The architecture of the spliceosomal U4/U6.U5 tri-snRNP.

U4/U6.U5 tri-snRNP is a 1.5-megadalton pre-assembled spliceosomal complex comprising U5 small nuclear RNA (snRNA), extensively base-paired U4/U6 snRNAs and more than 30 proteins, including the key components Prp8, Brr2 and Snu114. The tri-snRNP combines with a precursor messenger RNA substrate bound to U1 and U2 small nuclear ribonucleoprotein particles (snRNPs), and transforms into a catalytically active spliceosome after extensive compositional and conformational changes triggered by unwinding of the U4 and U6 (U4/U6) snRNAs. Here we use cryo-electron microscopy single-particle reconstruction of Saccharomyces cerevisiae tri-snRNP at 5.9 Å resolution to reveal the essentially complete organization of its RNA and protein components. The single-stranded region of U4 snRNA between its 3' stem-loop and the U4/U6 snRNA stem I is loaded into the Brr2 helicase active site ready for unwinding. Snu114 and the amino-terminal domain of Prp8 position U5 snRNA to insert its loop I, which aligns the exons for splicing, into the Prp8 active site cavity. The structure provides crucial insights into the activation process and the active site of the spliceosome.

The protein coding sequences of most eukaryotic genes are interrupted by non-coding segments called introns. Introns are removed from pre-mRNAs and the flanking coding segments (exons) are spliced together to form mRNAs by two successive trans-esterification reactions within a dynamic multi-megadalton protein-RNA complex known as the spliceosome. This complex comprises five canonical subunits, namely U1, U2, U4, U5 and U6 small nuclear ribonucleoprotein particles (snRNPs) and numerous non-snRNP factors[1]. Each snRNP contains an snRNA, 7 Sm or LSm proteins and a number of snRNP-specific proteins. During the initial stages of spliceosome assembly, U1 and U2 snRNPs recognize the pre-mRNA 5′ splice site (5′SS) and branch point (BP) forming pre-spliceosomal A complex. The subsequent binding of the preassembled U4/U6.U5 tri-snRNP allows formation of the fully assembled spliceosomal B complex, which is converted to the catalytically active B* complex through extensive structural and compositional remodeling. During this process, U4 and U6 snRNAs, extensively base-paired in tri-snRNP are unwound, U1 and U4 snRNPs are released and many new proteins join the spliceosome[1-3]. This leads to the formation of a highly structured RNA network between U2, U5 and U6 snRNAs and the 5′SS and BP sequences in the pre-mRNA. The extensively base-paired U2–U6 snRNAs harbour catalytic magnesium ions[4] and position the BP and 5′SS for the first trans-esterification reaction, which produces exon 1 and lariat intron–exon 2 intermediates. Further remodeling to C complex enables U5 snRNA loop 1 to align exons 1 and 2 for nucleophilic attack of exon 1 at the 3′-splice site (3′SS), yielding spliced mRNA and lariat intron products[5,6]. Finally the spliceosome is disassembled before the next round of splicing.

U4/U6.U5 tri-snRNP is the largest preassembled spliceosomal complex, containing U5 snRNA, extensively base-paired U4-U6 snRNAs and over 30 proteins[7,8] (Extended Data Table 1). Three key proteins, Prp8, Brr2 and Snu114, play crucial roles in activation of the spliceosome and formation of the active site[1]. Prp8 forms cross-links with 4-thiouridine introduced at key positions within U5 and U6 snRNAs and the substrate pre-mRNA, showing that Prp8 is involved in substrate positioning and closely associated with the catalytic RNA core[9,10]. Brr2 helicase, whose activity is regulated by the GTPase Snu114 (ref 11-13), catalyses the unwinding of the U4-U6 snRNA duplex[14,15]. Interactions between tri-snRNP proteins have been investigated by yeast two-hybrid and in vitro binding assays[16,17]. Electron cryo-microscopy (cryoEM) reconstruction of cross-linked human tri-snRNP at 21 Å resolution revealed a tetrahedral overall shape with no clear domain separation[18]. Negative stain microscopy of cross-linked yeast tri-snRNP revealed a tri-angular shape with maximum dimension of 30-34 nm[19]. The highly biased orientation of tri-snRNP on carbon films precluded full three-dimensional analysis whilst the projection structure revealed three extruding domains termed head, foot and arm; the arm domain adopts variable positions with respect to the rest. Some key proteins were localised within the projection structure using genetically introduced tags [19]. Brr2 and U4/U6 snRNP were attributed to the head and arm domains, respectively. Based on this the authors proposed that Brr2 may engage with U4/U6 snRNA for unwinding when Snu114 - mapped in the hinge region - brings the arm and head domains closer.

Extended Data Table 1

Components and modeling of yeast U4/U6.U5 tri-snRNP

	protein	total residues	M.W.	Domain		PDB code
US snRNP	Prp8	2413	279,299	N-terminal domain	1-884	α-helices modeled
				RT-like	885-1251
				Thumb/X	1257-1375
				Linker	1376-1649	4143
				Endonuclease	1653-1824
				RNaseH-like	1839-1029
				Jab1/MPN	2150-2396	4BGD
	Brr2	2163	246,125	N-terminal domain	1-441	not modeled
					442-478
				N-termnal helicase cassette	478-1309	4BGD
				C-terminal helicase cassette	1330-2163
	Snu114	1008	114,025	N-terminal domain	1-114	not modeled
				G domain	120-443
				domain II	446-580
				domain III	603-671	homology model (1NOV)
				domain IV	675-853
				domain V	856-990
	Dib1	143	16,774	thioredoxin-like		not modelled
	SmB	196	22,403	Sm fold
	SmD3	110	11,229	Sm fold
	SmD1	146	16,288	Sm fold
	SmD2	110	12,856	Sm fold		4WJZ(human model)
	SmE	94	10,373	Sm fold
	SmF	96	9,659	Sm fold
	SmG	77	8,479	Sm fold
	U5 snRNA-L	214	68,847	Loop 1	92-102	not modeled
				Stem 1	84-91;103-110	A-form double helix
				IL 1	75-83;111-113	not modeled
				VSL	41-74	A-form double helix
				Stem 2	28-40;114-125	A-form double helix
				IL2	13-27;126-135	not modeled
				Stem 3	4-12;136-144	A-form double helix
				3′ SL	185-212	not modeled
U4/U6 snRNP	Snu 13	126	13,570			2ALE
	Prp31	494	56,305	N-terminal domain		α-helices modeled
				coiled-coil domain		2OZB (human model)
				Nop domain
				C-termal domain		α-helices modeled
	prp3	469	55,877	N-terminal domain		α-helices modeled
				Ferredoxin-like domain		Model obtained from Robetta prediction
	prp4	465	52,425	N-terminal domain		α-helices modeled
				β-propeller domain	166-465	3MXX from Robetta prediction
	SmB	196	22,403	Sm fold
	SmD3	110	11,229	Sm fold
	SmD1	146	16,288	Sm fold
	SmD2	110	12,856	Sm fold		4WJZ (human model)
	SmE	94	10,373	Sm fold
	SmF	96	9,659	Sm fold
	SmG	77	8,479	Sm fold
	Lsm2	95	11,164	Sm fold
	Lsm3	89	10,020	Sm fold
	Lsm4	172	20,304	Sm fold
	Lsm5	93	10,415	Sm fold		4M77
	Lsm6	86	9,396	Sm fold
	Lsm7	115	13,010	Sm fold
	Lsm8	109	12,385	Sm fold
	U4 snRNA	160	51,390	Stem I	57-64	A-form double helix
				Stem II	1-17	A-form double helix
				5′ SL	20-53	homology model (2OZB)
				central domain	65-80	partial homology model (2P6R)
				3′SL	91-142	partial modei 4WJZ (human model)
	U6 snRNA	112	36,088	5′ SL	1-25	not modelled
				Stem I	55-62	A-form double helix
				Stem II	64-80	A-form double helix
tri-snRNP specific	Prp6	899	104,234		1-191	not modelled
				TPR domain	192-899	α-helices modeled
	Snu66	587	66,426			not modeled
	Prp38	242	27,957			not modeled
	Snu23	194	22,682	C2H2 zinc finger-like		not modeled
	Spp381	291	33,764			not modeled

See Methods for details.

The development of high-speed direct electron detectors[20,21] and powerful maximum likelihood algorithms for classification and particle alignment[22] have made it possible to determine the structure of macromolecular assemblies at near atomic resolution by cryoEM[23]. By applying these new methods we obtained a map of native unstained yeast tri-snRNP at an overall resolution of 5.9 Å in which protein α-helices and RNA double helices are readily discernible. This enabled us to fit the double-stranded helices of U5 snRNA and U4/U6 snRNA as well as previously determined crystal structures or homology models of nearly all the proteins. The structure accounts for a wealth of biochemical and genetic data from yeast and human spliceosomes, and suggests a possible mechanism for B complex formation and the activation of the spliceosome.

Results

CryoEM of the tri-snRNP complex

U4/U6.U5 tri-snRNP was purified from yeast by a gentle procedure without cross-linking, and the sample was subjected to cryoEM analyses (Methods, Extended Data Fig. 1). Using a combination of statistical classification and movie processing[22,24] (Extended Data Fig. 2), we obtained a density map with an overall resolution of 5.9 Å by the “gold standard” Fourier shell correlation (FSC) = 0.143 criterion[25] with local resolution ranging from 5.0 Å to 20 Å (Extended Data Fig. 3 and 4; Methods). The map revealed clear densities for double-stranded RNA, with protein helices appearing as long tubes and β-sheets as flat densities (Supplementary Video). The density for the Lsm proteins in the flexible arm domain became clearer after using a new multi-body refinement method (Methods; Extended Data Fig. 3).

Extended Data Figure 1

U4/U6.U5 tri-snRNP sample used for this study

a, Coomassie-blue stained SDS-PAGE gel showing protein composition of the purified tri-snRNP. U5-, U4/U6- and tri-snRNP-specific proteins are labeled in blue, red and teal, respectively. Sm proteins present in both U5 and U4/U6 are in black. b, Toluidine-blue stained denaturing acrylamide (9%) gel showing RNA compositions. c, Electron cryo-micrograph of tri-snRNP where the carbon coated grid was discharged in N-amylamine. d, and e, Reference-free two-dimensional class averages of a data set collected on a grid discharged in air and N-amylamine, respectively.

Extended Data Figure 2

Classification and refinement procedures used in this study

A total of 367,327 particles were subjected to reference-free 2D classification. A subset of 347,241 particles from good 2D classes was selected for 3D classification using an initial model obtained from SIMPLE-PRIME[53], which was low-pass filtered to 60 Å. The data was divided into four 3D classes, two of which (a total of 179,079 particles) showed better features and were combined for refinement. This resulted in a 7.6 Å reconstruction. To further improve the reconstruction, these particles were subjected to beam-induced motion correction (particle polishing)[24]. Refinement of these polished particles with a soft mask around the rigid part of the map (as indicated by the red envelope) yielded a 5.9 Å reconstruction while refinement with a mask around the whole map yielded a 6.4 Å reconstruction. The polished particles were also subject to further 3D classification with a finer angular sampling of 1.8°. The most populated class (47,674 particles), which also has the best rotational accuracy, was refined with a soft mask around the whole density. This resulted in a 7.0 Å reconstruction. In this study, the 5.9 Å reconstruction was used for subsequent biological interpretation. All steps were performed in RELION[22] unless otherwise stated.

Extended Data Figure 3

CryoEM maps and tilt-pair validation

a, CryoEM density of the whole tri-snRNP at 5.9 Å resolution by “gold standard” Fourier Shell Correlation (FSC) of 0.143 criterion at two different contour levels. The high contour map (gold) shows well-resolved densities for protein and RNA helices and flat densities for beta-sheets. The low contour map (silver) shows densities for the more flexible head and arm. The map was sharpened by a B-factor of −214 Å2 and low-pass filtered to 5.9 Å as determined by RELION. b, The unsharpened full map of tri-snRNP. c, The map resulting from multi-body refinement, in which tri-snRNP is divided into four parts: the head, body, arm and foot. This resulted in better density for the arm domain (indicated by red circles), which is at 20 Å resolution. d, Tilt-pair validation plot for tri-snRNP. This was obtained from 1196 particles from 32 micrograph pairs, imaged at 0° and 10° tilt angles. The position of each dot represents the direction and the amount of tilting for a particle pair in polar coordinates. Blue dots correspond to in-plane tilt transformations; red and purple dots correspond to out-of-plane tilt transformations. Blue dots cluster in the same region of the plot at a tilt angle of approximately 10° as indicated by the red circle.

Extended Data Figure 4

Resolution estimation of tri-snRNP map

a, Local resolution of the tri-snRNP map estimated by ResMap using the color scheme shown in panel c. b, Local resolution of the tri-snRNP map calculated by “gold-standard” FSC. For each component of the map that we modeled protein/RNA components, a soft mask (with a 30-pixel soft-edge) surrounding the region of interest was prepared and used for FSC calculations. Convolution effects of the masks on the FSC curves were corrected using high-resolution noise substitution[55]. Resolution was estimated at FSC=0.143. Local resolution for the unmodeled region of the map (in red) was not estimated. c, Local resolution of model versus map. The map of each modeled component was extracted from the map using a soft mask (with a 5-pixel soft-edge) surrounding the component. The model was converted into density by EMAN[57]. FSC of model versus map was calculated using Xmipp[56]. The map is colored according to resolution estimates based on a FSC threshold of 0.25. The lower resolution estimates from the FSC of model versus map compared to the estimates from ResMap and the gold-standard FSCs are explained by the nature of our models. Because of the limited resolution of our map, we did not perform full atomic refinement, but placed known crystal structures and homology models as rigid bodies in the map. d, Gold-standard FSC curves for the whole tri-snRNP map and some of its components calculated as described in b. e, FSC curves of model versus map for the whole model and some of the components. f, The full tri-snRNP map in which portions of the structure produced from crystal structures, homology modeling and de-novo building or unmodeled are colored as indicated.

Overall structure

Yeast U4/U6.U5 tri-snRNP has an overall Y-shape with a maximum dimension of approximately 300 Å (Fig. 1). The large domain of Prp8 (residues 885-1824) consisting of the Reverse Transcriptase-like (RT), Linker and type II Endonuclease-like domains[10] is located near the centre of the assembly and its crystal structure was fitted into the map as a rigid body[10] (Fig. 2; Extended Data Fig. 5a). The orientation of the RNaseH-like domain with respect to the large domain is inverted in tri-snRNP as compared with the Prp8-Aar2 complex[10] (Fig. 2d). Three segments of clear double stranded RNA density extending from Prp8 to the foot domain are assigned to coaxially stacked stems I and II, variable stem-loop and stem III of U5 snRNA (Fig. 3; Extended Data Fig. 6) connected to the U5 Sm core (Extended Data Fig. 5e). Snu114 shows a significant sequence similarity to eukaryotic translation elongation factor 2 (EF2)[11,26] comprising Domains I-V (Fig. 4; Extended Data Fig. 5c and 7). Homology models of each domain of Snu114 (residues 120-1008) were fitted individually into the density adjacent to Prp8 and U5 snRNA revealing a contact between Domain III of Snu114 and the RT domain of Prp8 (Fig. 4). The structure of the N-terminal 884 residues of Prp8 is still unknown. The N-terminal helix of the RT domain of Prp8 (RTα1)[10] extends further in tri-snRNP towards a bundle of four long helices. Another cluster of long helices, which makes close contact with the coaxially stacked stems I and II of U5 snRNA, is found in the vicinity (Figure 3b; Extended data Fig. 6c). The region containing residues 420-542 of Prp8 is known to interact with the N-terminal half of Snu114 (ref. 27) and 4-thiouridine introduced at C79 of U5 snRNA cross-links with both Prp8 and Snu114 (ref. 28), suggesting that the density adjacent to Snu114 and U5 snRNA is part of the Prp8 N-terminus. At the tip of stem I the density assigned to the U5 loop I extends towards the RT Thumb/X domain of Prp8 and makes close contact with a thioredoxin-like fold of Dib1 (Fig. 2c; Extended Data Fig. 5f) (ref. 29). This is in good agreement with the fact that a 16 kDa protein is cross-linked to 4-thiouridine incorporated at U97 in the U5 snRNA loop I (ref. 28). The binding of Dib1 is further stabilized by the N-terminal helices of Prp8 (Fig. 2c).

Figure 1

Overview of the U4/U6.U5 tri-snRNP structure with its protein and RNA components modeled into cryo-EM density

a, Front view, facing concave surface; b, back view; c, top view. d, 2D class average showing the different domains of tri-snRNP: head, body, arm and foot.

Figure 2

Prp8 in tri-snRNP

a, Domain organization of Prp8. The structure of the N-terminal domain (residues 1-884) is unknown. RT, Reverse transcriptase-like domain; X, Thumb/X; L, linker, RH, RNaseH-like; JM, Jab1/MPN domains. b, The large domain of Prp8 is located at the centre of tri-snRNP. The Jab1/MPN domain is bound to Brr2 (ref. 31,32). c, Inset, Loop 1 of U5 snRNA is inserted to the active site cavity and in contact with Dib1. d, Prp8 in the Prp8/Aar2 complex[10] is shown with its large domain in the same orientation as in b. In tri-snRNP, the RNaseH domain is inverted while the Jab1/MPN domain in complex with Brr2 is located at the opposite end of the large domain.

Extended Data Figure 5

Fitting of protein components into tri-snRNP map

a, Prp8885-2413 crystal structure[10] (PDB ID: 4I43, green) and additional helices built de novo assigned to the N-terminus of Prp8 (blue). b, Brr2-Jab1/MPN complex[31] (PDB ID: 4BGD). c, Snu114 homology model based on EF2 (ref. 26). d, The Prp6 TPR motifs built into the tri-snRNP map. e, U5 Sm proteins (grey) with Sm site (blue) based on the human U4 Sm structure (PDB ID: 4WZJ). f, Dib1 (ref. 29) (PDB ID: 1QGV). g, i, Prp31. ii, Comparison between the crystal structure of human Prp3178-333 (ref. 33) (PDB ID: 2OZB, grey) and that in tri-snRNP (yellow and blue). The coiled-coil domain (yellow) rotates by 60° in tri-snRNP with respect to the Nop domain (grey). Additional helices (blue) that extend from the N- and C-termini were built. h, U4 Sm proteins with part of U4 snRNA (blue) based on the human U4 Sm structure. i, Prp3 model. The ferredoxin-like domain was obtained from homology modeling while the extra helices were built de novo. j, Prp4 WD40 homology model with the extra helices built de novo. k, Snu13 (ref. 64) (PDB ID: 2ALE). l, U6 LSm proteins[67] (PDB ID: 4M77).

Figure 3

The snRNA components of U4/U6.U5 tri-snRNP

a, Secondary structures of U4/U6 and U5 snRNAs. b, Double-stranded regions of U4/U6 and U5 snRNAs modeled into the cryo-EM map.

Extended Data Figure 6

Fitting of the RNA components in tri-snRNP map

a, and c, The sequences and predicted secondary structures of U4/U6 snRNA and the long version of U5 snRNA, respectively. b, and d, The maps of the fitted parts of U4/U6 snRNA and U5 snRNA, respectively. Unmodeled density assigned to U5 snRNA was also shown in d.

Figure 4

Structure of Snu114 in tri-snRNP

a, Location of Snu114 in the U4/U6.U5 tri-snRNP. b, Arrangement of Domains (I-V) in Snu114 (see Extended Data Fig. 7). c, Domain arrangement in EF-G bound to the ribosome[44]. d, The interface between the N-terminal domain of Prp8 and Snu114. Some of the uninterpreted density at the interface may be attributed to the unmodeled switch I loop. e, The interaction of the switch region of EF-G with the sarcin-ricin loop[44] for GTPase activation.

Extended Data Figure 7

Sequence alignment of yeast and human Snu114 with yeast and human elongation factor 2 (EF-2)

The secondary structures of our homology model for yeast Snu114 and the yeast EF-2 (ref. 26) (PDB ID: 1N0V) are shown on the top and bottom of the alignment, respectively. Important sequence elements are also shown.

Brr2 forms a stable complex with the Jab1/MPN domain of Prp8 (ref. 30-32) and its characteristic shape was recognized in the less ordered head domain (Fig. 1 and 5; Extended Data Fig. 5b). Although this part of the map is lower in resolution the individual domains of yeast Brr2 were fitted into the density together with the Jab1/MPN domain[31]. This revealed a widening of the gap between the two RecA domains of the N-terminal cassette (Fig. 5c). Coaxially stacked stems I and II of U4/U6 snRNAs and the 5′-stem-loop (5′SL) of U4 snRNA branching from the 3-way junction are unambiguously identified near the N-terminal helicase cassette of Brr2 and Prp8 (Fig. 1 and 3; Extended Data Fig. 6). Snu13 and the Nop domain of Prp31 bind to the kink-turn at the tip of the U4 snRNA 5′SL[33]. Initially the crystal structures of Snu13 and Prp31 were fitted individually around the kink-turn but the density clearly showed that the coiled-coil domain of Prp31 is rotated by approximately 60° with respect to the Nop domain in tri-snRNP (Fig. 1; Extended Data Fig. 5g). The four-helix bundle of the coiled-coil domain is in contact with the RT domain of Prp8 (Fig. 1). Furthermore, Prp31 used for crystallization was a truncated form [33] and in our map clear α-helical density extends from both N- and C-termini (Extended Data Fig. 5g). The C-terminal helix extends from the 5′SL to the three-way junction, in agreement with previous foot-printing experiments[34] (Fig. 1a). Prp4 contains seven WD repeats at the C-terminus[35] and its characteristic seven-bladed β-propeller domain is packed against Snu13 and the junction between the Nop and coiled-coil domain of Prp31 (Fig. 1 and 5, Extended Data Fig. 5j). Four additional α-helices, likely belonging to the N-terminus of Prp4, were built into the rod-like density on top of the β-propeller. Prp3 is predicted to have a ferredoxin-like domain at its C-terminus[36] and interacts with Prp4 and U4/U6 stem II (ref. 37). Density sandwiched between the Prp4 WD40 domain and U4/U6 stem II, two long helices lying along U4/U6 stem II and a number of connected helices nearby are likely to belong to Prp3 (Fig. 1; Extended Data Fig. 5i). The U4 core domain is wedged between the tandem helicase cassettes of Brr2 (Fig. 1a, 1c and 5b; Extended Data Fig. 6a). The 3′SL of U4 snRNA contacts the helix-loop-helix domain of the N-terminal helicase cassette (Fig. 5b) which contains several lysine/arginine residues close to the RNA backbone of the 3′SL of U4 snRNA. Based on previous labeling data[19] U6 LSm proteins are fitted into the flexible arm region in the multi-body refined map (Fig. 1a, Extended Data Fig. 3c and 5l).

Figure 5

Brr2 mode of unwinding

a, Domain organisation of Brr2 N-terminal helicase cassette (NHC). WH, Winged-helix; HLH, helix-loop-helix; and FN3, fibronectin3-like domains. The inactive C-terminal helicase cassette (CHC) has the same domain organisation. b, U4/U6 di-snRNP and its interaction with Brr2 in tri-snRNP. The domains of Brr2 NHC are coloured as in a. The single-stranded RNA between U4/U6 stem I and U4 3′SL is already loaded in the active site of Brr2. When the Hel308 structure[43] is overlaid onto the NHC of Brr2, its 10-nucleotide DNA substrate coincide with the density in the Brr2 active site, which extends to U4 snRNA 3′SL (red dotted line). The helix-loop-helix domain of Brr2 interacts with U4 snRNA 3′SL (inset). c, Superposition of the RecA1 domain of Brr2 in the crystal structure[31] (PDB ID 4BGD, in grey) and in tri-snRNP (domains coloured as in a) shows the opening of the gap between the RecA1 and RecA2 domains (indicated by the red arrow) to accommodate the RNA substrate.

A striking elongated curved α-solenoid density bridging the RNaseH-like domain of Prp8 and the WD40 domain of Prp4 is assigned to the tetratricopeptide (TPR) motifs of Prp6 (Fig. 1b-c; Extended Data Fig. 5d). Prp6 is required for the accumulation of tri-snRNP[38] and is proposed to act as a bridge between U5 and U4/U6 snRNPs[16,39]. Prp6 contains up to 19 predicted TPR motifs, each comprising a helix-loop-helix motif[39] and 37 connected idealized poly-Ala helices were built into the map. Nine canonical tandem TPR motifs at the C-terminus of the protein form a highly curved alpha-helical solenoid-like structure, which contacts Snu13, U4 snRNA 5′SL and the Prp4 WD40 domain in tri-snRNP (Fig. 1b-c). This is consistent with the fact that antibodies against the C-terminal fragment of human Prp6 immunoprecipitate U5 snRNP but not tri-snRNP, as the C-terminal domain in our structure is in close contact with U4/U6 snRNP, which presumably occludes the epitope[39].

Central role of Prp8 in tri-snRNP assembly

Our single particle cryoEM reconstruction of yeast U4/U6.U5 tri-snRNP has revealed a nearly complete organization of its RNA and protein components although some densities remain unassigned, and Snu66, Snu23, Prp38 and possibly substoichiometric Spp381 are yet to be located (Extended Data Table 1; Extended Data Fig. 4f and 8d-e). Prp8 positioned at the centre of the assembly functions as a hub of protein-protein and protein-RNA interactions holding the whole assembly together (Fig. 1 and 2b). In yeast, a stable Prp8-Snu114-Aar2-U5 core domain complex is imported into the nucleus[40], where Brr2 replaces Aar2. The Jab1/MPN and RNaseH-like domains, held tightly onto the Prp8 large domain by Aar2 (Fig. 2d), are released in tri-snRNP wherein the Jab1/MPN domain forms a stable complex with Brr2 as in the crystal structure[31,32] (Fig. 1b and 2b; Extended Data Fig. 5b). The tri-snRNP structure provides the first glimpse of the interaction between Snu114, the U5 core domain and the N-terminal domain of Prp8 which holds the co-axially stacked stems I and II, and variable stem-loop of U5 snRNA (Fig. 3). On the opposite side of Prp8, Snu13 and Prp31 firmly bound to 5′SL of U4 snRNA[33], the U4 Sm protein assembly, the Brr2-Jab1/MPN domain complex, the Prp3/Prp4 complex, the RNaseH domain of Prp8 and the U4/U6 snRNA duplex assemble together (Fig. 1 and 2).

Extended Data Figure 8

The effect of ATP on Brr2-TAPS purified tri-snRNP

a, Ethidium-bromide stained native agarose gel (0.5%) showing the effects of ATP addition to Brr2-TAPS purified tri-snRNP used in this study. Upon ATP addition either without or with GTP/GDP, tri-snRNP fell apart (lanes 1-4). Under the same conditions, the addition of ADP or the non-hydrolysable ATP-analogue, AMPPNP had no effects on the complex (lanes 5-6). b, and c, The effect of ATP addition observed by negative stain microscopy. When ATP was not present, tri-snRNP particles could be observed. When ATP was added to the sample prior to grid preparations, tri-snRNP particles fell apart as observed by many small components on the micrograph rather than tri-snRNP particles. d, Tri-snRNP model where U4/U6 snRNP proteins are not shown. In tri-snRNP, Brr2/Prp8Jab complex is loosely associated to the remaining of U5 snRNP components including Prp8large, Prp8RNaseH, Prp8Nterm, Snu114, Dib1, U5 Sm proteins and U5 snRNA. After U4/U6 snRNA unwinding by Brr2, Brr2/Prp8Jab could be repositioned within the spliceosome. e, A schematic showing the arrangement of tri-snRNP protein and RNA components.

Prp8 has a surface on which are exposed the 5′SS-binding ACAGAGA sequence of U6 snRNA, the U6 sequences which pair with U2 snRNA, and U5 snRNA loop I which interacts with exon 1 and exon 2. This surface is partly occluded by a highly conserved protein Dib1 (Fig. 2c and 6), suggesting its potential role in regulating the incorporation of RNA components into the active site cavity during spliceosome assembly and activation. When U4 and U6 snRNAs are unwound, releasing U4 snRNA from the spliceosome together with Snu13, Prp31, Prp3/Prp4 (ref. 3), Brr2/Jab1MPN are no longer held in place and could be repositioned during catalysis and spliceosome disassembly (Fig. 5; Extended Data Fig. 8d).

Figure 6

Insights into activation mechanism and the active site of the spliceosome

a, Mapping of the U4-cs1 suppressor mutations on the surface of Prp8. Three clusters of mutations are found in close proximity to the key elements of spliceosomal activation: Prp31/U4snRNA 5′SL, Snu114, and ACAGAGA box of the U6 snRNA. b, A model of the catalytic core of group II intron docked into the active site cavity by superposition of the EBS1 stem of group II intron (PDB ID 3IGI) and stem I of the U5 snRNA.

Brr2 mode of action during activation

The unwinding of the U4/U6 snRNA duplex is an essential step in spliceosomal activation and is catalyzed by Brr2 (ref. 14,15). Like other Ski2-like helicases, Brr2 unwinds any RNA duplex with 3′ overhangs[41]. The U4/U6 snRNA duplex has 3′ overhangs on both ends (Fig. 3a; Extended Data Fig. 6a) and it has been suggested that Brr2 binds to the single stranded region of U4 snRNA and translocates along U4 snRNA[41,42]. Our structure shows that Brr2 is pre-loaded onto the single-stranded region between U4 snRNA 3′SL and stem I of the U4/U6 duplex, showing definitively that it translocates along U4 snRNA. The gap between the two RecA domains is widened in tri-snRNP and the prominent separator β-hairpin is located adjacent to stem I of the U4/U6 snRNA duplex (Fig. 5b-c). Our purified U4/U6.U5 tri-snRNP disintegrates upon addition of ATP regardless of the presence of GTP or GDP but remains intact after addition of ADP or AMPPNP, a non-hydrolyzable ATP analogue (Extended Data Fig 8). This shows that Brr2 is in an active state in our purified U4/U6.U5 tri-snRNP in perfect agreement with the structure. In vitro the RNaseH domain binds to the forked region of the U4/U6 snRNA duplex adjacent to stem I and inhibits Brr2 binding to the substrate RNA[41]. In tri-snRNP the RNaseH domain fails to prevent substrate loading onto Brr2 or unwinding. The helix-loop-helix domain of the Brr2 N-terminal helicase cassette interacts with the 3′SL of U4 snRNA and this interaction may be important for positioning U4 snRNA in the Brr2 active site[41,42]. Because of the limited resolution we cannot model the single-stranded region of U4 snRNA de novo. However, when we superpose the N-terminal helicase cassette of Brr2 onto the Hel308 structure with partially unwound DNA duplex[43], ten nucleotides of the substrate DNA coincide with the extra density in the Brr2 active site and six additional nucleotides can be accommodated in the density extending further to the 5′ end of the U4 3′ SL (Fig. 5b).

Role of Snu114 in spliceosome activation

Snu114 shows substantial sequence similarity to EF2 (Fig. 4; Extended Data Fig. 7) suggesting that it might induce conformational change in the spliceosome upon GTP binding or hydrolysis and regulate spliceosomal activation[11-13]. EF-G, bacterial counterpart of EF2, enters the ribosome in the GTP-bound form. Its GTPase is activated when switch regions I and II are remodeled upon interacting with the sarcin-ricin loop of 23S rRNA[44] (Fig. 4e) and GTP hydrolysis leads to translocation[45]. The activation process of the spliceosome has not been dissected in detail and it is not known at what stage GTP is hydrolysed or how Snu114 GTPase is activated. Snu114 and EF2 share highly similar switch I and II sequences, including the critical His residue, which places a water molecule adjacent to the γ-phosphate in EF-G. Unassigned density connecting the junction between stem I and II of U5 snRNA and the switch I and II loops coincides with the position of the sarcin-ricin loop (Fig. 4d). This is likely to be the N-terminal domain of Prp8 which may play a role in the activation of GTP hydrolysis.

Prior to the unwinding of the U4/U6 duplex, the 5′SS sequence pairs with the ACAGAGA sequence in U6 snRNA[2]. The U4-cs1 cold sensitive mutation, which extends U4/U6 stem I at the restrictive temperature and sequesters the ACAGAGA box from the 5′SS, stalls the spliceosome prior to unwinding[46] (Extended Data Fig. 6a). A suppressor of U4-cs1 has a duplication of the ACAGA sequence in U6 snRNA[47]. This shows that pairing of the ACAGAGA sequence with the 5′SS is a checkpoint to ensure proper assembly of Complex B prior to the unwinding of the U4/U6 snRNA duplex. Interestingly suppressors of U4-cs1 in Prp8 form three clusters on the surface of the large domain of Prp8 (Fig. 6a; Extended Data Table 2)[10]. In tri-snRNP one of these clusters is located at the interface between the RT domain and Domain III of Snu114, and another is at the interface with Prp31 and the junction between the RT and N-terminal domains of Prp8, showing this checkpoint can be bypassed when these subunit interfaces are tampered with (Fig. 6a). This suggests that the interactions between these components undergo allosteric changes, which possibly couple the guanine-nucleotide binding state of Snu114 and the pairing between 5′SS and the ACAGAGA sequence to the activation of the U4/U6 duplex unwinding. Understanding the activation process will require extensive interplay between structural and biochemical work; the tri-snRNP structure provides an important structural framework for further investigation of this process.

Extended Data Table 2

U4−CS1 suppressors

region	mutations	domains	locations	contact
region a	R236G	N-terminal domain	unknown
	L261P
	L280P
region b	K611R	N-terminal domain	unknown
	E624G
	N643S
	V644A
	D651G or N
	H659P
	K684E
region c	E788G or V	N-terminal domain	unknown
	N796S
	W856R
	E860K
	Q861R
cluster 1	D1094A or N or V	RT domain	loop in 4 stranded β-sheet	Interface with Prp31
	M1095T		loop in 4 stranded β-sheet
	V1098D		near in 4 stranded β-sheet
	N1099K		near in 4 stranded β-sheet
	I1104M		near in 4 stranded β-sheet
	R1105L		near in 4 stranded β-sheet
cluster 2	P1191L or 5 or T		within loop following α12	interface with Snu114 domain III
	D1192Y		within loop following α12
	N1194D		within loop following α12
cluster 3	L1624M	endonuclease	top surface
	L1634F		top surface
	L1641F		top surface
	T16851		top surface
	P1688L or R		top surface
	A1754V		side surface
	N1809D		side surface
region f	F1851L	RNaseH	on inner surface
	V1860D or N		β-finger
	T1861P		β-finger
	V1862A or D or Y		β-finger
	I1875T		β-finger

All suppressor mutants are described in Kuhn & Brow[37] and Kuhn et al.[75].

The structural resemblance between the Group II intron active site[48] and the catalytic RNA core of the spliceosome[4,49] endorsed the hypothesis that they evolved from a common ancestor[50]. Based on the similarity of the domain architecture between the group II intron encoded protein (IEP) and Prp8, we proposed that Prp8 evolved from IEP and recruited more domains and interacting proteins to assemble spliceosomal snRNAs[10], which derived from fragmented group II intron[50]. We placed the catalytic core of Group II intron RNA in the tri-snRNP structure by superimposing its exon binding stem-loop[48] onto stem I of U5 snRNA. The Group II intron catalytic core fits neatly into the active site cavity after removal of Dib1, which is absent from activated spliceosomes (Fig. 6b), and with small rearrangement it can make contacts with the Thumb/linker region of Prp8, which cross-links with the catalytic RNA core in the spliceosome [9,10]. The structure of tri-snRNP illustrates beautifully how the Prp8 domains and other spliceosomal proteins come together to assemble snRNAs and insert their functional segments into the active site cavity of the spliceosome[10].

Methods

Brr2-TAPS tagging for yeast U4/U6.U5 tri-snRNP purification

Primers specific for 55 nucleotides of the C-terminus and 3′ UTR of BRR2 were used to PCR-amplify the TAPS-tag cassette together with the KanMX6 gene from pFA6a-TAPS-kanMX6, a modified version of pFA6a-TAP-kanMX6 in which the Calmodulin-binding peptide tag is replaced by 2 tandem copies of the StrepII tag[51]. The PCR product was used to transform yeast strain BCY123 [MATa pep4::HIS3 prb1::LEU2 bar1::HIS6 lys2::GAL1/10-GAL4 can1 ade2 trp1 ura3 his3 leu2-3, 112] by homologous recombination, selecting for G418-resistance. C-terminal TAPS-tagging of Brr2 was confirmed by PCR analysis of genomic DNA and DNA sequencing.

Sample preparation

The Brr2-TAPS-tagged yeast cells (72 Liters) were grown in YEPD medium to OD600 of 3.5, harvested and resuspended in lysis buffer (100 mM HEPES KOH pH 8.0, 200 mM KCl, 2 mM Mg(OAc)2 and 10% w/v glycerol). The cells were frozen and lysed by a Freezer Mill 6870 (SPEX CertiPrep). The crude lysate was centrifuged at 45krpm for 1 hour. The resulting supernatants were incubated with IgG sepharose overnight at 4 °C. The resin was washed with TAPS wash buffer (20 mM HEPES KOH pH 7.9, 150 mM KCl, 1 mM Mg(OAc)2 and 0.1% NP40) and incubated with TAPS wash buffer in the presence of TEV protease at 4 °C overnight. The flow-through was collected and incubated with Streptactin resin (GE Healthcare) for 3 hours. The resin was washed with TAPS wash buffer and particles were eluted with Strep elution buffer (20 mM HEPES KOH pH 7.9, 150 mM KCl, 1 mM Mg(OAc)2, 0.1% NP40, 5 mM desthiobiotin). The eluate was subsequently applied to a 10-30% v/v glycerol gradient centrifuged at 210,000g at 4 °C in a SWTi60 rotor. The fractions from the gradient were analysed by SDS-PAGE for protein composition. Glycerol was removed from the peak fractions containing tri-snRNP by dialysis against B150 buffer (20 mM HEPES KOH pH 7.9, 150 mM KCl, 1 mM Mg(OAc)2) prior to EM sample preparation (Extended Data Fig. 1a and 1b).

Electron microscopy

For cryo-EM analysis, 3.5 μl of the tri-snRNP sample was applied to Quantifoil R2/2 or R1.2/1.3 grids which were previously coated with a 6 nm thick layer of homemade carbon film and glow-discharged (Extended Data Fig. 1c). The grids were blotted for 2 s at 4 °C and plunged into liquid ethane using an FEI Vitrobot MKIII. The grids were loaded onto a Tecnai F30 Polara transmission electron microscope operated at 300 kV. Images were collected manually in low-dose mode at a calibrated magnification of 79,096×. The micrographs were recorded on either a Falcon II or a (ultra back-thinned) Falcon III detector at the same calibrated pixel size of 1.77 Å in movie mode at 17 frames/s. A total dose of 40e/ Å2 over 2.5s, and a defocus range of 2-4 μm were used.

Data processing

Most steps of data processing were performed in RELION[22] unless otherwise stated. The 42 movie frames for each micrograph were corrected for whole-image drift using MOTIONCORR[21], and contrast transfer function (CTF) parameters were estimated from the resulting micrographs using CTFFIND3 and CTFFIND4 (ref. 52). A subset of 5000 particles was picked manually and extracted with a 2802 pixel box, followed by reference-free 2D classification to obtain initial 2D class averages, which were then used as references for automatic particle picking. Particles resulting from the first round of automatic picking were extracted with a 2802 pixel box for reference-free 2D classification to obtain better references for the next round of automatic particle picking. All particles from the second round of automatic particle picking were manually checked before extracting them for reference-free 2D classification (Extended Data Fig. 1d, e). Prior to both autopicking runs, the templates were low-pass filtered to 20 Å to prevent high-resolution noise bias. A total of 347,241 particles from 2035 micrographs were selected from good 2D classes, and these particles were used for subsequent 3D processing (Extended Data Fig. 2).

A subset of 18,000 particles from only the best 2D classes of one of the datasets was used for ab initio 3D reconstruction by SIMPLE-PRIME[53] to obtain an initial model of the complex, which was low-pass filtered to 60 Å for 3D classification. 3D classification with four classes was run for 25 iterations, using an angular sampling of 7.5° and a regularization parameter T of 4. This resulted in two classes with much better reconstructed features than the others. These classes were combined into a subset of 179,079 particles. Auto-refinement of these particles resulted in a 7.6 Å reconstruction. These particles were subsequently used for particle based beam-induced movement correction. For these calculations, we only used the first 30 frames of each micrograph with running averages of seven frames and a standard deviation of 1 pixel for translational alignments. We used the new “particle-polishing” approach, which fits linear tracks through the optimal translations for all running averages and takes into account the movements of neighbouring particles on the micrographs, to further improve the accuracy of the particle-based movement corrections[24]. The B-factors for the resolution and dose-dependent model for radiation damage were estimated using reconstructions from running averages of 3 frames.

Auto-refinement of the movement-corrected particles with a soft mask (with 12-pixel fall-off) around the entire map resulted in a map at 6.4 Å resolution while refinement with a similar soft mask around the more rigid part resulted in a map at 5.9 Å resolution (Extended Data Figure 2 and 3a). The 5.9 Å map was used for interpretation. Our map was also validated by a tilt-pair test (Extended Data Fig. 3d). Local resolution analysis showed a wide range of resolution from 5.0 Å to 20 Å (Extended Data Fig. 4a), indicating flexibility within some parts of the structure. Further 3D classification with a finer angular sampling interval of 1.8° and local angular search range of 10° revealed conformational heterogeneity of the head, body and foot domains of the structure and did not improve the overall resolution of the map (Extended Data Fig. 2).

We used a modified refinement approach in RELION, which we term “multi-body refinement”, to improve the density for the flexible arm domain (Extended Data Fig. 3b and c). In this approach, we used masks to divide the reference map into four “bodies”, approximately corresponding to the body, head, foot and arm domains. In each iteration of the auto-refine procedure, we independently aligned every experimental particle image against projections from the four distinct bodies. To minimize errors in these alignments, prior to the alignment against a given body we subtracted projections from the other three bodies from the experimental particle. Because we assume that the four bodies may adopt different relative orientations in each particle, we kept track of the most likely orientation for each of the four bodies for every particle during the course of the refinement. Thereby, subtraction of the other bodies should become ever more accurate, and this resulted in four sets of relative orientations for each particle. To express our expectation that the relative movements between the different bodies were limited, we used only local orientation searches (±22.5°) in the multi-body refinement, and centered the local searches around the orientations determined for the unmasked auto-refinement mentioned above. Details of this methodology will be described elsewhere (S.H.W.S. unpublished results), and its implementation will be made available through incorporation into RELION.

All refinements used gold standard Fourier shell correlation (FSC) calculations[25] and reported resolutions are based on the FSC = 0.143 criterion. The FSC curves are calculated using a soft spherical mask (Extended Data Fig. 4d). Prior to visualization, all maps were corrected for the modulation transfer function of the detector and sharpened by applying a negative B-factor.

Local resolution analyses

Local resolution analyses were performed by Resmap[54] and compared with that calculated by us for each protein/RNA component of the map (Extended Data Fig. 4a-c). For the latter calculations, FSC curves are calculated using a soft spherical mask (with a 30-pixel fall-off) around each protein/RNA component of interest. Convolution effects of the masks on the FSC curves were corrected using high-resolution noise substitution[55]. Resolution was estimated at FSC=0.143. These calculations were performed for each of the following components: Prp8 Large Domain, Prp8 RNaseH domain, Prp8 Jab domain, Prp8 N-terminal domain, Brr2, U4 Sm with U4 3′ SL, U5 Sm with Sm site, Snu114, Dib1, Prp6, Prp3, Prp4, Snu13, Prp31, LSm, U5 snRNA and U4/U6 snRNA. Extended Data Fig. 4d shows some representative curves from these calculations.

FSC curves of model versus map were calculated using the Xmipp package[56] and the reported resolutions were based on the FSC = 0.25 criterion. FSC curves of model versus map were calculated for not only the entire model of all components but also different parts of the maps. The map of each modelled component was extracted from the tri-snRNP map using a soft mask (with a 5-pixel soft-edge) surrounding the component. A map of each model was created by the program pdb2mrc within the EMAN package[57]. Some proteins/domains that are close together were grouped together for these calculations, including Prp8 N-terminal domain/Dib1, Brr2/Prp8-Jab domain, Prp3/Prp4 and Prp8 N-terminal domain/Dib1. Extended Data Fig. 4e shows some representative curves from these calculations.

Model fitting and building

Locations of available X-ray or homology models were fitted initially by visual inspection of the tri-snRNP map and low-pass filtered maps (to 10 Å) were generated for each model in Chimera followed by fit optimization in Chimera[58]. For the LSm proteins that are in the flexible arm region, the map resulting from multi-body refinement (Extended Data Fig. 3c and 5l) was used for fitting. Further rigid-body fitting was performed in Coot[59]. The homology model for Snu114 was prepared by I-TASSER web server[60] based on the crystal structure of the yeast Elongation Factor 2 (ref. 26; PDB ID: 1N0V) (Fig. 4, Extended Data Fig. 5c). The model was manually inspected and the disordered regions were removed. The model for the ferredoxin-like domain of Prp3 is available at the Yeast Genome Center (http://www.yeastrc.org), which contains its structure predictions[61]. The model with the highest Mammoth Confidence Metric (MCM) score was selected for fitting. For Prp4, the protein sequence was input into Robetta Beta Full-chain Protein Structure (http://robetta.bakerlab.org), which yielded a model for the C-terminal part of Prp4 based on the structure of the WDR5 protein[62] (PDB ID: 3MXX). Double stranded RNA helices and idealized poly-alanine helices were built into the masked map in Coot when possible. U4 snRNA 5′SL was modeled based on the structure of the human hPrp31-15.5K-U4snRNA complex[33] (PDB ID: 2OZB) using ModeRNA modeling tool[63]. Yeast Snu13 structure[64] (PDB ID: 2ALE) was fitted into the map. U4 snRNA 3′SL partial model was adapted from the structure of the human U4 snRNP core domain[65] (PDB ID: 4WZJ). The short and long forms of U5 snRNA[66] are present in our sample (Extended data Fig. 1b) but no density for 3′ stem-loop (3′SL) was observed, presumably because 3′SL attached to the Sm site with a long single stranded stretch is disordered or the particle population with the long U5 snRNA is classified out during classification. U5 snRNP Sm core with only the Sm site was also adapted from the human U4 snRNP core domain. The LSm proteins[67] (PDB ID: 4M77) were placed in the low-resolution arm region of the map with the flat surface of the LSm complex facing the entrance side of U6 snRNA. The register of the LSm proteins cannot be accurately determined. Human Dib1 structure[29] (PDB ID: 1QGV) was used for fitting. Extended Data Table 1 and Extended Data Fig. 4f summarise all the details of tri-snRNP components and modeling in our study. The active site cavity of Prp8 was described previously[10] and defined by cross-links with crucial elements of U5 snRNA, U6 snRNA and pre-mRNA[9] and suppressors of defective splice site mutations[68-74]. U4-cs1 mutants have been described[46,75]. Extended Data Table 2 summarises all the U4-cs1 mutants and their locations in tri-snRNP.

Map and model visualization

Maps were visualized in Chimera[58]. Map segmentation was performed in Chimera using each of the fitted models and the “zone-masking” function (Fig. 1, Extended Data Fig. 5, 6b and 6d). The LSm protein density was obtained from multi-body refinement and low-pass filtered to 20 Å. For all the remaining components, the sharpened tri-snRNP map (B= −214 Å2) low-pass filtered to 5.9 Å was used. Figures were generated using either Chimera[58] or Pymol (www.pymol.org) and the video was made in Chimera[58].

ATP assays

Purified tri-snRNP from glycerol gradient (~25 nM) was incubated at 30 °C for 30 min in the presence of either no nucleotide or with each of the following nucleotide combinations: ATP, ATP/GTP, ATP/GDP, ADP and AMPPNP (1 mM each). The samples (10 μl) were loaded onto a native agarose gel (0.5% in TB buffer supplemented with 1 mM MgCl2) and run at 75 V at 4 °C for 2.5 hours. The gel was stained with ethidium bromide for 1 hour before being imaged by a Syngene UV imager (Extended Data Fig. 8a). For negative staining, the sample was also treated similarly and stained with 2% uranyl acetate prior to imaging on a Tecnai T12 transmission electron microscope operated at 120 kV (Extended Data Fig. 8b and c).

The architecture of the spliceosomal U4/U6.U5 tri-snRNP.

The video sequences showing the cryoEM density at two different contour levels; tri-snRNP map with all modeled components; fitting of available crystal structures into the cryoEM density: Brr2-Jab1/MPN (Prp8) complex[31], Prp8 RNase H and large domains[10], U4 and U5 Sm core domains[65], Lsm core domain[67] fitted into the multi-body map, Snu13 (ref. 64), human Prp31 (ref. 33) with remodeling, human Dib1 (ref. 29); fitting of homology models: Snu114 based translation factor EF2 (ref. 26), WD40 domain of Prp4, ferredoxin-like domain of Prp3 (ref. 36), TPR domain of Prp6; fitting of double helical RNA of the U4/U6 snRNA duplex and U5 snRNA; fitting of α-helices attributed to the N-terminal domain Prp8, Prp3 and Prp4; near complete pseudo-atomic structure of the yeast U4/U6.U5 tri-snRNP.