Crystal structure of human U1 snRNP, a small nuclear ribonucleoprotein particle, reveals the mechanism of 5′ splice site recognition

1Removal of introns from pre-messenger RNA (pre-mRNA) is an essential step in eukaryotic gene expression. This process is catalysed by a large and dynamic RNA-protein assembly called the spliceosome, which consists of five small nuclear ribonucleoprotein particles (U1, U2, U4, U5 and U6 snRNPs) and numerous non-snRNP proteins. U1 snRNP recognizes a short sequence at the 5′-splice site (5′SS) of pre-mRNA through basepairing between the 5′-end of U1 snRNA and the 5′SS sequence and promotes an ordered assembly of the four remaining snRNPs to form the spliceosome, which then undergoes extensive conformational and compositional remodelings to become catalytically active. During activation, the interaction between U1 snRNP and the 5′SS is disrupted by RNA helicase Prp28 and then the 5′SS intron sequence base-pairs with part of the ACAGAGA box in U6 snRNA; whilst the 5′ exon interacts with U5 snRNA loop I for the first trans-esterification reaction. U1 snRNP is also an important regulator of mRNA 3′ end cleavage and polyadenylation (; reviewed in ).

2Human U1 snRNP comprises U1 snRNA, seven Sm proteins (SmB/SmB′, SmD1, SmD2, SmD3, SmE, SmF and SmG) and three U1-specific proteins (U1-70K, U1-A and U1-C). We reported the structure of the functional core of U1 snRNP based on an experimental electron density map at 5.5 Å resolution to which we fitted previously-determined structures of protein components. The most striking feature of the structure is the N-terminal region of U1-70k, which extends from its RRM through a long α-helix and wraps around the Sm protein assembly so that its N-terminus makes contact with U1-C protein, thus accounting for the requirement of U1-70k for U1-C binding. In this crystal the 5′-end of U1 snRNA pairs with its symmetry-related counterpart, mimicking the binding of the 5′SS of pre-mRNA to U1 snRNP. The Zn-finger domain of U1-C is located adjacent to this RNA duplex but the low-resolution map was insufficient for analysis of the RNA-protein contacts in atomic detail and hence it was not clear how U1-C contributes to the recognition of the 5′SS. The structure of U1 snRNP from HeLa cells, treated with chymotrypsin, was subsequently reported. Although this crystal had a DNA oligonucleotide with the 5′SS consensus sequence bound to the 5′-end of U1 snRNA, the N-terminal end of U1-70k together with U1-C protein were lost by protease treatment. Hence neither of these structures revealed molecular details of 5′SS recognition by U1 snRNP.

3In order to gain crucial insight into the mechanism of 5′SS recognition we continued our attempts to grow crystals of U1 snRNP diffracting to high resolution but this proved unsuccessful because the inherent mobility of long RNA helices arranged as a 4-way junction prevented the formation of well-diffracting crystals. Hence we designed two sub-structures of U1 snRNP, with exclusively human sequences, based on our 5.5 Å resolution structure and determined their crystal structures at high resolution.

4Yeast U1 snRNP, when compared to the human particle, contains a larger and more complex snRNA, which is associated with many protein factors (Prp39, Snu71, Prp40, Prp42, Nam8, Snu56, Urn1 and Prp5), which have no counterparts in human U1 snRNP. However, despite these differences, the sequence of the 5′-single stranded region of U1 snRNA (nts 1–10) is invariant from yeast to human (http://rfam.sanger.ac.uk/) and the amino acid sequence of the Zn-finger of U1-C (yeast Yhc1) is also highly conserved. Hence the 5′SS of pre-mRNA is expected to make exactly the same contacts with the 5′ end of U1 snRNA and U1-C in human and yeast U1 snRNPs. However, some positions of the 5′SS have quite different nucleotide bias in yeast and human genes. In human the 5′SS sequences processed by major spliceosomes are degenerate but show significant overall complementarity to the sequence of the 5′ end of U1 snRNA. In contrast the 5′SS intron sequence of yeast pre-mRNA is stringently conserved to be GUAUGU. Upon activation of the yeast spliceosome, the intron sequence, UGU (+4, +5 and +6), pairs with ACA within the ACAGAGA sequence in U6 snRNA and hence these nucleotides are selected to be nearly invariant. Cross-linking studies revealed interaction of the same regions of U6 snRNA and the 5′SS sequence in human but the sequence requirement is less obvious. Prp8 is also known to influence the selection of the 5′SS nucleotide (reviewed in; ). Furthermore in humans constitutive or alternative splicing factors facilitate the binding of U1 snRNP to weak splice sites. As discussed above, the 5′SS is subjected to multiple selections which differ in yeast and human and give rise to different nucleotide biases at 5′SS. The strength of variant 5′SS sequences is assessed by relative usage of competing 5′SS which is determined not only by the affinity of 5′SS to U1 snRNP but also by multiple factors.

5Our two new crystals together reveal the structures of the substantial parts of U1 snRNP at high resolution and provide crucial insights into the mechanism of pre-mRNA recognition by U1 snRNP. In particular, we find that U1-C makes no base-specific contacts with the 5′SS sequence. Also, by measuring the intrinsic affinity of recombinant U1 snRNP for various 5′SS sequences we disentangle the role played by the U1 snRNP from the other complexities of 5′SS recognition, and assess the relative contributions of U1 snRNA and U1-C protein in light of our crystal structure.

6A minimal U1 snRNP consisting of seven Sm proteins, the N-terminal peptide of U1-70k, U1-C and a truncated U1 snRNA was designed based on the 5.5 Å resolution structure (Figure 1A; Figure 1—figure supplements 1 and 2). A large portion of RNA attached to Helix H through the 4-way junction was replaced by a kissing-loop to facilitate crystal contacts (Figure 1—figure supplement 2). However, this eliminated the binding site (stem-loop I) for the U1-70k RRM (RNA-recognition motif) and thus weakened the binding of the U1-70k N-terminal peptide to the snRNP core domain. In order to stabilize its binding we fused the N-terminal 59 residue peptide of U1-70k to SmD1 via a Gly–Ser linker (70kSmD1F, Figure 1—figure supplement 1). The reconstituted complex was stable and yielded crystals diffracting to 3.3 Å (Table 1). The second crystal (U1A70kF-RNA) contains residues 60–216 of U1-70k and the entire U1 snRNA stem-loop I. In order to promote interaction of the long α-helix of U1-70k with the RNA stem we fused it to the U1-A RRM and capped the RNA with the apical loop of stem-loop II such that the α-helix is anchored to the RNA stem (Figure 1—figure supplements 1 and 2). The crystal structure of this complex was determined at 2.5 Å (Figure 1B; Table 1). Phases for both crystal structures were determined by molecular replacement.

7The structure of the minimal U1 snRNP fits well into the 5.5 Å electron density except for a slight tilt of the duplex between pre-mRNA and the 5′-end of U1 snRNA (Figure 1C). In the previous structure the fortuitous interaction between the 5′-end of U1 snRNA from two symmetry-related complexes could have distorted the orientation of this RNA duplex. Hence we believe that the minimal U1 snRNP, with the consensus 5′SS oligonucleotide bound to the 5′-end of U1 snRNA, represents the 5′SS-U1 snRNP interaction in the whole U1 snRNP. The overall structure of the core domain is very similar to that of the U4 snRNP core domain, which consists of seven Sm proteins and U4 snRNA and the Sm folds of the seven Sm protein assemblies of the two structures (PDB code: 4PJO and 4WZJ) superimpose with rmsd of 0.55 Å. However, there are some noteworthy differences (Figure 2 and Figure 2—figure supplement 1). In the minimal U1 snRNP, helix H0 of SmD2 points into the minor groove of helix H (Figure 1A), buttressing it, a feature already evident at 5.5 Å. The nonamer Sm site sequence (AAUUUGUGG in U1 snRNA and AAUUUUUGA in U4 snRNA) has been reported as a minimal RNA oligonucleotide to promote Sm core domain assembly. In U1 snRNP each base of the first seven nucleotides of the nonamer, AAUUUGU (A125 to U131), interacts one-to-one with SmF-SmE-SmG-SmD3-SmB-SmD1-SmD2 in the pockets formed by four key residues at equivalent positions in the L3 and L5 loops of the Sm fold except for the interaction of G130 with SmD1 (Figure 2; Figure 2—figure supplement 1). In the protease-treated U1 snRNP structure, G132 was placed in the nucleotide-binding pocket in SmF but our structure unambiguously shows that A125 occupies this pocket. The U4 core snRNP structure has now been refined to an Rfree of 22.4% (Li, Leung, YK, CO and KN, manuscript in preparation; PDB code: 4WZJ) and the new model shows that whereas previously the G equivalent to G132 was also incorrectly placed in SmF the Sm site nucleotides bind in a similar manner in both minimal U1 and U4 core domain, except for SmD1. The fourth U of the U4 Sm site sequence is replaced by G in U1 snRNA in the majority of species including humans (http://rfam.sanger.ac.uk/). A large guanine base cannot be accommodated in SmD1 and hence it lies above His37 of SmB outside the central hole (Figure 2; Figure 2—figure supplement 1F). In the Sm site sequence the phosphate groups of A126, U130 and U131 come close and are stabilized by a hydrated Mg++ ion (Figure 2). The last two nucleotides of the nonamer, G132 and G133, fit into the binding pocket formed by SmD1-SmD2 and SmF-SmE, respectively (Figure 2—figure supplement 2).

8In the original 5.5 Å resolution map we roughly modelled the first 60 residues of U1-70k using seleno-methionine derivatives of mutants of U1-70K (L9M, I19M, E31M, I41M, E49M, E61M and I75M). In the 3.3 Å structure almost the entire length of the U1-70k N-terminal peptide is well ordered and its interaction with the core domain is revealed in detail (Figure 3). Residues 49–58, containing a type II proline helix (Pro55, Pro56 and Pro57), interact exclusively with SmD2 and its binding to SmD2 is stabilized by salt-bridges and hydrogen bonds (Figure 3A). The U1-70k residues 39–49 are wedged in the crevice between SmD2 and SmF (Figure 3B). Then the U1-70k peptide crosses the central hole where it interacts with the backbone and bases of stem IV (A135 to G138) of U1 snRNA (Figure 3C). The 3′ stem of U4 snRNA (stem II) blocks the path of the U1-70k peptide in the U4 core domain and hence the binding of this peptide is snRNA dependent. The U1-70k residues 10–31 make extensive contacts as an extended peptide with SmD3 and the β4 strand of SmB (Figure 3D; Figure 4A). The residues 6–12 of U1-70k form a 310 helix and together with its N-terminal region contact the long helix B of U1-C to stabilize U1-C binding (Figure 4A). Apart from the interaction with U1-70k, U1-C interacts exclusively with SmD3. Helix A and β1 of the Zn-finger and part of long Helix B of U1-C sit on the concave surface of SmD3 made by the N-terminal helix 1, β2 strand and loop 2 and a short α-helix made by the C-terminal tail. The interface between U1-C and SmD3 is not tightly packed and is stabilized by salt-bridges and hydrogen bonds (Figure 4A). The R21Q mutation in U1-C weakens its binding to U1 snRNP by about 10-fold.

9Natural U1 snRNA has a tri-methyl-guanosine cap attached to the 5′-end but our U1 snRNA has a 5′-triphosphate resulting from in vitro transcription instead (Figure 1—figure supplement 2A,B). It also has uridines at positions 5 and 6 instead of the pseudo-uridines found in natural U1 snRNA. Pseudo-uridines at these positions are conserved from yeast to human U1 snRNA. Only poorly diffracting crystals were obtained with pseudo-uridines at these positions. In the minimal U1 snRNP crystal, nucleotides 3 to 10 of U1 snRNA form Watson-Crick basepairs with the AG/GUAAGU sequence of the pre-mRNA strand (Figure 4B–C) where / marks the splice junction. The minor groove of the RNA duplex between the 5′SS and the 5′-end of U1 snRNA interacts with U1-C on the surface made by Helix A and the β1 strand. Several amino-acid side chains and main chain atoms of U1-C form hydrogen bonds with the 2′OH groups and phosphate oxygen atoms of both strands of RNA near the splice junction (Figure 4B–C). However, U1-C makes no contacts with the RNA bases (Figure 4B). At an early stage of spliceosomal assembly, U1 snRNP binds to the 5′SS and U2 auxiliary factor (U2AF) and splicing factor 1 (SF1) bind to the 3′ splice site and branch point resulting in the formation of E complex. A double mutant (R28G, K29S) of U1-C fails to enhance the formation of E complex. The side-chains of Arg28 and Lys29 are located close to the phosphate backbone of U1 snRNA or 5′SS RNA but their side chains have no density in our map. The 5′-triphosphate of U1 snRNA is close to Glu32 and Asp36 of U1-C and their electrostatic repulsion may prevent stable interaction of the tri-methylguanosine cap with U1-C, thus allowing the pre-mRNA strand to gain access to bases of the 5′-end of U1 snRNA to pair.

10The first 215 residues of U1-70k are conserved well from yeast to human (48% sequence similarity), suggesting that this region has an evolutionarily conserved essential function, whereas the C-terminal region, predicted to be poorly structured, has diverged considerably. Some alternative splicing factors are known to bind to this region. The U1A70kF-RNA crystal structure (Figure 1B) reveals the interaction between stem-loop I and U1-70k in detail, illustrating a new mode of RRM-RNA interaction. The canonical RRM domain is known to bind with RNA through RNP1 and RNP2 motifs as first observed in the U1A- stem loop II complex structure. The RRMs of U1-A and U1-70k have very similar structures (the β-strands and α-helices of the two RRMs superimpose with an rmsd of 0.70 Å) (Figure 5A–B) whereas the RNA loops bound to these RRMs have strikingly different structures (Figure 5C–D). The U1-A bound RNA loop has an open structure with ten nucleotide bases splayed out (Figure 5B,D). Bases of the first seven loop nucleotides show stacking interactions, either with adjacent bases or with protein side chains, while the last three nucleotides are poorly ordered. In contrast, the U1-70k bound RNA loop 1 with 11 loop nucleotides (Figure 1—figure supplement 1C) is stabilized by base stacking interactions and basepairing of nucleotides within the loop and hence it is effectively a five-nucleotide loop (Figure 5A–C). In U1-A the polypeptide loop between β2 and β3 (loop 3) protrudes through the RNA loop, stabilising it in an open conformation (Figure 5B) whereas loop 3 of U1-70k forms a β-turn and embraces C33 and G34 (Figure 5A). The bases of C33 and G34 stack with each other and are sandwiched between the side chains of Arg191 and Lys138, which form salt-bridges with the phosphate groups of C33 and G34 (Figure 6A). The second residue of RNP2 motif (Phe106) and the fifth residue of RNP1 motif (Phe148) show stacking interactions with the bases of C31 and A32 (Figure 6B) as commonly observed in the RRM-RNA complexes. The RNA loop is closed by a trans WC/Hoogsteen base pair formed between A29 and A36 (Figure 6C). The base of G28, instead of forming a base pair with G37, flips out from the RNA helix and is sandwiched between the side chains of Arg172 and Tyr112, while the guanidinium group of Arg200 fills the gap (Figure 6D). The stacking interaction between G28 and Tyr112 accounts for the UV-crosslinking of these residues in U1 snRNP. On the opposite strand the bases A35, A36, G37 and G38 continuously stack (Figure 6C–D). U30 is packed against the side chain of Leu175 and forms a hydrogen bond with the side chain of Asp177 and the exocyclic amino group of the adjacent C31 (Figure 6E). The stacking interaction between U30 and Leu175 is consistent with the UV-crosslinking of these residues.

11The most unusual feature of the U1-70k complex is that the regions flanking the RRM fold, which have no apparent secondary structural elements, make extensive interactions with the RNA loop bound on the surface of the β-sheet of the RRM, almost completely burying the RNA loop (Figure 5E–F). The C-terminal region following the RRM folds onto the RNA and runs along the shallow minor groove by forming an extensive network of hydrogen bonds. Similarly, the region between helix 0 and the RRM folds onto the RNA.

12The crystal structure of the minimal U1 snRNP has revealed in detail molecular contacts between U1 snRNP and a 5′SS RNA with the consensus sequence. The role of U1-C in stabilising the 5′SS binding was first shown by using 172 nucleotide pre-mRNA. We measured binding of U1 snRNP to a [32P]-labelled 5′SS oligonucleotide by filter-binding assay (Figure 7A). The affinity of U1 snRNP without U1-C (U1 snRNP[ΔU1-C]) for the wild type 5′SS oligonucleotide increases by between three and fourfold on addition of U1-C (Figure 7A). In order to assess the contribution of the molecular contacts revealed by the crystal structure in 5′SS sequence selection we next assayed binding of variant 5′SS oligonucleotides to U1 snRNP containing uncapped, but otherwise fully authentic, U1 snRNA.

13Within the 5′SS of human genes processed by the major spliceosome, the most frequently observed nucleotides at each position of pre-mRNA from −3 to +6 form a Watson-Crick basepair with a nucleotide of the 5′-end of U1 snRNA (Figure 4; Figure 7B). The observed frequency tends to be higher in the middle and taper off towards both ends because mismatches affect the stability of the duplex less when they are further away from the middle. The first two intron nucleotides, which pair with C8 and A7 of U1 snRNA, are nearly invariantly GU (Figure 4C; Figure 7C). The nucleotides at +3 and +4 pair with Ψ6 and Ψ5 of U1 snRNA, respectively, and A/G are found at +3 but A is found predominantly at +4. G is preferred at positions −1 and +5 (Figure 7B), which pair with C9 and C4 of U1 snRNA, whilst U is preferred at +6. As the filter-binding assay described above requires large amounts of U1 snRNP, we studied binding of different 5′SS sequences to U1 snRNP by competition assay using a short, fluorescently-labelled oligonucleotide as a reference (Table 2). We found that a labelled consensus sequence oligonucleotide bound too tightly to be competed off by the weaker competitor oligonucleotides at concentrations that could be feasibly achieved in the experiment. Hence we used a mismatched labelled oligonucleotide (5SS-F). This does not influence the experiment's capacity to show whether one oligonucleotide competes better or worse than another. Single nucleotide substitutions at highly conserved positions substantially reduce the affinity of the 5′SS oligonucleotide for U1 snRNP (Figure 7C). For example, substitution of G at −1, +1 and +5 with C drastically reduces the binding of the competitor oligonucleotide (Figure 7C) as C does not form a stable basepair with C. Furthermore, substitution of +2U with an A severely reduces the binding affinity, presumably because an A–A mismatch distorts and destabilizes the duplex itself (Figure 7C). The affinity of +2C 5′SS RNA is higher than that of +2A (Figure 7C) presumably because cytosine could form a wobble basepair with a protonated A. This is consistent with the fact that C is found at +2 position, although very infrequently. G at +3 and +4 could form a wobble basepair with Ψ6 and Ψ5, respectively, and substitution of A with G at +3 or +4 position has only a moderate effect on the affinity (Figure 7D). The 5′SS with a G − Ψ wobble basepair at +4 position (+4G) has higher affinity than the one at +3 position (+3G) (Figure 7C) whilst +3G occurs more frequently in human genes (Figure 7B) than +4G. reported the effect of three single base substitutions in the 5′SS on the melting temperature of the duplex between 5′SS oligonucleotides and an oligonucleotide representing the 5′-end of U1 snRNA. +1A and +2C both had severe effects on melting temperature, in agreement with our results. Most other mutations they studied introduced a bulged nucleotide. Here we did not include such 5′SS oligonucleotides in our analysis, as it is hard to predict the effect of the bulged nucleotide on the molecular contacts observed in our crystal structure. We next studied binding of the same set of the 5′SS oligonucleotides to U1 snRNP in the absence of U1-C (U1 snRNP[ΔU1-C]) to see how U1-C contributes to the selection of 5′SS oligonucleotides (Figure 7E,F). To a first approximation these 5′SS mutants show very similar affinities relative to the wild type both in the absence and presence of U1-C, enforcing the notion that 5′SS are selected primarily by U1 snRNA. In the absence of U1-C, the mutant 5′SS oligonucleotides with +3G or +4G compete well with the wild type as in the presence of U1-C but other oligonucleotides with +3U, +4U, +1C, −1C or +5C mutations compete better in the presence of U1-C than in its absence. This shows that U1-C fine-tunes the affinity of these mismatched oligonucleotides, in most sequences studied here, to provide extra stabilisation relative to the wild type. For example, in the presence of U1-C the oligonucleotides with +3U or +4U, which introduce U–U mismatches, compete better with the wild type than they do in the absence of U1-C. Different types of U–U basepair have been reported. Our crystal structure showed that U1-C forms hydrogen bonds exclusively with sugar-phosphate backbone between −2 to +3 positions of pre-mRNA and at C9 and G11 of U1 snRNA (Figure 4B–C). When extra hydrogen bonding groups are provided by U1-C, the U–U pair could easily switch from one type to another. It is conceivable that these amino acid residues of U1-C could provide extra stabilization for these non-canonical basepairs, possibly by altering helical geometry and/or minor groove width.

14It has been proposed that U1 snRNP selects the same 5′SS sequence even in the absence of the 5′-end of U1 snRNA, and that U1C in isolation can recognize the 5′SS sequence. This conclusion is mainly based on a SELEX experiment, which may have been prone to an artefact caused by incomplete removal of the 5′-end of U1 snRNA. We observed no binding of 5′SS oligonucleotide by human U1-C protein, which was shown to be properly folded by NMR. investigated the role of conserved basic and hydrophilic residues in yeast U1-C: they mutated residues that form hydrogen bonds with the backbone atoms of either 5′SS (Thr11, Thr14, Tyr12 and His24) or U1 snRNA (His15 and Ser19), form a salt-bridge with SmD3 (Arg21) or lie near the phosphate backbone (Lys22 and Arg28 [Lys28 in yeast]) in our crystal structure (Figure 4). Substitution of any one of these residues with Ala had no effect on yeast growth at any temperature indicating that these mutations do not have a major influence on 5′SS selectivity, in agreement with our conclusion.

15The structures of two sub-domains of U1 snRNP reported here have revealed an intricate network of interactions between the components of U1 snRNP. U1-70k N-terminal peptide binds to the subunit interfaces between SmD2 and SmF, and between SmD3 and SmB, and hence only the fully-formed core domain can induce its binding, which in turn enables U1-C to bind the core domain. Our structure also revealed the molecular contacts between U1 snRNP and the 5′SS of pre-mRNA and provided new insights into the molecular mechanism of 5′SS selection. U1-C makes no contacts with nucleotide bases and U1 snRNP selects 5′SS sequences primarily by thermodynamic stability of the RNA duplex between the 5′-end of U1 snRNA and the 5′SS. However U1-C fine-tunes the affinity to stabilize the binding of some mismatched 5′SS oligonucleotides relative to the canonical 5′SS.

16The SmE/SmF/SmG trimer, the SmD1/SmD2 dimer, the SmB/SmD3 dimer, U1-A, U1-70k and U1-C were prepared as described previously. All proteins are based on human sequences, unless otherwise indicated. The coding sequence of thioredoxin, (His)6-tag, tobacco etch virus (TEV) protease cleavage site, and a U1-70K fragment (residues Thr 2–Arg 59) was PCR-amplified from the U1-70K expression vector and inserted at the initiation codon of SmD1 in the SmD1/SmD2 coexpression vector to create the U1-70kSmD1/SmD2 expression vector (Figure 1—figure supplement 1). A (Gly–Ser)3 sequence was included in the PCR primer to link the U1-70k fragment to SmD1. Escherichia coli BL21 (DE3) pLysS cells were transformed with the expression vector. The cells were grown at 37°C in 2xTY medium with ampicillin (100 μg/ml) and, when A600 nm = 0.4 − 0.8, protein expression was induced by addition of 0.5 mM IPTG and the culture was continued at 15°C overnight. Cell pellets were resuspended in Ni-A buffer (20 mM Tris-Cl pH7.4, 1 M NaCl, 1 M urea, 10 mM 2-mercaptoethanol) supplemented with EDTA-free protease inhibitor cocktail (Roche, Basel, Switzerland). The cells were lysed by sonication and clarified lysate was loaded onto a Ni-NTA column. The protein was eluted with a gradient of imidazole to 300 mM. The thioredoxin and (His)6-tags of the protein were cleaved by His-tagged TEV protease during dialysis against Ni-A buffer before passing through the Ni-NTA column again. The flowthrough fractions were diluted fivefold with Na-0 buffer (20 mM Tris-Cl pH7.4, 1 M urea, 10 mM 2-mercaptoethanol) and loaded onto a HiTrap heparin column (GE Healthcare, Little Chalfont, UK) equilibrated with heparin-A buffer (20 mM Tris pH7.4, 200 mM NaCl, 1 M urea, 10 mM 2-mercaptoethanol). The U1-70kSmD1/SmD2 heterodimer was eluted by a NaCl gradient and peak fractions were pooled, concentrated to ∼300 µM, rapidly frozen in liquid nitrogen and stored at −80°C.

17A coding sequence for U1A70kF, (Figure 1—figure supplement 1B) consisting of a (His)6-tag, TEV-protease cleavage site, residues 2–111 of U1-A protein, a linker of six Gly–Ser repeats and residues 60–216 of U1-70k protein, was constructed from PCR fragments and synthetic oligonucleotides, and ligated into the pET13 vector. E. coli BL21 (DE3) pLysS cells were transformed with the expression vector and cells were grown in 2xTY medium with ampicillin (50 μg/ml) and chloramphenicol (34 μg/ml) at 37°C until A600 nm was approximately 0.7. Then, the temperature was lowered to 20°C and protein expression induced by addition of 0.5 mM IPTG. After 8–10 hr cells were harvested by centrifugation and resuspended in lysis buffer (20 mM Na+-Hepes pH 7.5, 25 mM imidazole, 0.5 M NaCl, 0.5 M urea). The cells were lysed by sonication, clarified and loaded onto a Ni-NTA column. The U1A70kF protein was eluted by a linear gradient of 25–500 mM imidazole in the same buffer. The His-tag was cleaved off with His-tagged TEV protease and the uncleaved protein was removed by passing the protein through a second Ni-NTA column. The protein was loaded onto a heparin-Sepharose column and eluted with a linear NaCl gradient (120–1000 mM) in 20 mM Na+-Hepes pH 7.5, 25 mM Imidazole, 60 mM Na Phosphate pH 7.4, 1 M urea. Peak fractions were concentrated by ultrafiltration, buffer exchanged into 20 mM Na.Hepes, 25 mM imidazole, 0.3 M NaCl, pH 7.5, rapidly frozen in liquid nitrogen and stored at −80°C.

18DNA templates for in vitro transcription (Figure 1—figure supplement 2) were assembled by ligating overlapping oligonucleotides, which included the T7 promoter, into pUC18 vector. All RNAs are based on human sequences, unless otherwise indicated. Genes for full-length and minimal U1 snRNAs lacking the first 10 nucleotides were also cloned into pUC18 together with T7 promoter. These truncated RNAs were transcribed in the presence of 2 mM GMP to facilitate ligation of modified 5′ end oligonucleotides. The 5′ fragment of U1 snRNA with post-transcriptional modifications (5′-AmUmACΨΨACCU-3′ or 5′-AmUmACUUACCU-3′ where Ψ = pseudo-uridine, Am, Um = 2′-O-methyl nucleotides) were purchased from Dharmacon (GE Healthcare, Little Chalfont, UK) and ligated to the truncated U1 snRNAs by splint-assisted ligation with T4 DNA ligase. The plasmid template preparation, in vitro transcription and purification of RNA were carried out as described.

19The full-length U1 snRNA at 8 µM was incubated in 50 mM KCl at 80°C for 2 min and annealed by snap cooling on ice. Three Sm protein sub-complexes were diluted to 100 μM each in reconstitution buffer (RB: 250 mM KCl, 20 mM K+-HEPES pH 7.5, 5 mM DTT). RNA was mixed with Sm proteins so that the solution contained 4 μM of RNA, 6 μM of each of the Sm proteins, 10 mM of DTT, and 40 units/ml of RNasin (Promega, Fitchburg, Wisconsin, USA) in RB. The mixture was incubated at 30°C for 30 min, and then at 37°C for 15 min. U1-70k was diluted to 12 μM with RB. The Sm protein-RNA complex was mixed with an equal volume of the U1-70k solution and incubated on ice for 15 min. U1-A protein was added to 4 μM final concentration. The final reconstitution mix contained 2 μM of RNA, 3 μM each of the Sm proteins, 4 μM of U1-A, and 6 μM of U1-70k protein in RB. The solution was incubated on ice overnight and applied to a monoQ column (GE Healthcare, Little Chalfont, UK). The reconstituted complex was eluted with a gradient of KCl from 250 mM to 1 M in RB buffer. U1-C protein was added to U1 snRNP by buffers supplemented with 1.5 μM of U1-C protein to ensure an excess of U1-C in binding assays.

202 µM of SmKCm RNA (Figure 1—figure supplement 2B) was mixed with 1.5-fold molar excess of U1-70kSmD1/SmD2, SmD3/SmB and SmE/SmF/SmG sub-complexes in RB supplemented with 2 M urea. After incubation at 37°C for 1 hr, the sample was dialyzed against RB buffer at 4°C overnight. U1-C protein was added to a final concentration of 6 µM and incubated at 30°C for 15 min and at 4°C for at least 1 hr. The complex was purified on a monoQ column as described for full-length U1 snRNP.

21U1A70kF protein was added slowly to 10 μM SL1·SL2 RNA in 0.4 M NaCl, 40 mM Na+-Hepes pH 7.5, 50 mM imidazole at room temperature for a final protein concentration of 15 μM. After 20 min the mixture was diluted with an equal volume of water and incubated for a further 10 min. The complex was purified on a monoQ column equilibrated with 20 mM Na+-Hepes pH 7.5, 150 mM NaCl and eluted with a NaCl gradient (150 mM–1 M). Peak fractions were pooled, concentrated to 6 mg/ml with a centrifugal concentrator, and buffer exchanged into 0.2 M NaCl, 20 mM Na+-Hepes pH 7.5, 25 mM imidazole.

22Crystals of the SmKCm complex were obtained by sitting-drop vapour diffusion at 277K. The purified U1 snRNP complex was mixed with 1.2-fold molar excess of 5′SS RNA oligo (5′-AGGUAAGUCC-3′) purchased from Dharmacon (GE Healthcare, Little Chalfont, UK) and 0.15 mM of polyamine-9. The complex solution (4 mg/ml in 300 mM KCl, 20 mM K+-Hepes, pH 7.5, 5 mM MgCl2, 1 mM DTT) was mixed with an equal volume of reservoir solution (7% MPD, 180 mM KCl, 5 mM MgSO4, 50 mM Na+-Hepes pH 6.4). The crystals were improved by streak-seeding from another crystallization drop using a feline whisker 3 min after mixing the drops. Crystals suitable for data collection appeared after 1 month, reaching maximum dimensions of 0.35 × 0.05 × 0.05 mm3. Crystals were transferred to cryo-protection buffer (20% MPD, 200 mM KCl, 5 mM MgSO4, 50 mM Na+-Hepes pH 6.4, 25% PEG4000) in three steps with 30 min equilibration after each transfer and flash frozen in liquid nitrogen.

23The U1A70kF RNA complex was mixed with an equal volume of 40% MPD, 0.15 M NaCl, 0.1 M sodium acetate pH 4.2 and equilibrated with the same solution by sitting drop vapour diffusion at 20°C. Crystals of type I (wedge-shaped) grew within 16–48 hr whereas crystals of type II (plate-like) grew in the same drops after several weeks. Only the latter type diffracted to sufficient resolution for a structure determination.

24Diffraction data were collected at Diamond Light Source beamlines I02, I03, I04 and I04-1. Data were integrated in iMosflm and scaled with Scala. For the minimal U1 snRNP (SmKCm complex) an initial molecular replacement solution was found in Phaser using the U4 snRNP core domain proteins. These form similar ring structures in both U1 and U4 snRNP, and have a total mass of ∼70 kDa, which is more than half the mass of the minimal U1 snRNP (∼120 kDa). Phases were improved by density modification using Parrot. This allowed us to build the remainder of the model using Coot. The N-terminal region of U1-70k was built de novo and we manually placed the zinc finger domain of U1-C protein and HIV1 kissing loop structure into electron density. The structure was refined by Refmac (Table 1). For the U1A70kF-RNA complex, an initial molecular replacement solution was found in Phaser using part of the structure of U1-A complex with SL2 RNA and a homology model of the U1-70k RRM, which was built with Modeller using multiple templates and manually refined alignment as an input. The remainder of the model was built into the MR map using Coot, and the structure was refined by Refmac (Table 1). Figures of molecular structures were drawn using Pymol (www.pymol.org).

25Pre-mRNA oligonucleotide substrates containing 5′ splice-site sequences were purchased from Dharmacon (GE Healthcare, Little Chalfont, UK) (Table 2). All the components were diluted in binding assay buffer (BAB: 10 mM K+-Hepes pH7.5, 200 mM KCl, 2 mM MgCl2, 0.5 mM DTT, 100 µg/ml tRNA, 50 µg/ml BSA), and 1.5 µM U1-C protein was added to BAB to ensure saturation of U1 snRNP with U1-C protein under assay conditions. In competition assays, each reaction (16 μl) containing 1 nM fluorescein-labeled reference RNA oligo (5ss-F) and 35 nM U1 snRNP reconstituted with U1 + Ψ RNA (Figure 1—figure supplement 2), was titrated with non-labelled competitor 5′SS RNA oligos. Binding curves were measured by fluorescence polarization using Pherastar (BMG Labtech, Ortenberg, Germany). The assays were carried out in triplicate. The values of fluorescence polarization in the absence of competitor RNA oligo were plotted at 1 pM for reference.

26In the filter-binding assay, a 5′SS RNA oligonucleotide (5′-32P- CAAAGGUAAGAUGGA-3′, 10 fmol), was mixed with various concentrations of U1 snRNP, with or without U1-C protein, in Filter-Binding Buffer (FBB: 200 mM KCl, 2 mM MgCl2, 0.5 mM DTT, 10 mM Hepes, pH 7.5, 100 µg/ml tRNA, 50 µg/ml BSA) in a final volume of 40 µl and incubated at 22°C for 2 hr. The binding reaction was passed, under vacuum, through a Schleicher & Schuell (Dassel, Germany) NC45, 25 mm diameter filter that had been pre-wetted with FBB. The filter was washed with 0.5 ml FBB and then dried. The proportion of radiolabeled oligonucleotide retained on the filter was determined by scintillation counting. The experiment was performed in triplicate.