To reconstitute the RuvAB-HJ complex, we prepared a synthetic HJ by annealing four 55-bp DNA strands ( Figure 1A ; Supplementary Table S1 ). The recombinant PaRuvA and PaRuvB proteins were separately expressed and purified from bacterial cells ( Supplementary Figure S1 ). We first tested whether the purified recombinant proteins were active in HJ branch migration. In the gel-based DNA unwinding assay, PaRuvA in combination with PaRuvB, but not each alone, could unwind synthetic HJs into two linear DNA duplexes in the presence of ATP ( Figure 1B ). To quantitatively monitor this reaction in real time, we then performed a fluorescence resonance energy transfer (FRET) assay using the Cy3/BHQ2 modified-HJ as substrate ( Figure 1A ). At a fixed amount of PaRuvA (400 nM), the fluorescence intensity was gradually enhanced with increasing concentrations of PaRuvB, and reached a plateau at 1000 nM ( Figure 1C ). These data thus suggest that the recombinant PaRuvAB complex is active in HJ branch migration.

We next sought to reconstitute the PaRuvAB-HJ complex for cryo-EM. For more than 30 years efforts, researchers have determined the structures of RuvAB (Yamada et al., 2002), RuvA-HJ (Rafferty et al., 1996; Hargreaves et al., 1998; Ariyoshi et al., 2000), but have not obtained the atomic structure of RuvAB-HJ tripartite complex, presumably due to the labile nature of the reconstituted complexes. Until recently, while we were preparing this manuscript, Wald et al. broke the stalemate by making a hetero-RuvAB complex (RuvA from Salmonella typhimurium and RuvB from Streptococcus thermophilus), which turned out to be more stable than the homo-complexes (Wald et al., 2022). In this study, we first incubated PaRuvA with HJ DNA at 4°C for 15 min with a molar ratio of 8:1.2 to allow the formation of RuvA-HJ core complex. Then, the mixture was further incubated with excess PaRuvB proteins (RuvB:RuvA = 2:1) at 4°C for another 15 min. The resulting mixture was further purified through the superose 6 size-exclusion column. The peak eluted at ~13.9 ml contained both PaRuvA and PaRuvB proteins and HJ DNA ( Figure 1D ). However, the molecular weight (MW) of this complex was much smaller than the predicted MW (712 kDa) based on the retention times of a gel filtration standard ( Figure 1D ), indicating that this might be an intermediate complex. Since we could not reconstitute the intact RuvAB-HJ complex, we therefore focused on the intermediate complex for cryo-EM structure determination.

We next performed cryo-EM experiment using the above purified RuvAB-HJ tripartite complex. After imaging processing and 2D classification, about 551,160 particles of dumbbell shape were selected for 3D reconstruction and refinement ( Figure 2A ; Supplementary Figure S2 ). As a result, a cryo-EM map of the PaRuvAB-HJ intermediate complex was obtained at an overall resolution of ~ 6.0 Å, and the local resolutions for the PaRuvA-HJ core and PaRuvB-HJ arms were estimated to be ~ 3.0 Å and 7.0 Å, respectively ( Supplementary Figure S2 , S3 ). In addition, the results of 3D variability analysis of cryo-EM map revealed multiple conformations in the RuvB moiety, implying that there might be variable catalytic states in the sample ( Supplementary movie S1 ).

The PaRuvA-HJ model was built based on the crystal structure of Escherichia coli (E. coli, Ec) RuvA-HJ complex (PDB ID: 1C7Y)(Ariyoshi et al., 2000). It clearly shows two RuvA tetramers sandwiching one HJ DNA molecule ( Figure 2B ), as described in the previous crystal structures of RuvA-HJ complexes (Roe et al., 1998). For PaRuvB-dsDNA complex model building, we determined a 2.1-Å crystal structure of PaRuvB in complex with ADP ( Supplementary Table S2 ). The resulting PaRuvA-HJ and PaRuvB-dsDNA structures were subsequently used to composite the model of PaRuvAB-HJ tripartite complex. Owing to the resolution limits of current optimization methods, the final model only shows single RuvB motor ( Figure 2B ), although the particles selected for map reconstruction clearly showed two RuvB motors within a single complex ( Figure 2A ). Moreover, because RuvAB catalyzes HJ branch migration in a non-sequence-specific manner, and the density for the nucleobases is the averaging result of all four possibilities, no sequence was assigned to HJ DNA in the structures.

The cryo-EM structure of PaRuvA-HJ region was determined at 3.0 Å with well-defined electron density ( Supplementary Figure S4A ). In agreement with the previous crystal structures of HJ-bound octameric RuvA from Mycobacterium leprae (M. leprae, Ml) (Roe et al., 1998), each PaRuvA subunit is composed of three distinct domains: D1, D2 and D3 ( Supplementary Figure S4B ). The HJ molecule is sandwiched by two PaRuvA tetramers that dimerize through a pair of antiparallel helices, involving at least four pairs of electrostatic interactions and one pair of hydrophobic interaction ( Figure 3A ). A triple mutation (E122K/V126A/D130K) at the dimerization interface significantly diminished the branch migration activity of PaRuvAB complex ( Figure 3A and Supplementary Figure S4C ), implying that the formation of RuvA octamer is required for efficient branch migration. The two tetramers together enclose a cross-shaped tunnel of ~ 23 Å in diameter, which encircles ~ 10 bp of HJ arm ( Figure 3A ; Supplementary Figure S4D, E ). The protein-DNA contacts at each HJ arm are mainly mediated by the RuvA^D2 and DNA phosphodiester backbone ( Figure 3B ). Interestingly, all RuvA^D2 in tetramer 1 insert into the DNA major groove, whereas the equivalent domains in tetramer 2 all bind the minor groove ( Figure 3B ). As a result, the two tetramers do not form a perfect two-fold symmetry but are misaligned by ~18° ( Supplementary Figure S4E ).

The junction center is devoid of base pairs but occupied by a conserved ‘RED’ motif (R54, E55 and D56) in RuvA^D1 ( Figures 3C–F ). In the motif, R54 binds to the ribose-phosphate backbone of the branch site with its positively charged guanidino group, while the carboxyl groups of E55 and D56 may provide strong negatively electrostatic repulsion for strand separation ( Figures 3D, E ). Of note, the base pairs at the branch sites of RuvB-free HJ arms are disrupted and interact with D56, whereas the equivalent base pairs at RuvB-bound arms remain intact ( Figures 3C–E ). Alanine substitution of R54 completely abolished the HJ branch migration activity, while the E55A and D56A mutations caused ~ 50% reductions ( Figure 3G ). These observations thus suggest that base pairs passing through the acidic center would be disrupted through the transient interactions with RED motifs, and after that, new base pairs would form once the DNA duplex are pumped out by RuvB motor through the exit gates ( Figure 3H ). This is the basis of DNA strand exchange at the HJ branch migration phase.

Structural comparison between the current cryo-EM structure of PaRuvAB-HJ complex and the previous MlRuvA-HJ complex (PDB ID: 7OA5)(Roe et al., 1998) revealed a prominent rearrangement of RuvA^D3 ( Figures 4A, B ). In the absence of RuvB, MlRuvA^D3 packs against MlRuvA^D2 of the neighboring subunit, resulting in severe steric clashes with HJ duplex arms ( Figure 4A ). In the current structure, however, all the eight PaRuvA^D3 spring out from the junction center and each individually captures a PaRuvB subunit ( Figures 4B–D ). As a result, the HJ DNA molecule can be perfectly docked onto RuvA without steric hindrance ( Figure 4B ). Consistent with these observations, while RuvA and HJ each alone had no effects on the ATP hydrolysis by RuvB, a combination of the two significantly stimulated its activity ( Figure 4E ), suggesting that RuvB alone can not for a functional closed hexameric ring onto HJ.

To reconstruct the 3D model of PaRuvB-dsDNA complex, we determined a 2.1-Å crystal structure of PaRuvB in complex with ADP. The crystal structure of PaRuvB-ADP complex is essentially identical to the Thermotoga Maritima (T. Maritima, Tm) RuvB (PDB ID: 1IN4), which consists of three consecutive domains (N, M and C) and contains canonical AAA+ ATPase fold at domain N ( Supplementary Figure S5 ). In the cryo-EM structure of PaRuvB-dsDNA complex, the four PaRuvB subunits pack against with each other using a binding interface similar to that used for crystal packing ( Figures 5A–C ). The intersubunit interaction involves E40 and R52 on α2 of one subunit and D233 and R238 on α11 of the other subunit ( Figure 5D ). In particular, the arginine finger R175 (equivalent to R174 of EcRuvB) is located on this interface and approaches to the β-phosphate of ADP in the neighboring subunit ( Figure 5D ). Charge-reversal mutations of D233/R238 or alanine substitution of R175 eliminated the branch migration activity ( Figure 5E ). These results thereby demonstrate that the intersubunit interface of RuvB observed here is functionally related.

The four PaRuvB subunits form an asymmetric spiral and encircle 22 bp of HJ arm, and two of them intimately contact the DNA minor groove with their β-hairpins in domain C, involving resides R314, R317 and R319 ( Figures 5A , 6A, B ). Alanine substitution of R314, R317 or R319 on the β-hairpin abolished the branch migration activity of PaRuvAB complex ( Figure 6C ), suggesting that they are involved in the interactions with DNA. Furthermore, because the β-hairpin is in close proximity to both DNA strands, we propose that R317 may bind a backbone phosphate from the first strand, while R314 and R319 may interact with two consecutive phosphates from the second strand ( Figure 6B ). This suggests that each PaRuvB translocates 2 bp of DNA per catalytic step, and for a full cycle within the intact RuvAB-HJ complex that containing 6 RuvB, 12 bp of DNA is translocated through the hexameric ring.

Structural analysis revealed distinct subdomain orientations of the four PaRuvB subunits. The angle between domains N and C defined by the Cα atoms of G165-V186-T314 increased from 34° for RuvB3 to 52° for RuvB1 ( Figure 6D ). The conformation for RuvB4 could not be unambiguously defined due to the poor electron density on the N-terminal part. Such gradual conformational changes have been correlated to different nucleotide states in other ring helicases (Enemark and Joshua-Tor, 2006; Thomsen and Berger, 2009; Jean et al., 2020). Due to the limited resolution of current cryo-EM structure, the nucleotide state for each PaRuvB subunit could not be immediately assigned. Despite that, structural comparisons showed that the overall conformations of RuvB3 and RuvB4 are close to the crystal structure of ADP-bound RuvB ( Figure 6E ), implying that these two DNA-free subunits are probably in ADP-bound state. Furthermore, the subunits involved in nucleic acid substrate binding are generally bound by ATP in the reported structures of hexameric helicases (Enemark and Joshua-Tor, 2006; Thomsen and Berger, 2009; Jean et al., 2020). Thus, the two subunits RuvB1 and RuvB2 are likely in ATP-bound state. We propose that DNA binding would trigger ATP hydrolysis and subdomain motion in RuvB, which in turn drives DNA translocation from domain N to domain C.

The cDNA encoding PaRuvA (GenBank ID: 879609) and RuvB (GenBank ID: 882028) proteins were synthesized at GenScript (Nanjing, China) and subcloned into pGEX-6P-1 bacterial expression vector containing an N-terminal glutathione S-transferase (GST) tag. The recombinant vectors containing RuvA and RuvB genes were transformed into E. coli strain Rosetta (DE3) and BL21 (DE3) competent cells, respectively. Cells were grown in the LB media at 37°C until the OD₆₀₀ reached around 0.8. Protein expressions were induced with 0.5 mM isopropyl-β-D-1-thiogalactopyranoside (IPTG) at 17°C overnight. The purification of RuvA and RuvB proteins was performed as previously described(Lin et al., 2019). Briefly, cells were resuspended with lysis buffer (50 mM HEPES pH 7.4, 200 mM NaCl, 5% (v/v) glycerol and 1 mM dithiothreitol (DTT)) and disrupted by high pressure homogenizer (Union Biotech, China). The GST-tagged proteins in the supernatant of cell lysates were pooled through a GST Sepharose column (Senhui Microsphere Technology, China). The GST tags were subsequently removed by home-made PreScission proteases and the untagged proteins were further purified using Source 15Q and Superdex 200 10/300 GL columns (GE Healthcare Life Sciences) in running buffer containing 20 mM HEPES pH 7.4, 200 mM NaCl, 15 mM reduced glutathione and 1 mM DTT. The purified proteins were concentrated, aliquoted and stored at -80°C. The mutant variants of RuvA and RuvB were expressed and purified similarly.

All of the oligonucleotides used in this study ( Supplementary Table S1 ) were synthesized and HPLC-purified by SunYa Biotechnology (Fuzhou, China), and dissolved in annealing buffer (100 mM HEPES pH 7.4, 50 mM NaCl). For HJ DNA annealing, the four HJ strands were divided into two groups, A/B and C/D. Each group of DNA strands were subjected to heating at 95°C for 5 min, followed by gradual cooling to room temperature. Thereafter, the annealed products of both groups were mixed and incubated at 37°C for 30 min, and then cooled down to room temperature. For DNA imaging, one strand of HJ was labeled with 6-carboxyfluorescein (FAM). For FRET-based measurement, two complementary strands of HJ were labeled with Cyanine 3 (Cy3) and Black Hole Quencher^® 2 (BHQ2), respectively.

The ATPase activity of RuvB was determined by a firefly luminescence assay as described previously (He et al., 2022). Briefly, in 100 μl reaction, various concentrations of RuvB proteins were mixed with reaction buffer containing 50 mM Tris-HCl pH 7.4, 100 mM NaCl, 10 mM MgCl₂, 1 mM DTT, 0.1% Tween20. Reactions were initiated by the addition of 10 μM ATP, and further incubated at 37°C for 1h. Subsequently, the reactions were stopped by addition of Kinase-Glo reagent (Promega), and the residual amounts of ATP were measured by luminescence on the SpectraMax-L microplate reader (Molecular device).

The branch migration assay was performed as previously described (Hishida et al., 1999; Dawid et al., 2004). Briefly, 20-μl of reactions containing 250 nM FAM-labeled HJ DNA substrates (HJ1, Supplementary Table S1 ) and various concentrations of RuvA and RuvB proteins in branch migration buffer (50 mM Tris pH 8.0, 10 mM MgCl₂, 20 mM NaCl, 1 mM DTT, 1mM ATP, 0.1 mg/ml BSA) were incubated at 37°C for 60 min. The reactions were terminated by adding 2 μl of proteinase K (2 mg/ml) and 5 μl of 5× stop buffer (100 mM Tris pH 8.0, 250 mM EDTA, 10% (w/v) SDS, 5% (v/v) glycerol and 0.1‰ (w/v) bromophenol blue), followed by incubation at 37°C for 15 min. The products were analyzed by 10% TBE-native PAGE (110 V, 40 min) and visualized with the ChemDoc™ touch imaging system (Bio-Rad, USA). For FRET-based measurement of branch migration activities, in 50-μl aliquots, 250 nM Cy3/BHQ2-labeled HJ DNA substrates were mixed with 800 nM RuvA, 1200 nM RuvB, or indicated protein concentrations, in branch migration buffer. Reactions were initiated with 1 mM ATP and incubated at 37°C. The fluorescence was recorded every 2 min using a SpectraMax i3x plate reader (Molecular Devices, USA) with λ_ex/λ_em = 542 nm/567 nm.

For crystallization of RuvB-ADP complex, RuvB (10 mg/ml) was pre-incubated with 10 mM ADP in 20 mM HEPES pH 7.4, 200 mM NaCl, 10 mM MgCl₂, and 1 mM DTT. The protein solution was mixed with an equal volume of precipitant solution containing 0.2 M L-Proline, 0.1 M HEPES pH 7.5, 10% (w/v) polyethylene glycol 3350 (PEG3350) and 10 mM Trimethylamine hydrochloride. Crystals were grown under 18°C using the sitting drop vapor diffusion method. Diffraction-quality crystals were fast-frozen in liquid nitrogen with reservoir solution supplemented with 15-25% (v/v) glycerol. X-ray diffraction data were collected at beamline BL19U1 at Shanghai Synchrotron Radiation Facility (SSRF) with a wavelength of 0.978 Å, and processed with HKL3000(Minor et al., 2006). The crystal structure of RuvB-ADP complex was solved by molecular replacement with the program Phaser-MR in the PHENIX package (Liebschner et al., 2019), using the structure of T. Maritima RuvB (PDB ID: 1IN4) as a searching model. Iterative manual model building and structural refinements were conducted using the programs of Phenix (Liebschner et al., 2019) and WinCoot (Emsley et al., 2010). The final model was validated by MolProbity (Chen et al., 2010). Data collection and structure refinement statistics were summarized in Supplementary Table S2 .

To reconstitute RuvAB-HJ complex, HJ DNA substrates (HJ2, Supplementary Table S1 ) in 50 mM NaCl, 20 mM HEPES pH 7.4, 1 mM ATPγS, 10 mM MgCl₂ and 1 mM DTT were pre-incubated with RuvA at 4°C for 15 min, and then further incubated with RuvB for another 15-min, with a final molar ratio of RuvA:RuvB:HJ = 8:16:1.2. The RuvAB-HJ complex was further purified through a Superose 6 10/300 GL column (GE Healthcare Life Sciences) equilibrated in buffer containing 20 mM HEPES pH 7.6, 100 mM NaCl, 10 mM MgCl₂ and 1 mM DTT. The corresponding fractions were pooled and concentrated to about 4 mg/ml. Samples were subjected to negative staining for homogeneity analysis prior to cryo-EM experiments. The cryo-EM grids were prepared by applying 2 μl aliquot of the sample to glow-discharged Quantifoil™ R1.2/1.3 300 mesh gold grids. The grids were blotted for 3 s at force-3 before being plunged into liquid ethane using a Vitrobot Mark IV (Thermo Fisher Scientific) operated at 4°C and 100% humidity. Grids were transferred to an FEI Titan Krios (Thermo Fisher Scientific) operated at 300 kV high tension and images were collected semi-automatically with SerialEM 4.0 software under low-dose mode at a nominal magnification of 130,000×, with a resulting pixel size of 0.83 Å. A Gatan K3 Bioquantum summit direct electron detector was used under super-resolution mode for image recording with defocus values ranged between -3.0 and -1.0 µm, and a total accumulated dose of 60 e^-/Ų on the specimen. Detailed information was summarized in Supplementary Table S3 .

Image processing was performed using the software packages RELION 3.1 (Zivanov et al., 2020) and cryoSPARC 3.1 (Punjani et al., 2017), and summarized in a flowchart in Supplementary Figure S2 . A total of 5,535 movie stacks were subjected to beam-induced motion correction using MotionCor2 in Relion 3.1 (Zheng et al., 2017), and their contrast transfer function (CTF) parameters were estimated with CTFFIND4 in cryoSPARC (Rohou and Grigorieff, 2015). About 2,000 particles from different views were manually picked and 2D classified. Using these 2D classes as references, a total number of 584,587 particles were automatically picked and extracted (box size 480 pixels) from denoised images with Topaz in cryoSPARC (Bepler et al., 2019; Bepler et al., 2020). All the particles were then 2D classified and the unclassified were removed. After 2D classification, 551,160 particles appeared as dumbbell shape were selected and sorted into four classes through Ab-Initio reconstruction and heterogeneous refinement. Subsequently, 143,397 particles with the coordinate information were selected and re-extracted by Relion 3.1 and further improved using 3D refinement, post-processing, Bayesian polishing, and homogeneous refinement, resulting in an overall density map with a resolution of 3.21 Å. The refined particles and a soft mask focused on RuvA region were used to build the final density map of RuvA region with a resolution of 3.01 Å, as determined by Fourier shell correlation (FSC) using the correlation cutoff value of 0.143 (Chen et al., 2013). To probe the conformational dynamics and improve the resolution of RuvB region, 101,792 Bayesian polished particles corresponding to RuvAB-HJ complexes were subjected to another round of 3D focused classification without alignment in Relion 3.1 and 3D variability analysis in cryoSPARC (Punjani and Fleet, 2021), respectively. Particle subsets with improved density in RuvB region were selected after 3D skip alignment classification (25,482 particles) and 3D variability analysis (29,607 particles), and the duplicate particles (16,258 particles) were removed by particle sets tool in cryoSPARC. After another round of heterogeneous refinement, 20,536 particles were selected for final homogeneous refinement with the application of a soft mask, yielding an overall density map for RuvAB-HJ complex with a resolution of 6.18 Å. After local refinement with a mask of RuvB region and map sharpening with a negative B factor of -300 Ų, the resolution of density map within the RuvB region defined by the soft mask is estimated to be 7.02 Å.

The RuvA-HJ model was built with a 3.01-Å-resolution map using the crystal structure of E. coli RuvA-HJ (PDB ID 1C7Y) (Ariyoshi et al., 2000) as template. The model of RuvB-dsDNA complex was built with a cryoEM map of 7.02 Å, using the crystal structure of RuvB-ADP complex determined in this study. Each subunit was manually fitted into the density map in UCSF Chimera 1.15 (Pettersen et al., 2004), and then refined using Phenix(Liebschner et al., 2019) and WinCoot (Emsley et al., 2010) and the ISOLDE flexible atomic model refinement package in UCSF ChimeraX 1.2.5 (Croll, 2018). The resulting models of RuvA-HJ and RuvB-dsDNA complexes were used to build a composite model of RuvAB-HJ complex using UCSF Chimera 1.15, which was refined with a similar procedure mentioned above. The final models were finely adjusted for geometry violations guided by electron density. Cryo-EM data collection, reconstruction and refinement statistics and the final refinement statistics were summarized in Supplementary Table S3 . All structural figures in this paper were generated with UCSF ChimeraX 1.2.5 (Goddard et al., 2018) and PyMOL (http://www.pymol.org/).