2–C19L1. These starting models were aligned to the human ILS2 (PDB 61D1), truncated as appropriate and manually adjusted into their corresponding local refinement maps in ISOLDE⁶¹ and ChimeraX⁶², and confidence-weighted reference restraints and secondary structure restraints from AlphaFold2 models where side-chains were not unambiguously resolved, as implemented in ISOLDE.

To identify the SYF1 ‘tether’, we performed an AlphaFolf2 Multimer screen of DHX15-C19L1(1-277) against the top 200 most abundant in Ce spliceosome sample identified by MS, which identified the SYF1 C-terminus as protein element interacting with both DHX15 and C19L1. The AF model was then manually adjusted into the density.

To model snRNAs, used the human ILS structure (PDB 6ID1) as a template and adjusted the sequence. Compared to their human counterparts, the U6 snRNA and U5 snRNAs are highly conserved (94% and 93 % sequence identity, respectively), and contain only minor differences. The C. elegans U5 snRNA contains a three-nucleotide insertion after position 20, which were resolved in the cryo-EM map and could be unambiguously modelled, and a three-nucleotide deletion in the 3’ stem loop. The U6 snRNA contains two deletions (three nucleotides after position +3 and one nucleotide after position +10, using the human numbering), making the 5’ stem loop slightly shorter. The U2 snRNA is slightly more divergent (61% sequence identity), but all nucleotides that pair with U6 or the intron RNA branch site are perfectly conserved. Due to a divergent sequence of the U2 snRNA 3’ stem loop, we generated a model using RNAcomposer⁶³ and fitted it into the cryo-EM density. Due to limited resolution, we truncated bases and refrained from assigning a sequence register for the U2 snRNA 5’ stem loop.

After adjusting the model in ISOLDE, the model geometry was refined in phenix (1.20.1-4487) against the composite map. We generated base-pair and stacking restraints in phenix to stabilize nucleic acid geometry, and further used the input model to generate reference model restraints. The nonbonded weight parameter was set to 2000 and rotamers were fitted for sidechains with poor density that were also outliers. These settings gave yielded excellent refinement statistics and real-space correlation values (Extended Data Table 2a-c).

Composite map generation

Composite maps were generated using UCSF ChimeraX⁶². To optimally preserve high-resolution information from each focused refinement, we first fitted each focused refinement into the locally filtered consensus map of the corresponding state. We then manually segmented our atomic model and visually identified the map that shows the best quality for each model region (Supplementary Table 1). Maps were then zoned around these atoms with a 10 Å distance cutoff. Finally, we combined all zoned focused maps and scaled them relative to each other using the ChimeraX ‘volume max’ command.

Human spliceosome cryo-EM data analysis

Pre-processing

Both human ILS datasets were pre-processed using cryoSPARC³⁷ live with default settings for gain and motion correction using ’Patch Motion’ and CTF estimation using ‘Patch CTF’. Particle picking was done in WARP v1.09 using a custom BoxNet2Mask neural network. For the first dataset 303,126 particles were picked from 12.344 micrographs. For the second dataset 1,388,112 particles were picked from 27,699 micrographs. The particles were extracted using a box size of 550 pixels and binned to 2.65 Å/px for initial classification.

Processing

The cryo-EM data was processed as indicated in Supplementary Data Fig. 4. Briefly, the two acquired datasets were individually classified using heterogenous refinements. As reference volumes we chose two ab initio models derived after initial 2D classification obtained from the first 100,000 particles of which 32,295 particles were used to generate two ab-initio models. As second reference we supplied low pass filtered maps of the C.elegans spliceosome homogenous refinement before classification and the human ILS2 density (emdb:9647).

After three rounds of heterogenous refinements without a mask we combined the highest quality particles from each dataset and applied a non-uniform refinement using a dynamic mask, yielding a map at resolution 3.53 Å with 116 770 particles. A second application of three consecutive rounds of heterogenous refinement using the low pass filtered published ILS structures (PDB: 6ID1 and 6ID0) as reference volumes yielded a second particle set (particle set II), which after non-uniform refinement yielded Map 1, at 3.38 Å resolution with 87,951 particles.

Local refinements

We subsequently applied several masks for focused classifications and refinements to improve local densities. Particle set II was used for focused refinements with masks around the core, PRP8, the U5snRNP and the IBC/AQR to generate Maps 2,3,7,11. To yield better results for the sub-regions of PRP8, the U5 Sm ring and U5 associated proteins two focused refinements with a wider and then smaller mask around the region of interest were applied to generate Maps 4,6,8. For the peripheral more flexible region of the NTC, we additionally applied 3D classification of particle set I without a mask and chose only the classes showing best densities for the region of interest. Following a focused refinement of the remaining 52,490 particles with a mask around the NTC yielded Map 9 at 6.08 Å resolution.

Similarly, we used the particles of Map 10 (IBC) as input for 3D classification with a mask around DHX15. Subsequent homogenous reconstruction of the best class with 9,321 particles yielded Map 13 at 8.06 Å resolution.

3D Variability analysis (3DVAR) with a mask around the heterodimer of TFIP11/PAXBP1 was applied to particle set I. The 23,887 particles of the best class (showing density for the disassembly factors) were further classified using 3D classification, yielding 14,635 particles. Focused refinement with a mask encompassing TFIP11/ PAXBP1 yields Map 12 at 4.78 Å. The particles were further used for focused refinement of TFIP11 only. Therefore, we applied 3D classification with a second smaller mask focused on TFIP11 on particle set II and combined these particles with the particles of Map 12 for focused refinement of TFIP11 only, yielding Map 14 from 26,103 particles at 5.79 Å resolution. For more details see Extended Data Table 2, Supplementary Table 1, and Supplementary Fig. 5.

Model building

To prepare the integrative model of the human ILS, we started out with the already published model of the human ILS (PDB:6ID1) (see Supplementary Fig. 6). From this model we deleted the U2 snRNP, part of the U2 snRNA (residues 54-184), the intron (residues 25-30, 118-135), part of the U5 snRNA (residues 7,8,70-84) and SYF2 (residues 114-126, 196-243), which were not resolved in our densities. Additionally, we deleted chains corresponding to SYF1, SYF3, SNU114, AQR, SPF27, PRP19, CDC5L (residues 251-270), CWC15, the U5 Sm ring and rebuilt them using AlphaFold2 Multimer prediction. In addition, we built extensions for CWC15 (residues 36-105), SNW1 (residues 1-46), C19L2 (residues 54-75), SNRPB (residues 99-117), PLRG1 (residues10-52) and SYF2 (residues 20-116). Finally, we added chains for CCDC12, ESS2, PAXBP1, TFIP11, ISY1, DHX15, SDE2, C19L1. All newly modelled chains are based on AlphaFold2 Multimer predictions. To see which subunits are best resolved in which map, see Supplementary Table 1.

For the NTC we predicted combinations of SPF27, PRP19, CDC5L, PLRG1 and rigid body fitted them into the density of Map 6, followed by adjustments using ChimeraX and ISOLDE. For the IBC we predicted combinations of SYF1, SYF3, CCDC12, ISY1, AQR, SYF2, SDE2. We used ChimeraX and ISOLDE to fit SYF1 and SYF3 into Map 10. We used the low pass filtered consensus refinement Map 1 for rigid body fitting of the predicted models of CCDC12, SYF2, ISY1, AQR and SDE2.

TFIP11 (residues 726-837) was predicted with SNU114 and fitted into Map 14. The TFIP11/PAXBP1 dimer was predicted using AF2 and fitted into Map 12. PAXBP1 was trimmed according to the homologues domains which were resolved in the C. elegans ILS” model. The helicase DHX15 was predicted with the G-patch domain of TFIP11 (residues 146-211) and placed in Map 13; its orientation was compared to the C. elegans ILS” model. In the integrative model of the human ILS, we modelled the RNA network based on the PDB 6ID1 and trimmed the U2 snRNA, the intron and the U5 snRNA. However, we observe that the RNA in the human structure is very mobile. The final coordinate model was real space refined using Phenix as described for the C. elegans ILS models (Extended Data Table 2). After refinement, we trimmed lowly resolved protein chains to a minimal backbone model for subunits SYF1, SYF3, C19L2 (54-75), CWC15, CCDC12, CDC5L, SYF2, ESS2, AQR, PAXBP1, PRP19, ESS2, DHX15, TFIP11, ISY1, C19L1, SPF27, SNW1.

Modelling of a revised human P complex structure

To generate the revised model of the human P complex, we used the published P complex model (PDB 6QDV), mass spectrometry data and cryo-EM densities from ref.³⁷ (EMD-4530, EMD-4525, EMD-4532, EMD-4526, EMD-4539). We first compared the deposited EM densities with PDB 6QDV to identify unmodelled densities, and then used AlphaFold2 models of PPWD1, TLS1, ESS2, CDC5L, Cxorf56, FAM50A, NKAP, and NOSIP to obtain candidate fits to these densities. Subsequently we used AlphaFold2 Multimer to predict combinations of the new factors with spliceosome proteins adjacent to unmodelled densities (PRP8, BRR2, SNW1, DHX8, CACTIN). AlphaFold2 Multimer models were aligned to PDB 6QDV and adjusted into the maps using rigid body fitting and ISOLDE in UCSF ChimeraX (Supplementary Data Fig. 1). In addition, we used models of SRRM2 and PPIE which have been assigned in previous structures of the human B^act, C, and C* complexes^22,37–39, and extended them using respective AlphaFold2 models. In addition, we used the NTR/NTC AF2 models, which we prepared for the Hs ILS” to improve models of PRP19, SPF27, CDC5L, SYF3, SYF2, SDE2, ISY1, CCDC12, SNW1). The final coordinate model was refined in real space using Phenix as described for the C. elegans ILS models, with the exception that no composite map was used. Instead, refinement was performed with the consensus map EMD-4525. After refinement, we removed the sidechains located in lower resolution areas of the map. These include sidechains for: ESS2, FAM50A, Cxorf56, PRP8 (residues 2067-2335), ISY1, SYF1, SYF3, CDC5L (residues 518-801), SYF2, NOSIP, SRRM2 (residues 56-99), DHX8, SDE2, PRP19, NKAP (residues 376-415), EIF4A3, MAGOH, RBM8A, CWC22 (residues: 149-406), AQR, SPF27, PPWD1, TSSC4, CCDC12, PPIE. The revised P complex coordinate model extends the previously deposited structure by over 2,100 residues.

Protein purification

DHX15

Recombinant Ce DHX15 was expressed in insect cells using a pGB10 plasmid containing 10x-His-FLAG-3C-ceDXH15. The plasmid was electroporated into DH10EMBacY cells to generate bacmids⁶⁴ that were then transfected into Spodoptera frugiperda Sf9 cells to generate a V0 virus. The V0 virus was further amplified in Sf9 cells to yield V1 virus. Then, we expressed this new construct in Hi5 insect cells using baculovirus. Insect cell pellets were resuspended in buffer A (50 mM HEPES pH 7.0, 400 mM NaCl, 20 mM Imidazol, 10% (v/v) glycerol, 2 mM MgCl₂, 2 mM beta-mercaptoethanol), and lysed by sonication. The lysate was cleared by centrifugation (first for 30 min at 18,500 rpm, then for 1 h at 40,000 rpm in a Ti45 rotor). The supernatant was filtered through 0.45 μm filters and applied to a HisTrap HP 5 ml column, previously equilibrated in buffer A. The column was washed with buffer A and washed with 5 % Buffer B (50 mM HEPES pH 7.9, 400 mM NaCl, 10% Glycerol, 2 mM MgCl₂, 500 mM imidazole, 2 mM beta-mercaptoethanol) and eluted with 60% buffer B. Peak fractions were pooled, diluted to 100 mM NaCl, and further purified via a Heparin chromatography column, using buffer C (50 mM HEPES pH 7.9, 10% Glycerol, 2 mM MgCl₂, 2 mM DTT) and buffer D (50 mM HEPES pH 7.9, 1 M NaCl, 10% Glycerol, 2 mM MgCl₂, 2 mM DTT). After loading, the column was washed with 5 column volumes (CV) of 5% buffer D and then eluted with a linear gradient from 5% to 100% buffer D over 20 CV. Peak-fractions were concentrated and further purified via gel filtration, using a HiLoad S200 16-60 column in gel filtration buffer (25 mM HEPES pH 7.9, 250 mM NaCl, 5% Glycerol, 2 mM DTT). Peak fractions were concentrated to 8.6 mg/ml and flash-frozen in liquid nitrogen.

C19L1 full-length protein

Recombinant Ce C19L1 was expressed in insect cells using a pGB10 plasmid containing 10x-His-MBP-3C-C19L1. Virus generation, expression, lysis and Ni-NTA chromatography were performed as described for DHX15. Peak fractions from Ni-NTA chromatography were diluted to 100 mM NaCl and applied to a 5 ml HiTrapQ FF anion exchange column, using buffer C and buffer D as described for Heparin chromatography for DHX15. The column was washed with 5% buffer D (100 mM NaCL) for 5 CV and developed with a linear gradient from 5 to 40% buffer D over 30 CV, followed by a step gradient at 100% buffer D (2M NaCl). Peak fractions were either directly concentrated and further purified via SEC on a HiLoad 200 16-60 column in gel filtration buffer (25 mM HEPES pH 7.9, 250 mM NaCl, 5% Glycerol, 2 mM DTT), of first cleaved with 3C protease for 2h. 3C protease and the His-MBP tag were removed via Ni-NTA chromatography prior to SEC. Peak fractions were concentrated to ~20 mg/mL and flash-frozen in liquid nitrogen.

C19L1(1-277)

The Ce C19L1(1-277) was cloned into pOPINB+ vector with an N-terminal 10x-His-MBP-3C tag and expressed in LB medium overnight at 18 °C. Pellets were frozen and resuspended in buffer A (25 mM HEPES pH 7.9, 500 mM NaCl, 5% Glycerol, 20 mM imidazole), lysed by sonication and cleared by centrifugation (18 000 rpm for 45 minutes at 4 degrees). The lysate was filtered through 0.45 μm pores and loaded on HisTrap column. The column was washed with 5% buffer B (25 mM HEPES pH 7.9, 500 mM NaCl, 5% Glycerol, 500 mM imidazole) and then eluted with a linear gradient to 60% buffer B over 10 CV. The protein was further purified using anion exchange chromatography as described for the C19L1 full-length protein. Tag cleavage and SEC were performed as described for C19L1 full-length protein but using a Superdex 75 16/60 column. Proteins were concentrated to ~20 mg/mL and flash-frozen in liquid nitrogen.

TFIP11 G-patch domain

The Ce TFIP11 G-patch domain (residues 117-221) was cloned into pOPINB+ vector with an N-terminal 10x-His-MBP-3C tag and expressed in LB medium for 3h at 37 °C. Pellets were frozen and resuspended in buffer A (25 mM HEPES pH 7.9, 500 mM NaCl, 5% Glycerol, 20 mM imidazole), lysed by sonication and cleared by centrifugation (18 000 rpm for 45 minutes at 4 degrees). The lysate was filtered through 0.45 μm pores and loaded on HisTrap column. The column was washed with 5% buffer B (25 mM HEPES pH 7.9, 500 mM NaCl, 5% Glycerol, 500 mM imidazole) and then eluted with a linear gradient to 60% buffer B over 10 CV. Peak fractions were pooled and digested with 3C protease to cleave the tag for 3h and then diluted to 50 mM NaCl and further purified via cation exchange chromatography on a MonoS column, concentrated to 2 mg/ml and flash frozen.

Peptide synthesis for pulldowns

Peptides were synthesized in-house on a Liberty Blue peptide synthesizer (CEM) using standard Fmoc chemistry. For each amino acid cycle, 4 min coupling with DIC/Oxyma was performed. N-term was acetylated on resin with 5% Acetic anhydride/2,5% DIPEA in DMF for 20min. Peptides were purified on a Phenomenex Luna C18(2) using a 2-45% in 45 min 0,1%TFA/ACN+0,1%TFA gradient. Peptides were synthesized with an N-terminal fluorescein and a C-terminal biotin modification. The identity and quality of peptides were confirmed using MALDI-MS (4800 MALDI TOF/TOF, Sciex). The following peptides were synthesized for pull-downs (Fluo = Fluorescein):

ceC19L2(75-106)-wildtype -Fluo-EDEKNKLSAKILKAEMKGDTDLVKKLKRKLESM-

biotin;

ceC19L2(75-106)-scrambled: Fluo-SKVMKEELKLEDASRNGKATEILDKLKKMLDKK-biotin;

ceSYF1(788-818)-wildtype: Fluo-SMNKGNISFVRGAGKTVQQNTTENPDEIDLD-biotin; ceSYF1(788-818)-scrambled Fluo-QAIITSDMLETRENDPNDQTNVNKVGKGSGF-biotin.

Peptides were dissolved in 100 mM HEPES pH 7.9 to a concentration of 1 mM.

Peptide pulldown assay

C19L2(α1-α2) versus C19L1

30 μl of High Capacity Neutravidin Agarose (ThermoScientific) beads were equilibrated in binding buffer (50 mM KCl, 25 mM HEPES pH 7.9, 2 mM MgCl₂, 10 % glycerol). Next, we added saturating amounts of either wildtype or scrambled Ce C19L2(75-106)-biotin peptide (12.5 μL of 1 mM peptide solution) and incubated at 4 °C for 1h on a rotating wheel. Complete saturation of the neutravidin beads with peptides was indicated by strongly visible fluorescence in the supernatant after 1h.

Beads were washed three times with 500 μL wash buffer (binding buffer + 0.05% NP-40) and the supernatant removed. We then added 15 μL of full-length, untagged C19L1 at a concentration of 20 μM in 50 mM NaCl, 25 mM HEPES, 10% Glycerol and allowed to bind for 1h at 4 °C. Unbound protein was removed by washing 5 times in 200 μL wash buffer. The beads were eluted with 30 μL elution buffer (200 mM Glycine, pH 2.5) and the eluted proteins visualized with SDS-PAGE.

Ce DHX15, C19L1(1-277) versus SYF1(788-818)

The pulldown was performed as described for C19L2(α1-α2)-C19L1 but with the following modifications. The binding buffer contained 25 mM HEPES pH 7.9, 100 mM NaCl, 0.05 mM ZnCl₂, 20 % glycerol, 0.05% NP-40 and 2 mg/mL BSA. The wash buffer contained 25 mM HEPES pH 7.9, 100 mM NaCl, 0.05 mM ZnCl₂, 20 % glycerol, 0.05% NP-40. Fluo-SYF1(788-818)-biotin (wildtype or scrambled) were immobilized at saturating concentrations as before. Untagged C19L1(1-277) and FLAG-DHX15 were used at a final concentration of 10 μM each.

In vitro ILS disassembly assay

For the in vitro spliceosome disassembly assay, 500 μl of extract were used for a PRP19-3xFLAG IP using 20 μL of magnetic M2-FLAG beads. The beads were equilibrated prior to extract addition in wash buffer 1 (20 mM Hepes pH 7.9, 100 mM KCl, 2 mM MgCl₂, 0.05% NP-40, 10% glycerol, 1 mM TCEP). After extract addition, the beads were incubated for 1 hour at 20 °C with gentle agitation. After this incubation, the beads were washed three times in 500 μL wash buffer 1. The washed beads were then resuspended in 300 μl of disassembly buffer (wash buffer containing either 10 mM ATP and 20 mM additional MgCl₂ or no ATP) and incubated for 1h at 20 °C with agitation. The beads were then washed 3 times with 500 μL wash buffer containing 250 mM KCl, and further washed four times with1 mL of detergent-free wash buffer (20 mM HEPES, pH 7.5, 150 mM KCl), changing tubes midway to prevent detergent carry-over. Three triplicates for each condition were performed in parallel.

Mass spectrometry analysis

Co-immunoprecipitated proteins coupled to magnetic beads from the ILS disassembly assay were digested with LysC on the beads, eluted with glycine followed by trypsin digestion. The nano HPLC system (UltiMate 3000 RSLC nano system, Thermo Fisher Scientific) was coupled to an Exploris 480 mass spectrometer equipped with a FAIMS pro interface and a Nanospray Flex ion source (Thermo Fisher Scientific).

Peptides were loaded onto a trap column (PepMap Acclaim C18, 5 mm × 300 μm ID, 5 μm particles, 100 Å pore size, Thermo Fisher Scientific) at a flow rate of 25 μl/min using 0.1% TFA as mobile phase. After loading, the trap column was switched in line with the analytical column (PepMap Acclaim C18, 500 mm × 75 μm ID, 2 μm, 100 Å, Thermo Fisher Scientific). Peptides were eluted using a flow rate of 230 nl/min, starting with the mobile phases 98% A (0.1% formic acid in water) and 2% B (80% acetonitrile, 0.1% formic acid) and linearly increasing to 35% B over the next 120 min. This was followed by a steep gradient to 95% B in 5 min, stayed there for 5 min and ramped down in 2 min to the starting conditions of 98% A and 2% B for equilibration at 30 °C.

The mass spectrometer was operated in data-dependent mode, performing a full scan (m/z range 350-1200, resolution 60,000, normalized AGC target 300%) at 3 different compensation voltages (CV-45, -60, -75), followed each by MS/MS scans of the most abundant ions for a cycle time of 0.9 (CV -45, -60) or 0.7 (CV -75) seconds per CV. MS/MS spectra were acquired using HCD collision energy of 30%, isolation width of 1.2 m/z, Orbitrap resolution of 30.000, normalized AGC target of 200% and minimum intensity threshold of 2.5E4. Precursor ions selected for fragmentation (include charge state 2-6) were excluded for 45 seconds. The monoisotopic precursor selection (MIPS) filter and exclude isotopes feature were enabled.

Proteomics data analysis of the ILS disassembly assay and of gradient-purified Ce spliceosomes ILS

Raw MS data was loaded into Proteome Discoverer (PD, version 2.5.0.400, Thermo Scientific). All MS/MS spectra were searched using MSAmanda v2.0.0.16129 (ref.⁶⁵). Trypsin was specified as a proteolytic enzyme cleaving after lysine and arginine (K and R) without proline restriction, allowing for up to 2 missed cleavages. Mass tolerances were set to ±10 ppm at the precursor and fragment mass level. Peptide and protein identification was performed in two steps. An initial search was performed against the databases ID1242_Flag.fasta (1 sequences; 22 residues), ID1242_PRP19.fasta (1 sequences; 492 residues), tags_v11.fasta (28 sequences; 2,153 residues), uniprot_reference_C_elegans_2023-05-15.fasta (19,823 sequences; 8,134,158 residues) and PD_Contaminants_TAGs_v20_tagsremoved.fasta. Here, Beta-methylthiolation of cysteine was searched as fixed modification, whereas oxidation of methionine, deamidation of asparagine and glutamine were defined as variable modifications. Results were filtered for a minimum peptide length of 7 amino acids and 1% FDR at the peptide spectrum match (PSM) and the protein level using the Percolator algorithm⁶⁶ integrated in Proteome Discoverer. Additionally, an Amanda score of at least 150 was required.

A sub-database of proteins identified in this search was generated and used for a second search, where the RAW-files were searched using the same settings as above plus considering additional variable modifications: Beta-methylthiolation on cysteine was set as a fixed modification, oxidation on methionine, phosphorylation on serine, threonine and tyrosine, deamidation on asparagine and glutamine, pyro-glu from q on peptide N-terminal glutamine, acetylation on protein N-Terminus were set as variable modifications. The localization of the post-translational modification sites within the peptides was performed with the tool ptmRS, based on the tool phosphoRS⁶⁷. Identifications were filtered using the filtering criteria described above, including an additional minimum PSM-count of 2 per protein in at least one sample. The identifications were subjected to label-free quantification using IMP-apQuant⁶⁸. Proteins were quantified by summing unique and razor peptides and applying intensity-based absolute quantification (iBAQ⁶⁹) with subsequent normalization of each replicate using PRP19, based on the MaxLFQ algorithm⁷⁰. Identified proteins were filtered to contain at least 3 quantified peptide groups. Statistical significance of differentially expressed proteins was determined using limma⁷¹.

RNA helicase assay

We used a fluorogenic RNA duplex for the helicase assay, as previously described⁷². For this, an AlexFluor-488 labeled RNA oligo (5′-AF488-UAGUACCGCCACCCUCAGAACCUUUUUUUUUUUUUU-3) was mixed with an equimolar amount of quenching strand (5′-GGUUCUGAGGGUGGCCCUACUA-BHQ-3’) containing a ‘black hole quencher’ modification at a concentration of 500 nM in annealing buffer (50 mM NaCl, 5% glycerol, 1 mM TCEP, 50 mM HEPES) and heated to 95 °C for five minutes and then cooled to 12 °C over 2 hours to anneal the strands. We then added a fivefold molar excess of unlabeled competitor DNA (5’-TAGTACCGCCACCCTCAGAACC-3’). For the helicase assay, the RNA duplex and competitor strand (50 nM scaffold and 250 nM competitor) were mixed with DHX15 (0.5 μM), TFIP11 G-patch (1 μM) or both DHX15 and TFIP11-G-patch in helicase buffer (50 mM NaCl, 5% glycerol, 1 mM TCEP, 50 mM HEPES, with or without 2 mM ATP, 2 mM MgCl₂) and incubated for 30 minutes at 25 °C.

The reaction was then analyzed using a PheraStar plate reader. For this, we first digested proteins with proteinase K at 37 °C for 40 minutes to mitigate fluorescence quenching effects, and then measured fluorescence in 384 well black bottom plates (Greiner), using 10 μL sample per well.

Fluorescence values were background subtracted and normalized to the highest fluorescence value in the experiment. Statistical significance was tested using unpaired t-tests in the GraphPad Prism software.

Extended Data

Extended Data Figure 1. Cryo-EM analysis of C. elegans spliceosomes.

a. Schematic of purification of spliceosomes from C. elegans. The endogenous locus of PRP19 was tagged with am N-terminal FLAG-tag using CRISPR/Cas9 and extract was prepared from ~12 million adult worms. After immunopurification (IP) and elution with FLAG peptide, spliceosomes were further purified via a sucrose gradient.

b. Coomassie-stained SDS-Poly-Acrylamide Gel (SDS-PAGE) of gradient-purified Ce spliceosomes. This experiment was performed seven times.

c. Denoised cryo-EM micrograph of gradient-purified and crosslinked Ce spliceosomes imaged on a Titan Krios with a K3 detector.

d. 2D class averages from the dataset.

e. Abundance of ILS subunits in gradient-purified sample measured by mass spectrometry. For this analysis we quantified absolute protein abundances by integrating the protein peptide peaks and normalizing to the protein length using iBAQ⁶⁹, which were then normalized to PRP8. The labels next to the bars indicate how many peptides were identified for each subunit and which percent of the sequence was covered.

f. Schematic of the data analysis pipeline. Stringent classification of ~4 million single particle images revealed the ILS’ (~85-90% of ILS particles) and ILS” (~10-15% of ILS particles, see Supplementary Data Fig. 1) as the major PRP19-containing spliceosomes populations in Ce extract. Extensive focused refinements of each state yielded a total of 27 maps, revealing the ILS’ and ILS” in unprecedented detail and facilitating the building of high-quality structural models. For details, see Extended Data Fig. 2 and Supplementary Data Figs 1,2.

g. Sequence conservation plot of ILS subunits between human and C. elegans (Ce), and human and Saccharomyces cerevisiae (Sc) shows a highly conserved ILS protein composition between human and Ce.

Extended Data Figure 2. Comparison of the complete C. elegans ILS” to a partial human ILS2.

a.-b. Side-by-side comparison of the Ce ILS” cryo-EM density map with the deposited human ILS2 map. Top: Overview, with cryo-EM density colored by subunits. For the human ILS2 (EMD-9647), a low pass filtered map (gaussian filter with a width of three standard deviations) is shown in addition (transparent white surface). Bottom: Zoom-ins to the spliceosomes core reveal nearly indistinguishable densities where high-resolution density is available for both Ce and Hs ILS.

c. Coordinate model statistics for Ce ILS” and Hs ILS2, listing number of residues included as full sidechain models or backbone models, respectively. Numbers in brackets indicate completeness relative to the sum of all residues calculated from deposited sequences for the full-length proteins.

d. ILS subunit diagrams indicating which residues are included in coordinate models of the Ce ILS” or Hs ILS2 as full side chain models (solid fill), backbone models (semi-transparent fill with stripes), or not modelled (transparent fill). Asterisks indicate severe register error in deposited human ILS2 models in SYF1 (register error of up to 120 residues) and SYF3 (register error of ~20 residues).

Extended Data Figure 3. Yeast and metazoan ILS architectures are poorly conserved.

a. Comparison of disassembly factors observed in available baker’s yeast (S. cerevisiae, Sc), fission yeast (S. pombe, Sp), human (Hs) or nematode (Ce) ILS structures.

b. Cartoon representation of the Ce TFIP11-PAXBP1 heterodimer.

c. Cartoon representation of the Sc TFIP11-PAXBP1 homolog Ntr1-Ntr2, with the G-patch factor Ntr1 aligned to its homolog TFIP11.

d. Side views of the ILS from Sc, Sp, Hs and Ce, with disassembly factors shown in ribbon representations and the ILS core shown as a transparent white surface.

e. Yeast Sc Ntr1-Ntr2 (transparent ribbons) overlayed on the Ce ILS”, revealing substantially different binding sites on the ILS. Structures were aligned on PRP8.

Extended Data Figure 4. Conformational and compositional changes from ILS’ to ILS”.

a. Close-up view of protein-protein interactions between the PRP8 RNase H (RH) domain, TFIP11, PAXBP1 and BRR2 in the ILS’. Protein elements thar are mobile in the ILS’-to-ILS”-transition are shown as ribbons, whereas elements thar are static are shown in addition as transparent surfaces.

b. Close-up view of protein-protein interactions between the PRP8 RNase H (RH) domain, TFIP11, PAXBP1 and C19L2 in the ILS”. C19L2 binding requires repositioning of the PRP8 RH domain and TFIP11–PAXBP1, which displaces the BRR2–PRP8 JAB1/MPN domains from PAXBP1. C19L2 recruits C19L1 by binding its C-terminal CWFJ domain.

c. Overlay of TFIP11–PAXBP1 in the ILS’ and ILS” in a 90° rotated view. Yellow arrows connect identical residues in both states.

d. Overview of the ILS’. DHX15 (transparent) is likely tethered via the TFIP11 G-patch domain but cannot dock onto its target.

e. Overview of the ILS”. C19L1 and C19L2 binding allows docking of DHX15, and the associated conformational change in TFIP11–PAXBP1 and PRP8 displaces BRR2. Circled numbers 1 and 2 indicate regions of zoom-ins in panels f, i and j.

f. Close-up view of the ILS” U6 snRNA 3’ end, with DHX15 and the TFIP11 G-patch removed for clarity. The oligo-uridylated and single-stranded U6 snRNA 3’ end is the ideal substrate for DHX15, and SDE2 and SYF2 shield the U2/U6 helix II.

g. RNA cryo-EM density in the ILS’. After dissociation of ligated mRNA and catalysis-specific splicing proteins, the RNA active site is more mobile.

h. RNA cryo-EM density in the ILS”. Compared to the ILS’, RNA densities are better defined in the ILS”, presumably due to binding of C19L2 (see panel j).

i. Continuous cryo-EM density between U2-U6 helix II and the DHX15 active site reveals U6 snRNA as the target for ILS disassembly.

j. C19L2 binds the active site RNA network near the branch helix and contacts U2 snRNA, intron-lariat RNA, U5 snRNA, and U6 snRNA.

Extended Data Figure 5. AlphaFold2 Multimer predictions support disassembly factor interactions.

a. AlphaFold2 Multimer prediction of full-length Ce C19L2–C19L1.The prediction is shown colored by subunit (left) or AlphaFold2 confidence score (per residue local difference distance test, plDDT, right). The prediction supports the binding of the C19L1 CWFJ domain to C19L2. The C19L1 MMP domain (rendered transparent) is predicted to be collapsed onto the structure, however our experimental cryo-EM density shows that in the ILS” the C19L1 MMP domain is distant from the C19L1 CWFJ–C19L2 complex. Note that the C19L1 CWFJ–C19L2 interaction is predicted with low confidence and in only 2 of 5 models (panel c).

b. plDDT scores of the 5 models plotted over the amino acid number. Scores for the 5 models are overlayed.

c. Predicted aligned error (PAE) plot of the 5 models, sorted from highest ranked prediction (left) to lowest ranked prediction (right).

d.Pull-down experiment with immobilized C19L2(α1-α2) peptide and recombinant C19L1. C19L1 binds the wildtype C19L2(α1-α2) peptide but not a C19L2(α1-α2) scrambled peptide control. Small insets underneath the lanes show fluorescent images of the beads with immobilized fluorescently labelled C19L2 peptides in the fluorescein channel to show equal loading of the wildtype and scrambled C19L2 peptides. This experiment was performed once.

e. Overview (left) and close-up (right) view of the Ce C19L1–DHX15–SYF1 interfaces in the ILS”. C19L1 and SYF1 jointly bind a conserved hydrophobic pocket in the DHX15 CTD. The close-up panel shows the DHX15 surface colored by molecular hydrophobicity potential, with white colors indicating hydrophobic surfaces.

f. The same close up as in panel e (right), but colored by sequence conservation. Residue conservation scores were obtained from the ConSurf server⁷³.

g. AlphaFold2 Multimer prediction of Hs DHX15 with C19L1 and SYF1 suggests a conserved binding mode and conserved hydrophobic residues in the DHX15 CTD, the C19L1 ‘loop 2’, and the SYF1 ‘tether’.

h. Pull-down experiment with immobilized SYF1(788−818) peptide and recombinant C19L1 and DHX15. C19L1 and DHX15 both bind the SYF1(788−818) ‘tether’ peptide but not a scrambled peptide control. C19L1 and DHX15 can also simultaneously bind to the wildtype but not the scrambled peptide. Small insets underneath the lanes show images of the beads with immobilized fluorescently labelled SYF1 peptides in the fluorescein channel to show equal loading of the wildtype and scrambled SYF1 peptide. This experiment was three times.

i. Predicted aligned error plots of the Ce and Hs DHX15–C19L1–SYF1 AlphaFold2 Multimer predictions.

j. Predicted local distance difference test (plDDT) plots of the AlphaFold2 Multimer predictions from panel i.

k. SYF3 might assist with positioning the C19L1 MMP domain in the ILS”. Zoom-in of an AlphaFold2 Multimer prediction between Hs SYF3 and C19L1, highlighting the interface between a SYF3 C-terminal β-hairpin (residues 780-805) that extends the C19L1 MMP domain central β-sheet. The SYF3 β-hairpin might be flexible relative to the HAT (half a tetratricopeptide repeat) domain through movement around a hinge residue (indicated with an arrow).

l. As panel k, but for the Ce proteins.

m. The Ce SYF3–C19L1 AlphaFold2 Multimer prediction overlayed onto C19L1 in the Ce ILS” cryo-EM structure. Cryo-EM density is shown as a transparent surface. Weak density is visible at the position predicted for SYF3 by AlphaFold2 Multimer, and also near the end of the SYF3 HAT domain (circled with a dashed line), indicating that the putatively assigned SYF3 β-hairpin might alternate between a C19L1-bound and C19L1-unbound conformation.

n. and o. AlphaFold2 Multimer PAE and pLDDT plots for the predictions shown in k and l.

Extended Data Figure 6. Cryo-EM analysis of human ILS spliceosomes.

a. Schematic of purification of TFIP11-bound spliceosomes from human cells. GFP-TFIP11 was overexpressed in K562 suspension cells and TFIP11-bound spliceosomes were purified from 30 L of suspension cell culture. After immunoprecipitation (IP) and elution with 3C protease, spliceosomes were further purified via a sucrose gradient.

b. Coomassie-stained SDS-Poly-Acrylamide Gel (SDS-PAGE) of the TFIP11-GFP IP. Bands in the gel are labelled according to the molecular weight of the ILS subunits. This experiment was performed four times.

c. Denoised micrograph of gradient-purified and crosslinked Hs ILS, imaged on a Titan Krios G4 with a Falcon 4i detector.

d. 2D class averages from the dataset.

e. Composite cryo-EM density, obtained from 14 local refinements and filtered by local resolution. Transparent density in the background shows a local refinement map (focused on PAXBP1) low-pass filtered with a gaussian filter with a sigma of three standard deviations.

f. Model of the human ILS”, with disassembly factors shown as ribbons and spliceosome core proteins shown in addition as a transparent surface. A difference density, calculated by subtracting simulated model density (low-pass filtered to 20Å) resolution) from experimental density (ILS consensus refinement map low-pass filtered with a gaussian filter with a sigma of three standard deviations) reveals additional density at the Ce ILS” C19L1 CWFJ position.

Extended Data Figure 7. Release of mRNP and spliceosome proteins from the post-catalytic spliceosome unmasks binding sites for the disassembly factors.

a. Overview cartoon, placing the depicted structures into context of the spliceosome disassembly pathway.

b. Structural comparison of the structures of the P complex (Model of a Ce P complex based on the updated human P complex, this work), the intron lariat spliceosome immediately after mRNP release (modelled) and the ILS’ (Ce structure, this work). Proteins thar are exchanged in the transition are labelled. Numbers indicate regions for zoom ins in panels c-e.

c. Overlay and close-up view of the P-complex structure with the ILS’ reveals a clash of TFIP11 with the EJC (EIF4A3 subunit) and with CWC22, NOSIP, and SRRM2 in the P complex. This clash would occur both in the ILS’ and ILS”. Clashing proteins are outlined in black.

d. Overlay and close-up view of the P complex structure with the ILS’ reveals a clash between PPWD1 and PAXBP1 on BRR2.

e. Overlay and close-up view of the P complex structure with the ILS” reveals a clash between C19L2 and the path of the ligated exons in the P-complex.

Extended Data Figure 8. The ILS” is competent for disassembly upon ATP addition.

a. Comparison between the structures of DHX15 bound to the G-patch domains of TFIP11 (this study), NKRF1 (ref.³⁰) (PDB 6SH7), and SUGP1 (ref.⁴¹) (PDB 8EJM). All G-patch domains show an identical binding mode, but additional residues are observed in the TFIP11 G-patch in the Ce ILS”.

b. Sequence alignments of the G-patch domains shown in a.

c. Schematic of a fluorescence-based helicase assay as described in ref.⁷² in which an RNA substrate with 3’ overhang and a 5’ fluorophore label (AlexaFluor588) is annealed to a complementary RNA that carries a fluorescence quencher (black hole quencher, BHQ) at its 3’ end. When the RNA strands are annealed, fluorescence is quenched. Upon separation of the RNA duplex by a helicase, fluorescent signal is increased. To prevent re-annealing, an excess of an unlabeled DNA strand complementary to the AF588-labeled RNA is added (not shown in schematic). Sequences for RNA and DNA used are as in ref.⁷².

d. The Ce TFIP11 G-patch stimulates DHX15 helicase activity. Helicase assay was performed as shown in panel c. N=4 replicates were measured. Fluorescence intensities were measured in a plate reader, background corrected, and normalized to the highest value. Error bars show the standard deviation of the mean. P-values from pairwise two-sided t-tests are indicated.

e. Denaturing PAGE analysis of the RNAs from the helicase assay shown in panel d after incubation of the RNAs with proteins and ATP. No degradation of the RNA was observed. This experiment was performed three times.

f. DHX15 is not bound to ATP in the Ce ILS” cryo-EM structure. Comparison of DHX15-RNA structure in the ILS” (left) and a crystal structure of the highly conserved Chaetomium thermophilium (Ct) PRP43 (PDB ID 5LTA, ref.³¹, 60 % sequence identity between Ce DHX15 and Ct Prp43) bound to RNA and the ATP mimic ADP-BeF₃ (middle). In presence of ADP-BeF₃, DHX15 adopts a closed conformation, compressing the RNA so that nucleotide +5 (n+5) is flipped outwards and no longer forms a stacking interaction with the neighboring bases. Right: Overlays of the Ce ILS” RNA density (transparent red) with the modelled U6 snRNA 3’ end, or the poly-U RNA conformation of Prp43 in presence of ADP-BeF₃ indicates that in the Ce ILS” the RNA is relaxed and DHX15 is ATP-unbound. This is further confirmed by the lack of density in the DHX15 ATP binding pocket in the Ce ILS” (not shown).

g. In vitro ILS disassembly assay. Left: schematic of the assay. Spliceosomes where immobilized on beads via the PRP19-3xFLAG tag, washed, and incubated with ATP. Upon ILS disassembly, components of the Nineteen core complex (NTC core), the Nineteen related complex (NTR) and the U5 snRNP should remain immobilized, while the disassembly factors and U2 snRNP proteins should be depleted from the beads. The bead bound fraction was then analyzed by mass spectrometry. Right: Volcano plot showing differential abundance of proteins with or without ATP treatment. Consistent with the ILS” structure, the disassembly factors TFIP11, PAXBP1, DHX15, C19L1, C19L2, the NTR subunit SDE2, and the U2 snRNP subunits U2A’ (RU2A) and U2B’’ (RU2B) are depleted upon ATP treatment, suggesting that the Ce ILS”, which constitutes a minor fraction of spliceosomes in Ce extract according to cryo-EM particle classification (Supplementary Data Fig. 1), is competent for in vitro disassembly. ILS subunits are indicated by large circles and color-coded according to subcomplex. The horizontal line at p=0.05 indicates the commonly used statistical significance cutoff.

h. Fold reduction of ILS subunit abundance after incubation with ATP and PRP19-3xFLAG IP as determined by mass spectrometry in panel g.

Extended Data Figure 9. Genetics in C. elegans support roles of SYF2.

a. Ce SYF2 and Ce SDE2 bind the U2-U6 helix II in the ILS”. U2 snRNA, U6 snRNA, SYF2 and SDE2 are shown as ribbons and DHX15 is shown as an outline.

b. An AlphaFold2 Multimer prediction of human SDE2 with SYF2 and SYF1 suggests an identical binding mode of Hs SDE2, however it was not observed in the experimental density due to limited local resolution.

c. Sequence alignments of Ce and Hs SDE2 and SYF2.

d. Schematic of syf-2 mutant alleles generated by CRISPR-Cas9 in C. elegans.

e. Viability of syf-2 mutant animals. Single worms of the indicated genotypes were placed on individual plates at the L3/L4 stage and grown at 20 ºC for 96 hours. Animals with deletion of helix 1 (Δanchor) are viable as homozygotes, but animals with deletion of helices 1 and 2 (Δanchor+wedge) are only viable as heterozygotes; homozygous mutants are thus progeny of heterozygous mothers. Sterility was scored as the inability to produce numerous progeny that developed into L4 larvae. A few sterile animals still produced <10 embryos or early larvae but these did not develop further.

f. Viability of wild-type or syf-2 Δanchor mutant strains treated with empty vector (e.v.) or anti-syf-2 RNAi, at standard (20 ºC) or low (15 ºC) culture temperatures. Worms were synchronized as L1 larvae, placed on RNAi plates and grown at the corresponding temperatures. Viability was assessed as the total number of F1 progeny that reached the L4 stage. N=3 animals were analyzed for assays at 15 °C and N=5 animals were analyzed for assyays at 20 °C. P-values from pairwise two-sided t-tests are indicated.

g. Measurement of the synthetic effect of RNAi against sde-2 on syf-2 Δanchor mutant viability. Viability was measured as described in e at both 15 ºC and 20 ºC. RNAi against mog-7/PAXBP1 was used as a positive control as an essential splicing protein. N=3 animals were analyzed for assays at 15 °C and N=5 animals were analyzed for assyays at 20 °C. P-values from pairwise two-sided t-tests are indicated.

Extended Data Table 1. Proteins contained in human P complex and ILS spliceosomes and their homologs in Ce and S. cerevisiae yeast.

Orthologs of splicing proteins relevant to this work across S. cerevisiae, C. elegans, and H. sapiens. The protein color code as used throughout. Orthologs were assigned using The Alliance of Genome Resources (https://www.alliancegenome.org), as well as The Spliceosome Database (http://spliceosomedb.ucsc.edu). Though no discrepancies were identified between these two sources, not all orthologs were predicted with equal confidence. The table above does not provide information on the confidence of ortholog prediction for each protein. For additional details, please see Supplementary Data Table 2.

	Bakers yeast	Human	C. elegans	Color		Bakers yeast	Human	C. elegans	Color
U5snRNP	Snu114	SNU114	eftu-2		NTC	Syf1	SYF1	syf-1
	Prp8	PRPF8	prp-8			Syf2	SYF2	syf-2
	Brr2	BRR2	snrp-200			Clf1	SYF3	syf-3
	unknown	SNR40	snrp-40.1			Isy1	ISY1	isy-1
						Cef1	CDC5L	cdc-5L
U2snRNP	Lea1	RU2A	mog-2			unknown	SDE2	sde-2
	Msl1	RU2B	rnp-2/rnp-3			Prp19	PRPF19	prp-19
						Snt309	SPF27	bcas-2
U5/U2 smring	Sm D3	SMD3	snr-1
	Sm B1	RSMB	snr-2		NTR	Ntc20	CCDC12	cede-12
	Sm D1	SMD1	snr-3			Ecm2/Cwc2	RBM22	rbm-22
	Sm D2	SMD2	snr-4			Prp46	PLRG1	plrg-1
	Smx3	RUXF	snr-5			Cwc15	CWC15	ewe-15
	Sm E1	RUXE	snr-6			Bud31	BUD31	bud-31
	Smx2	RUXG	snr-7			unknown	AQR	emb-4
						unknown	PPIL1	cyn-12
P Complex	Cwc22	CWC22	let-858			unknown	PPIE	cyn-13
	elF4A/Fal1	EIF4A3	eif-4A3			Cdc40	CDC40	prp-17
	unknown	RBM8A/Y14	rnp-4			Prp45	SNW1	skp-1
	Mnh1	MAGOH	mag-1			Prp17	CDC40
	unknown	SRRM2	rsr-2
	unknown	PRKRIP1/PKRI1	F37A4.2		Dissassembly	Ntr1/SPP382	TFIP11	stip-1
	unknown	FAM32A	KO1G5.8			Ntr2	PAXBP 1	mog-7
	unknown	CACTIN	cacn-1			Prp43	DHX15	ddx-15
	Slu7	SLU7	sluh-7			Drn1	C19L1	cwf-19L1
	Prp22	HRH1/DHX8	mog-5			unknown	C19L2	cwf-19L2
	CPR1/CPR3	PPWD1	cyn-15

Extended Data Table 2. Cryo-EM model refinement, data collection, and map statistics.

a. Cryo-EM data collection and focused refinement map statistics for the Ce ILS’ and Ce ILS”.

b. Cryo-EM data collection and focused refinement map statistics for the Hs ILS”.

c. Refinement statistics for the coordinate models of the Ce ILS’, Ce ILS”, Hs ILS”, and the revised Hs P complex. Refinement statistics were calculated using Phenix⁷⁴.

a	Cryo-EM map statistics Crye-EM maps (Ce ILS”)
Map number	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23	24	25	26	27
State	celLS’	celLS	celLS	celLS	celLS	celLS	celLS	celLS’	celLS’	celLS	celLS	celLS	celLS’	celLS’	celLS	ceILS”	ceILS”	celLS’	celLS’	celLS”	celLS”	celLS”	celLS’	celLS”	celLS”	celLS”	celLS’
Data collection
Defocus range (μm)	-0.72 to -2.1	-0.72 to -2.2	-0.72 to -2.3	-0.72 to -2.4	-0.72 to -2.5	-0.72 to -2.6	-0.72 to -2.7	-0.72 to -2.8	-0.72 to -2.9	-0.72 to -2.10	-0.72 to -2.11	-0.72 to -2.12	-0.72 to -2.13	-0.72 to -2.14	-0.72 to -2.15	-0.72 to -2.1	-0.72 to -2.2	-0.72 to -2.3	-0.72 to -2.4	-0.72 to -2.5	-0.72 to -2.6	-0.72 to -2.7	-0.72 to -2.8	-0.72 to -2.9	-0.72 to -2.10	-0.72 to -2.11	-0.72 to -2.12
Voltage (kV)	300	300	300	300	300	300	300	300	300	300	300	300	300	300	300	300	300	300	300	300	300	300	300	300	300	300	300
Electron dose (e-/A2)	60	60	60	60	60	60	60	60	60	60	60	60	60	60	60	60	60	60	60	60	60	60	60	60	60	60	60
Reconstruction (CryoSPARC)
Resolution	2.9	3.29	2.99	3.14	5.51	3.31	6.35	3.09	3.69	2.73	3.03	4.77	6.4	4.24	3.61	3.06	2.97	2.82	3.14	3.13	6.33	3.13	3.62	3.92	3.94	3.91	3.88
Map-sharpening B-factor (Å2)	79.9	102	92	75.1	537.7	70.5	559.5	66.3	111.1	68.9	67.6	225	656.4	1312	23.8	62.6	63.1	67.9	69.8	67.6	671.1	76.6	104.6	102.4	103.9	10.7	132.9
Pixel size (Å/px)	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3	1.3
Number of Particles	879,523	1,131,513	1,131,513	1,131,513	599,066	349,902	60,258	879,523	879,523	879,523	879,523	148,075	148,075	148,075	78,663	247,908	57,951	247,908	247,908	247,908	247,908	247,908	247,908	69,968	84,004	26,170	84,004
Data deposition (EMDB)	EMD-50447	EMDB-50449	EMDB-50450	EMDB-50451	EMDB-50452	EMDB-50453	EMDB-50454	EMDB-50455	EMDB-50456	EMDB-50457	EMDB-50458	EMDB-50459	EMDB-50460	EMDB-50461	EMDB-50462	EMD-50463	EMD-50464	EMD-50465	EMD-50466	EMD-50467	EMD-50468	EMD-50469	EMD-50471	EMD-50472	EMD-50473	EMD-50475	EMD-50474

b	Crye-EM maps (Hs ILS")
Map number	1	2	3	4	5	6	7	8	9	10	11	12	13	14
State	hs ILS”	hs ILS”	hs ILS”	hs ILS”	hs ILS”	hs ILS”	hs ILS”	hs ILS”	hs ILS”	hs ILS”	hs ILS”	hs ILS”	hs ILS”	hs ILS”
Data collection
Defocus range (pm)	0 to 3.57	0 to 3.57	0 to 3.57	0 to 3.57	0 to 3.57	0 to 3.57	0 to 3.57	0 to 3.57	0 to 3.57	0 to 3.57	0 to 3.57	0 to 3.57	0.01 to 1.78	0 to 3.57
Voltage (kV)	300	300	300	300	300	300	300	300	300	300	300	300	300	300
Electron dose (e-/A2)	50	50	50	50	50	50	50	50	50	50	50	50	50	50
Reconstruction (CryoSPARC)
Resolution	3.41	3.22	3.13	3.39	3.23	3.51	3.14	3.24	6.08	7.32	5.7	4.78	8.06	5.79
Map-sharpening B-factor (Å2)	12	49.3	63.1	68.4	56.3	56.7	58.4	50.7	299.7	817	266.1	37.9	566.8	274.1
Pixel size (Å/px)	0.945	0.945	0.945	0.945	0.945	0.945	0.945	0.945	0.945	0.945	0.945	0.945	0.945	0.945
Number of Particles	87951	87951	87951	87951	87951	87951	87951	87951	52490	87951	87951	14635	9321	26103
Data deposition	EMD-50477	EMD-50478	EMD-50479	EMD-50480	EMD-50481	EMD-50482	EMD-50483	EMD-50484	EMD-50485	EMD-50486	EMD-50487	EMD-50488	EMD-50489	EMD-50490

c		Coordinate model statistics
		CeILS’	CeILS”	HsILS’	HsP
	Model compos ition
	Protein residues	14323	14055	11764	16748
	Nucleotide residues	308	323	277	365
	Ligands	3	3	2	3
	Refinement (PHENIX)
	Map CC (aroundatoms)	0.71	0.76	0.62	0.54
	RMS deviations
	Bond lengths (Å)	0.007	0.007	0.007	0.010
	Bond angles	1.448	1.417	1.458	1.457
	Validation
	MolProbity score	1.42	1.29	1.38	1.91
	All-atom clach score	1.67	1.63	1.29	4.37
	Rotamer outliers (%)	1.8	1.36	1.75	2.97
	C-beta deviations (%)	0.04	0.04	0.03	0.02
	Ramachandran plot
	Outliers (%)	0.041	0.31	0.28	0.24
	Allowed (%)	4.06	3.73	4.61	4.51
	Favoured(%)	95.53	95.96	95.10	95.25
	Data deposition
	PDB-ID	8RO0	8RO1	8RO2	9FMD
	EMDB deposition (composite map)	EMD-19397	EMD-19398	EMD-193979	ref.[37]

Data availability

Three-dimensional cryo-EM composite density maps of the Ce ILS’ and ILS” have been deposited in the Electron Microscopy Data Bank under the accession numbers EMD-19397 to EMD-19398. The individual maps 1-27 have been deposited under the accession numbers EMD-50447, EMD-50449 to EMD-50569, and EMD-50471 to EMD-40475. Three-dimensional cryo-EM composite density map of the human ILS” have been deposited in the Electron Microscopy Data Bank under the accession numbers EMD-19399. The individual human ILS” maps 1-14 have been deposited under the accession numbers EMD-50477 to EMD-50490. The coordinate files of the Ce ILS’, Ce ILS”, the revised human P complex, and the human ILS” have been deposited in the Protein Data Bank under the accession numbers 8RO0, 8RO1, 9FMD, and 8RO2.