Structural Insight into the MCM double hexamer activation by Dbf4-Cdc7 kinase

The Dbf4-dependent kinase Cdc7 (DDK) regulates DNA replication initiation by phosphorylation of the MCM double hexamer (MCM-DH) to promote helicase activation. Here, we determine a series of cryo electron microscopy (cryo-EM) structures of yeast DDK bound to the MCM-DH. These structures, occupied by one or two DDKs, differ primarily in the conformations of the kinase core. The interactions of DDK with the MCM-DH are mediated exclusively by subunit Dbf4 straddling across the hexamer interface on the three N-terminal domains (NTDs) of subunits Mcm2, Mcm6, and Mcm4. This arrangement brings Cdc7 close to its only essential substrate, the N-terminal serine/threonine-rich domain (NSD) of Mcm4. Dbf4 further displaces the NSD from its binding site on Mcm4-NTD, facilitating an immediate targeting of this motif by Cdc7. Moreover, the active center of Cdc7 is occupied by a unique Dbf4 inhibitory loop, which is disengaged when the kinase core assumes wobbling conformations. This study elucidates the versatility of Dbf4 in regulating the ordered multisite phosphorylation of the MCM-DH by Cdc7 kinase during helicase activation.

I n eukaryotes, DNA replication initiates from multiple origins along each chromosome 1,2 . Every initiation event is tightly regulated to ensure that it occurs only once per cell cycle. This regulation is achieved through a temporal separation of replication licensing from origin firing by the oscillating kinase activities during the cell division cycle 3,4 .
At early G1 phase, when the DDK (Dbf4-dependent Cdc7 kinase) and S-CDK (S-cyclin dependent kinase) kinase activities are low, the origin recognition complex (ORC) together with Cdc6 and Cdt1 load the Mcm2-7 helicases onto origin DNA to form an inactive MCM double hexamer (DH) encircling dsDNA to license replication [5][6][7][8][9] . Upon entering S phase, the rise of these kinase activities drives origin firing to initiate DNA replication 4 . During this process, DDK and S-CDK, act in concert with various origin-firing factors to convert the DH into two active Cdc45-Mcm2-7-GINS (CMG) helicases which then serve as scaffolds for the assembly of bidirectional replisomes [10][11][12] . Extensive genetic, biochemical, and structural studies have been carried out on the DDK and S-CDK, but little is known about their mechanistic actions on their substrates partly because the structures of these kinases in complex with the DH and other targets have not been determined.
Cdc7 is a serine/threonine kinase whose activity is dependent on its association with Dbf4 13 . The protein level of Dbf4 fluctuates during the cell cycle, peaking in late G1 and disappearing during mitotic exit to maintain a low DDK activity in G1 phase [14][15][16] . Although DDK can act on diverse substrates implicated in various events related to cell cycle progression [17][18][19][20][21] , its major target is the Mcm2-7 complex 22,23 . DDK prefers to phosphorylate Mcm2/4/6 subunits that have been incorporated into the DH at origin DNA 24,25 . Compelling evidence in budding yeast has also shown that the essential role of DDK in DNA replication is to relieve an inhibitory effect imposed on replication initiation from the N-terminal serine/threonine-rich domain (NSD) of Mcm4 (residues 74-174), as NSD deletion can bypass DDK function 26 . Furthermore, DDK function can also be bypassed by the P83L (bob1) mutation in Mcm5, which is not a substrate of DDK, suggesting that DDK phosphorylation may affect the overall structure of the MCM-DH 22 . However, previous structural studies at low resolutions did not detect an obvious conformational change in the core of the DH after DDK phosphorylation 7,27 . In response to DNA damage in S phase, Rad53 antagonizes DDK through inhibiting its kinase activity, keeping it away from unfired origins [28][29][30] .
Cdc7 orthologs share similar sequence signatures with many canonical serine/threonine kinases but with variable kinase insertions (KIs) intervening the conserved kinase elements 31 . In contrast, Dbf4 contains only three short regions, motifs N, M, and C, with limited homology among orthologs while the remaining parts are predicted to be largely unstructured. Nevertheless, the function of DDK in regulating replication initiation is highly conserved among eukaryotes.
Crystal structures of human Cdc7-Dbf4 kinase complex have been determined by removing most flexible motifs or insertions to optimize crystallization 32,33 . In these structures, Cdc7 assumes a bilobed configuration with its kinase active site sitting in a deep cleft in between the N-and C-lobes. Motifs M and C from Dbf4 each positions against one of the bilobed surfaces of Cdc7 to stabilize the kinase complex in an active state. These truncated Dbf4 structures, though informative, do not provide a complete picture about Cdc7 activation by Dbf4. It is also unclear how DDK is recruited onto the DH to target its substrates for phosphorylation and what impact DDK has on the conformation of the DH that signals helicase activation.
Cdc7's role in regulating cell proliferation is most evident in cancer cells 34 . Cdc7 overexpression is often observed in human cancer cell lines and tumor tissues, and cancer cells with Cdc7 knockdown undergo apoptosis in a p53-independent manner 35 . Thus, Cdc7 has emerged as an attractive drug target for cancer therapy 36 . At present, a number of Cdc7 inhibitors are being developed or evaluated in clinical or preclinical trials. Notably, all of these inhibitors are ATP competitors with a potential to produce undesirable off-target effects through non-specific binding to other kinases. To achieve selective killing of only cancer cells for chemotherapy, it is important to develop highly specific DDK inhibitors that are capable of antagonizing unique features of the kinase. Therefore, understanding the detailed structural and molecular mechanisms of Cdc7 activation by Dbf4, the docking of the DDK onto the DH, and the activation of DH by DDK would be invaluable for developing anti-cancer drugs.
In this work, we assemble the full-length DDK onto the endogenous MCM-DH purified from G1 chromatin of budding yeast and determine the cryo-EM structures of the kinasesubstrate complexes. The structures present snapshots of the Dbf4-Cdc7 kinase engaging with the DH in an intricate process that regulates DDK substrate targeting and helicase activation at origin DNA.

Results
Structural determination of DH-DDK complexes. The DDK has been shown to specifically target MCM subunits within the assembled DH 24 . To investigate how DDK performs this function, we isolated the full-length Cdc7-Dbf4 heterodimer from yeast cells, and assembled the kinase complex onto the DH in the presence of ATP or ATP-γ-S for structural analysis. Initially, we incubated DDK in excess with the DH at a molar ratio of 10:1 before subjecting the mixture to cryo sample preparation to capture the kinase-substrate complex in different functional states. However, the population of the DDK-bound DH (referred to as DH-DDK) was less than 10% of the total particles for both the ATP and ATP-γ-S treated samples ( Supplementary Fig. 1a-c). Similar results were observed when the kinase complex was incubated with two mutant DHs, DH-Mcm4ΔN140 (referred to as DHΔN140) and DH-Mcm4ΔN174 (referred to as DHΔN174) ( Supplementary Fig. 1d, e), in which the NSD was either partially or completely truncated from Mcm4. These results indicate that the interaction between DDK and the DH is relatively weak and transient.
To stabilize the assembled DH-DDK complexes, we subjected the kinase-substrate complex (treated with ATP-γ-S) to crosslinking by grafix, which involved a mild fixation with glutaraldehyde (GA) during ultracentrifugation 37 . Using the samples collected from grafix for cryo-EM analysis, we found that the percentage of DDK-bound DH particles was greatly increased to 88.7% of the total particles ( Supplementary Fig. 3a-e), indicating that the GA crosslinking effectively preserved the integrity of the unstable kinase-substrate complexes. As evident from 2D classification averaged images, many clearly show densities for bound DDK ( Supplementary Fig. 2c). Subsequent initial 3D reconstruction showed that while the DH is relatively rigid, DDK exhibits high flexibility (Supplementary Fig. 2d). Considering that the two binding sites of DDK on a single DH is symetrical, we used symmetry expansion to double the particle number and performed focused classification on the DDK region ( Supplementary Fig. 2d). The first round of 3D classification identified two major conformational subsets for DDK on the DH. Based on the occupancy of the two symmetrical DDK binding sites on the DH and the general stability of DDK (stable DDK vs stable Dbf4-NTD) ( Supplementary Fig. 2d), the particles of these two subsets could be divided into six groups ( Supplementary Fig. 3), representing major compositional and conformational populations of DH-DDK assemblies in the reaction. Consequently, the maps of these six groups were determined at resolutions ranging from 3.4 to 4.0 Å ( Supplementary Fig. 3g). Groups I-III, representing DHs with two DDKs bound, account for 44.1% of the total particles (Supplementary Fig. 3a-c). In these groups, the two DDKs are well separated from each other on the DH, showing no detectable interaction between the kinase complexes. Groups IV and V, showing one DDK bound to the DH, account for 44.5% of the total particles ( Supplementary Fig. 3d, e); whereas Group VI (11.3%) represents DHs without DDK bound ( Supplementary Fig. 3f). Together, these DH-DDK structures present snapshots of the docking of two DDKs onto the DH to activate the MCM complexes.
In the map produced from particles of the first subset, both Dbf4 and Cdc7 are stably attached to the DH (State I) (Fig. 1, Supplementary Fig. 2d, and Supplementary Movie 1), whereas in the second subset only the N-terminal half of Dbf4 is visible in the derived maps. Particles of the second subset were then pooled and subjected to another round of focused 3D classification, which identified several additional states for DDK (States II-IV). In the maps of States II-IV, the N-terminal part of Dbf4 are stably bound at the same site, while the kinase core of Cdc7 is in different wobbling positions ( Supplementary Fig. 2d).
The final density map of State I has a global resolution of 2.9 Å (Supplementary Fig. 2d-f) with most residues at the DH-DDK interfaces clearly resolved ( Supplementary Fig. 2g). However, the local resolution of the kinase core region, which is away from the DH-DDK interface, was still relatively low. Based on signal subtraction focusing on the DDK region, we further improved the map of the kinase core region of DDK to a resolution of 3.8 Å ( Supplementary Fig. 2d-f), enabling atomic modeling for~50% of the sequence of Dbf4 and~90% of Cdc7.
The role of Dbf4 in docking Cdc7 onto the DH. A striking feature of the DH-DDK structure is that Cdc7 has almost no physical contact with the DH (Fig. 1) Dbf4 is a very flexible protein (704 residues) containing three long disordered segments, and only those segments interacting with MCM subunits or Cdc7 are resolved in our structure (Fig. 2a, b). These binding motifs spread over the entire length of the protein (Fig. 2b). At the N-terminus, motif N of Dbf4 (residues 105-221) containing a BRCT domain clasps onto the NTD-A of Mcm2' (Figs. 1a, d and 2a). Consistent with the crystal structure of the human DDK complex 33 , motif M exclusively interacts with Cdc7 (Figs. 1b, c, e, f, and 3a). In contrast, motif C of Dbf4 at its very C-terminus is sandwiched between Cdc7 and the NTD-A of Mcm4 (Fig. 1c, f), appearing to stabilize DDK on the DH. Of note, motif C adopts different orientations relative to Mcm4-NTD-A in States I and II of the DH-DDK (Fig. 2e, f). In addition to these known motifs, two other MCM-interacting When motif N is removed from Dbf4, the ability of DDK to phosphorylate MCM complex is largely suppressed (Supplementary  The structure of DDK on the DH. Overall, the structure of the kinase core of yeast DDK matches very well with its human counterpart (Fig. 3d) despite their limited Dbf4 homology 31 . Their structural similarity indicates that DDK employs a highly conserved mechanism to target its substates on DHs across species. In the DH-DDK structure, the yeast Cdc7 adopts a bilobed configuration with the active center located in a deep cleft between its N-and C-lobes (Fig. 3a), a common feature shared among most serine/threonine kinases 40 . Similar to the human DDK structure 32 , the N-and C-lobes of Cdc7 are positioned against motif M and motif C of Dbf4, respectively (Fig. 3a), suggesting that Dbf4 may have a role in modulating the conformation of Cdc7 kinase. Indeed, motif M plays an important role in stabilizing two large kinase-insertion loops (KI-2 and KI-3) on the Cdc7 C-lobe (Fig. 3a, b). Specifically, as also seen in the human DDK structure 33 , a β-hairpin (β2-3) of Dbf4 motif M pairs up with the N-terminal end of the KI-3 of Cdc7; and another short β-strand (β1) of the motif M complements a short β-strand from the KI-2 in an anti-parallel fashion (Fig. 3b). This special arrangement stabilizes the KI-2 and KI-3 in a stretched conformation, contributing to a structured activation loop of the kinase in a configuration resembling that observed in the active human DDK 33 . In addition, two newly identified α helices, which extend from the C-terminus of motif M, encircle the C lobe of Cdc7 for more than a half turn to associate with most of the α helices present in this domain (Fig. 3a), establishing a strong coupling between Dbf4 and Cdc7. Together, the extensive interactions of Dbf4 with the two lobes of Cdc7 as well as its special engagement with the KI-2 and KI-3 stabilize, shape as well as activate the Cdc7 kinase.
More importantly, in our DH-DDK (State I) structure, a unique loop from Dbf4 is found residing in the substrate-binding pocket of Cdc7 (Fig. 3a), in an equivalent position of the peptide substrate in the human DDK structure 33 (Fig. 3d). This loop (residues 504-519) extends from the N-terminus of the M4NB motif, and interacts with the KI-2 in several positions (Fig. 3c). The N-terminal residue of this loop, Thr506, is situated close to the γ-phosphate of the ATP-γ-S nucleotide in the active center of Cdc7 (Fig. 3d, e). This observation suggests that this loop (refered as regulatory loop) likely has a regulatory role to limit the accessibility of the stubstrate binding pocket.  Fig. 4g). These results suggest that motif C of Dbf4 plays a critical role in enabling the kinase activity.
More interestingly, while the motifs N and M6NB of Dbf4 remain stably bound to the analogous surfaces on the NTD-As of Mcm2' and Mcm6, the kinase core of DDK takes up several distinct positions, in two relatively stable states (I and II) (Fig. 4), two less stable states (III-IV) and more highly dynamic states in different subpopulations of the DH-DDK particles (Supplementary Figs. 2d, 5). In State I, the kinase core makes an intimate contact with the Mcm4-NTD-A via motif C of Dbf4 (Fig. 4a-c). In State II, the kinase core still engages with the NTD-A of Mcm4 but undergoes a rotation, which relocates the ZF motif from motif C of Dbf4 to a different surface on Mcm4 (Fig. 4d- Fig. 5). Notably, the kinase core from these two less stable states is completely disengaged from Mcm4 such that the main body of Cdc7 is turned away from Mcm4, and positioned towards the NTD-A of Mcm2 where the motif N of Dbf4 binds ( Supplementary Fig. 5c, d, g, h, k, l). In State IV, the EM density of the kinase core makes a direct contact with motif N of Dbf4 ( Supplementary Fig. 5d, h). These different conformations of DDK illustrate a dynamic association between the kinase core and the NTD-A of Mcm4, and explain the ability of DDK to target various sites within the NTEs of Mcm2, Mcm4, and Mcm6 on the DH.
In State I of the DH-DDK, the regulatory loop (residues 504-519) of Dbf4, which is upstream of the M4NB (Fig. 3a), is embedded in the substrate binding site of Cdc7 (Fig. 3a, d, e). Similar confirmation of this loop can also be observed in both DHΔN140-DDK and DHΔN174-DDK structures (Supplementary Fig. 6). This feature suggests a regulatory function for this loop on the activity of Cdc7 kinase. Intriguingly, this putative inhibitory loop does not associate with the kinase in State II of the DH-DDK, whereas its downstream M4NB remains stably attached to the NTD-A of Mcm4 (Fig. 3d, e). These comparisons suggest that the M4NB serves as a relatively fixed anchor point for attaching DDK to the NTD-A of Mcm4, whereas the regulatory loop is a dynamic element that might impact the substrate preference of Cdc7. Indeed, upon removal of the regulatory loop, the mutant DDK appears to be more active towards Mcm6 and the phosphorylation of Mcm4 is also evidently delayed (Supplementary Fig. 4e) DDK dislodges Mcm4-NSD on the DH. Previous study showed that the essential role of DDK is to relieve the inhibitory effect of Mcm4-NSD upon origin firing 26 . To understand the role of DDK phosphorylation in this process, we analyzed the structures of the NSD on the DH surface before and after DDK binding. In two published cryo-EM maps of the yeast DH, some extra densities associating with the NTD-A of Mcm4 were visible but not analyzed because of their poor quality for subunit identification and sequence assignment 7,9 . In our structures of the free DH (more than 90% of all particles) from the two large datasets of noncrosslinked DH-DDK treated with ATP-γ-S, similar EM densities were also observed. After multiple rounds of local 3D classification of this region ( Supplementary Fig. 1a, b), we finally obtained a density map for the DH with an obviously structured fragment nested on the NTD-A surface of Mcm4. The map was resolved at an overall resolution of 3.2 Å ( Supplementary Fig. 1a, b, f), enabling us to unambiguously assign residues 132-153 of the Mcm4-NSD to this density piece (Fig. 5a, e, h). The NSD fragment, largely in a loop conformation, wraps around the NTD-A of Mcm4 and extends to the subunit interface between Mcm4 and Mcm6. Its N-terminal residues before residue 131 appear highly disordered (Fig. 5a, e, h). The N-terminus of the structured NSD (residues 132-141) forms a U-turn loop at the juncture of the OB and NTD-A subdomains of Mcm4. The C-terminus of this fragment (residues 145-153) contains a short β-strand which is parallel to the N-termimal β-strand of the Mcm4-NTD-A (residues 176-181), leaving the connecting residues of 154-176 in a highly flexible loop.
To further confirm this assignment of the Mcm4-NSD, we next examined the structures of the two mutant DHs (DHΔN140 and DHΔN174) (Supplementary Fig. 1d, e). As shown in Fig. 5c, the EM densities for the residual NSD region (residues 148-153) in the map of the DH-ΔN140 is also evident (Fig. 5c, f, i). In contrast, when the entire NSD is truncated from Mcm4, the densities corresponding to the structured NSD completely disappear from the map of the DH-ΔN174 (Fig. 5d).
In the previous cryo-EM map of DH after DDK phosphorylation (EMD-3833) 7 , the entire NSD motif becomes unstructured and is no longer visible on the DH altogether. Consistently, in our dataset of non-crosslinked DH-DDK treated with ATP, where DDK phosphorylation is active, we did not identify any 3D classes with a structured Mcm4-NSD ( Fig. 5b; Supplementary  Fig. 1c). As removal of the NSD from Mcm4 bypasses the requirement of DDK for helicase activation 26 , we believe that Mcm4-NSD either deleted or phosphorylated by DDK achieves the same goal of remodeling NSD conformation to alleviate its inhibitory effect on helicase activation.
In addition, the dynamic behavior of Mcm4-NSD is best illustrated by comparing the DH structure before and after DDK binding. In the DH-DDK structure from the ATP-γ-S treated sample, upon DDK binding, Mcm4-NSD becomes flexible and completely invisible. Instead, the M4NB of Dbf4 (residues 520-534) associates with the Mcm4-NTD-A by displacing Mcm4-NSD from its binding site in the free DH (Fig. 5e-j), suggesting that the M4NB and Mcm4-NSD compete for the same binding surface on Mcm4. The phosphorylation of NSD likely acts to maintain this dislodged conformation after DDK is released from the DH.
Temporal order of DDK phosphorylation of the MCM subunits. The structural features of the DH-DDK complex suggest that Dbf4 helps to establish a hierarchy for Cdc7 kinase to phosphorylate the multiple substrates located on the MCM complex. To test this hypothesis, we first examined the phosphorylation status of the MCM subunits by DDK. As DDK is very active in phosphorylating MCM subunits 27 , kinase assay was performed with purified DDK and DH on ice to slow down the reaction. Under this condition, a clear separation of the temporal phosphorylation patterns between Mcm4 and Mcm6 can be observed ( Supplementary Fig. 4b). Phosphorylation of Mcm4 by DDK started immediately after the kinase was mixed with the DH in the presence of ATP. In sharp contrast, obvious phosphorylation of Mcm6 could only be detected at much later time points, and even up to 32 min a large portion of Mcm6 remained unphosphorylated ( Supplementary Fig. 4b). Interestingly, when either the regulatory loop or M4NB of Dbf4 is removed, this temporal order of DDK in targeting its substrates is largely abolished, and phosphorylation of Mcm6 is much enhanced (Supplementary Fig. 4e, f). Together, our structural and biochemical data indicate that Dbf4 mediates the docking of DDK onto the DH to initiate an orchestrated series of phosphorylation events that begins with the NSD of Mcm4 containing the only essential phosphorylation targets for DDK in helicase activation.

Discussion
The head-to-head dimeric MCM helicases encircling dsDNA are activated simultaneously to enable bidirectional DNA synthesis by passing each other along opposite single DNA strands to form the bidirectional replication forks. A previous working model proposed that the dimerization of two DDK complexes upon binding to the DH coordinates the simultaneous phosphorylation of the coupled MCM complexes 30 . However, in our DH-DDK structures, the DDKs are positioned apart from each other without direct contacts ( Supplementary Fig. 3a-c). We did not capture any interactions formed between the assembled kinases or allosteric changes induced by one DDK to recruit the other, suggesting that individual DDKs might dock onto the DH in a stochastic manner to operate independently. However, we cannot rule out the possibility that two DDKs are loaded cooperatively through weak and/or transient interactions between them or through mediation of other factors. Indeed, forkhead transcription factor Fkh1 is known to recruit Dbf4 to a subset of early replication origins for helicase activation 41 . This example highlights the importance of specific DNA element(s), such as Fkh1 binding sites, in facilitating DDK docking onto the DH especially in cells where Dbf4 level is limited. It is also possible that phosphorylation of the DH by the first DDK might pave the way for recruiting the second DDK. These possibilities need to be further investigated by other approaches. Consistent with previous data that the DH structures with or without DNA bound appeared almost identical 7-9 , our structures, at much-improved resolution, confirmed that DDK binding and phoporylation did not induce obvious structural changes to the DH. It is likely that at least under in vitro condition, dsDNA bound with the DH does not play a major role in regulating DDK recruitment or its activity on the DH. An immediate consequence of DDK phosphorylation is the recruitment of Sld3-Sld7 and Cdc45 onto the DH 11,42 . Previous studies showed that the Sld3 and Sld7 heterodimers form a quaternary complex through dimerization of the CTD of Sld7 43 . It is likely that simultaneous activation of the two MCM hexamers is initiated by the recruitment of the Sld3-Sld7 dimer to the fully DDK-phosphorylated DH and not before.  only essential substrate 24,26 . The docking of Cdc7 onto the DH is mediated exclusively by Dbf4 through engaging with the NTD-As of Mcm2 from one hexamer and Mcm4 and Mcm6 from the opposite hexamer ( Figs. 1 and 2). This arrangement ensures DDK to preferentially target MCM subunits assembled into the DH, and also positions the kinase core in a strategic location to capture the NSD of Mcm4 with a higher probability (Figs. 1 and 4, Supplementary Fig. 4b). Specifically, the M4NB motif of Dbf4 displaces the NSD from its binding surface on Mcm4 through competitive binding to facilitate an immediate capture of NSD by the kinase. Importantly, the kinase core of DDK can assume multiple wobble conformations (States II-IV) when disengaging from Mcm4. In fact, a large population of the DH-DDK particles exhibit a highly dynamic kinase core that cannot be modeled while motif N and M6NB of Dbf4 remain stably associated with Mcm2 and Mcm6. This dynamic feature enables the kinase to reach out for its substrates from Mcm2 and 6. After DDK phosphorylation, the NSD motif becomes highly flexible and loses its affinity for Mcm4-NTD-A. This is highly consistent with the observation that NSD deletion can bypass the requirement of DDK for helicase activation 26 . Our observation also agrees with a previous study which showed that a phosphomimetic mutant of Mcm4 with seven DDK-dependent phosphorylation sites mutated to D or E, can support growth of a cdc7-4 strain at nonpermissive temperature 44 . In fact, two of these DDK sites, T140 and S141, are located right in the middle of the structured Mcm4-NSD (residues 132-153) (Fig. 5h). It is likely that DDK phosphorylation of these two sites are important for preventing the NSD from binding to the NTD-A of Mcm4 and facilitates its displacement by Dbf4-M4NB. These results suggest that the inhibitory effect of the NSD on helicase activation may modulate the accessibility of particular surfaces on Mcm4 to helicaseactivating factors before S phase entry. In further support of this notion, Tof1, a component of fork protection complex (FPC), binds to these same surfaces on Mcm4 in the replisome in addition to docking on the NTD-As of Mcm2 and Mcm6 (Supplementary Fig. 7a-d) 45 . It is also important to note that Sld3 binds to the phosphorylated NTEs of Mcm4 and Mcm6 to promote the recruitment of Cdc45 onto the DH for helicase activation upon DDK phosphorylation 42 . Further investigation is needed to determine how Sld3-Sld7 complex and/or other helicase-activating factors such as Cdc45 are recruited onto the DH through the NSD protected surfaces, as well as the potential roles of these phosphopeptides from Mcm4 and Mcm6.
A recent in vitro study showed that Mcm2-NTE (1-127) plays a critical role in recruiting DDK onto the DH as removal of this motif from Mcm2 largely disrupted this process 30 . In addition, the NTE of Mcm2 is also found to be important for mediating the interaction between Mcm2 and Dbf4 in a yeast two-hybrid analysis 46 Fig. 7). Although the overall structures are very similar among the six NTD-As, each NTD-A unit exhibits features specific for its binding partner(s) (Supplementary Fig. 7). As all MCM subunits have their own unique NTEs 48 , these flexible extensions may orchestrate the binding pattern for the helicase-activating factors to promote CMG formation and replisome assembly. Remarkably, though Tof1 and Dbf4 share the same binding surfaces on Mcm4 and Mcm6 ( Supplementary  Fig. 7b-d), their docking sites on the NTD-A of Mcm2 do not overlap ( Supplementary Fig. 7i-l). This selection of docking sites may have important implications for the function of DDK in fork progression. During S phase, while the binding sites of Dbf4 on Mcm4 and Mcm6 are occupied by Tof1, DDK could still retain its association with the replisome mainly through its engagement with the NTD-A of Mcm2. With the help of Tof1 20 , DDK may target its substrates at the replication fork (Fig. 6). Indeed, DDK is required at the replication fork to coordinate DNA replication with other processes, such as DNA repair 18 and meiotic recombination 20 .
Another striking feature of the DDK structure is the regulatory loop of Dbf4 embedded in the kinase center of Cdc7 to limit the accessibility of the activity center (Figs. 3a, c-e, and 4b). This special arrangement locks the kinase into an autoinhibitory state that effectively prevents premature capture of DDK substrates by Cdc7 to ensure a temporal order of DDK phosphorylation. A recent study reported that a similar autoinhibitory loop was also observed at the active center of CDK1 in the structure of CDK1cyclin B1-CKS1 in complex with separase 49 . This recurring feature in CDK1 and DDK suggests that these cell cycle-dependent kinases may use a similar strategy to regulate substrate accessibility to their kinase active centers. Interestingly, this inhibition from Dbf4 regulatory loop is relieved when the kinase core disengages from Mcm4 and dislocates the inhibitory loop of Dbf4 from Cdc7 while remaining associated with Mcm4 through its downstream M4NB. Importantly, Thr506 and Ser507 from the regulatory loop of Dbf4 within the kinase center are situated very close to the ATP-γ-S nucleotide bound by Cdc7 (Fig. 3d, e). We propose that auto-phosphorylation of these sites by DDK may release the inhibitory loop of Dbf4 from the kinase center. This tantalising hypothesis merits further investigations.
In response to DNA damage in S phase, the checkpoint kinase Rad53 inhibits DNA replication initiation through targeting Dbf4, first, by phosphorylating Dbf4 to inhibit DDK kinase activity and, second, by binding to Dbf4 to keep DDK away from unfired origins 28,29 . Our DDK structure supports both of these independent mechanisms. So far, 26 potential Rad53 phosphorylation sites have been identified in Dbf4 28 . Four of them crucial for maintaining DDK activity under stress condition are clustered within the M4NB of Dbf4. We surmise that the inhibitory phosphorylation of this region by Rad53 might interfere with the binding of Dbf4 to the NTD-A of Mcm4. When these four sites were changed to alanine with key Rad53 phosphorylation sites in Sld3 also mutated, firing of late origins was observed in the mutant cells when challenged by hydroxyurea 28 . This result reinforces the importance of Dbf4-M4NB in displacing Mcm4-NSD to promote helicase activation. It has also been shown that the binding of Rad53 to DDK sterically hinders DDK from docking onto the DH independent of Rad53 kinase activity 30 . To understand the structural basis for this steric interference, we superimposed the DDK-DH structure with the crystal structure of Rad53-FHA1 bound to Dbf4-BRCT 50 ( Supplementary Fig. 7g, h). It appears that the FHA1 domain of Rad53 and the NTD-A of Mcm2 bind to different surfaces of the BRCT domain of Dbf4 and that Rad53 does not directly compete for the docking site on Dbf4-motif N for Mcm2. Rather, the bulkiness of the Rad53-DDK complex may block Dbf4 access to the NTD-A of Mcm2, or Rad53 may remodel DDK into a state incompetent for docking onto the DH.
In summary, this study unravels a multifaceted role of Dbf4 in modulating Cdc7 kinase activity on substrates of the DH (Fig. 6): (1) Dbf4 docks Cdc7 onto the DH; (2) Dbf4 situates the kinase core in a position to phosphorylate the NSD with prescribed priority; (3) Dbf4 displaces the NSD motif from its binding sites on Mcm4 to facilitate NSD phosphorylation by DDK; (4) the binding of Dbf4 to Cdc7 activates its kinase activity; and (5) Dbf4 regulates the accessibility of the kinase center to ensure an orderly targeting of its substrates. Furthermore, Dbf4 exerts another level of regulation for Cdc7 as a target of Rad53 to inhibit DDK activity upon DNA damage. Although the sequence homology of Dbf4 between yeast and human is limited, their structures appear very similar. Given the fact that the MCM proteins are highly conserved, we believe that human DDK would act on the human DH in a manner similar to its yeast counterpart for helicase activation. As DDK is an obvious target for cancer therapy, information about the unique features of Dbf4 in regulating Cdc7 activities provides the necessary details for developing novel anti-cancer drugs that selectively kill cancer cells with low toxic off-target side effects.

Methods
Yeast strains. All yeast strains used in this study were of the W303 (leu2-3, -112 ura3-1 his3-11 trp1-1 ade2-1 can1-100) genetic background. Mutagenesis was performed using PCR-based methods 51 . Proteins were purified from yeast strains containing integrated expression constructs (see Supplementary Table 3 for details). One-step PCR-based approach with pTF272 (pFA6a-TEV-6xGly-3xFlag-HphMX, Addgene) as DNA template was used to generate MCM7-TEV-3xFLAG or MCM2-TEV-3xFLAG tagging modification in the W303-1a background strain. Strain ySDK8 (a gift from J. Diffley) was used for DDK purification. Strains used for purification of DDK mutants were constructed by transforming ySDK8 with linearized expression vectors using standard genetic techniques (Supplementary Table 3). Strain used for purifying the mutant DHs was generated using one-step PCR based opproach with pYM-N7 (pFA6a-natNT-tADH1) as DNA template for N-terminal truncation of MCM4 gene at its genomic locus.
DH-DDK complex assembly. For each round of DH purification, 30 liters of logphase G1 yeast cells were collected and cells were resuspended in 0.5 pellet volume of Benzonase Buffer (50 mM HEPES, pH = 7.6, 100 mM K-Glu, 1 mM EDTA, 2 mM NaF, 1 mM NaVO 3 , 8 mM MgCl 2 , 3 mM ATP, 1x protease inhibitor (Roche)). The suspensions were frozen drop-wise in liquid nitrogen before being processed in a Freezer/Mill ® High Capacity Cryogenic Grinder with 25 cycles at a crushing rate of 15. The resulting powders were thawed and resuspended in 3 volumes of Benzonase Buffer. The crude chromatin was then recovered and resuspended in Benzonase buffer for Benzonase (1U μl −1 ; Merck Biosciences 71206-3) treatment for 10 min at 37°C to solubilize chromatin fractions. The suspension was then centrifuged for 20 min at 25,000 g. The clear phase was recovered, and subjected to anti-Flag immunoprecipitation with 1 ml bed volume of washed anti-Flag M2 agarose (Sigma, A2220) at 4°C for 2 h. Beads were recovered, and washed extensively with benzonase buffer. For elution, we added 0.5 mg/ml (final concentration) 3×Flag peptide and incubated the beads at 4°C for 25 min. The beads were washed once with an equal volume of benzonase buffer. Eluates were combined and concentrated using Amicon centrifugal filter concentrators (30-kDa molecular weight cutoff; Millipore). The concentrated complexes were then applied to the top of a 20-40% glycerol gradient in benzonase buffer with protease inhibitors. The gradients were centrifuged in a TLS-55 rotor (Beckman Optima TLX ultracentrifuge) at 105,000 g for 13 h. The fractions containing the DH were pooled and aliquoted for storage.
Full-length Dbf4-Cdc7 kinase complex was expressed and purified as described previously 10  Electron microscopy. For negative staining, the samples were stained with 2% uranyl acetate and examined using an FEI Tecnai T20 electron microscope operated at 120 kV to determine sample quality and estimate relative concentration of samples used for cryo-grid preparation (about 10 times higher).
For cryo-grid preparation, aliquots (4 μL) of samples were applied to glowdischarged C-flat or Quantifoil Au grids (R1.2/1.3, 400 mesh) and blotted using an FEI Vitrobot IV. Cryo grids were screened using an FEI Talos Arctica operated at 200 kV. High-quality grids were recovered and transferred to a Titan Krios microscope (FEI) operated at 300 kV for data collection. For DH-DDK samples without Grafix, images were collected using a GIF K2 camera (Gatan) with SerialEM 3.6.11 52 in the super-resolution counting and movie mode, at a nominal magnification of 130,000× and defocus range of −1.0 to −3.0 μm. A total of 32 frames were collected for each micrograph, with dose rate at 10.2 e − Å −2 s − and a total exposure time of 6.4 s. For the crosslinked DH-DDK and the two mutant DH-DDK samples, images were collected using a GIF K3 camera (Gatan) in the superresolution counting and movie mode, at a nominal magnification of 81,000× and defocus range of −1.0 to −2.5 μm. A total of 40 frames were collected for each micrograph, at a dose rate of 7.0 e − Å −2 s − and a total exposure time of 6.5 s.
Image processing. For the DH-DDK (ATP-γ-S) samples without cross-linking, two datasets were collected (4253 and 3644 movie stacks) and processed in similar procedures ( Supplementary Fig. 1a, b). Drift-correction and electron doseweighting were applied to the movie stacks using MotionCor2 53 , which generates summed images with or without dose weighting. Images without dose weighting were used to evaluate the parameters of contrast transfer function (CTF) by CTFFIND4 v4.1.13 54 . Summed images and CTF power spectra were screened manually to select high-quality images for further processing. For each dataset, after 2D and global 3D classifications using RELION3.1 55 , high-quality particles were selected and subjected to one round of 3D refinement with C2 symmetry imposed. The particles were then duplicated using the command "relion_particle_symmetry_expand" in RELION3.1 and a mask-based 3D classification (on DDK region) with "--skip alignment" option (also with particles recentered at DDK region) was performed. According to DDK occupancy in the resulting 3D classification maps, particles with and without DDK bound were estimated to be 7.9% and 92.1%, respectively ( Supplementary Fig. 1a). Particles without DDK bound (empty DH) were pooled and re-centered at Mcm4-NSD region, and subjected to another round of focused 3D classification. Those classes with Mcm4-NSD stably anchored to Mcm4-NTD from both datasets of the DH-DDK (ATP-γ-S) samples were combined and refined to generate a global density map for the DDK-free DH at an overall resolution of 3.2 Å (Supplementary Fig. 1a, b).
The dataset of the DH-DDK (ATP) samples (3055 movie stacks) was similarly processed (Supplementary Fig. 1c). From the DDK-free DH particles (90.3% of all particles), focused classification on Mcm4-NSD region did not reveal any class with structured Mcm4-NSD.
For the mutant DH-DDK (ATP-γ-S) samples, 3938 and 4027 movie stacks were collected for the DHΔN140-DDK and DHΔN174-DDK datasets respectively. They were also similarly processed as the WT DH-DDK datasets ( Supplementary Fig. 1d, e). After the first round of 3D classification, DDK-bound particles were selected for a second round of focused 3D classification to improve the occupancy and conformational homogeneity of DDK region, and the final maps for the DH-DDK complexes were refined to 3.8 Å and 3.9 Å for the two datasets respectively. The DDKfree particles from the DHΔN140-DDK dataset were subjected to another round of focused 3D classificastion on the Mcm4-NSD region. A class of particles containing density of truncated NSD was clearly separated, which was refined to a global resolution of 3.8 Å (Supplementary Fig. 1d, f).
For the crosslinked DH-DDK (ATP-γ-S) samples, a total of 5184 qualified movie stacks were selected for image processing ( Supplementary Fig. 2). Around 1461 K particles were auto-picked, and screened by a cascade of 2D and global 3D classifications, resulting in a dataset of 679 K particles for further processing. Similar as non-crosslinked samples, particles were first refined with C2 symmetry imposed. Next, after C2 symmetry expansion, particles were re-centered at DDK region and subjected to one round of focused classification on DDK region. Among the five resulting classes, two of them display stable binding of both Dbf4 and Cdc7. These two classes (208 K particles) were combined for high-resolution refinement using dose-weighted particles. After CTF refinement and Bayesian polishing, the final density map for a stable DH-DDK complex (State I) was resolved at an overall resolution of 2.9 Å (Supplementary Fig. 2d, f). Further refinement of the kinase core region of DDK using density substraction improved the resolution of the kinase core region to 3.8 Å (Supplementary Fig. 2d, f). To investigate other possible conformations of DDK on the DH, two classes from the first round focused classification were combined (both only display stable binding for the N-terminal half of Dbf4) for another round of focused classification, from which three representative conformations (State II to IV) for DDK were identified.
For ease of data presentation, we also divided the particles of the crosslinked DH-DDK into six groups according to the results from the first round of focused 3D classification (Supplementary Figs. 2d and 3). The grouping standards were based on the occupancy of the two symmetrical DDK-binding sites and the general stability of DDK (Stable DDK vs Stable Dbf4-NTD). Groups I, II, and III represent the DHs with both sites occupied; Groups IV and V represent the DHs with one DDK site occupied, and Group VI represents the DHs free of DDK binding. These groups of particles were refined to resolutions of 3.3-4.0 Å, representing major compositional and conformational populations of DH-DDK in the reaction.
The resolution estimation was based on gold-standard FSC at the cutoff of 0.143. The local resolution maps were generated using ResMap 56 . Structural analysis was done with UCSF Chimera 1.11.2 57 and Pymol (http://pymol.org).
Model building. Modeling was performed using the DH-DDK maps from crosslinked datasets. The initial templates used for modeling were from a 2.5 Åresolution cryo-EM structure of the yeast DH (PDB: 7W8G) generated with the free DH particles from the non-crosslinking DH-DDK sample treated with ATP-γ-S (Supplementary Table 1), crystal structures of human DDK (PDB: 6YA7) and yeast Dbf4 BRCT domain (PDB: 5T2F). Each domain of the DH-DDK subunits was manually fitted into the density map in Chimera, followed by manual rebuilding using Coot 0.9.5 58 . Secondary structure prediction of Cdc7 and Dbf4 subunits was performed using PSIPRED 59 . The nucleotide moieties in the nucleotide-binding pockets of MCM subunits were modeled as ATP-γ-S or ADP. The models were refined against the 2.9-Å density map and 3.8-Å local density map of DDK region using Phenix.real_space_refine 60 with secondary structure restraints and geometry restraints imposed. The final models were evaluated by MolProbity 1.19.2 61 , and statistics are presented in Supplementary Table 1.
Kinase assays. For in vitro kinase assay, both WT and mutant DDKs and DH were incubated on ice in kinase buffer (50 mM HEPES pH7.5, 100 mM K-Glutamate, 1 mM EDTA, 8 mM MgCl 2 , 3 mM ATP, 4 mM NaF, 2 mM NaVO 3 ), and 1x protease inhibitor. Reactions were terminated at indicated time points by adding 2x Laemmli sample buffer and then boiled immediately at 100°C for 5 min followed by SDS-PAGE (6%) and western blotting analysis. DDK autophosphorylation was performed with both WT and mutant DDK preparations in kinase buffer at 30°C for 30 min, and the protein samples were analyzed by SDS-PAGE (7.5%) and visualized by Coomassie blue staining.
Quantification and statistical analysis. The orientation distribution of the particles used in the final reconstruction was calculated using RELION 3.1 55 . The local resolution map was calculated using ResMap v1.1.4 56 .
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.