How Pol α-primase is targeted to replisomes to prime eukaryotic DNA replication

Summary During eukaryotic DNA replication, Pol α-primase generates primers at replication origins to start leading-strand synthesis and every few hundred nucleotides during discontinuous lagging-strand replication. How Pol α-primase is targeted to replication forks to prime DNA synthesis is not fully understood. Here, by determining cryoelectron microscopy (cryo-EM) structures of budding yeast and human replisomes containing Pol α-primase, we reveal a conserved mechanism for the coordination of priming by the replisome. Pol α-primase binds directly to the leading edge of the CMG (CDC45-MCM-GINS) replicative helicase via a complex interaction network. The non-catalytic PRIM2/Pri2 subunit forms two interfaces with CMG that are critical for in vitro DNA replication and yeast cell growth. These interactions position the primase catalytic subunit PRIM1/Pri1 directly above the exit channel for lagging-strand template single-stranded DNA (ssDNA), revealing why priming occurs efficiently only on the lagging-strand template and elucidating a mechanism for Pol α-primase to overcome competition from RPA to initiate primer synthesis.


In brief
By determining cryo-EM structures of budding yeast and human replisomes containing the Pol a-primase complex, Jones et al. reveal a conserved mechanism for the coordination of nascent-strand priming in the eukaryotic replisome. The mechanism explains why priming by Pol a-primase is highly efficient on the lagging-strand but not leading-strand template.

INTRODUCTION
Following replisome assembly and template unwinding at bi-directional origins of DNA replication, Pol a-primase is recruited to the two advancing replisomes where it primes the lagging-strand template. 1 To start coupled leading-strand replication, these primers are extended across the origin by the main lagging-strand polymerase, Pol d, 2 before a polymerase switch transfers the nascent strand to the principal leading-strand polymerase, Pol ε. 1,[3][4][5][6] As replication forks progress primers are synthesized every few hundred nucleotides to support discontinuous lagging-strand replication. If leading-strand synthesis is interrupted due to DNA damage, biochemical reconstitution experiments have demonstrated that S. cerevisiae (budding yeast) replisomes continue lagging-strand replication but do not frequently reinitiate leading-strand replication due to a failure to support efficient primer synthesis on this strand. [7][8][9][10] Collectively, these observations indicate that the replisome efficiently targets Pol a-primase to the lagging-strand template but not the leading-strand template, which likely explains why some eukaryotes, including humans, encode a second primase-polymerase, PRIMPOL, to restart leading-strand replication. [11][12][13][14] Currently the mechanistic basis underlying the preference of Pol a-primase for lagging-rather than leading-strand priming is unknown.
Pol a-primase is a constitutive heterotetramer composed of a dimeric primase (PRIM1 and PRIM2 in H. sapiens [human], Pri1 and Pri2 in budding yeast) and a dimeric DNA polymerase (POLA1 and POLA2 in human, Pol1 and Pol12 in budding yeast) ( Figure 1A). Primase synthesizes 8-10 nucleotides (nt) of RNA that are transferred to the Pol a DNA polymerase for limited extension to a total primer length about 20-35 nt. [15][16][17][18] In vitro, the ability of budding yeast 7 and human [19][20][21] Pol a-primase to initiate primer synthesis on single-stranded DNA (ssDNA) templates is blocked when the template is saturated with RPA, indicating that a mechanism exists to target primase to ssDNA at the eukaryotic replication fork. Consistent with this idea, Pol a-primase interacts with several core components of the replisome including AND-1 (Ctf4 in budding yeast), MCM10, GINS, and CMG. [22][23][24][25][26][27] In budding yeast, Pol a-primase is tethered to replisome progression complexes (RPCs) via interaction between its Pol1 subunit and Ctf4, a trimeric scaffold protein that binds directly to CMG. [28][29][30][31] Similarly, human Pol a-primase associates with AND-1 but does so primarily via the N-terminal domain (NTD) of POLA2, which binds a C-terminal HMG-box in 33 However, considerable evidence indicates that AND-1/Ctf4 does not provide a pivotal link between Pol a-primase and the replisome to support priming: Ctf4 is a non-essential protein in budding yeast; disruption of the Pol1:Ctf4 interaction does not result in obvious DNA replication defects in vivo 34 ; Pol a-primase localizes to replisomes in yeast cells lacking Ctf4 35 ; depletion of AND-1 in DT40 cells does not prevent the completion of bulk replication 36 ; AND-1 and Ctf4 are dispensable for lagging-strand synthesis in DNA replication reactions reconstituted with purified proteins. 27

OPEN ACCESS
human replisomes perform lagging-strand replication in the absence of MCM10. 27,39,40 Accumulating evidence suggests that Pol a-primase is recruited to the replisome for priming via direct interactions with CMG. Yeast Pol a-primase can execute lagging-strand replication when functioning only with CMG and RPA. 39 We recently found that minimal human replisomes consisting of CMG, Pol a-primase, Pol ε, CTF18-RFC, PCNA, and RPA support lagging-strand replication and that human Pol a-primase comigrates with CMG in glycerol gradient sedimentation experiments. 27 However, because there are no structures of Pol a-primase in the eukaryotic replisome, we do not know how Pol a-primase binds to CMG, how these putative interactions contribute to lagging-strand DNA replication, how Pol a-primase might overcome competition with RPA for exposed ssDNA, where in the replisome Pol a-primase is positioned, and why Pol a-primase efficiently primes the lagging-strand but not leading-strand template. To address these questions, we have determined cryoelectron microscopy (cryo-EM) structures of budding yeast and human replisomes bound to Pol a-primase and replication fork DNA.

RESULTS
Yeast Pol a-primase replisome structure To assemble budding yeast replisomes associated with Pol a-primase for cryo-EM analysis, CMG was bound to a model replication fork containing a 39 nt 3 0 flap, onto which CMG loads, and a 60 nt 5 0 ssDNA flap to mimic the unwound lagging-strand template ( Figure S1A). Fork-bound CMG was incubated with Pol a-primase and the replisome accessory factors Mrc1 and Tof1-Csm3. Tof1-Csm3 binds to the leading edge of CMG where it engages and stabilizes the parental DNA duplex and fork junction, 30 which we reasoned might be important to aid the visualization of template DNA in cryo-EM reconstructions. After glycerol gradient sedimentation, complexes were isolated containing all replisome and Pol a-primase subunits and used to prepare grids for cryo-EM data collection and analysis (Figures S1 and S2; Table 1).
Three-dimensional (3D) reconstructions revealed well resolved cryo-EM density for CMG and Tof1-Csm3. Multiple additional densities were also apparent extending from the N-tier face of CMG atop Mcm3  (G) Schematic illustrating the organization of Pol a-primase in the budding yeast replisome. The path of lagging-strand template ssDNA visualized in the structure immediately following strand separation is illustrated (solid pink line). The putative path of the lagging-strand template between the Pri1 and Pol1 active sites is also illustrated (dashed pink line).  Figure 1B). The resolution of these densities was typically lower than for CMG and displayed considerable variability ( Figure S1E), indicating large conformational flexibility. Nonetheless, following extensive focused classification and refinement (see Figure S2), these densities could be unambiguously attributed to Pol a-primase (Figures 1B, S1E-S1O, and S3B-S3D), enabling us to construct a model of a DNA engaged yeast replisome encompassing CMG, Tof1-Csm3, several small sections of Mrc1 ( Figure S3A) and regions of all four Pol a-primase subunits ( Figures 1C and 1D; Video S1). The Pol a-primase model comprises the primase catalytic subunit Pri1 aside from its flexible N and C termini, the N terminus (residues 1-5) and NTD (residues 44-177 and 181-299) of the primase accessory subunit Pri2, the C-terminal domain (CTD) of the Pol a catalytic subunit Pol1 (Pol1 CTD ) (residues 1,271-1,468) and the majority of the Pol a accessory subunit Pol12 (residues 1-79, 203-582, and 604-705) (Figures 1C, 1D, and S3B-S3D). Pri1 is positioned close to the incoming parental doublestranded DNA (dsDNA) above a channel between the Mcm3 and Mcm5 zinc-finger (ZnF) domains, through which laggingstrand template ssDNA is extruded after strand separation (Figure 1E). [41][42][43][44] It is localized to the replisome through its interaction with the Pri2 NTD (Pri2 NTD ), which sits on the periphery of the MCM N-tier straddling Mcm3 and Mcm5 (Figures 1C and 1D). Pol a (Pol1 CTD and Pol12) is situated between the CMG N-and C-tiers, close to Mcm3 and GINS (Figures 1C and 1D) and is coupled to primase via an interaction between Pol1 CTD and Pri2 NTD . 45 Pol12 interacts extensively with Pol1 CTD and is anchored to the MCM C-tier through an interface involving its flexibly tethered NTD and Mcm3 ( Figure 1C). Although Ctf4 was not included in replisome reconstitutions, we identified a 3D class containing both Pol a-primase and Ctf4 ( Figure S3E). The presence of Ctf4 likely resulted from endogenous Ctf4 copurifying with CMG due to the extensive interface between the two complexes. 30,31 Comparison of the structures with and without Ctf4 revealed no substantial changes in the conformation of Pol a-primase ( Figures S3E and S3F).

Article
Clear densities for the Pol1 exonuclease-catalytic (exo-cat) domain (Pol1 exo-cat ) and Pri2 CTD (Pri2 CTD ) were not observed in our consensus refinement ( Figures 1B and S1E), indicating that neither domain adopts a single stable conformation when Pol a-primase is bound to the replisome. This behavior contrasts with the crystal structure of human apo Pol a-primase, where both domains were well ordered. 46 However, we recovered several rare 3D classes with low-resolution densities of the appropriate shape and volume to accommodate Pol1 exo-cat and Pri2 CTD , although the precise orientation of each domain could not be assigned (Figures S3G and S3H). In these reconstructions, Pri2 CTD is adjacent to Pri1 close to the primase active site, while Pol1 exo-cat sits above Pri2 NTD on the periphery of the replisome adjacent to the Pri2 CTD . This configuration more closely resembles the architecture of human Pol a-primase bound to CST (CTC1-STN1-TEN1) and telomeric ssDNA 47 (Figure S3I), and a very recent structure of a human Pol a-primase elongation complex, 48 than the human Pol a-primase apo structure. 46 This led us to consider that Pol a-primase conformation in the yeast replisome might be modulated by protein-protein inter-actions and/or DNA engagement. Because DNA binding was heterogeneous across the dataset, we obtained a 3D replisome reconstruction lacking DNA ( Figure S3J). Here, the positioning of Pol1 exo-cat and Pri2 CTD resembled the human apo crystal structure 46 ( Figure S3K), indicating that Pol a-primase undergoes DNA-dependent conformational changes when associated with CMG in the budding yeast replisome.
To further explore the putative DNA engagement state of Pol a-primase, we performed additional classification focusing on regions of primase close to the replication fork junction (Figure S2). Strikingly, this strategy revealed a 3D class with continuous density extending from the parental DNA duplex at the point of strand separation, through the channel between the Mcm3 and Mcm5 ZnF domains and alongside Pri1 in the direction of the primase active site ( Figure 1F). The density between the Mcm3 and Mcm5 ZnF domains is in an equivalent position to the previously identified path of the lagging-strand template following strand separation in the human replisome, 41,42 strongly suggesting that it corresponds to lagging-strand template ssDNA. Moreover, the close proximity of the density to Pri1 and its continuation beyond the Mcm3-Mcm5 ZnF channelwhich has not been observed in prior human and yeast replisome structures lacking Pol a-primase-indicate that Pri1 engages lagging-strand template ssDNA in the yeast replisome structure. We hypothesize that this configuration functions to ensure a minimal length of ssDNA is required for the lagging-strand template to reach the primase active site, thereby enabling primase to outcompete RPA for access to the template to initiate primer synthesis. Moreover, the positioning of Pri1 and Pol1 exo-cat arranges the primase and DNA polymerase catalytic centers in synthesis order along the template ( Figures 1G and S3L), suggesting a possible mechanism for transfer of the RNA primer to the Pol a DNA polymerase as the replisome advances, similar to the mechanism proposed for human Pol a-primase during telomere C strand fill-in. 47 Pol a-primase replisome interactions Four small interaction sites, labeled sites a-d in Figure 2A, tether Pol a-primase directly to CMG and position primase to engage lagging-strand template ssDNA (Video S1). Pri2 NTD forms electrostatic interfaces with both the Mcm5 ZnF domain (site a) and the Mcm3 N-terminal helical domain (site b) ( Figure 2B). The interface with the Mcm5 ZnF is mediated by a small insertion in Pri2 NTD that appears confined to a subset of fungal species ( Figure S4A), while the interface with Mcm3 involves three flexible loops within the Pri2 NTD (between helices a3-4, a4-5, an a6-7) that are positioned to interact with conserved surfaceexposed charged residues on the first alpha helix (a1) of Mcm3 (Figures 2B,2C,and S4B). 3D variability analysis 49 shows that Pri2 NTD adopts a continuum of rotational states with respect to Mcm3 while remaining engaged, likely due to the electrostatic nature of the interface ( Figure S4C).
The remaining two interfaces between Pol a-primase and CMG (sites c and d) involve regions of the Pri2 and Pol12 subunits situated at the ends of regions of polypeptide predicted to be unstructured, 50 indicating that they form flexible tethering points ( Figures 1A, 1D Figures 2D and S4D). During ATP-dependent DNA translocation the MCM C-tier adopts multiple conformational states dependent on nucleotide occupancy and DNA engagement. 30,44 While we observe only one C-tier conformation when Pol12 NTD is bound to Mcm3, the Pol12 NTD can be docked without clashes onto Mcm3 via the same interface through a range of C-tier configurations, indicating it might remain associated with Mcm3 throughout active replication ( Figure S4E). Site d involves the N-terminal 5 amino acids of Pri2 (Pri2 Nterm ), where Pri2-F2-invariant in fungal species-docks into a surface-exposed hydrophobic pocket on the GINS subunit Psf2 (Figures 2E-2G). Pri2 Nterm is connected to Pri2 NTD via a 40 aa linker and, although this linker is predicted to be unstructured, at low map thresholds, continuous density is visible between the last modeled residue of Pri2 Nterm (S5) and the first modeled residue of the Pri2 NTD (S44), indicating that a section of the linker might adopt a structured conformation ( Figure S4F).
In reconstructions containing Ctf4, local refinement revealed the presence of Pol a-primase-dependent density on the surface of the C-terminal a-helical domains of Ctf4 at the previously identified Pol1 binding site, 29 Figure 2A), presumably because the CIP box is linked to Pol1 exo via $200 aa of largely unstructured polypeptide.
Although we obtained 3D reconstructions where Pol a-primase was bound to CMG at all 4 sites, a substantial fraction of the dataset lacking Ctf4 displayed binding at just the Mcm5 ZnF and GINS interfaces (sites a and d), demonstrating that only a subset of binding sites are necessary to anchor Pol a-primase to CMG ( Figures S4I-S4K). Inspection of the cryo-EM density in reconstructions where only sites a and d were engaged reveals that Pri2 NTD and Pol1 CTD -Pol12 are less well resolved, indicating that these regions are stabilized by the binding of Pri2 NTD and Pol12 NTD to Mcm3 (sites b and c, respectively).
These data indicate that Pol a-primase can utilize only a subset of interaction sites for replisome association, which might be important to permit conformational changes during the priming cycle. 18,46 Pol a-primase interaction mutants To examine the contributions of the Pol a-primase:CMG interfaces during DNA replication, we purified Pol a-primase mutants and truncations designed to disrupt the Pri2:Mcm5 (Pri2-5A), Pol12:Mcm3 (Pol12-DN) and Pri2:GINS (Pri2-D2-8) interfaces (sites a, c, and d, respectively) and a Cdt1-Mcm2-7 charge reversal (CR) mutant (Mcm3-CR) designed to disrupt the Pri2 NTD :Mcm3 binding site (site b) ( Figures 3A and S5A). We also purified a Pol a-primase complex in which the Pol1 CIP box was mutated to abrogate its interaction with Ctf4 29 (Pol1-4A) (targeting sites e i and e ii ) ( Figures 3A and S5A). Figure S5B shows that all Pol a-primase mutants displayed similar priming and DNA synthesis activities to the wild-type protein on ssDNA templates. Origin-dependent DNA replication reactions that (E-G) Diploid budding yeast cells of the indicated genotype were sporulated and the resulting tetrads were dissected and grown on YPD medium for 3 days at 25 C. Dissections that displayed abnormal segregation patterns were cropped from plate images.

OPEN ACCESS
Article generate leading-and lagging-strand products were reconstituted with purified budding yeast proteins on a 10.1 kbp linear DNA template with the origin positioned roughly at its center (Figure 3B). 1,37,38 In reconstituted replication reactions in which the lagging-strand maturation machinery is omitted, the length distribution of lagging-strand products is dependent on Pol a-primase concentration, with less frequent priming resulting in the synthesis of longer lagging strands. 37,39,51 Replication with wild-type proteins produced a population of $4.5-6.5 kb leading strands and lagging strands of less than 0.6 kb ( Figure 3C, lanes 1 and 7). Figure 3C shows that all mutant proteins with a single binding site targeted were competent for leading and lagging-strand DNA replication, demonstrating that no single interface is essential for Pol a-primase function. However, the distribution of lagging-strand products varied considerably among mutants, indicating that each interface does not contribute equally to productive primer synthesis. Surprisingly, mutations designed to target the Pri2:Mcm5 interface (Pri2-5A, site a) did not affect the length of lagging-strand products ( Figure 3C To gain further insight into the hierarchy of Pol a-primase:replisome interactions during DNA replication, we purified additional complexes harboring combinations of mutations/truncations ( Figures S5A and S5B). Figures 3D, S5C, and S5D show that, in almost all cases, disrupting multiple Pol a-primase binding sites resulted in further increases in the length of lagging-strand products. Notably, lagging-strand synthesis was all but abolished in a reaction where both the Pri2:Mcm3 and Pri2:GINS interfaces (sites b and d) were disrupted, and there was a reduction in intensity and subtle lengthening of leading-strand products ( Figure 3D, lane 6), which is indicative of delayed synthesis of the primers used to start leading-strand replication. 1 8 and S5D, lane 9), indicating that the interaction between Pri2 Nterm and GINS is sufficient to support the necessary priming to start leading-strand replication. These data demonstrate that four distinct interfaces between Pol a-primase and the replisome contribute to nascent-strand priming and that collectively they are essential for efficient in vitro DNA replication. Importantly, the data also indicate that the contribution of each interface is not equal: disruption of the interface between Pri2 Nterm and GINS (site d) is most deleterious for laggingstrand replication followed by the interface between Pri2 NTD and Mcm3 (site b), whereas the interactions between Pol1 and Ctf4 (sites e i /e ii ) and Pol12 NTD and Mcm3 (site c) make more minor contributions.
Pol a-primase mutants in vivo Priming at replication forks is an essential function of Pol a-primase and therefore the key interactions we have identified should be critical for cell growth. To test this, we generated budding yeast strains with mutations targeting the Pri2:GINS (Pri2-AAA) and Pri2:Mcm3 (Mcm3-CR) interfaces ( Figure 2A, sites b and d). In the Pri2-AAA allele, amino acids F2, R3, and Q4 are substituted to alanine. Figure S5E shows that Pol a-primase complexes containing Pri2-AAA and Pri2-D2-8 displayed almost indistinguishable behavior in in vitro replication assays. Colony growth of both pri2-AAA and mcm3-CR cells was comparable to control cells, indicating that priming was occurring at sufficient levels to permit relatively normal DNA replication ( Figure S5F). We therefore combined the pri2-AAA and mcm-CR mutations. This resulted in a profound reduction in colony size relative to control cells, consistent with these cells having DNA replication defects ( Figures 3E and S5G), which is concordant with the near absence of lagging-strand products in in vitro reactions when these interfaces are disrupted (Figures 3D,lane 6 and S5E,lanes 5 and 6).
Although the lack of obvious growth defects for pri2-AAA and mcm3-CR cells was somewhat surprising, previous work has shown that budding yeast are reasonably tolerant of reduced Pol a-primase levels. 52 Moreover, colony growth of pol1-F1463A cells in which the interaction between primase and the Pol1 C terminus is disrupted, was comparable to control cells. 45 However, pol1-F1463A is synthetic lethal with deletion of the gene encoding the apical checkpoint kinase Mec1, the ortholog of ATR, indicating that these cells do in fact have DNA replication defects. 45 We therefore wondered if pri2-AAA and mcm3-CR might have subtle DNA replication defects that render cells dependent on checkpoint activation. Figure 3F shows that deletion of MEC1 in combination with mcm3-CR had minimal effect on colony growth. In contrast there was a notable reduction in colony size when mec1D was combined with pri2-AAA (Figure 3G), revealing that tethering of Pol a-primase to CMG via the Pri2:GINS interface is essential for unperturbed DNA replication in budding yeast.
Human Pol a-primase replisome structure Because priming is fundamental for genome duplication, we considered it likely that key features of the mechanism targeting Pol a-primase to prime DNA synthesis were conserved. To examine this directly we determined the cryo-EM structure of a  Table 1). Similar to the yeast replisome, in addition to well resolved cryo-EM density for CMG, TIMELESS-TIPIN, and AND-1, poorly resolved density extended from the N-tier face of CMG atop MCM3 ( Figures S6E-S6J). Following focused classification and refinement ( Figure S6C) this density could be unambiguously assigned to Pol a-primase enabling assignment of PRIM1 (residues 9-349 and 386-408), the N terminus (residues 1-5) and NTD (residues 17-252) of PRIM2, the CTD of POLA1 (residues 1,279-1,445 and 1,448-1,458) and the majority POLA2 (residues 96-114 and 170-598) ( Figures 4A, 4B, and S7A-S7C; Video S2).
The conformation of human Pol a-primase and its positioning in the replisome are remarkably similar to yeast (Figures S7D and S7E), with the catalytic PRIM1 subunit again positioned above the mouth of the lagging-strand template exit channel ( Figures 4A-4C). Similar to yeast, POLA1 exo-cat is invisible when CMG is bound to replication fork DNA but is visualized stably engaging Pol a-primase in reconstructions lacking DNA, where it adopts the conformation observed for apo human Pol a-primase 46 ( Figures S7F and S7G). This suggested that the conformation of Pol a-primase in the human replisome might represent a DNA engaged state. To investigate this further, we repeated our human replisome sample preparation and cryo-EM analysis as before, but with a replication fork containing a  Table 1). In the resulting 3D reconstructions, clear density is observed for the POLA1 exo-cat domain, both when CMG is bound to the replication fork and when CMG is not engaging DNA (Figures 5, S8I, and S8J). In both situations, Pol a-primase adopts the same conformation as observed for the human apo structure, 46 strongly suggesting that Pol a-primase is bound to laggingstrand template ssDNA when the human replisome is assembled on a replication fork with a 60 nt 5 0 flap.

Conservation of Pol a-primase tethering
Human Pol a-primase is tethered directly to CMG via three small interfaces that are all occupied independently of DNA engagement state (Figures 6A-6D and S9A-S9C; Video S2). PRIM2 binds to MCM3 (site b) and GINS (site d) in an analogous manner to Pri2 in the budding yeast replisome ( Figures 6B-6D, S9D, and S9E), consistent with these binding sites being the most important for priming in yeast ( Figures 3C and 3D). The PRIM2:MCM3 interface involves charged residues on a4 and the a3-4 linker of the PRIM2 NTD that form electrostatic contacts with four conserved residues on a1 of MCM3 ( Figures 2C, 6B, and S9D). Binding of PRIM2 to GINS is mediated by the N terminus of PRIM2, with amino acids M1 and F3-invariant in Metazoa-projecting into a surface-exposed hydrophobic pocket on PSF2 in a comparable manner to Pri2-F2 in yeast ( Figure 6D). Continuous cryo-EM density links the last modeled residue of PRIM2 Nterm (G5) and the first modeled residue of PRIM2 NTD (Q17) indicating this interface spatially constrains the position of the PRIM2 NTD and PRIM1 ( Figure S9E). In addition to sites b and d, a flexibly tethered helix in POLA2 (residues 96-114) interacts with PSF1 and SLD5 (site g, Figures 6A, 6C, and S9F). Although this helix is predicted to be absent from Pol12, we note the presence of low-resolution Pol a-primase-dependent density on the surface of Psf1 in our budding yeast replisome maps, suggesting a similar binding site could be present in the yeast replisome ( Figure S9G). We also note that binding of the POLA2 helix to PSF1 and SLD5 will localize the POLA2 NTD-that binds to the C-terminal AND-1 HMG-box 32 (labeled site f in Figure 6A)-to this region of the replisome because the POLA2 helix and NTD are separated by a short (12 aa) linker. In contrast to the budding yeast replisome, we find no evidence of PRIM2 NTD binding to the MCM5 ZnF, indicating it is not a conserved mode of interaction and perhaps explaining why the yeast Pri2-5A mutant did not display a lagging-strand replication defect ( Figure 3C). Finally, consistent with reports that AND-1 binds the unstructured POLA1 N terminus (residues 151-171), 32,33 we observe a small region of Pol a-primase dependent density on the C-terminal a-helical domain of each AND-1 monomer (labeled sites e i-iii , Figures S9H and S9I), indicating that Pol1 can access all available binding sites on the AND-1 trimer.
The conservation of the PRIM2:GINS and PRIM2:MCM3 interfaces ( Figure 6E) suggested they would be important for priming in the human replisome. To test this directly we purified a Pol a-primase complex lacking the PRIM2 N terminus (PRIM2-D2-7) and a CMG complex where four conserved residues on helix a1 of MCM3 were mutated to alanine (MCM3-4A) (Figures S9J-S9L) and analyzed them in an in vitro human DNA replication system that we recently developed ( Figure 6F). 27 Here, replisomes assembled around purified CMG at model replication forks perform leading and lagging-strand DNA replication at rates comparable to those measured in cultured human cells. Figures 6G and S9M show that lagging-strand products distributed around $0.6 kb in length were synthesized with wild-type proteins. These products were substantially longer when the PRIM2:GINS interface (PRIM2-D2-7) was disrupted ( Figure 6G, lanes 2 and 6). While disruption of the PRIM2:MCM3 (MCM3-4A) interface was less severe, there was still a notable increase in the length of lagging-strand products, which were longer compared with when the AND-1 HMG :POLA2 interface was abolished (AND-1-DHMG) ( Figures 6G and S9M). Disruption of multiple interfaces in the same reaction further compromised lagging-strand replication

OPEN ACCESS
Article compared with single site mutants ( Figure S9M). Collectively, these data indicate that key anchor points for attaching Pol a-primase to CMG to facilitate primer synthesis at replication forks are structurally and functionally conserved between yeast and human.

DISCUSSION
By determining cryo-EM structures of budding yeast and human replisomes that are poised to initiate primer synthesis, we have elucidated a conserved mechanism for targeting Pol a-primase to replication forks for priming. The positioning of the catalytic Pri1/PRIM1 subunit at the mouth of the exit channel for lagging-strand template ssDNA explains how Pol a-primase functions so efficiently on this template strand and reveals a mechanism for primase to overcome competition with RPA for access to the DNA template. By contrast, the unwound leading-strand template exits CMG $150 Å away from Pri1/PRIM1 on the opposite side of the replisome, which is presumably not conducive for leading-strand priming, thus explaining why the core yeast replisome cannot efficiently restart leading-strand replication by repriming downstream of DNA damage 7,8 or secondary structures 10 and why lagging-strand primers are used to start leading-strand replication. 1,3,4 Pol a-primase is targeted to the lagging-strand template for priming via a complex multisite interaction network involving several direct interactions with CMG. These interactions explain why Pol a-primase tethering by Ctf4/AND-1 is dispensable for DNA replication. 27,34,36,37 The primary function of Ctf4/AND-1dependent tethering of Pol a-primase is to facilitate the transfer of parental histones to the lagging strand. 34,53 It will be interesting to discover why Ctf4/AND-1-dependent tethering is required for this activity and why the interactions between Pol a-primase and CMG that we have identified cannot fulfill this role. In both the yeast and human replisomes, the majority of Pol a-primase docking sites-including the crucial interaction between Pri2/PRIM2 and GINS-are mediated by regions of Pol a-primase situated at the end of, or within, unstructured

OPEN ACCESS
Article linker regions, thereby providing flexible tethering points. Of the interactions that contribute to priming, only binding of the Pri2/ PRIM2 NTD to MCM3 involves the association of two large rigid bodies. However, this interface is frequently disengaged and, due to its electrostatic nature, permits considerable motion between the two domains. This suggests that, although the positioning of Pol a-primase at the mouth of the lagging-strand template exit channel is crucial for priming, it is also important that primase is not rigidly fixed in this location. We hypothesize that flexible tethering of Pol a-primase in the replisome is required to allow other proteins access to key binding sites on CMG. For example, the E3 ubiquitin ligase that regulates replisome disassembly (Cul2 LRR1 in human and SCF Dia2 in budding yeast) binds across the lagging-strand template exit channel 42 and this binding site is inaccessible when Pri2/PRIM2 is bound to MCM3. Flexible tethering may also function to enable Pol a-primase to remain associated with the replisome while it undergoes conformational changes during the primer synthesis reaction.
Our structures indicate that incorporation of Pol a-primase into the replisome does not induce conformational changes in CMG that are likely to modulate helicase activity. Consequently, concomitant primer synthesis and template unwinding will result in increasing lengths of ssDNA being formed between the primase/DNA polymerase active sites and the point of template unwinding, thereby generating what has been termed a ''priming loop.'' 54,55 Currently, we do not know whether Pol a-primase remains fully engaged with CMG throughout the priming reaction, or whether the multiple docking sites are utilized dynamically. We consider it likely that Pol a-primase remains associated with CMG via at least one docking site for the entirety of the priming cycle given the prolonged replisome association kinetics that have been observed in single molecule experiments. 35,56 The conformational dynamics of Pol a-primase and its utilization of docking sites during primer synthesis are interesting subjects for future investigation that we anticipate will also influence the disposition of priming loops in the replisome.
Considerable recent progress has been made in delineating the mechanisms of primer synthesis including the molecular basis for DNA primer initiation, 57 how Pol a-primase activity is coordinated by the CST complex during telomeric C strand fillin. 47,58,59 Our work represents another important advance by revealing a conserved mechanism for targeting Pol a-primase to replisomes to prime eukaryotic DNA replication and also provides a platform to visualize additional key intermediates during this fundamental process.
Limitations of the study Our structures of budding yeast and human Pol a-primase bound to the replisome likely only represent a small subset of conformations that Pol a-primase adopts during the priming cycle. Moreover, while our data strongly support the conclusion that Pol a-primase is engaging ssDNA in both the yeast and human replisomes, it is not possible to determine precisely which step of the priming cycle the structures represent. Although the structures provide important insights into how Pol a-primase is targeted to the replisome for priming, including identifying key protein:protein interaction sites, additional proteins that were not included in our replisome preparations might also modulate Pol a-primase activity at replication forks. For example, subunits of Pol a-primase have been reported to interact directly with Mcm10, RPA, and Pol d. Therefore, an important future goal will be to determine structures of more complete replisomes performing lagging-strand replication to visualize intermediates along the primer synthesis pathway, the handoff of primers from Pol a-primase to Pol d, and gain insights into how proteins such as RPA modulate the activity of Pol a-primase in the context of the replisome.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

ACKNOWLEDGMENTS
We thank K. Labib for sharing budding yeast strains. We thank J. Shi for operation of the LMB baculovirus facility; LMB media preparation for budding yeast plates; S. Chen, G. Sharov, G. Cannone, A. Yeates, and B. Ahsan for smooth running of the MRC LMB EM facility; and J. Grimmett, T. Darling, and I. Clayson for maintenance of scientific computing facilities. We are grateful to F. Abid-Ali and members of the Yeeles lab for comments on the manuscript. This work was supported by the MRC as part of UK Research and Innovation (MRC grant MC_UP_1201/12 to J.T.P.Y.).

Article AUTHOR CONTRIBUTIONS
Investigation-performed all cryo-EM experiments, data analysis, and model building, methodology-purified proteins, writing -original draft, review & editing, M.L.J.; investigation-performed preliminary DNA replication assays and Pol a-primase:CMG interaction studies, methodology-generated expression vectors and yeast strains and purified proteins, V.A.; investigation-performed human DNA replication assays, methodology-purified proteins, Y.B.; conceptualization, supervision, funding acquisition, investigation-performed biochemistry and genetic experiments, methodology-DNA template preparation, protein purification, yeast strains, writing -original draft, review & editing, J.T.P.Y.

DECLARATION OF INTERESTS
The authors declare no competing interests.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS
S. cerevisiae strains constructed for genetic experiments were based on the W303 genetic background. Comprehensive information regarding the genotypes of these S. cerevisiae strains can be found in Table S4.

METHOD DETAILS
Protein expression and purification Details of protein expression plasmids and strains made during this study can be found in Tables S1 and S2. An overview of the purification strategy for each protein is provided in Table S3. All wild type budding yeast proteins were expressed and purified as described previously. 30,37,38,42 Cdt1-Mcm2-7 and Pol a-primase mutants / truncations were purified using the same procedure as for the wild type proteins. Human proteins were expressed and purified as described previously. 27,41,42 Human Pol a-primase (Pri2-D2-7) was expressed and purified as described for the wild type protein. 27 Human CMG (MCM3-4A) was expressed by coinfecting Hi5 cells at a density of 1 x 10 6 cells/ml with four viruses (generated as previously described 42 )

TIMELESS-TIPIN purification
Cells from a 1L culture were resuspended in lysis buffer (25 mM HEPES-KOH pH 7.2, 150 mM KCl, 5% glycerol, 0.5 mM TCEP, 0.01% NP-40-S) + protease inhibitors (cOmplete, EDTA-free, one tablet per 50 ml buffer) and lysed by dounce homogenization. Insoluble material was cleared by centrifugation (235,000 g, 4 C, 45 min) and 0.5 ml Strep-Tactin XT superflow high capacity resin was added to the lysate. Following a 30 min incubation at 4 C the resin was collected in a 20-ml column and was washed with 50 ml lysis buffer. The resin was resuspended in $ 2 ml lysis buffer and TEV protease was added to 100 ug/ml. The sample was incubated at 4 C overnight with gentle rotation. The sample was collected and applied to a 1 ml HiTrap Q HP column (GE Healthcare) equilibrated in 25 mM HEPES-KOH pH 7.2, 150 mM KCl, 5% glycerol, 0.5 mM TCEP, 0.01% NP-40-S. Proteins were eluted with a 20 column volume gradient from 150 to 1,000 mM KCl and peak fractions containing TIMELESS-TIPIN were pooled, concentrated to $ 500 ml in an Amicon Ultra-15 30 kDa MWCO concentrator and applied to a Superdex 200 Increase 10/300 gel filtration column (GE Healthcare) equilibrated in 25 mM Tris-HCl pH 7.2, 5% glycerol, 0.01% NP-40-S, 1 mM DTT, 150 mM NaCl. Peak fractions were pooled, frozen in liquid nitrogen and stored at À80 C.

Preparation of fork DNA for cryo-EM
To prepare forked DNA for cryo-EM sample preparation, leading and lagging strand oligonucleotides (Integrated DNA Technologies) were mixed at equimolar ratios in annealing buffer (25 mM HEPES-NaOH, pH 7.5, 150 mM NaOAc, 0.5 mM TCEP, 2 mM Mg(OAc) 2 ) and gradually cooled from 80 C to room temperature. Leading strand oligo: 5 0 -(Cy3)-TAGAGTAGGAAGTGAGGTAAGTGATT AGAGAATTGGAGAGTGTG(T) 34 T*T*T*T*T*T -3 0 * Denotes a phosphorothioate backbone linkage. 15 Nucleotide 5 0 -flap lagging strand oligo: 5 0 -GGCAGGCAGGCAGGCACACACTCTCCAATTCTCTAATCACTTACCACACTTCCTACT CTA -3 0 . 60 nucleotide 5 0 -flap lagging strand sequence: Replisome assembly for cryo-EM Reconstitution reactions were set up to yield a final volume of 300 ml, containing 150 nM CMG with a 1.5-fold molar excess of replisome proteins and fork DNA in reconstitution buffer (25 mM HEPES-NaOH pH 7.6, 150 mM NaOAc, 0.5 mM TCEP, 500 mM AMP-PNP, 10 mM Mg(OAc) 2 ). Firstly, CMG was incubated with fork DNA for 10 min on ice in an 80 ml reaction. Next, the additional proteins were added in the following order: Ctf4/AND-1, Tof1-Csm3/TIMELESS-TIPIN, Mrc1/CLASPIN and Pol a-primase. The reaction volume was then adjusted to 300 ml using reconstitution buffer before being incubated for 20 min on ice. Following incubation, 132 ml of the reconstitution reaction was loaded separately onto two 10-30% glycerol gradients, each containing crosslinker. The remaining 36 ml of the reconstitution reaction was diluted in reconstitution buffer to 132 ml and this sample loaded onto a glycerol gradient prepared in the absence of crosslinker. Glycerol gradients were prepared as previously described 30 : Buffer A (40 mM HEPES-NaOH, pH 7.5, 150 mM NaOAc, 0.5 mM TCEP, 10% v/v glycerol, 0.5 mM AMP-PNP and 3 mM Mg(OAc) 2 ) was layered on top an equal volume of Buffer B (Buffer A, except 30% v/v glycerol, 0.16% glutaraldehyde [Sigma] and 2mM bis(sulfosuccinimidyl)suberate (BS 3 , ThermoFisher Scientific)) in a 2.2 mL TLS-55 tube (Beranek Laborgerate) and gradients made using a gradient-making station (Biocomp Instruments, Ltd.) before cooling on ice. The sample was separated by centrifugation (200,000g, 4 C, 2 h) prior to manual fractionation. SDS-PAGE gel analysis was used to identify two peak fractions from each gradient containing crosslinker (total volume 368 ml) as previously described. 41 These fractions were then buffer exchanged and concentrated prior to being immediately used for cryo-EM grid preparation as previously described. 41 ll OPEN ACCESS Article Cryo-EM data collection Budding yeast replisome + Pol a-primase + 60 nucleotide 5 0 -flap DNA fork A total of 12,819 raw movies were acquired using a 300 keV Titan Krios microscope (FEI) equipped with a K3 direct electron detector (Gatan) operated in electron counting mode using the EPU automated acquisition software (ThermoFisher) with ''Faster Acquisition'' mode (AFIS) enabled. A slit width of 20 eV was used for the BioQuantum energy filter. Data were collected in super-resolution mode bin 2 at an effective pixel size of 0.86 Å /pixel over a defocus range of -1.8 to -3.5 mm. Movies were dose-fractionated into 39 fractions over a 4 s exposure, resulting in a total dose of 39.2 e -/Å 2 .
Human replisome + Pol a-primase + 60 nucleotide 5 0 -flap DNA fork A total of 7,355 raw movies were acquired using a 300 keV Titan Krios microscope (FEI) equipped with a K3 direct electron detector (Gatan) operated in electron counting mode using the EPU automated acquisition software (ThermoFisher) with ''Faster Acquisition'' mode (AFIS) enabled. A slit width of 20 eV was used for the BioQuantum energy filter. Data were collected at a pixel size of 0.86 Å / pixel using a defocus range of -2.1 to -3.5 mm. Movies were dose-fractionated into 39 fractions over a 4 s exposure resulting in a total dose of 47.4 e -/Å 2 .
Human replisome + Pol a-primase + 15 nucleotide 5 0 -flap DNA fork A total of 6,718 raw movies were acquired using a 300 keV Titan Krios microscope (FEI) equipped with a Falcon III direct electron detector (Thermo) operated in linear mode using the EPU automated acquisition software (ThermoFisher). Data were collected at a pixel size of 1.07 Å /pixel using a defocus range of -0.9 to -3.5 mm. Movies were dose-fractionated into 39 fractions over a 1 s exposure resulting in a total dose of 88.5 e -/Å 2 .

Cryo-EM data processing
Budding yeast replisome + Pol a-primase + 60 nucleotide 5 0 -flap DNA fork The data processing pipeline outlined here is schematised in Figure S2. Data were processed using either RELION-3 64-67 (henceforth referred to as RELION) or cryoSPARC-3 68 (henceforth referred to as cryoSPARC) unless otherwise stated. 12,819 39-fraction movies were aligned and dose-weighted (1.00513 e -/Å 2 /fraction, 5 x 5 patches, 150 Å 2 B-factor) using RELION's implementation of a MotionCor2-like program. 69 CTF parameters were estimated using CTFFIND-4.1 70 and 112 poor-quality micrographs excluded from future processing. Particles were picked using RELION's Laplacian-of-Gaussian (LoG) function providing a minimum diameter of 200 Å and maximum of 350 Å . 2,003,322 picked particles were extracted using a box size of 430 Å . During extraction the data were down-sampled to a pixel size of 3.44 Å /pixel. One round of RELION 2D classification was carried out and 1,623,209 particles were selected for further classification. Four successive rounds of RELION 3D classification (regularisation parameter, T = 4), each generating 6 classes, were carried out using a previously obtained map of the budding yeast replisome as a reference (EMD-10227). 30 Class selection was based upon the presence of secondary structure features within CMG. The first two round of 3D classification were performed with a 250 Å diameter circular mask to focus classification on the CMG, with the subsequent two rounds using a more dilated mask of 380 Å to incorporate signal from Pol a-primase. 202,655 particles containing poor density for dsDNA were selected for an additional two rounds of 3D classification in RELION. This resulted in the selection of 44,871 particles in classes displaying density for both Pol a-primase and CMG in the absence of dsDNA. These particles were refined using 3D auto-refinement in RELION generating a reconstruction, after post-processing, at 7.4 Å resolution. In order to enrich for replisomes stably bound by Pol a-primase in the absence of dsDNA, signal subtraction was carried out in RELION focussing on the interface between Pol a-primase and Mcm3. The refined reconstruction was low-pass filtered to 10 Å in UCSF Chimera 71 and a soft mask was generated covering density for both Pol a-primase and the Mcm3 N-terminal helical domain. Subtracted particles were re-centred on the mask and sub-classified using 3D classification without alignment in RELION. 18,412 particles in classes containing strong density for both Pol a-primase and the Mcm3 N-terminal helical domain were selected and reverted to the original (non-subtracted) particles prior to refinement using 3D auto-refine in RELION. This generated a reconstruction, following postprocessing, in RELION at 6.8 Å resolution ( Figure S3J).
Returning to the results of the fourth overall round of 3D classification, 884,301 particles were selected for further processing to enrich for replisome complexes bound by Pol a-primase engaged on DNA. These particles were re-extracted using an un-binned pixel size of 0.86 Å in a 450 Å box and submitted for refinement using RELION 3D auto-refine, yielding a reconstruction at 3.5 Å following postprocessing. These data were submitted for two rounds of iterative per-particle motion correction using dataset-trained particle polishing in RELION 72 and RELION CTF-refinement 64 (beamtilt and trefoil correction, anisotropic magnification correction, and per-particle defocus and astigmatism CTF correction). These data were refined to an improved resolution of 3.0 Å following postprocessing in RELION. At this stage of processing the density for CMG, Ctf4, Tof1-Csm3 and DNA was of high quality yet the Pol a-primase density was disordered and fragmented with only the Mcm5 ZnF contact preserved at appropriate map thresholds.
In order to identify classes in which Pol a-primase was stably engaged with the replisome, the strategy previously described to enrich for Pol a-primase stably engaged on replisomes in the absence of DNA was employed. Signal subtraction was carried out using a mask encompassing both Pol a-primase and the Mcm3 N-terminal helical domain. These subtracted particles were than sub-classified using 3D classification without alignment, resulting in 614,228 particles in classes with improved Pol a-primase density. These data were reverted to the original non-subtracted particles and refined using RELION 3D auto-refine. The subtraction and subclassification process was then iterated to select for 3D classes representing 588,597 particles with improved Pol a-primase density. These particles were imported into cryoSPARC and subsequently down sampled to a pixel size of 1.72 Å /pixel to boost the signal-to-noise for spatial frequencies describing secondary structure elements. These data were then classified in 3D via heterogeneous refinement using five different replisome reference maps (composition indicated in Figure S2). Classes in which Pol a-primase was only engaged at both the Mcm5 ZnF and Psf2 sites were selected representing 434,311 particles. These particles were refined using non-uniform refinement, 73 with a pixel size of 0.86 Å , to 3.0 Å resolution ( Figure S1N). To aid interpretation, all non-uniform and local refinements completed in cryoSPARC were subsequently locally filtered using their respective local resolution maps.
Returning to the previous heterogeneous refinement, classes with improved Pol a-primase density representing 140,426 particles were selected and further classified via heterogeneous refinement with six identical 3D references, the results of which are used as the input for processing strategies 1-4: Strategy 1: Classes representing 84,628 particles were selected based on the presence of both strong Ctf4 and Pol a-primase density. A soft mask was generated in UCSF Chimera 71 covering Pol a-primase and masked 3D classification without alignment carried out using five identical 3D references in cryoSPARC (target resolution 8 Å , initialisation mode PCA). Masks generated for use in cryoSPARC were binarised using the vop_threshold command in UCSF Chimera 71 and a soft padding width applied using the Volume Tools in cryoSPARC (mask softness=5*resolution(Å )/pixel-size(Å )). Classes representing 54,970 particles were selected based upon Pol a-primase Pri2 NTD engagement with the Mcm3 N-terminal helical domain. A consensus refinement for these unmasked particles was carried out using non-uniform refinement in cryoSPARC, with an binned pixel size, resulting in a reconstruction at 4.6 Å resolution in which the MCM C-tier adopts conformation II. This process was iterated to generate a reconstruction, with an un-binned pixel size of 0.86, at 3.5 Å resolution ( Figure S1M). In this reconstruction Pol a-primase engages the replisome via contacts with Mcm5 ZnF , Psf2, Mcm3 N-terminal helical domain, Mcm3 AAA + domain and Ctf4. Reconstructions in which these sites are all engaged (+/-Ctf4) we define as the ''fully engaged'' complex. Soft masks were generated covering both the visible regions of Ctf4/Cdc45/GINS in the resulting map, and the remainder of the density. These maps were used to subtract the non Ctf4/Cdc45/GINS density from the particle images and subsequently carry out masked local refinement of the Ctf4/Cdc45/GINS region in cryoSPARC. Local refinement was carried out using the default parameters and a fulcrum point defined by the mask centre, resulting in a reconstruction at 3.3 Å resolution for Ctf4/Cdc45/GINS ( Figure S1H). Strategy 2: Classes representing 73,884 particles were selected following heterogeneous refinement displaying strong Pol a-primase density in the absence of Ctf4 with the Mcm2-7 C-tier in conformation II. Masked 3D classification without alignment using ten 3D references was carried out in cryoSPARC, as described in strategy 1, to enrich for classes where the Pol a-primase Pri2 NTD is engaged with the Mcm3 N-terminal helical domain. Classes representing 53,964 particles were selected for two non-uniform refinement jobs in cryoSPARC using either a pixel size of 0.86 Å or 2.27 Å , resulting reconstructions at both 3.5 Å and 4.6 Å resolution respectively for the fully engaged complex in the absence of Ctf4 ( Figure S1E). Signal subtraction and masked local refinement was carried out in cryoSPARC using a pixel size of 0.86 Å , as described in Strategy 1, focussing on both the Pol12/ Pol1 CTD ( Figure S1I) and Pol12/Pol1 CTD /Pri2 NTD ( Figure S1J) regions of the complex, yielding reconstructions at 4.8 Å and 5.0 Å resolution respectively. In order to improve the noisy density adjacent to Pri1, a soft mask covering this region and Pri1 was generated in UCSF Chimera 71 based upon a low-pass filtered map derived from the consensus non-uniform refinement at 3.5 Å resolution. Masked 3D classification of this region (10 classes, target resolution 12 Å ) resulted in the selection of 39,555 particles that were unmasked and refined using non-uniform refinement, resulting in a reconstruction at 4.6 Å resolution at a pixel size of 2.27 Å ( Figure S1K). The improved resolution in this region resulted in its assignment as the Pri2 CTD . Strategy 3: Classes representing 100,178 particles with strong density for both the Mcm2-7 C-tier in conformation II and Pol a-primase regardless of Ctf4 occupancy. These data were refined using non-uniform refinement in cryoSPARC, using a pixel size of 0.86 Å , to a resolution of 3.5 Å . Signal subtraction and masked local refinement was carried out in cryoSPARC using a pixel size of 0.86 Å , as described in Strategy 1, focussing on the Mcm2-7 C-tier yielding a reconstruction at 3.2 Å resolution ( Figure S1F). Strategy 4: Classes representing 84,350 particles with strong density for Tof1-Csm3/dsDNA and Pol a-primase regardless of Ctf4 occupancy. These data were refined using non-uniform refinement in cryoSPARC, using a pixel size of 0.86 Å , to a resolution of 3.4 Å . Signal subtraction and masked local refinement was carried out in cryoSPARC using a pixel size of 0.86 Å , as described in Strategy 1, focussing on Tof1-Csm3/DNA yielding a reconstruction at 3.9 Å resolution ( Figure S1G).
In order to enrich for particles in which the Pri2 CTD was well resolved, the 588,597 particle subset initially imported into cryoSPARC following processing in RELION was re-processed using a 3D reference obtained from refinement of the 39,555 particle subset containing density for the Pri2 CTD . Four rounds of iterative heterogeneous refinement were carried out using four 3D references containing Pri2 CTD domain density resulting in an 87,540 particle subset. The non-Pol a-primase density was subtracted from these particles and 3D classification without alignment was carried out within a mask encompassing all of the visible regions of Pol a-primase. Classes were selected based on clear Pri2 CTD domain density resulting in a 45,111 particle subset. The corresponding un-subtracted particles were then refined using non-uniform refinement to a resolution of 4.6 Å prior to being further classified using 3D variability analysis 49 using 3 modes and a filter resolution of 9 Å . The results were displayed using clustering analysis using 6 clusters and a filter resolution of 9 Å . Inspection of the resulting reconstructions revealed the presence of additional density corresponding to the Pol1 exo/cat domain. An additional round of 3D variability analysis was carried out using the same parameters with a new mask additionally encompassing the Pol1 exo/cat domain density. This procedure identified a subset of 9,633 particles which were then refined to a resolution of 4.6 Å that displayed density for both the Pri2 CTD and the Pol1 exo/cat domains ( Figures S3G and S3H).

Article
Human replisome + Pol a-primase + 60 nucleotide 5 0 -flap DNA fork The data processing pipeline is illustrated by the schematic in Figure S6C. 7,355 39-fraction movies were aligned and dose-weighted (1.22 e-/Å 2 /fraction, 5 x 5 patches, 150 Å 2 B-factor) using RELION's implementation of a MotionCor2-like program. 69 CTF parameters were estimated using CTFFIND-4.1 70 and 44 poor-quality micrographs excluded from future processing. Particles were picked using RELION's Laplacian-of-Gaussian (LoG) function providing a minimum diameter of 180 Å and maximum of 330 Å . 1,535,548 particles were extracted using a box size of 380 Å . During extraction the data were down-sampled to a pixel size of 3.78 Å /pixel. Two successive rounds of RELION 3D classification (regularisation parameter, T = 4), using 6 classes, were carried out using a previously obtained map of the human core replisome as a 3D reference (EMD-13375). 41 Class selection was based upon the presence of secondary structure features within CMG. 3D classification was performed with a 250 Å diameter circular mask to focus classification on CMG. A resulting 550,340 particles were subsequently refined using 3D auto-refinement in RELION generating a reconstruction at 7.6 Å resolution following postprocessing. These data were submitted for per-particle motion correction using dataset-trained particle polishing in RELION 72 using a pixel size 0.86 Å in a 450 Å box. RELION CTF-refinement 64 (beamtilt and trefoil correction, anisotropic magnification correction, and per-particle defocus and astigmatism CTF correction) was then carried out and the data refined to an improved resolution of 3.6 Å following postprocessing in RELION. A further round of 3D classification was carried out with a dilated circular mask of 380 Å and classes with significant Pol a-primase density representing 359,677 particles selected for subclassification. Signal subtraction and masked 3D classification without alignment was carried out in RELION as described in the budding yeast data processing methods to enrich for replisomes stably associated with Pol a-primase using a mask covering both Pol a-primase and the MCM3 N-terminal helical domain.
3D classes in which Pol a-primase adopted the previously reported autoinhibited primosome conformation (PDB:5EXR) 46 were selected, representing 74,940 particles. Following reversion to the original non-subtracted particles.star file these data were imported into cryoSPARC and refined using non-uniform refinement 73 to 3.6 Å resolution ( Figure S7G). Density for DNA was not observed within this reconstruction.
Returning to the previous masked 3D classification without alignment in RELION, classes comprising 174,696 particles were selected in which Pol a-primase adopted a conformation distinct from that of the primosome. Following reversion to the original non-subtracted particles.star file these data were refined via 3D auto-refinement in RELION and postprocessed to a resolution of 4.1 Å . These data were then submitted to an additional round of particle polishing and CTF-refinement in RELION using the same parameters as the previous round. Particles were imported into cryoSPARC and a consensus refinement carried out using non-uniform refinement generating a reconstruction at a resolution of 3.4 Å ( Figure S6E). Local refinements were carried out in cryoSPARC, as described in the budding yeast data processing methods, for regions encompassing both TIMELESS-TIPIN/DNA ( Figure S6F) and the MCM2-7 C-tier ( Figure S6H) resulting in reconstructions at 4.1 Å and 3.7 Å respectively.
In order to improve the quality of the AND-1 density, particle subtraction followed by masked 3D classification without alignment was carried out as described in the budding yeast data processing methods. 3D classification was carried out on the total dataset imported into cryoSPARC, in a conformation distinct from the primosome, focussing on the AND-1/CDC45/GINS region of the map. 3D classes were selected, consisting of 63,393 particles, based on the presence of continuous strong AND-1 density. These data were then locally refined to generate a reconstruction at 3.3 Å resolution ( Figure S6G).
In order to improve the quality of the Pol a-primase density, particle subtraction followed by masked 3D classification without alignment was carried out using a mask covering the POLA1 CTD /POLA2/PRIM1/PRIM2 NTD region of the map. 3D classes were selected, consisting of 148,103 particles which were locally refined to generate a reconstruction at 4.4 Å resolution ( Figure S6I). A single 3D class was selected from this procedure, containing 23,758 particles, with particularly strong PRIM1 density. The non-subtracted particles comprising this class were subjected to consensus non-uniform refinement generating a reconstruction at 4.3 Å resolution.
Human replisome + Pol a-primase + 15 nucleotide 5 0 -flap DNA fork The data processing pipeline is illustrated by the schematic in Figure S8D. 6,718 39-fraction movies were aligned and dose-weighted (2.27 e-/Å 2 /fraction, 5 x 5 patches, 150 Å 2 B-factor) using RELION's implementation of a MotionCor2-like program. 69 CTF parameters were estimated using CTFFIND-4.1 70 and 109 poor-quality micrographs excluded from future processing. Particles were picked using Gautomatch v0.56 (https://www2.mrc-lmb.cam.ac.uk/research/locally-developed-software/zhang-software/#gauto) leading to extraction of 724,557 particles using a box size of 380 Å and pixel size of 4.28 Å /pixel (raw pixel size 1.07 Å /pixel). Two successive rounds of RELION 3D classification (regularisation parameter, T = 4), using 6 classes, were carried out using a previously obtained map of the human core replisome as a 3D reference (EMD-13375). 41 Class selection was based upon the presence of secondary structure features within CMG. 3D classification was performed with a 250 Å diameter circular mask to focus classification on CMG. A resulting 584,362 particles were subsequently refined using 3D auto-refinement in RELION generating a reconstruction at 8.8 Å resolution. These data were submitted for per-particle motion correction using dataset-trained particle polishing in RELION 72 using a pixel size 1.02 Å in a 450 Å box. RELION CTF-refinement 64 (beamtilt and trefoil correction, anisotropic magnification correction, and per-particle defocus and astigmatism CTF correction) was then carried out and the data refined to an improved resolution of 3.4 Å following postprocessing in RELION. A further round of 3D classification was carried out with a dilated circular mask of 380 Å . 28,202 particles in 3D classes lacking DNA were imported into cryoSPARC and refined using non-uniform refinement to 3.6 Å resolution. 492,011 particles in 3D classes with significant Pol a-primase and DNA density were selected for further subclassification. Signal subtraction and masked 3D classification without alignment was carried out in RELION as described in the budding yeast data processing methods to enrich for replisomes stably associated with Pol a-primase using a mask covering both Pol a-primase and the MCM3 N-terminal helical domain. The signal subtraction followed by 3D classification without alignment procedure was iterated resulting in the selection of 258,339 particles in 3D classes with strong Pol a-primase density. These data were imported into cryoSPARC and refined using non-uniform refinement to 3.3 Å resolution.

Cryo-EM model building
Budding yeast replisome + Pol a-primase + 60 nucleotide 5 0 -flap DNA fork To begin model building, structures of the budding yeast core replisome (PDB:6SKL) 30 with the MCM C-tier removed and the MCM C-tier in conformation II (PDB:6SKO), 30 were rigid body docked into the cryo-EM map of the budding yeast replisome fully engaged by Pol a-primase in the absence of Ctf4 at 4.6 Å resolution (binned pixel size of 2.27 Å ) using ChimeraX. 74 The atomic model for Ctf4 contained within 6SKO was manually removed at this stage. The structure of the human primosome (PDB:5EXR) 46 was then rigid body docked into the region of the cryo-EM map that remained unassigned, guided by the presence of secondary structure features within density for the Pri2 NTD . Inspection of the fit-to-density following 5EXR docking revealed a lack of strong cryo-EM density corresponding to both the POLA1 exonuclease and catalytic domains and the PRIM2 CTD , therefore these were subsequently removed from the model. The quality of the fit-to-density for the remaining human Pol a-primase model was then improved via manual manipulation followed by automated docking for both the PRIM1 and PRIM2 NTD domains and the POLA1 CTD /POLA2 module respectively. The positions of these human Pol a-primase subunits provided a reference to which the AlphaFold 50 models for budding yeast Pri2 NTD (residues S44-T299), PRIM1 (residues S12-D402) and the crystal structure the Pol1 CTD /Pol12 dimer (PDB:3FLO) were aligned prior to being rigid body fit into the density. An AlphaFold multimer 75 model for the Pol12 (residues 203-705) / Pol1 CTD (residues 1260-1468) complex was aligned to the Pol12 subunit of 3FLO, rigid-body fit into the cryo-EM density and the most N-terminal residue of Pol1 trimmed to I1271. The Pol1 C-terminus was remodelled based on an AlphaFold-Multimer 75 result indicating complex formation between the Pol1 C-terminus and Pri2 NTD , in an analogous fashion to the POLA1 C-term in the primosome structure 5EXR.
The quality of the fit for the model into the fully engaged cryo-EM map lacking Ctf4 at 4.6 Å resolution was optimised via an all-atom simulation in ISOLDE, 27 using adaptive distance restraints for the dsDNA model (kappa=100). The resulting model was further refined using Phenix 76 real-space-refine, utilising the input model as a reference to generate restraints with sigma=0.1 and global minimisation with nonbonded_weight=2000 and weight=0.5. Regions of the model that fit poorly to the density were the manually refined in Coot 77 using the local refinement and regularisation tools incorporating stereochemical restraints. The model for the Mcm2-ZnF (residues 338-378) was truncated due to the absence of well resolved density in this region of the map.
At this stage in the modelling process, regions of cryo-EM density that remained unmodelled were identified for further analysis to determine their identity. Density for a small four-helical bundle bound to the Mcm3 AAA + domain was assigned as the flexibly tethered Pol12 NTD (residues M1-I79). An AlphaFold-Multimer prediction for the Pol12 NTD interacting with the AAA + domain of Mcm3 was used to dock the Pol12 NTD into the cryo-EM density.
Two regions of disconnected helical density were identified in analogous positions to CLASPIN in the core human replisome (PDB:7PFO). 41 This led us to speculate that these represented regions of the budding yeast homologue of CLASPIN, Mrc1. Furthermore, AlphaFold-Multimer modelling predicted multiple high-confidence interactions between Mrc1 and the replisome. Predictions were validated by the presence of corresponding side-chain density for Mrc1, present in the un-binned cryo-EM reconstruction of the fully engaged Pol a-primase complex at 3.5 Å resolution. Mrc1 residues S339-K323 interact with the a-solenoid of Tof1, whilst D468-Q483 contacts Mcm2 in the N-tier and the Tof1 N-terminus. An additional region of disconnected density, on the same side of the replisome to the two modelled regions of Mrc1, was also assigned to Mrc1 based on AlphaFold-Multimer prediction and the presence of clear side-chain density. Alphafold-Multimer predicts an interaction between Mrc1 residues L815-E858 spanning both Cdc45 and the Mcm2 N-tier, for which there is cryo-EM density present in the fully engaged reconstruction at 4.6 Å resolution. However, there is only cryo-EM density of sufficient resolution to enable unambiguous assignment, in the 3.5 Å resolution fully engaged map, for Mrc1 residues N842-E858, therefore these are the only residues deposited in the final model for this particular Mrc1 interface.
A region of unmodelled density bound to the GINS subunit Psf2 was ascribed to Pri2 Nterm (residues M1-S5). Assignment was based upon the close proximity of the otherwise most N-terminal modelled residue of Pri2 (S44) and the presence of continuous density between S5-S44 present at low map thresholds. Clear side chain density was present for residues M1-Q4. As the second residue in Pri2 is a large phenylalanine residue the first methionine will not be removed 78 and is likely to be acetylated. 79 At this stage the model was inspected residue-by-residue in Coot, 77 docked into the highest resolution cryo-EM map (consensus and focused refinements) for the corresponding region of the model. Focussed refinements were rigid-body docked into the consensus and resampled onto the same origin. The model was manually refined against the map in Coot 77 and both backbone Ramachandran and rotamer outliers corrected. A final global run in ISOLDE 80 was carried out to minimise the clash-score using distance and torsion restraints prior to Phenix 76 real-space-refinement using the same restraints as described above. Model validation was carried out using the Molprobity server, 81 Phenix 76 validation and the wwPDB OneDep validation server.
In order to model the budding yeast replisome fully engaged by Pol a-primase in the presence of Ctf4, the structure of trimeric Ctf4 (PDB:6SKL) 30 was rigid body docked into the corresponding density. The fit-to-density was subsequently refined via local ISOLDE 80 simulation followed by Phenix 76 real-space-refinement and manual adjustment in Coot. 77 Inspection of the locally refined Ctf4/GINS/ Cdc45 cryo-EM map revealed the presence of unmodelled density bound to the helical bundles of two Ctf4 monomers in the correct position to accommodate the Pol1 CIP-box (a.a. F140-S149). 29 The Pol1 CIP box was subsequently modelled into each discrete ll OPEN ACCESS Article density based on the crystal structure of the Ctf4 CTD -Pol1 CIP-box (PDB: 4C93). 29 It was not possible to sub-classify the Ctf4 density to generate reconstructions with only one-site occupied at any time, therefore two models were deposited to the PDB with the Pol1 CIP-box interacting with a different monomer of Ctf4 in each. Human replisome + Pol a-primase + 60 nucleotide 5 0 -flap DNA fork A previously determined structure of the core human replisome (PDB: 7PFO) 41 was rigid body docked into the locally filtered cryo-EM map of the human replisome fully engaged by Pol a-primase on a DNA fork containing a 60 nucleotide 5ʹ-flap at 3.4 Å resolution using ChimeraX. 74 The atomic model for Pol ε contained within 7PFO 41 was manually removed at this stage, as were MCM2 residues 324-368 which comprise its zinc-finger motif due to the lack of corresponding density. Using the same strategy employed for the modelling of the budding yeast Pol a-primase-replisome structure, a previously determined model for the human primosome (PDB:5EXR) was rigid body docked into the remaining unassigned density. The 5EXR model was then manually edited to remove both the POLA1 exonuclease and catalytic domains and the PRIM2 CTD due to a lack of corresponding cryo-EM density. The fit-to-density for the Pol a-primase subunits was improved by docking each module: PRIM1, PRIM2 and the POLA1 CTD /POLA2 dimer, individually as a rigid body. AlphaFold multimer 75 structure predications for the PRIM1-PRIM2 NTD complex and the POLA2-POLA1 CTD complex were aligned to the corresponding subunits derived from 5EXR, replacing the crystal structure subunits. Using this strategy atomic models were generated for the following regions of sequence: PRIM2 residues Q17-H252, PRIM1 residues M9-T349 and T386-G408, POLA1 residues Q1279-G1445 and E1448-C1458 and POLA2 residues I96-T114 and V170-I598.
The quality of the fit-to-density was optimised via an all-atom simulation in ISOLDE, 80 using adaptive distance restraints for the dsDNA model (kappa=100). The resulting model was further refined using Phenix 76 real-space-refine, utilising the input model as a reference to generate restraints with sigma=0.1 and global minimisation with nonbonded_weight=2000 and weight=0.5. Regions of the model that fit poorly to the density were then manually refined in Coot 77 using the local refinement and regularisation tools incorporating stereochemical restraints.
Inspection of the cryo-EM density following the modelling procedure outlined revealed a short region of unmodelled helical density adjacent to the PSF1 subunit. AlphaFold-Multimer analysis predicted a helix within the flexible N-terminus of POLA2, residues I96-T114, to bind at the location of the unmodelled density. Furthermore, clear side chain density for POLA2 Y113 and L109 corroborated the prediction in addition to stereo-chemically favourable contacts formed. This region of POLA2 was subsequently incorporated into the final model.
A region of unmodelled density bound to the GINS subunit PSF2 was assigned to the PRIM2 Nterm , residues M1-G5, and incorporated into the final model. Assignment was based upon the close proximity of the otherwise most N-terminal modelled residue of Pri2, Q17, and the presence of continuous density between G5-Q17 present at low map thresholds. Clear side chain density was present for residues M1-S4. Furthermore, the PRIM2 Nterm -PSF2 contact was predicted by AlphaFold-Multimer at high confidence.
At this stage the model was inspected residue-by-residue in Coot, 77 docked into the highest resolution cryo-EM map (consensus and focused refinements) for the corresponding region of the model. The model was manually refined against the map in Coot 77 and both backbone Ramachandran and rotamer outliers corrected, followed by real-space refinement in both ISOLDE 80 and Phenix. 76 Each focussed refinement was rigid-body docked into the consensus refinement (EMD-15341) and resampled onto the same map origin. Only density at the AND-1 -CDC45/GINS interface was used to align the CDC45/GINS/AND-1 local refinement (EMD-15342) to the consensus (EMD-15341). However, due to subtle differences between the density at this interface between the local and consensus refinements it was not possible to build a model that perfectly satisfied both maps. A final global run in ISOLDE 80 including distance and torsion restraints was carried out to minimise the clash-score prior to Phenix 76 real-space-refinement using the same restraints as described above into the consensus refinement with additional reference model restraints. Model validation was carried out using the Molprobity server, 81 Phenix 76 validation and the wwPDB OneDep validation server. Following this procedure, the model for CLASPIN residues E299-E310 was removed due to the lack of corresponding cryo-EM density.

QUANTIFICATION AND STATISTICAL ANALYSIS
No quantification or statistical analysis were performed in this manuscript. (B) Focused view of the interface between PRIM2Nterm and PSF2. Models overlaid were derived from reconstructions on fork DNA with either a 15 nt or 60 nt 5ʹ-flap, in addition to models derived from these respective datasets where DNA engagement was not observed. Models were aligned on PSF2.
(C) Focused view of the interface between POLA296-114 and GINS subunits PSF1 and SLD5. Models overlaid were derived from reconstructions on fork DNA with either a 15 nt or 60 nt 5ʹ-flap, in addition to models derived from these respective datasets where DNA engagement was not observed. Models were aligned on PSF1.
(D) Detailed view of the atomic model for the interface between the PRIM2NTD (green) and the MCM3 N-terminal helical domain (cyan) (site b). Selected residue sidechains positioned to form inter-protein contacts are labelled.
Residue sidechains are displayed as truncated stubs as the corresponding cryo-EM density is of insufficient resolution to determine their conformation.
(E) Pol α-primase associated replisome model docked into a cryo-EM map in which continuous low-resolution density is visualised between the most C-terminal modelled residue of the PRIM2Nterm (G5) and the next modelled residue (Q17). The connecting density was manually coloured green using ChimeraX.
(F) Alternative views of the atomic model for the interface between the POLA2 N-terminal helix (residues 96-114) (green) and GINS subunits PSF1, PSF3 and SLD5 (brown). Residue sidechains positioned to form interprotein contacts are labelled. Residue sidechains are only displayed when the corresponding cryo-EM density is of sufficiently high-resolution to determine their conformation, otherwise sidechains are displayed as truncated stubs.
(G) Overlay of cryo-EM density for reconstructions of budding yeast replisomes containing (transparent grey, this study) and lacking (black mesh, EMD-10227) Pol α-primase [S6] . The associated atomic model is docked into the density (yellow). The approximate position of the human POLA2 helix (residues 96-114) bound to PSF1 and SLD5 is overlaid in green and unmodelled density in the vicinity in yeast reconstructions is circled in red.
(H) Model for AND-1 in complex with three copies of the POLA1 AND-1 interaction motif (residues 151-171), docked into a locally refined reconstruction.
(I) Focused views of the cryo-EM density for the helical bundle of each AND-1 monomer bound by POLA1 (transparent grey) overlaid with cryo-EM density for a reconstruction of the human replisome in the absence of Pol α-primase (EMDB-13376) [S9] (black mesh). An AlphaFold-Multimer model for the interface between POLA1151-171 and the helical bundle of AND-1 was aligned to each AND-1 monomer and the fit optimised using rigid-body docking into the cryo-EM map.