Mechanism of Concerted RNA-DNA Primer Synthesis by the Human Primosome*

The human primosome, a 340-kilodalton complex of primase and DNA polymerase α (Polα), synthesizes chimeric RNA-DNA primers to be extended by replicative DNA polymerases δ and ϵ. The intricate mechanism of concerted primer synthesis by two catalytic centers was an enigma for over three decades. Here we report the crystal structures of two key complexes, the human primosome and the C-terminal domain of the primase large subunit (p58C) with bound DNA/RNA duplex. These structures, along with analysis of primase/polymerase activities, provide a plausible mechanism for all transactions of the primosome including initiation, elongation, accurate counting of RNA primer length, primer transfer to Polα, and concerted autoregulation of alternate activation/inhibition of the catalytic centers. Our findings reveal a central role of p58C in the coordinated actions of two catalytic domains in the primosome and ultimately could impact the design of anticancer drugs.

In eukaryotes, the primosome, a tight complex of DNA primase and DNA polymerase ␣ (Pol␣), 4 synthesizes primers for both leading and lagging strands in a highly coordinated fashion (1,2). The primosome is indispensable for initiation of replication and has a large impact on genome stability (3)(4)(5)(6). RNA primer synthesis by primase involves three steps: initiation, elongation, and termination (7,8). During the rate-limiting initiation step, primase binds the DNA template and two ribonucleotide triphosphates (NTPs) and catalyzes the formation of a dinucleotide (9,10). Further synthesis of the RNA primer is much faster but restricted, because of the intrinsic property of primases to count the primer length and terminate synthesis after incorporation of 8 -10 nucleotides (7). Next, the mature so-called "unit length" RNA primer is intramolecularly translocated to Pol␣ for the subsequent extension by dNTPs, and the primase became inhibited by an unknown mechanism (9,11,12). Orchestration of all these steps requires changes in primosome conformation (13).
Structural characterization of eukaryotic primosome is limited to the low-resolution electronic microscopy data (13) and the x-ray structures of its parts. This includes the crystal structures of the intact human primase (21) and its truncated variant without p58 C (22), as well as the high resolution structures of p180core (31), p180 C -p70 (14), p58 C (25,30), p49 (33), and their yeast orthologs (15,29,32,33). The interaction of eukaryotic primases with substrates has never been structurally characterized except for the binding site for an incoming nucleotide triphosphate (NTP) (22,33). None of these data provide clear clues on the details of RNA-DNA primer synthesis including initiation, elongation, accurate counting of RNA primer length, primer transfer to Pol␣, and coordinated action of two catalytic centers. The comprehensive understanding of these mechanisms requires crystal structures of primosome in complex with a variety of substrates. Crystallization of such complexes is extremely challenging due to the significant mobility of the functional domains. We elected to obtain mechanistic insight into the primosome transactions by focusing on solving two key structures: the primosome in apo-form, and the C-terminal domain of the human primase large subunit (p58 C ) bound to the DNA template/RNA primer (D/R), p58 C -D/R. The results of these efforts, in combination with functional analysis, lead to delineation of all the steps of RNA primer synthesis and its transfer to Pol␣.
Crystallization of Human Primosome-Primosome was concentrated to 10 mg/ml in buffer containing 10 mM Tris-HCl (pH 7.7), 0.1 M KCl, 1% glycerol, and 1 mM DTT and then flashfrozen in aliquots. The aliquots were defrosted and centrifuged at 40,000 ϫ g to remove the aggregated molecules, and the sample monodispersity was verified with dynamic light scattering. Screening was performed with the sitting-drop vapor diffusion method at 295 K by mixing 1 l of protein solution with 1 l of reservoir solution. The reservoirs contained 2-fold FIGURE 1. Crystallization of the primosome and p58 C -D/R. A and B, photomicrographs of the primosome crystals before and after optimization, respectively. C, a single crystal of primosome scooped in a nylon-fiber loop. D, analysis of the content of primosome crystals by 7% SDS-PAGE. Lane 1, molecular weight markers; lanes 2 and 3 correspond to the crystals shown in A and B, respectively. E and F, photomicrographs of p58 C -D/R crystals in the form of plates and needles, respectively. G, analysis of the content of a needle-like p58 C -D/R crystal by 5% native PAGE. Gel was stained first with ethidium bromide (left panel) and then with Coomassie R-250 (right panel).
diluted Crystal Screen, Crystal Screen 2, and Natrix (Hampton Research) solutions with the addition of 2 mM tris(2-carboxyethyl)phosphine (pH 7.5). Within 1-2 months the oil-like spherulites in the 17th condition of the Crystal Screen became covered by barely visible needles (Fig. 1A). After optimization and multiple screenings of additives, thin plates in the form of parallelograms appeared and grew on the precipitates in 3-5 days (Fig. 1B). The best reservoir solution contained 0.2 M lithium sulfate, 50 mM Tris-HCl (pH 8.5), 2 mM tris(2-carboxyethyl)phosphine (pH 7.5), 11.2% (w/v) PEG 4,000, 3% (v/v) ethanol, 0.5% (v/v) polypropylene glycol P400, and 0.2 mM EDTA. The crystals grew over 1 mm in length but were extremely thin and fragile, easily breaking into pieces due to the movement of solution around them. The resolution of the x-ray diffraction was time-dependent, indicating that crystals with larger volume (i.e. thicker) may have improved diffraction at shorter exposures. Multiple rounds of seeding produced crystals with a thickness of 2-4 m that were diffracting anisotropically up to 3.2-Å resolution. However, the rate of success of seeding was too low; in addition to difficulties in peaking and washing the fragile single crystals, most seeds had a propensity to grow as layers and began to be covered by needle-shaped crystals within 30 -40 h before completion of growth. Only single-appearing crystals were picked for the next round of seeding. We obtained over 50 crystals by seeding, but only two were diffracting as single crystals. One crystal diffracting up to 3.6 Å was used for the optimization of data collection conditions at the synchrotron, and the second, better diffracting crystal (Fig. 1C) was used for the complete data set collection. The washed and dissolved crystals revealed the presence of all subunits as in the original sample (Fig. 1D). We also crystallized the primosome with a longer p180 subunit (residues 292-1462). The growth conditions and shape of the crystals were similar as in primosome containing p180 residues 335-1462. However, the crystals were not diffracting.
Crystallization of p58 C -D/R-Based on our recent biochemical results (26), we designed the optimal substrate for p58 C with a 12-mer DNA template T8 annealed to a 6-mer RNA primer P1 containing a 5Ј-triphosphate. Purified p58 C was exchanged into buffer containing 5 mM Tris-HCl (pH 7.9), 20 mM NaCl, 0.5% glycerol, and 1 mM DTT using a Superose 12 column (GE Healthcare). Next, the 6-bp DNA/RNA duplex T8-P1 was added at 10% molar excess; then the complex was concentrated to 12 mg/ml and flash-frozen in aliquots. Screening was performed as described for primosome, with the exception that the screen solutions were not diluted. Small thin plates appeared in the 15th condition of the Crystal Screen. The optimized condition (0.125 M ammonium sulfate, 0.1 M sodium cacodylate, pH 6.5, 0.02 M tris(2-carboxyethyl)phosphine, and 28.5% (w/v) PEG 8,000) produced slow-growing, well shaped rectangular plates in 3-4 weeks (Fig. 1E). These crystals diffracted poorly, with the diffused reflections extended up to a resolution of only 12-15 Å. In an attempt to obtain a new crystal form with better packing, we varied the type of salt and buffer and added a variety of compounds from the Hampton Research Additive screen. Eventually, we succeeded with a combination of 0.18 M lithium sulfate, 50 mM sodium citrate, 2 mM tris(2-carboxyethyl)phosphine, 0.1 M sodium malonate, 0.1 M Hepes-NaOH (pH 7.0), 19.7% PEG 8,000, and 0.02 M magnesium chloride that produced relatively fast-growing (within 2-4 days) needle-like prisms, with a square shape in crosssection (Fig. 1F). The crystals contained both p58 C and DNA/ RNA components (Fig. 1G) and diffracted to 2.2-Å resolution.
Data Collection-For diffraction data collection, the crystals were soaked in a cryoprotectant solution for a few seconds, scooped in a nylon-fiber loop, and flash-cooled in a dry nitrogen stream at 100 K. The cryoprotectant for the primosome crystals was prepared by increasing the PEG 4,000 concentration in the reservoir solution to 14% (w/v) and inclusion of 18% (v/v) ethylene glycol. In the case of p58 C -D/R crystals, the PEG 8,000 and ethylene glycol concentrations were 21 and 18%, respectively. Because the latter crystals were grown in the presence of Mg 2ϩ , we refer to them as p58 C -D/R-Mg. To exchange Mg 2ϩ to Mn 2ϩ , we soaked crystals for 5 min in a reservoir solution containing 2 mM manganese chloride instead of magnesium chloride. The resulting crystals are referred to as p58 C -D/R-Mn. The cryoprotectant was also prepared with 2 mM manganese chloride. Preliminary crystal characterizations and full dataset collection from p58 C -D/R-Mn crystals were performed on a Rigaku R-AXIS IV imaging plate using Osmic VariMax TM HR mirror-focused CuK␣ radiation from a Rigaku FR-E rotating anode operated at 45 kV and 45 mA. Complete diffraction data sets for primosome and p58 C -D/R-Mg crystals were collected using synchrotron x-rays on the Argonne National Laboratory Advanced Photon source beamline 24ID-C with the Pilatus 6MF pixel-array detector. Continuous vector scan along the crystal length was applied to reduce the radiation damage ( Fig.  1, C and F). All intensity data were indexed, integrated, and scaled with DENZO and SCALEPACK from the HKL-2000 program package (36). The crystal parameters and data processing statistics are summarized in Table 2.
Structure Determination-Only one complete native dataset from a single crystal was available for the phasing and refinement of the human primosome structure. Twinning and nonisomorphism among heavy-atom derivative crystals prevented merging data from several smaller crystals that were collected with a rotation of 10°. To apply the molecular replacement, we needed initial models; therefore, we started and completed structural studies of the three largest parts of primosome: p180core with bound substrates and the inhibitor aphidicolin, full-length primase, and p180 C -p70 (PDB accession numbers 4QCL (31), 4Q5V (31), 4RR2 (21), and 4Y97 (14), respectively). An initial search was performed with the p180core coordinates from PDB entry 4Q5V, because, in complex with aphidicolin, the fingers were in an open conformation similar to apo-form. Clear solutions were obtained for both independent molecules only when we removed the thumb and palm domains. Both independent positions of p180 C -p70 and their locations relative to p180core were also clearly resolved by molecular replacement. However, the search results were not clear with models containing the palm and thumb domains of p180core, or primase (with or without p58 C ), or any one of p49, p58 N , and p58 C . The density modifications and averaging using the phases from the model containing Pol␣ without palm and thumb domains of p180 revealed positions of the latter domains and primase subunits that were manually placed into the electron density maps. Anomalous difference maps confirmed the locations of two zinc ions in p180 C and one in p49, along with the location of the 4Fe-4S cluster in p58 C for both independent primosome molecules. To avoid model bias, we verified the position of each domain by omitting its coordinates separately in each molecule as well as simultaneously from both molecules. After positioning of all the domains, we verified the structures of the loops and terminal residues. Zonal scaling (37) and bulk solvent corrections were applied to structure factors. Rigid-body refinement was performed after each cycle of model building. Positional and B-factor refinements were applied only after the completion of model building. Noncrystallographic 2-fold symmetry constraints were applied during the initial stages of refinement; these were replaced by restraints of backbone atoms during the final stages of refinement. The structure was refined at 3.6-Å resolution to an R cryst of 26.9% and an R free of 32.8% and contains amino acids 1-283, 289 -360, and 379 -412 of p49; 22-455 of p58; 338 -672, 680 -808, 842-882, 898 -1258, and 1266 -1456 of p180; and 155-598 of p70.
The structure of p58 C -D/R-Mg was solved by the molecular replacement method using the coordinates of p58 C from the structure of human primase (PDB accession number 4RR2 (21)) as a search model. Clear solutions were found for two molecules in an asymmetric unit. Initial electron density maps revealed a 6-bp DNA/RNA duplex and part of the template 3Ј-overhang for both molecules. Their addition into the structure further clarified the packing of the remaining residues for the template 3Ј-overhang. One molecule contained all template residues that were stabilized by crystal packing, whereas no density was found for the dAde Ϫ6 and dThy Ϫ5 in another molecule. Residues 354 to 363 were rearranged in the p58 C -D/R-Mg structure due to interaction with template DNA. No density was found for N-terminal residues 266 -269. Non-crystallographic 2-fold symmetry restraints were applied during the initial stages of model refinement. After the addition of solvent molecules, the model was refined at 2.2-Å resolution to an R cryst of 22.3% and an R free of 25.8%. The electron density peak close to the GTP phosphates of the primer was assigned as an Mg 2ϩ . For additional confirmation that the triphosphate moiety coordinates a divalent metal, the Mg 2ϩ was exchanged with a heavier Mn 2ϩ . A difference F o Ϫ F c Fourier map revealed a significantly stronger peak at the same position with a contour level of over 8 (Fig.  7B), confirming the metal binding. The structure of p58 C -D/ R-Mn was refined at 3-Å resolution to an R cryst of 22.5% and an R free of 27.6%. The final refinement statistics for all structures are provided in Table 2. Inspections of the electron density maps and the building/adjustment of the structures were performed with Turbo-Frodo. All crystallographic computing was performed with CNS version 1.1 (38). Supplemental Movie S1 and the figures containing molecular structures were prepared with PyMol (39).

Results
Crystal Structure of Human Primosome-The structure of the apo-form of primosome has been determined at 3.6-Å resolution (Table 2). It acquires an extended shape with the dimensions of 148 ϫ 119 ϫ 107 Å 3 and packs as three massive parts: p180core, p180 C -p70, and primase (Fig. 2B). p180 C -p70 associates with p180core by wedging the Zn2 module of p180 C and OB domain of p70 into the substrate-binding area between the exonuclease and thumb domains of p180core and displacing the tip of the thumb up to 32 Å in comparison with a substrate-bound form (Fig. 3). As a result, the Pol␣ active center is blocked, which explains the inhibitory role of p180 C -p70 on Pol␣ activity (35). An elongated primase locates on p180core and stretches from its thumb to the N-terminal domain. Total buried surface area between the p180core and the rest of the molecule is 6117 Å. The distance between the primase and Pol␣ catalytic centers is 57 Å.
Substantial conformational changes are essential for the seamless accomplishment of all the steps of primer synthesis; therefore, the relative positioning of the moving domains and the linkers between them play the crucial role in primosome transactions (Fig. 4A). The 180core-p180 C linker is only partially defined in the structure. Some electron density was found for linker residues 1251-1258 that are packed as a random coil, but no density was found for the remaining portion of the linker (1259 -1265). The residues 1243-1250 preceding this linker form a well stabilized ␣-helix in the primosome structure (Fig. 4B). However, electron density for these residues was missing in the p180core structures (31), indicating that they could become a part of the 180core-p180 C linker in the case of p180core dissociation from the rest of primosome. The p58 N -p58 C linker has a well defined density (Fig. 4C). The position of p58 C relative to p58 N is different in the primosome and primase (21), supporting the idea of flexibility of this linker (Fig. 4D). The third linker, which connects the p58 N -interacting C-terminal peptide of p180 (1447-1456) with the rest of p180 C , has been proposed to provide the flexible connection between p180 C -p70 and primase (22). Interaction of p180 C -p70 with primase rearranges the folding of p180 residues C-terminal to Thr-1436 and the surrounding residues 197-210 of p70, making this part of p180 C -p70 much more stabilized and resolved in the primosome structure than in p180 C -p70 (14) (Fig. 5, A-C). The folding of residues 1437-1442 into the ␣-helix reduces the size of the linker, placing p180 C -p70 and primase in close proximity and restricting their movement relative to each other (Fig.  5D). As a result, the substructure p49-p58 N -p180 C -p70 of primosome has a defined elongated shape, whose basic architecture is not influenced by interactions with p180core or p58 C and resembles the low-resolution electron microscopy images of the yeast ortholog (13). Therefore, p49-p58 N -p180 C -p70 can be considered as a platform that holds p180core and p58 C stationary by docking in the inactive form or flexibly by linkers during various stages of primosome transactions. Both linkers are closely packed with a distance of 9 Å between the points of their attachment to the p49-p58 N -p180 C -p70 platform (Fig. 4A).
Crystal Structure of p58 C -D/R-Currently, there is no structural data for any primase in the complex with a templateprimer. We recently defined the elements of the DNA template and the RNA primer that are important for interaction with the human primase and found that all of them interact with p58 C (26). These findings allowed us to design the optimal substrate and prepare a stable p58 C -D/R complex for crystallographic studies. The structures of p58 C -D/R in complex with Mg 2ϩ or Mn 2ϩ have been determined at 2.2-and 3.0-Å resolutions, respectively ( Table 2). The only notable difference between these two structures is the nature of the divalent metal ion. The p58 C -D/R-Mg complex has a bean shape where the RNA-primed DNA template binds to the positively charged depressions on the surface of p58 C with a primer 5Ј-triphosphate on one side and a template 3Ј-overhang on other side (Fig. 6, A and B). Furthermore, the electrostatic interactions are extended by the DNA backbone of the DNA/RNA duplex. This structure explains why the elimination of 5Ј-triphosphate and/or 3Ј-overhang abolished the interaction of DNA/RNA with p58 C as well as with fulllength primase in our recent study (26).
Only the ␤and ␥-phosphates of the triphosphate moiety are involved in interactions with p58 C , via six hydrogen bonds with the side chains of His-300, Arg-302, Arg-306, Tyr-345, and the main chain of His-303 (Fig. 6C). Arg-306, which interacts with both the ␤and ␥-phosphates, has been shown to be important for RNA synthesis (22,24,27). The triphosphate coordinates the Mg 2ϩ or Mn 2ϩ ion that provides additional stability (Fig. 7,  A and B). The template forms 13 hydrogen bonds with the protein: eight with the 3Ј-overhang and the rest with the duplex portion (Fig. 6D).
There are no dramatic conformational changes in p58 C induced upon DNA/RNA binding apart from stabilization of the loop comprised by residues 354 -366; it is partially disordered in the structure of apo-p58 C (25) or acquires different crystal packing-induced conformations as seen in the structures of full-length primase (21) and primosome (Fig.  7C). However, p58 C affects the conformation of the DNA/ RNA substrate: it maintains the B-DNA conformation of the template deoxyriboses, except at positions 4 -6 that are in the A-DNA conformation.
Primase Initiation Complex-p58 C -D/R structure provides an intriguing and long-awaited mechanism elucidating the docking of the initiating GTP (GTP 1 ) (Fig. 6). The well defined interactions of the template phosphates at positions Ϫ2, Ϫ1, 1, and 2 with p58 C help to orient its bases at positions Ϫ1, 1, and 2 for a duplex formation. The dCyt 1 base conformation is additionally stabilized via a hydrogen bond with a side chain of Asn-348. As a result, the dAde Ϫ1 base stacks between the hydrophobic side chain of Met-307 and dCyt 1 base. Interestingly, the side chain of His-303 mimics the RNA base that is paired with the template dAde Ϫ1 via a hydrogen bond and supports the stacking of the initiating GTP. The structure of p58 C -D/R together with the structures of the full-length human primase (21) and the primase-UTP complex (22) provides all necessary components for the modeling of the primase initiation and elongation complexes (Fig. 8). Superimposition of the second nucleotide of a primer in p58 C -D/R with the incoming UTP in primase revealed a well defined shape complementarity between the catalytic p49 subunit and p58 C domain (Fig. 8A). Unlike the elongated substratefree primase (21), the initiating complex acquires a globular shape. It reveals the formation of a "mini duplex" (Fig. 8B) that is comprised of three template nucleotides at positions Ϫ1, 1, and 2 on one strand and His-303 of p58, initiating GTP, and incoming GTP on the other strand. The stacking of the DNA bases is extended with the side chains of p49 Tyr-54 from one side and p58 Met-307 from the other side. The duplex is partially buried in the middle of the complex, but the channel for loading an incoming NTP remains open (Fig.  8A). The formation of the initiation complex results in 16 additional hydrogen bonds: six between DNA phosphates and p49, and eight between p49 and p58 C , one between p49 A, primosome has two long linkers connecting p180core and p58 C to the platform composed of p49-p58 N -p180 C -p70. Primase linker (colored red) connects p58 C and p58 N ; Pol␣ linker (colored green) connects p180core and p180 C . B, analysis of the electron density map for p180 residues 1243-1250 confirms their packing as ␣-helix. Nitrogens, oxygens, and carbons are colored blue, red, and orange, respectively. C, the carbons in p58 N , p58 C , and the linker are colored cyan, gray, and magenta, respectively. D, the relative positions of p58 C in the primase (PDB ID 4RR2) and the primosome structures (only p58 N was used for alignment). Arg-163 and the ␣-phosphate of initiating GTP, and one between p49 Asp-306 and O2Ј of initiating GTP (Fig. 9). Moreover, the O3Ј of the initiating GTP coordinates with an Mg 2ϩ ion that is also bound to Asp-111 of p49 and the ␣-phosphate of the incoming GTP (Fig. 9C). The interaction between the O2Ј of initiating GTP and Asp-306 of p49 stabilizes the ribose during dinucleotide synthesis, which explains the strict preference for ribonucleotides at this step. The abundance of additional interactions is supportive for the accuracy of the model and also explains why substitutions of the participating residues, Tyr-54, Arg-56, and Arg-163 of p49, resulted in the loss of primase activity (22,40).
The Mechanism of Primer Length Counting by Primase-There are two alternative models for counting: one is based on the restriction of the distance between the two lobes of the large primase subunit by the linker (21,41), and another is based on steric hindrance (7,25). The obtained structures finally allow us to discover an exact mechanism of primer length counting.
The p58 N -p58 C linker acquires the most extended conformation in the initiation complex, with a distance of 43 Å between its N and C termini (Figs. 8C, panel 2, and 10A). Further RNA elongation up to a 9-mer gradually reduces this distance to 19 Å (Figs. 8C, panel 10, and 10A). Considering the capacity of the linker to be extended by over 60 Å, it can easily tolerate the extension of the primer well beyond the unit length, and hence cannot be the reason for primase counting. Indeed, when the linker was artificially shortened, the primase activity was reduced but the size of the products did not change (21).
Inspection of elongation complexes with 2-to 7-bp DNA/ RNA duplexes points to the absence of any steric hindrance between p58 C and the helical bundle of p58 N (Fig. 8C, panels  3-8). Some noticeable steric hindrance between them is observed for an 8-bp duplex (Figs. 8C, panel 9, and 10B). However, the helical bundle of p58 N can move nearly perpendicular to its axis and acquire a range of positions that accommodate the structure with an 8-bp duplex and incoming NTP (Fig. 10B). Consequently, the reaction can proceed efficiently to form a 9-bp duplex. Unlike the case with an 8-mer primer, p58 C with a 9-mer primer produces a steric hindrance at the top of the helical bundle in a direction that is parallel to its axis, and hence the clash cannot be relieved by flexibility in p58 N (Fig. 8C, panel  10). As a result, the pre-catalytic alignment of the 3Ј-OH of a primer and the ␣-phosphate of incoming NTP is compromised, FIGURE 6. Structure of p58 C in complex with the RNA-primed DNA template. A, specific recognition of the template/primer junction at the 5Ј terminus of the primer by p58 C . B, the surface electrostatic potential of p58 C shows docking of the template and the 5Ј-triphosphate of the primer on two separate positively charged areas (colored blue). C, hydrogen bonds (dashed lines) between p58 C and the ␤and ␥-phosphates of GTP 1 . D, hydrogen bonds between p58 C and the template.   (58). C, models of the initiation and elongation complexes. The crystal structures of p58 C -D/R, human primase (PDB code 4RR2), and p49-p58 N -UTP complex (PDB code 4BPW) were used to build these models. Panel numbers correspond to the length of the growing RNA primer plus incoming NTP in the modeled complexes. Panel 2 corresponds to the initiation complex resulting in dinucleotide formation and also shows the position of p58 C in apo-primase. The clash area is depicted by dotted circles (colored green or red to indicate avoidable or unavoidable steric hindrance, respectively), whereas the arrows show the direction at which p58 C is pushing p58 N . Close-up view of the complex in panel 9, explaining how steric hindrance is avoided, is provided in Fig. 10B. The image on the right shows the typical pattern of primase-catalyzed reaction products.
which explains the dramatically reduced efficiency of the 10th nucleotide incorporation. Addition of an 11th NTP would be impossible (panel 11). Therefore, interdomain steric hindrance is a cause of exact primer counting.
To verify the mechanism of counting, we examined the proficiency of primase in extension of 6-, 9-, and 10-mer primers (Fig. 11A). Primase extends a 6-mer primer very efficiently (lanes 2 and 3), whereas extension of 9-mer (lanes 8 and 9) and especially 10-mer primers (lanes 4 -7) is dramatically decreased. The result means that primase not only terminates primer synthesis when its length is nine nucleotides, but also cannot efficiently extend the unit length primers.
The Mechanism of the Primase-Pol␣ Switch-According to earlier data, the DNA/RNA duplex does not dissociate from primosome after synthesis of the unit length RNA primer, which allows for intramolecular substrate transfer between the two active sites (9,11). However, the mechanism of switch and involved domains remained unknown. Recently we found that p58 C makes a tight complex with a 7-bp DNA/RNA duplex with almost the same affinity for the template/primer as a whole primase (26). Moreover, the final and intermediate products of primase reaction have similar affinity for primase (Fig. 11B). Altogether, these data suggest that p58 C holds the template/ primer during its transition from primase to Pol␣, which results in concurrent binding of the unit length primer by p58 C and Pol␣ during the switch. We created a model of the switch complex, which contains p180core and p58 C sharing the 9-bp DNA/RNA duplex, the optimal substrate for Pol␣ FIGURE 9. p49-DNA and p49-p58 C interaction interfaces in the model of initiation complex. A, interaction of the DNA template with p49 and p58 C . The hydrogen bonds between DNA and p58 C or p49 are depicted by pink or gray dashed lines, respectively. B, the potential p49-p58 C interaction interface shows its plasticity due to the absence of hydrophobic contacts. C, p49 interacts with a ribose and ␣-phosphate of the initiating GTP. Second magnesium ion was modeled as described in the legend to Fig. 8B. recognition (31) (Fig. 12). This model illustrates good complementarity between surfaces of these two domains, indicating that the 3Ј-end of 8-mer or shorter primers will not reach the catalytic site.
To verify the model and define the precise position of the switch (i.e. the length of the primer at which the 3Ј-end of RNA primer becomes accessible for the Pol␣ catalytic site), we analyzed the products of primosome-catalyzed extension of a 6-mer primer in reactions with various composition of the rNTPs/dNTPs pool (Fig. 13A). The template sequence directs incorporation of UTP/dTTP at the 8 -10th positions and CTP/ dCTP at the 11th position of the primer. The incorporation of dNMPs (versus rNMPs) could be distinguished because of slightly higher mobility of the corresponding products on the gel. In the presence of ribo-and deoxy-NTPs, the primase incorporates only ribonucleotides (Fig. 13A, lane 1). Analysis of the products generated by the primosome reveals that the insertion of dNMPs begins from the 10th position (lanes 2 and  3). This means that Pol␣ can gain access to the 3Ј-end of the primer when its length is 9 nucleotides or longer. The exclusion of dCTP (lanes 5 and 6) or both CTP/dCTP (lanes 8 and 9) from the reaction, which inhibits the processive primer extension by Pol␣, demonstrated that 10-mer primers mainly contain the 3Ј-dTMP. In the absence of dTTP, the primase to Pol␣ switch with a 9-mer primer is abrogated and we can observe the switch with a 10-mer primer (lanes 11 and 12). These results clearly show that the switch mainly happens when the primer length is 9. If Pol␣ is unable to pick up the 9-mer primer, it can easily start with a 10-mer. Incorporation of dTMP was not observed using templates T3 and T2 that were designed to stop the primosome after synthesis of the 8and 9-mer primers, respectively (Fig. 13B). These results confirm that Pol␣ cannot start extension of primers shorter than 9-mer (when p58 C holds them) and support the structural model (Fig. 12). FIGURE 10. Conformational changes in human primase during RNA primer synthesis. A, position of the p58 N -p58 C linker in apo-primosome (Fig. 2B), in the initiation complex (Fig. 8A) and in the primase complex with a DNA template primed by a 9-mer primer and the incoming UTP (Fig. 8C, panel 10). The alignment is performed using p58 N . B, close-up view of the primase-DNA/RNA-CTP elongation complexes containing an 8-mer primer. The complexes were modeled based on the structures of primosome (p58 N is colored gray), primase (p58 N is colored cyan; PDB code 4RR2), and p58 C -D/R. The alignment was performed using p49 subunits. In the primosome, only p58 N with the p58 N -p58 C linker is shown for clarity. Clash area between the two domains in the primase-based model is depicted by the green dotted circle; no clash is observed in the primosome-based model.

Discussion
Our study provides structural insight at the atomic level into eukaryotic primosome and DNA primase in complex with the DNA/RNA substrate that revealed the exact structure of the initiation site of the primase with bound GTP. The linkers connecting p180core and p58 C to the p49-p58 N -p180 C -p70 platform provide significant freedom in their movement relative to each other and other subunits. Such flexibility allows p58 C to be close to p49 during initiation, to move away together with the 5Ј-end of the primer during elongation, and, finally, to intramolecularly transfer and load the 9-mer primer to the Pol␣ active site. To perform these duties, p58 C conforms to several topological requirements: it is complementary to p49 during initiation and to p180core during switch, and makes a clash with p58 N to terminate RNA synthesis at the defined position.
The p58 C -D/R structure, elongation complex models, and biochemical data (24,26,42) show that p58 C has a much stronger role in template/primer recognition than the catalytic p49 subunit. This is consistent with a distributive nature of RNA synthesis pointing to the high probability of p49 dissociation from the template/primer after incorporation of each nucleotide (Fig. 13A, lane 1). The unique mode of DNA/RNA substrate binding, with p58 C holding the 5Ј-end of primer and the weakly bound p49, allows Pol␣ to efficiently catch the 3Ј-end of RNA primer without dissociation of the template/primer from primosome. Biochemical data indicate that p58 C holds the DNA/RNA duplex when Pol␣ begins primer extension, thus providing the negative regulation for primase (11,26).
With the structures of apo-primosome and p58 C -D/R, it became possible to visualize all key steps of the entire cycle of primer synthesis (Fig. 14 and supplemental Movie S1). The structure of apo-primosome suggests that Pol␣ is in an autoinhibitory conformation, because entry to the DNA-binding site is blocked by the Zn2 module of p180 C and OB-fold of p70; this is consistent with the increase of Pol␣ activity when p70-p180 C is absent (35). Interestingly, apo-primase has a space between p180core and p49 that is suitable for docking of p58 C during initiation (Figs. 14 and 15A). Furthermore, upon RNA primer elongation, p58 C moves toward p180core, pushing it to dissociate from the platform (Figs. 14 and 15B). As a result, the p180core became conveniently positioned to catch the p58 Cbound 9-mer primer after its release from p49 and rotation by about 100° (Figs. 14 and 15C). The presented structures and models, together with biochemical data, showing that p58 C remains bound to the template/primer during RNA primer  elongation and switch (9,11,26), finally revealed the structural basis for coordination of the primase and Pol␣ catalytic activities during all steps of primer synthesis.
The mode of primase interaction with the template/primer is unique and differs from the mode employed by RNA polymerases where the initiating nucleotide does not interact perma- FIGURE 14. Mechanism of RNA-DNA primer synthesis by the human primosome. At the first step (steps are labeled by roman numerals at the circle in the center), p58 C moves toward p49 (curved arrow) to initiate RNA synthesis. During the second step, p58 C is moving toward p180core by following the helical path of the growing DNA/RNA duplex (curved arrow) and pushes the p180core to dissociate from the platform (straight arrow). Additionally, when RNA primer length is nine nucleotides, p58 C makes a steric hindrance with p58 N , which prevents primer extension by p49. At the third step, p58 C rotates (curved arrow) and loads the template/primer to the Pol␣ active site. At the fourth step, Pol␣ extends the RNA primer with dNTPs, whereas p58 C is holding the 5Ј terminus of the primer. At the fifth step, primosome is replaced by Pol⑀ or Pol␦. Models of primosome at different steps of primer synthesis are based on the apo-primosome structure (Fig. 2) and primase models (Fig. 8). FIGURE 15. Models of the human primosome during key steps of RNA primer synthesis. A, at the initiation step, p58 C fits into a space between p180core and p49. B, at the RNA elongation step, p58 C is moving toward p180core and displaces it from the p49-p58 N -p180 C -p70 platform. C, flexibly tethered p180core binds DNA template with a 9-mer RNA primer presented by p58 C .
nently with the non-catalytic domain (43)(44)(45)(46). This mode might be applicable to primases from other species where an accessory DNA(NTP)-binding domain is also present (47)(48)(49)(50)(51). The lack of interaction between the human primase and the emerging RNA strand (Fig. 8C) explains the significant ability of DNA primases to perform translesion synthesis (52)(53)(54)(55). The data presented here have wide-ranging therapeutic implications, because they provide new perspectives for the design of specific primosome inhibitors ultimately aimed at arresting the onset of replication in targeted cancer cells. Moreover, these structures could be used as a reference for the detection of the differences between the human and bacterial or viral primases, thus facilitating the design of novel classes of antibiotics and antivirals.
Now that the mechanism of concerted primosome transactions in the synthesis of RNA-DNA primers has become clear, further study is needed to understand how primosome integrates into the replisome and how the primer is released and translocated to the main replicative DNA polymerases ␦ and ⑀ (3,6,56). Pol␣ switch to the autoinhibitory state would prevent its competition with Pol␦ and Pol⑀ for template/primer binding.