Mechanism for the initiation of spliceosome disassembly

Abstract

Precursor-mRNA (pre-mRNA) splicing requires the assembly, remodelling and disassembly of the multi-megadalton ribonucleoprotein complex called the spliceosome¹. Recent studies have shed light on spliceosome assembly and remodelling for catalysis^2-6, but the mechanism of disassembly remains unclear. Here we report 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 therefore controls both the start⁵ and end of pre-mRNA splicing. Taken together, our results explain the molecular basis of the initiation of canonical spliceosome disassembly and provide a framework to understand general spliceosomal RNA helicase control and the discard of aberrant spliceosomes.

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 iBAQ, 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 server. 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.) (PDB 6SH7), and SUGP1 (ref.) (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. 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.. 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., 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.

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 view. 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.

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.

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.

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.). 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.

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).