Dynamics of the DEAD-box ATPase Prp5 RecA-like domains provide a conformational switch during spliceosome assembly

ABSTRACT
The DEAD-box family of proteins are ATP-dependent, RNA-binding proteins implicated in many aspects of RNA metabolism. Pre-mRNA splicing in eukaryotes requires three DEAD-box ATPases (Prp5, Prp28 and Sub2), the molecular mechanisms of which are poorly understood. Here, we use single molecule FRET (smFRET) to study the conformational dynam- ics of yeast Prp5. Prp5 is essential for stable as- sociation of the U2 snRNP with the intron branch site (BS) sequence during spliceosome assembly. Our data show that the Prp5 RecA-like domains un- dergo a large conformational rearrangement only in response to binding of both ATP and RNA. Mutations in Prp5 impact the fidelity of BS recognition and change the conformational dynamics of the RecA- like domains. We propose that BS recognition during spliceosome assembly involves a set of coordinated conformational switches among U2 snRNP compo- nents. Spontaneous toggling of Prp5 into a stable, open conformation may be important for its release from U2 and to prevent competition between Prp5 re-binding and subsequent steps in spliceosome as- sembly.

INTRODUCTION
DEAD-box ATPases, named after the conserved tetrapep- tide ‘DEAD’ motif in their active sites, are ubiquitous in both prokaryotic and eukaryotic RNA metabolism includ- ing RNA processing. They generally possess weak helicase activity and are incapable of translocation along RNA (1,2). Nonetheless, they are essential for a wide-range of biologi- cal functions such as the melting of short RNA duplexes, remodeling of RNPs, strand annealing, or RNA binding (1). While only a few DEAD-box enzymes have been stud- ied in biochemical depth, experiments from many labora- tories have revealed that their RecA-like ATP-binding do- mains can exhibit a range of conformations upon crystal- lization (3). Many of these conformations represent ‘open’ structures in which the ATPase active site formed at the in- terface of the two RecA-like domains is disrupted. Forma- tion of an ATPase-competent closed complex can be cou- pled with binding of RNA (1,4). Several DEAD-box pro- teins are also known to interact with protein cofactors that stimulate or inhibit ATP hydrolysis and RNA unwinding (5–7). In the case of the prototypical Saccharomyces cere- visiae (yeast) DEAD-box ATPase eIF4A, stimulation can occur by perturbation of open/closed conformational tran- sitions upon interaction with the eIF4B and eIF4G cofac- tors (8,9).
The yeast spliceosomal DEAD-box protein Prp5 (DDX46 in humans) is required for stable association of the U2 snRNP with the intron BS (10). How Prp5 partic- ipates in this process is not clear. The ATPase activity of Prp5 is essential in the presence of the U2 snRNP protein Cus2 but not in cus2Δ yeast (11,12). This suggests that Prp5 plays a role in U2 snRNP remodeling. Biochemical and cross-linking experiments have identified a short duplex structure within the U2 snRNA, the BS stem loop (BSL), as a potential RNA binding site for Prp5 (13,14). In addition, Prp5 likely engages in protein-protein interactions with the U2 snRNP protein Hsh155 (SF3B1 in humans) (15,16).

Mutations in the Prp5 SAT sequence (conserved DEAD- box motif III) can decrease ATP hydrolysis by Prp5 in vitro and increase usage of introns containing weak BS (those with low complementarity to the U2 snRNA) in vivo (17). This finding has been interpreted as Prp5 proofreading duplex formation between U2 and the intron by coordi- nating the rate of pairing with an ATPase-dependent step (17). Alternatively, it has been proposed that proofreading originates from release or retention of Prp5 from the pre-spliceosome, independent of ATPase rate (13). The crys- tal structure of the helicase core of Prp5 in the absence of RNA revealed an unusual conformation of the RecA-likedomains in which they were twisted 183◦ relative to one an-other compared to the ATP-bound state of other DEAD- box proteins (Figure 1A) (18). Toggling of Prp5 into and out of this ‘twisted open’ structure may also play a role in BS recognition by U2 (18). However, it is not known if the ‘twisted open’ structure represents the predominant conformation of Prp5 in solution or if the protein under- goes a conformational change upon RNA and ATP bind- ing. We have used smFRET to directly observe structural dynamics of the Prp5 helicase core and their perturbation by ligands and mutations. Our results support formation of a stable ‘twisted open’ conformation in the absence of lig- ands. Transient closure of the RecA-like domains requires ATP and RNA. Mutations that alter BS usage in yeast change Prp5 dynamics and equilibria between the open and closed states. Our data suggest a mechanism in which a large conformational change of the Prp5 RecA-like do- mains is associated with U2 snRNP binding during spliceo- some assembly. Spontaneous toggling of DEAD-box pro- teins between open and closed conformations in responseto RNA and ATP binding may be used to promote progres- sion along other multi-step reaction pathways in addition to pre-mRNA splicing.Saccharomyces cerevisiae (yeast) strains used in these stud- ies were derived from yYZX02 [MATa ade2 cup1Δ::ura3 his3 leu2 lys2 prp5Δ::loxP trp1, pR316-PRP5(PRP5 URA3 CEN ARS)], a gift from Drs. Charles Query and Yong- zhen Xu (17).

Supplementary Tables S1 and S2 contain de- tailed lists of strains and plasmids. Yeast transformation and growth were carried out using standard techniques and media (19).Expression, purification, and fluorophore labeling of Prp5core proteinsA fragment of Prp5 corresponding to amino acids 206– 699 was cloned from yeast genomic DNA by polymerase chain reaction (PCR) and ligated into the pET28b expres- sion plasmid using NdeI and BamHI restriction sites by standard procedures. The resulting plasmid was sequenced and ultimately coded for Prp5core variants containing a N- terminal histag for initial affinity purification, a TEV pro- tease site for removal of the histag, and a N-terminal AP sequence for biotinylation by the Escherichia coli biotin lig- ase (9). A list of plasmids used in these studies can be found in Supplementary Table S1.Recombinant proteins were produced in BL21(DE3) E. coli by co-transforming the bacteria with the pET28b plas- mid expressing a Prp5core variant and a plasmid express- ing E. coli biotin ligase (pAAH0947). Plasmid-containing cells were selected by growth in liquid or on solid lysogeny broth (LB) media containing 70 µg/ml kanamycin and 100 µg/ml ampicillin. Protein expression was carried out in LB (70 µg/ml kanamycin, 100 µg/ml ampicillin) by first in-cubating the cells at 37◦C with shaking at 225 rpm un-til an OD600 of 0.6 was reached. At that time, proteinproduction was induced by addition of isopropyl β-D-1- thiogalactopyranoside (IPTG) to a final concentration of1 mM. Cells were then incubated with shaking for an addi- tional 16 h at 16◦C. After this time, cells were collected by centrifugation and pellets stored at -80◦C.Prp5core variants were purified by first resuspending thecells in lysis buffer (50 mM BisTris pH 6.5, 500 mM KCl, 5 mM imidazole; 1 ml/g cell pellet) along with 1 U/ml DNase I (RNase Free, Thermo Scientific). Lysozyme (100 mg/ml, Sigma) was then added to a final concentration of0.1 mg/ml.

The cells were then incubated for 1 h while rotat- ing at 23◦C before sonication on ice using a microtip (40% power level; 3 s on, 2 s off; 3 cycles of 1 min on-time). The lysate was centrifuged at 18 000 g for 15 min at 4◦C in a Beckman JA-20 rotor. The supernatant was next appliedto a Ni-NTA agarose (Qiagen) column (1 ml resin/5 g cell pellet) by gravity flow. The column was washed with high salt wash buffer (50 mM BisTris at pH 6.5, 500 mM KCl, 5 mM imidazole), followed by elution of the protein with im- idazole (50 mM BisTris at pH 6.5, 500 mM KCl, 500 mM imidazole).10844 Nucleic Acids Research, 2019, Vol. 47, No. 20TEV protease was added to the eluate (0.21 mg TEV/1 ml protein solution, purified in lab), and the solution (3– 5 ml) was placed into dialysis using a 20K MWCO Slide- A-Lyzer cassette (Thermo Fisher) for 16 h while stirring at4◦C in dialysis buffer (20 mM BisTris pH 6.5, 100 mM KCl,10% v/v glycerol, 1 mM TCEP). Following dialysis, the pro- tein was loaded onto a HiTrap™ SP FF cation exchange column (1 ml, GE Healthcare Life Sciences) equilibrated inlow-salt buffer (20 mM BisTris pH 6.5, 100 mM KCl, 10% v/v glycerol). Pure protein was eluted using a concentra- tion gradient of 0.1 1.0M KCl over 30 min at a flow rate of 1 ml/min using a NGC™ medium-pressure chromatog- raphy system (Bio-Rad). Prp5core-containing fractions werecombined, and the protein concentration was determined by measuring the absorbance at 280 nm and the calcu- lated extinction coefficient (ProtParam, https://web.expasy. org/protparam/).The purified protein was labeled with fluorophores by addition of Cy3 and Cy5 maleimide-derivatized dyes (dis- solved in dimethylformamide, GE Healthcare Life Sci- ences). Typically, protein was present at a concentration of 10 µM and fluorophores were added to a final concentra- tion of 100 µM Cy3 and 120 µM Cy5 in a final volume of 1 ml in labeling buffer (20 mM BisTris pH 6.5, 250 mM KCl,10% v/v glycerol). This solution was allowed to incubate for 16 h at 4◦C while rotating in the dark. After incuba- tion, excess dye was removed and the protein re-purified bycation exchange chromatography as above. Labeled protein was aliquoted (5 µl), frozen in liquid N2, and stored at – 80◦C.An RNA oligonucleotide corresponding to nucleotides 1– 90 of the yeast U2 snRNA (lacking post-transcriptionalmodifications) was purchased from IDT, resuspended in buffer, aliquoted, and stored at –80◦C. Poly(A) RNA was purchased from Sigma.

RNA concentration was deter-mined by UV-Vis spectroscopy.Mutations in Prp5 were generated using inverse PCR with Phusion DNA polymerase (New England Biolabs) (15), and all plasmids were confirmed by sequencing.Biotinylated, fluorescent Prp5 proteins were immobilized onto PEG/PEG-biotin derivatized quartz slides using streptavidin (20,21). smFRET experiments were carried out in imaging buffer, which replicates conditions that support yeast splicing in vitro and contains an oxygen scavenging system as well as triplet-state quenchers (10 mM HEPES– KOH pH 7.9, 100 mM potassium phosphate pH 7.3, 3%w/v PEG 8000, 2.5 mM MgCl2, 20 mM KCl, 1 mM cyclooc- tatetraene, 1 mM 4-nitrobenzyl alcohol, 1 mM propyl gal- late, 1 mM trolox, 10 mM protocatechuic acid and 1 U/ml protocatechuate dioxygenase) (22–25). When present ATP, ADP or ADPNP were added to the imaging buffer at a fi- nal concentration of 1 mM, and U2 RNA was added to afinal concentration of 1 µM. Poly(A) RNA was added to a concentration equivalent to 310 µM AMP.smFRET data were collected on a home-built, prism-based total internal reflection fluorescence (TIRF) microscope with laser excitation at 532 and 640 nm (20,26). Excitation by the 640 nm laser (22.5 mW, measured prior to entry into the prism) was used to localize the labeled Prp5 at the be- ginning and the end of each experiment. The 532 nm laser (10 mW, measured prior to entry into the prism) was used to detect FRET. Data were recorded continuously with a frame rate of 100 ms/frame for a time span of 100 s. Donor and acceptor emission fluorescence were collected simulta- neously using a DualView apparatus (DV2, Photometrics) and an EM-CCD camera (Andor). Images were recorded using Metamorph software (Molecular Devices).Raw images were analyzed using custom MATLAB (Mathworks) software as described (26–28).

Briefly, inte- grated fluorescence intensities for donor (ICy3) and acceptor (ICy5) spots of fluorescence were used to calculate apparent FRET efficiencies using the formula, EFRET ICy5/(ICy3 + ICy5) along with vbFRET software (29). Transition events and the EFRET values during which they occurred were iden- tified and HaMMy software (30). Lifetimes of the observed EFRET states were determined using maximum likelihood methods and fit to equations containing a single exponen- tial term as described (31) except that fitting was facili- tated using AGATHA software developed by our labora- tory (https://github.com/hoskinslab/AGATHA). Errors in the fits were determined by bootstrapping (22,27), also im- plemented in AGATHA (32). Histograms of EFRET values were fit to Gaussian functions using MATLAB.Yeast were grown to mid-log phase, adjusted to OD6000.5 and plated on YPD media in ten-fold serial dilutions. Plates were incubated for 2 days at 30◦C or 10 days at 16◦C prior to imaging. FOA selection assays were performed ina similar manner, except yeast were grown in liquid media (-His) and plated on solid media (-His) to maintain plasmid selection. When present, FOA (1 mg/ml) was used to selectfor URA3 loss. Images were taken after 2 days at 30◦C.ACT1-CUP1 reporter 2µ plasmids all contained the LEU2 selection marker and growth assays have been described previously (14,41). Briefly, yeast strains expressing WT or mutant Prp5 proteins and ACT1-CUP1 reporters were grown to mid-log phase in -Leu -His dropout media to maintain selection for the plasmids, adjusted to OD600 0.5 using fresh media, and equal volumes were spotted using a 48-pin inoculator onto -Leu -His dropout plates containing 0, 0.025, 0.05, 0.075, 0.1, 0.15, 0.2, 0.25, 0.3, 0.4, 0.5, 0.6,0.7, 0.8, 0.9, 1.0, 1.1, 1.2, 1.3, 1.4, 1.5, 1.6, 1.7, 1.8, 1.9, 2.0,2.25 or 2.5 mM CuSO4. Plates were scored for the presence or absence of growth after 3 days at 30◦C.ATP hydrolysis by Prp5 proteins was determined as de- scribed using a charcoal-binding assay (33) except that buffers used matched those in smFRET assays (10 mM HEPES–KOH pH 7.9, 100 mM potassium phosphate pH 7.3, 20 mM KCl, 3% w/v PEG 8000, 2.5 mM MgCl2).Each reaction contained 645 nM labeled Prp5core (WT or variants), 1 mM ATP, and 50 nCi/µl γ-[32P] ATP (Perkin Elmer). 10 µM U2 snRNA (1–90) or 5 µg polyA (equiva- lent to 310 µM AMP, Sigma) were included in each reac- tion as noted. Activated charcoal was prepared and resus- pended in 20 mM phosphoric acid (34). Reactions (20 µl) were incubated at room temperature for 90 min followed by the addition of 250 µl charcoal suspension. The reactions were chilled on ice for 10 min and then centrifuged to pel- let the charcoal. Free [32P]-phosphate was quantified using a scintillation counter.

RESULTS
To create single-molecule probes of Prp5 interdomain dis- tances, we incorporated fluorophores into the helicase core domain of Prp5 at positions in which they are predicted to have high FRET when the protein is in the ‘twisted open’ conformation and lower FRET based on a model of the ATPase-active, closed structure (Figure 1A) (18). First, we constructed a prp5 gene encoding a functional protein lack- ing cysteines capable of complementing a prp5∆ yeast strainat 30◦C (Prp5cysless, Figure 1B, C; Supplementary Tables S1–S3). We then replaced V268 and K666 with cysteines for flu-orophore attachment (Figure 1A). This derivative also com- plemented the prp5∆ strain (Prp5VK/CC, Figure 1B and C). We expressed and purified from E. coli a fragment of Prp5VK/CC containing the central RecA-like domains and flanking sequences corresponding to the fragment of Prp5 that was crystallized (Prp5core, Figure 1B; Supplementary Figure S1A–C and Table S3) (18). Prp5core also included a genetically encoded biotin acceptor peptide (AP) at the N-terminus of the protein. Co-expression with E. coli bi- otin ligase resulted in significant biotinylation of the protein (Supplementary Figure S1C), which was used to anchor the protein to microscope slides. Purified Prp5core was labeled with a mixture of Cy3- and Cy5-maleimide, and excess dyes were removed by chromatography (Supplementary Figure S1D). Consistent with previous studies of Prp5 (35), labeled Prp5core possessed very little ATPase activity in the absence of RNA (Figure 1D). When poly(A) RNA was added (310 µM AMP equivalents, 30-fold higher than reported Km values for Prp5 for this substrate (18)), the ATP hydrolysis activity was activated (Figure 1D).

Thus, labeled Prp5core isan RNA-stimulated ATPase. smFRET supports formation of an open and stable confor- mation of Prp5corecontaining both Cy3 and Cy5 fluorophores by illuminating the sample with a 640 nm laser to excite Cy5 and identify molecules containing at least one Cy5 fluorophore. We then monitored smFRET by illumination with a 532nm laser to excite Cy3. Post-acquisition inspection of Cy3- and Cy5- emission and photobleaching from each Cy5-containing molecule identified those containing both fluorophores. For example, molecules containing both Cy3 and Cy5 show an increase in Cy3 fluorescence upon Cy5 photobleaching when illuminated at 532 nm.Prp5core molecules were uniformly static, exhibiting a sta- ble FRET efficiency (EFRET) of 0.8–0.9 (Figure 2B). Since we failed to observe transitions, this suggests that confor- mational changes are either very infrequent or that the life- time of any alternate conformation is much less than our time resolution of 100 ms. Histogram analysis of Prp5core revealed a single peak, which could be fit to a Gaussiandistribution centered at EFRET = 0.89 (Figure 2C, Supple-mentary Table S4). Assuming R0 = 0.54 A˚ (36), this valueWe attached biotinylated and Cy3/Cy5-labeled Prp5core to biotin PEG-passivated quartz microscope slides with strep- tavidin and visualized single Prp5core molecules using a prism-type total internal reflection fluorescence (TIRF) mi- croscope (Figure 2A) (20). We analyzed molecules likelyyields an approximate distance of 38 A˚ compared to a 31 A˚ Cα–Cα distance between V268 and K666 observed in the ‘twisted open’ structure. These results are also in agreementwith previous studies of the bacterial YxiN and yeast eIF4A DEAD-box proteins, which both showed unimodal distri-butions of smFRET in the absence of ligands consistent with the RecA-like domains being open (8,37).

Our data in- dicate that Prp5core exhibits a homogenous, stable confor- mation in solution that likely corresponds to an open form similar to that observed by crystallography.RNA and ATP induce transient closure of Prp5coreWe next tested the impact of ligands on Prp5core conforma- tion. In the presence of either ATP or poly(A) RNA in- dividually, Prp5core molecules remained static and EFRET distributions continued to show only a single peak near EFRET 0.88 (Supplementary Figure S2A–D, Table S4). When poly(A) RNA and ATP were simultaneously added, molecules transitioned from high EFRET to a shorter-lived conformation with lower EFRET (Figure 2D). Histogram analysis revealed the appearance of a second, less-populated peak of EFRET at 0.52. The requirement for ATP and RNA to induce Prp5core conformational change is again very sim- ilar to observations made with YxiN and eIF4A, both of which require cooperative binding of ATP and RNA to ini- tiate closure of the RecA-like domains (4,8,37,38).We were able to fit distributions of dwell times for Prp5core in both EFRET states to maximum likelihood equations con- taining a single exponential term. This yielded characteris- tic lifetimes for the open (ropen) and closed (r closed) states of3.0.31 and 0.46 0.04 s, respectively (Supplementary Table S5). These lifetimes are similar to those reported for eIF4A, albeit only in the presence of protein cofactors that induce formation of an alternate half-open conformation of the ATPase (8). Our data are consistent with Prp5core re- versibly switching directly between a closed and open state. These transitions require RNA and ATP but, unlike eIF4A, are not dependent on the presence of other protein cofac- tors or formation of any detectable intermediate structure. We tested whether or not a fragment of the U2 snRNA (nt 1–90) containing the proposed Prp5 binding site (13) could also trigger Prp5core ATPase and conformational switching. This RNA was able to stimulate both activities, and Prp5core exhibits similar changes in smFRET when exposed to ei- ther the poly(A) RNA or U2 fragment (Supplementary Fig- ure S3). For both RNAs under these conditions, the life- time of the closed conformation is much shorter than that of the open conformation (Supplementary Table S4). Previ- ous work suggested that Prp5 exhibits a substrate preference for the U2 snRNA (35). While we did not test the substrate preference of Prp5core, conformational switching appears to be a general feature of Prp5core in the presence of RNA andATP.To further test whether changes in smFRET observed with Prp5core correspond to the proposed open/closed structural transitions (Figure 1A), we constructed an or- thogonal reporter in which the protein would exhibit low FRET when open and high FRET when closed (Supple- mentary Figure S4A).

A gene encoding Prp5 with cysteines only at T430C and K543C was also able to complement prp5∆ yeast (Prp5TK/CC, Supplementary Figure S4B and C). The corresponding Prp5 fragment (Prp5core2) exhibited a static distribution of low EFRET in the absence of ligands (Supplementary Figure S4D and E). The fitted EFRET value of 0.37 (Supplementary Table S4) corresponds to an ap-proximate distance of 59 A˚ , close to the predicted value of 66 A˚ obtained from the ‘twisted open’ crystal structure. Inthe presence of U2 RNA and ATP, Prp5core2 also sponta- neously switched between low and high EFRET likely corre- sponding open and closed states, respectively (Supplemen- tary Figure S4F). As with Prp5core, the closed conforma- tion of Prp5core2 was much less abundant than open state (Supplementary Figure S4G). Together data obtained with Prp5core and Prp5core2 support a stable conformation of Prp5 in solution consistent with the ‘twisted open’ crystal structure. Formation of a closed structure is transient un- der these conditions and requires the presence of both RNA and ATP.To test the dependence of the closed conformation on ATP, we incubated Prp5core with poly(A) RNA and ADP. Prp5core failed to exhibit any signs of conformational switching in the presence of ADP and only a single, static smFRET state was observed (Supplementary Figure S5A and B). This in- dicates that ADP cannot substitute for ATP for trigger- ing closed state formation and is consistent with the open structure of Prp5 bound to ADP (18). Non- or slowly hy- drolyzable ATP analogs such as ADPNP have been used to trap several DEAD-box proteins in closed conformations for biophysical or structural studies (39–43). We also tested whether ADPNP could similarly trap Prp5core in a closed state by incubating the protein with the analog and U2 RNA and observing changes in smFRET.

As expected, the lifetime of the low EFRET (closed) conformation of Prp5core was extended by ADPNP (Supplementary Figure S5C and Table S5). However, individual molecules of the protein still reversibly switched between higher and lower EFRET val- ues. The open conformation was still the predominantly observed species, and it exhibited an 3-fold longer life- time than the closed complex (Supplementary Figure S5D and Table S5). Reversible conformational changes in the presence of ADPNP have previously been reported with yeast eIF4A and are distinct from observations with YxiN, which could be trapped in the closed state (9,43). In the case of eIF4A, this has been interpreted as a less stable closed conformation relative to other DEAD-box family members and failure of ADPNP to provide enough binding energy to fully couple with protein conformational change (9). Prp5, eIF4A, and other DEAD-box proteins may share these properties.A mutation that decreases weak BS usage also destabilizes the open conformationThe Prp5 E235A mutation has been proposed to disrupt the protein interface that stabilizes the ‘twisted open’ struc- ture (18). This mutation also results in increased ATPase catalytic efficiency (kcat/Km) in vitro and decreased in vivo usage of BS containing mismatches to U2 snRNA (18). To directly test if E235A alters Prp5 conformation, we first in- corporated E235A into Prp5VK/CC in yeast to confirm that the mutation changes usage of weak BS in this background. We tested BS usage by monitoring Cu2+-resistance in yeast containing the mutation along with the well-characterizedACT1-CUP1 splicing reporter (Figure 3A) (44). In this assay, increased Cu2+-tolerance correlates with increased splicing of the reporter gene in vivo. As expected, Prp5, Prp5-E235A, Prp5VK/CC and Prp5VK/CC-E235A exhibited similar Cu2+-tolerances when using an ACT1-CUP1 re- porter containing a consensus BS (Figure 3B). The E235Amutation resulted in decreased Cu2+ tolerance compared to Prp5WT or Prp5VK/CC when reporters were used contain- ing the U257C weak BS consistent with decreased BS us- age and spliced product formation.

However, this effect was mitigated when the A258U reporter was used (Figure 3D) suggesting that phenotypes of at least some Prp5 mutants are dependent on the reporter RNA being used as well as the parental Prp5 protein sequence (Prp5WT or Prp5VK/CC). We incorporated the E235A mutation into Prp5core and then purified and labeled the protein. The E235A mutation significantly increased the ATPase rate of Prp5core (Figure 4D). We carried out smFRET experiments in the absence and presence of RNA and ATP. Molecules of Prp5core- E235A were static in the absence of ligands (Figure 3E and F). However, the distribution of smFRET signals was bi- modal with peaks near EFRET 0.80 and 0.53 (Figure 3G). Inspection of individual smFRET trajectories revealed that while the molecules of Prp5core-E235A were not transition- ing, they were apparently locked into one of two conforma- tions corresponding to the peaks in EFRET. Most molecules (89%) occupied the higher EFRET conformation while the remainder (11%) were found to be in the lower EFRET state. Molecules of Prp5core-E235A reversibly switched between these two states once poly(A) RNA and ATP were added (Figure 3H and I). These dynamic molecules were kineti- cally homogenous, meaning that we were unable to detectsub-populations that could have arisen from the two static populations observed in the absence of ligands. The E235A mutation still permits formation of open and closed struc- tures of Prp5core; however, it also introduces conformational heterogeneity in the absence of added ligands.Kinetic analysis of ATP- and RNA-stimulated confor- mational switching of Prp5core-E235A revealed that the mu- tation destabilized ropen 4-fold, while r close remained un- changed. This agrees with the structure of Prp5 in which E235A alters the domain interface involved in stabilization of the ‘twisted open’ conformation but is not part of the pro- posed interface formed upon domain closure (18). The in- creased ATPase catalytic efficiency of Prp5-E235A is largely due to a decrease in the Km for RNA (18). Destabilization of the open conformation could allow for more efficient cou- pling between RNA binding and closing of the RecA-like domains.While the E235A mutation decreases use of weak BS in yeast, several mutations of the DEAD-box SAT motif are known to do the opposite. Mutation of the SAT sequence (Prp5 amino acids 446–448) to TAG or GAG results in in- creased use of weak BS in vivo as well as decreased rates of ATP hydrolysis in vitro (17).

Mutation of this motif to GAR has a more dramatic effect and results in cold sen- sitivity (cs) and a >95% decrease in ATPase activity (17). These results are consistent with the amino acids of motif III helping to structure the ATP binding cleft formed by the two RecA-like domains in a catalytically competent confor- mation (1,17,18).In ACT1-CUP1 assays, yeast expressing the GAG or TAG mutants of Prp5WT or Prp5VK/CC grew similarly to the corresponding parental strain in the presence of a consen- sus BS reporter (Figure 3B). The TAG mutation increased Cu2+-tolerance of yeast containing either the U257C or A258U reporters in both Prp5WT and Prp5VK/CC back- grounds (Figure 3C and D). Previous studies reported that the GAG mutant causes a slight growth defect in the pres- ence of the WT ACT1-CUP1 reporter and improves growth of yeast relative to WT in the presence of the U257C ACT1- CUP1 reporter, albeit to a lesser extent than the TAG mu- tant. Using Prp5VK/CC, we were also able to observe the growth defect of the GAG mutant with the WT reporter (Figure 3B); however, we could not observe a significant change in U257C splicing (Figure 3C). Yeast containing the GAR mutation were cs in both Prp5WT and Prp5VK/CC (Supplementary Figure S6). The growth defects of strains with GAR mutations prevent direct comparisons with the GAG and TAG reporters in ACT1-CUP1 assays.We purified and labeled Prp5core GAG, TAG and GAR mutants. ATPase assays confirmed that the mutants de- creased the RNA-stimulated ATPase activity. The GAG mutation had the least impact on activity while the GAR mutation had the greatest. Our ATPase assay results with Prp5core differ from those reported for the full-length pro- tein. For full-length Prp5, the TAG mutant reduced Prp5 ATPase activity by 2-fold and was more active than the corresponding GAG mutant (17). For labeled Prp5core, theTAG mutant reduces ATPase activity 20-fold and is less active than the GAG mutant. However, our in vivo ACT1- CUP1 assays (Figure 3) agree with previous reports that the TAG mutant increases splicing of RNAs with nonconsen- sus BS more so than the GAG mutation (17).

Thus, in our experiments the TAG mutants have both a more reduced ATPase activity and greater stimulation of splicing using nonconsensus BS when compared to WT or GAG mutant proteins.We next used smFRET to test whether or not changes in ATPase activity due to SAT-motif mutations also reflect changes in protein conformational dynamics. In the absence of ligands, the GAG, TAG, and GAR mutants showed sim- ilar distributions of EFRET compared to Prp5core containing the native SAT-motif (Supplementary Figures S7 and S8). Like Prp5core, molecules containing these mutations were static with EFRET values close to 0.89. The GAR and GAG mutations resulted in slight broadening of histogram distri- butions towards lower EFRET values (Supplementary Fig- ure S8A and C); however, we were unable to fit these data to more than one Gaussian function. The SAT-motif mu- tations appeared to have little impact on the open structure of Prp5 consistent with observations that the SAT-motif is involved in binding of ATP at the domain interface after conformational rearrangement.Each of the Prp5core SAT-motif mutants transitioned between two EFRET states in the presence of ATP and poly(A) RNA (Supplementary Figure S7). However, his- togram analysis showed much less accumulation of the lower, closed EFRET state for the mutants relative to Prp5core (Figure 4A–C). Prp5core-TAG and GAR, in particular, showed very little occupancy of this state, and transitions were both infrequent and very short-lived. Differences in proteins due to SAT-motif mutation were much more ap- parent in the presence of ligands than in their absence, also in agreement with their location near the RecA-like domain interface.By analyzing ropen and r close for each Prp5core mutant (Supplementary Table S5), we were able to calculate the equilibrium constants of closing (Kclose) (Figure 4D). At identical concentrations of RNA and ATP, it is appar- ent that the E235A mutant favors the closed conformation much more than Prp5core or the SAT-motif mutants. The in- crease in Kclose for the E235A mutation largely stems from a shorter ropen relative to WT (1.09 versus 3.97 s, respectively; Supplementary Table S5). All of the SAT-motif mutations, on the other hand, decreased Kclose as well as the lifetime of the closed state relative to WT. The same trend was observed with both the poly(A) and U2 RNA substrates (Supplemen- tary Figure S9, Table S5). For the TAG and GAR mutants, the decrease in the closed state lifetime was accompanied by an increase in the lifetime of the open state (Supplementary Table S5). Together the data indicate that mutation of the SAT motif can destabilize the closed conformation of Prp5 as well as making it less likely for that closed state to form. Finally, we compared Kclose values for Prp5core proteins with ATPase measurements acquired at the same con- centrations of ATP and poly(A) RNA (Figure 4D). For the E235A mutation, both the ATPase rate and Kclose in- creased 3-fold. However, all of the SAT-motif mutants showed larger-fold changes in ATPase activity than in Kclose.We cannot determine whether or not any one particular FRET transition is associated with hydrolysis of one ATP molecule. However, it is possible that the differences in the magnitude of effects on ATPase and Kclose arise from de- coupling between conformational change and ATP hydrol- ysis. In other words, SAT mutants may toggle conformation without successfully hydrolyzing ATP. This would result in the smaller changes in Kclose relative to ATPase that we ob- serve. In support of this, we note that the GAR mutant has a larger Kclose than the TAG mutant (0.042 vs. 0.024, respec- tively) despite having a 30% lower ATPase activity.

DISCUSSION
By engineering a version of the yeast DEAD-box Prp5 con- taining only two cysteines within the ATPase core, we have been able to use smFRET to directly visualize and quan- tify the structural dynamics of the RecA-like domains. The yeast splicing machinery requires eight well-conserved AT- Pases for function, three of which are DEAD-box pro- teins (45). A number of mutations in these ATPases impact splicing fidelity in cells, presumably by altering spliceosome structure or regulating entry into discard pathways lead- ing to spliceosome disassembly (46,47). To our knowledge, this study represents the first real-time analysis of confor- mational dynamics for any spliceosomal ATPase. Our results provide direct evidence for the Prp5 RecA do- mains undergoing a large-scale, two-state conformational change upon ligand binding. This switch is triggered only by binding of RNA and ATP (or an ATP analog) (Figure 2). Once Prp5 closes, it can spontaneously re-form the open conformation with FRET values similar to those observed in the absence of ligands. This suggests that if Prp5 acts as a conformational switch turned ‘on’ by ligand binding then it can also reversibly transition back to the same ‘off’ con- formation observed in the apo enzyme. We also obtained information about the relative lifetimes of the closed and open states and how these are impacted by mutations. The E235A mutation was previously proposed to destabilize a protein-protein interface in the open state of Prp5, which in turn stimulates ATP hydrolysis. Our sm- FRET work supports these conclusions and shows that the primary effect of this mutation is on decreasing the open state lifetime without impacting that of the closed state. The impacts of the SAT-motif mutants are more variable. Each of them is able to decrease the lifetime of the closed con- formation while TAG and GAR also increase the lifetime of the open conformation. Together our data indicate that Prp5 mutations can potentially have different effects during splicing and at different stages. For example, SAT-motif mu- tants could inhibit steps during which Prp5 must transition from the open to closed state and/or promote steps in which Prp5 must transition from closed to open. We have not yet deconvoluted how one or all of these activities contribute to increased splicing of pre-mRNAs containing weak BS.

Association of the U2 snRNP with the BS is one of the least well-understood steps in spliceosome assembly in part be- cause it involves multiple, coordinated changes in U2 con- formation and composition. In addition to the toggling of Prp5 studied here, structural and biochemical data indicate that the large scaffold protein Hsh155 (SF3B1 in humans) must close to dock the U2/BS duplex into its RNA bind- ing cleft (15,48–51). The stem II region of the U2 snRNA must also be readied for spliceosome assembly by adopt- ing the IIa conformation, a process which is stimulated by Cus2 (11,26,52–54). The ATPase activity of Prp5 is dispens- able in the absence of Cus2 (12), implying that Prp5 might also promote Cus2 release during this stage (11). Intrigu- ingly, all of these factors are concentrated in one location on the snRNP: Hsh155 contains binding sites for Cus2 (11), Prp5 (15,16), and the U2/BS duplex (55,56). Correct associ- ation of U2 with a BS may require multiple conformational switches to be correctly activated on or near Hsh155.It is likely that protein-protein interactions between Prp5 and Hsh155 (15,16) help localize Prp5 near the U2 snRNA and its site of switching. This may provide a high local con- centration of RNA and/or favorable interactions to help Prp5 efficiently toggle to the closed conformation during assembly. During these stages, Prp5 is also known to play a role in BS fidelity and two competing models have been proposed based on ATP hydrolysis activity (17) or affinity of Prp5 for the pre-spliceosome (13). The reversible toggling that we observe between open and closed states could be accommodated by both of these models. It is possible that the open and closed forms of Prp5 have differing affinities for the pre-spliceosome and that spontaneous transition- ing from the closed to the open conformation is needed to drive Prp5 release (Figure 5). While ATP is needed to trig- ger conformational change, our model would suggest that Prp5 conformation rather than ATPase activity per se is the critical determinant of release or retention from the spliceosome. The Prp5 SAT-motif mutants may promote release and splicing by destabilization of the closed complex, while the E235A mutation could inhibit release and splicing by destabilization of the open complex (Figure 4D). We also note that spontaneous opening of Prp5 (Fig- ure 2D) during or after its release from U2 can provide an irreversible step during spliceosome assembly (Figure 5). Retention of Prp5 by U2 blocks tri-snRNP addition (13). Spontaneous formation of the conformationally sta- ble (Figure 2B), open Prp5 structure could prevent Prp5 re- association with U2 and interference with tri-snRNP bind- ing. The kinetic pathway for Prp5 closing/opening, ATP hy- drolysis, Cus2 release, and U2/BS duplex formation has not yet been determined. Ultimately, the structural and kinetic illumination of this pathway will be important for distin- guishing between these and alternate models.

While significant insights have been made into how DEAD- box ATPases bind RNA and promote duplex unwinding (1,2), much less is known about how these activities can be regulated either through the presence or absence of cofac- tors or by conformational changes. X-ray crystallographic structures of several DEAD-box proteins have revealed a wide range of conformations in the absence of ligands. It has been speculated that this might be an artifact of crystal- packing resulting in isolation of one particular conforma- tion of an otherwise flexible protein in solution (3). Our single-molecule results suggest that this is not the case for Prp5core, which exhibits a single, static EFRET state in solu- tion (Figure 2) consistent with the ‘twisted open’ conforma- tion observed by crystallography (18). Prp5, YxiN and eIF4A all possess stable open conforma- tions that fail to undergo domain closure in the absence of ligands (8,38,43). We were not able to find evidence for a half-open conformation of Prp5core similar to that previ- ously seen in eIF4A (9). This indicates that formation of an intermediate state with a detectable EFRET signature is not obligatory for domain closure. Both Prp5core and eIF4A are unable to form stable ternary complexes with RNA and ADPNP––consistent with instability of the closed confor- mations and unlike the stable complexes reported for YxiN (9,43). We speculate that different DEAD-box proteins may fall into distinct functional classes dependent on the stabili- ties of their open and closed states (4). Some, like Brr2 Inhibitor C9 eIF4A and Prp5, could depend on relatively unstable closed conforma- tions to limit unproductive RNA binding or ATP hydroly- sis. Such DEAD-box proteins could also couple large-scale conformational changes to other events in order to drive multi-step reactions like pre-mRNA splicing forward.