A conserved mechanism for regulating replisome disassembly in eukaryotes

1The eukaryotic replisome is assembled around the CMG helicase at replication origins during replication initiation. Once assembled, CMG remains stably associated with replication forks until two forks emanating from adjacent origins converge, or a single fork encounters the end of a linear chromosome or a template discontinuity, at which point replication terminates (Fig. 1a). Upon termination, the replisome is disassembled in two steps: first, CMG is ubiquitylated on its Mcm7 subunit by a cullin-RING E3 ubiquitin ligase (SCFDia2 in budding yeast, CUL2LRR1 in metazoa); second, ubiquitylated Mcm7 is unfolded by the Cdc48 ATPase (also known as p97 in higher eukaryotes), leading to disassembly of the replisome–,–. As there is no known mechanism for origin-independent CMG assembly in S phase, premature disassembly of CMG must be avoided, to prevent replication fork collapse and genome instability. CMG translocates on the leading-strand template while excluding the lagging-strand template from its central channel. It has been suggested that this ‘excluded’ DNA strand, which is lost upon termination (Fig. 1a), inhibits ubiquitylation of CMG at replication forks–. However, because there are currently no structures of terminated replisomes in complex with SCFDia2 or CUL2LRR1, how ubiquitylation of CMG is regulated to restrict replisome disassembly to termination remains a key unanswered question.

2To determine the molecular basis for the regulation of CMG ubiquitylation, we aimed to solve the structure of a terminated replisome, by adapting our system for reconstituting budding yeast replisomes for structural analysis. After convergence of two replication forks, CMG translocates onto nascent double-stranded DNA (dsDNA) produced by the converging replisome,, (Fig. 1a). To trap a replisome bound around dsDNA, we used a DNA substrate that lacked a 5′ flap and contained a short stretch methylphosphonate modifications embedded in dsDNA, which slow translocation of CMG (Extended Data Fig. 1a). This DNA substrate was incubated with CMG, the replisome factors Tof1–Csm3, Mrc1 and Ctf4, SCFDia2 (Hrt1–Cdc53–Skp1–Dia2), an E2–ubiquitinconjugate (Cdc34–Ub), and the leading-strand DNA polymerase Pol-ε, in the presence of ATP (Extended Data Fig. 1a). After glycerol gradient sedimentation, complexes containing all replisome and SCFDia2 subunits were isolated (Extended Data Fig. 1b). Cdc34–Ub did not associate with the complex, perhaps reflecting the absence of neddylation on the Cdc53 cullin subunit of SCFDia2 (refs.,).

3After gradient fixation, samples were prepared for cryo-electron microscopy (cryo-EM), yielding three-dimensional (3D) reconstructions at average resolutions of 3.2–4.0 Å (Fourier shell correlation (FSC) = 0.143 criterion; Extended Data Fig. 1c–h, Extended Data Table 1). DNA binding was heterogenous across the dataset, with the majority of particles still engaging single-stranded DNA (Extended Data Fig. 2). Nonetheless, we identified a subset of particles, which was subsequently subclassified into two conformations (conformations I and II), that had unambiguously translocated onto dsDNA, representative of bona fide termination intermediates produced after fork convergence (Fig. 1b, Extended Data Figs. 2, 3a–e). While the configuration of the MCM C-tier differed between conformations I and II (Extended Data Fig. 3), in both cases the incoming dsDNA was bent by approximately 46° between the MCM N-tier and C-tier, necessitating distortion of the DNA duplex within the N-tier (Extended Data Fig. 3f). For conformation I, the nucleotide occupancy and interactions with the phosphate backbone of the leading-strand template are similar to replication fork-bound CMG (Fig. 1c, Extended Data Fig. 3b, g), suggesting a shared mechanism for translocation of CMG over single-stranded DNA and dsDNA.

4Having identified particles that had translocated onto dsDNA, we were able to build an atomic model of a terminated replisome (Fig. 1d). The overall architecture of CMG, Ctf4, Tof1–Csm3 and the non-catalytic module of Pol-ε (Pol-εnon-Cat) was almost indistinguishable from previous structures,– (for details of the structure of Pol-ε, see Extended Data Fig. 4). We observed an additional, large region of density at the N-tier face of CMG beside Mcm3 and Mcm7, which closely approaches Csm3 and the dsDNA ahead of CMG, before extending away from the core of the complex, forming an elongated arm characteristic of the cullin subunit (Cdc53) of SCFDia2 (Fig. 1d, e). The resolution of the cullin arm is relatively poor (precluding model building for Cdc53–Hrt1), due to a large degree of flexibility in this region, as highlighted by comparison of 3D classes (Extended Data Fig. 5a). We predict that this flexibility is important for conjugating the long K48-linked polyubiquitin chains required for Cdc48-dependent replisome disassembly. Regardless, the orientation of SCFDia2 can be unambiguously defined, placing the Cdc53 C terminus and Hrt1 ~45–70 Å from the primary ubiquitylation site on Mcm7 (Lys29), (Extended Data Fig. 5a, b), consistent with previous structures of un-neddylated cullin-RING E3 ligases.

5Density corresponding to the E3 ligase substrate-recognition module (Skp1–Dia2) is adjacent to the N-tier face of CMG (Fig. 1d, e). The N-terminal tetratricopeptide repeat domain of Dia2, which binds Ctf4 and Mrc1 (refs.,,), was not visible in our structure. However, clearsecondary structure and side chain density enabled us to build a de novo atomic model for the remainder of Dia2, encompassing the F-box (residues 211–247), 15 tandem leucine-rich repeats (LRRs) (248–716) and a C-terminal tail (717–732), which folds back onto the concave surface of the horseshoe-shaped LRRs (Extended Data Fig. 5c–k). The C-terminal end of the LRR domain forms an extensive interface with the N-tier of the Mcm3, Mcm5 and Mcm7 subunits of CMG (Fig. 1d, Extended Data Fig. 5l–n; see text below for a detailed description), demonstrating that Dia2 binds directly to CMG bound around dsDNA, equivalent to the situation after convergence of two replication forks.

6When DNA replication terminates at the end of linear chromosomes, CMG is thought to dissociate from DNA, at which point the loss of the excluded strand triggers CMG ubiquitylation, (Fig. 1a). To establish how SCFDia2 engages the replisome following termination at chromosome ends, we repeated cryo-EM sample preparation as described above, except in the absence of DNA. This yielded a 3D reconstruction of an ‘off DNA’ replisome at 3.9 Å resolution (Fig. 1f, Extended Data Fig. 6). Notably, binding of the Dia2 LRRs across Mcm3, Mcm5 and Mcm7 is indistinguishable from complexes bound around dsDNA (Fig. 1g, Extended Data Fig. 3e). Furthermore, comparison of our dsDNA-bound and off-DNA complexes with a previous structure of a replication fork-associated replisome revealed no conformational changes in the region of the MCM N-tier to which Dia2 binds (Fig. 1g, Extended Data Fig. 3e). Therefore, we conclude that termination does not induce conformational changes in CMG that are important for the regulation of CMG ubiquitylation by SCFDia2 (ref. ), either following fork convergence or when CMG dissociates from DNA.

7The extensive interface between Dia2LRR and MCM is predominantly formed by the Mcm3 N-tier (helices α1 and α5 and the zinc-finger (ZnF) domain), which forms a cradle for the C terminus of Dia2LRR (Fig. 2a, Extended Data Fig. 5l–n). In addition, the N terminus of Mcm7 wraps around the ZnF domain of Mcm3 and becomes sandwiched between Mcm3 and Dia2, while the ZnF domain of Mcm5 interacts with the C-terminal end of Dia2LRR, at the periphery of the Dia2LRR–MCM interface. The details of the residues involved are illustrated in Extended Data Fig. 5n.

8To examine the significance of the Dia2LRR–MCM interaction for CMG ubiquitylation and replisome disassembly, we generated a series of point mutants targeting the Dia2LRR–MCM interface, in both Dia2 (Fig. 2a) and MCM. The majority of MCM mutants exhibited defects in the formation of the Mcm2-7–Cdt1 complex or in MCM loading (data not shown), probably because the Dia2 LRR binding site is positioned at the inter-hexamer interface in the MCM double hexamer. While this precluded analyses of Mcm7 ubiquitylation after convergence of two replication forks in vitro, we were able to purify a CMG complex containing mutations in Mcm3 and Mcm5, which, while being proficient for DNA replication, was defective for ubiquitylation of Mcm7 (Extended Data Fig. 7a–d). Dia2LRR mutants formed stable tetrameric SCFDia2 complexes and supported ubiquitylation of Ctf4 (Extended Data Fig. 7e–g). Importantly, with the exception of Dia2-3A, the Dia2LRR mutants were defective for ubiquitylation of Mcm7, both after replication fork convergence (Fig. 2b) and off DNA (Extended Data Fig. 7h), with Dia2-13A showing the most penetrant defect. Haploid yeast cells with the dia2-13A allele accumulated CMG in the G1 phase of the cell cycle (Fig. 2c, Extended Data Fig. 7i), reflecting a failure to disassemble CMG during replication termination in the S phase of the previous cell cycle. Furthermore, these cells exhibited a profound growth defect at 20 °C, indistinguishable from cells lacking Dia2 (ref. ) (Fig. 2d). Together, these data demonstrate that the Dia2LRR–MCM interface that we describe is essential for CMG ubiquitylation and replisome disassembly, both after fork convergence and when CMG dissociates from DNA.

9Ubiquitylation of CMG in metazoa is driven by CUL2LRR1 (LRR1–CUL2–ELOB–ELOC–RBX1)–. Although LRR1 displays no apparent sequence homology to Dia2, metazoan CUL2LRR1 ubiquitylates CMG on its MCM7 subunit– and is suppressed by the excluded DNA strand,, suggesting there might be common features of replisome association that are important for the regulation of both SCFDia2 and CUL2LRR1. To investigate this, we used our approach for human replisome assembly and a DNA substrate lacking a 5′ flap, to determine a high-resolution structure of CUL2LRR1 in the human replisome (Fig. 3a, b, Extended Data Figs. 8, 9).

10The overall architecture of human CMG, AND-1, TIMELESS–TIPIN and Pol-ε are indistinguishable from our previous structure lacking CUL2LRR1 (ref. ) (Fig. 3a, b, Extended Data Fig. 10a). LRR1 is positioned across the MCM N-tier, in close proximity to the parental DNA duplex. In addition, an elongated arm of lower-resolution density, into which thecrystal structure of ELOB–ELOC–CUL2–RBX1 could be unambiguously docked, projects from the MCM N-tier in an analogous manner to yeast Cdc53–Hrt1 (Fig. 3b). Although metazoan CUL2 and yeast Cdc53 are tethered to their respective substrate adaptors (ELOB–ELOC–LRR1 for CUL2, Skp1–Dia2 for Cdc53) via very different interactions, the cullin C terminus and RING-box protein are similarly located in both cases, ~45–70 Å from the primary ubiquitylation sites in Mcm7 (refs.,,) (Figs. 1d, 3b). Furthermore, like Cdc53, CUL2 displays considerable conformational variability, which is probably important for the conjugation of long polyubiquitin chains onto MCM7 (refs.,) (Extended Data Figs. 5a, 10b).

11The majority of LRR1 was well resolved in our cryo-EM map (Extended Data Fig. 9e, k–n), which enabled de novo modelling of an N-terminal pleckstrin homology domain and a C-terminal LRR domain, which are connected by a flexible linker that stretches perpendicularly across the parental dsDNA (Fig. 3c, d). The pleckstrin homology domain interacts with the ZnF domains of MCM2 and MCM6, parental dsDNA and the N-terminal region of the TIMELESS α-solenoid (Extended Data Fig. 10c, d), consistent with the reported role for TIMELESS–TIPIN in recruiting CUL2LRR1 to the replisome in Caenorhabditis elegans. The LRR domain comprises seven canonical and two irregular LRR motifs and forms a shallow arc, reaching from the parental dsDNA to the N-tier face of MCM3 and MCM5 (Fig. 3d, Extended Data Fig. 10e, f). The BC and CUL2 boxes, which link LRR1 to ELOB–ELOC–CUL2–RBX1, are situated between LRR repeats 8 and 9 (Fig. 3c, d, Extended Data Fig. 10e, f), and a two-stranded antiparallel β-sheet caps the LRR domain at its C-terminal end (Extended Data Fig. 10g). In addition, the C-terminal HMG box of AND-1 could be docked into a small region of density alongside ELOC and LRR1 (Extended data Fig. 10h, i), which was absent in 3D classes that lacked AND-1 (Extended Data Fig. 10j, k), indicating that AND-1 interacts with CUL2LRR1 in the human replisome.

12Remarkably, despite the very different architectures of the LRR1 and Dia2 LRR domains, they bind to the same region of the MCM N-tier, but do so via completely different modes of interaction. The LRR1 LRR domain interacts predominantly with the three-stranded antiparallel β-sheet of the ZnF domain of MCM3, which extends the shallow arc of the LRR1 β-sheet (Extended Data Fig. 10l). This interface is augmented on one side by interactions between the MCM7 N terminus and the tip of the ZnF domain of MCM3 and LRR1 repeats 8 and 9 (Extended Data Fig. 10l, m). On the other side, MCM3 residues 3–8 and 164–174 are significantly rearranged upon CUL2LRR1 binding, such that the N terminus of MCM3, now projecting between the ZnF domains of MCM3 and MCM5, stabilizes an interaction between MCM3 residues 164–174 and a loop and short helix preceding LRR1 repeat 9 (Extended Data Fig. 10n). Finally, charged residues immediately preceding the β-strands of LRR1 repeats 4–7 form multiple polar contacts with the tip of the ZnF domain of MCM5 (Extended Data Fig. 10m). Further details are illustrated in Extended Data Fig. 10m, n.

13Ubiquitylation of CMG by both SCFDia2 and CUL2LRR1 is suppressed by the excluded DNA strand at replication forks–; our discovery that Dia2 and LRR1 bind directly to a common site across the ZnF domains of MCM3 and MCM5 suggested that this region of MCM might be important for the regulation of ubiquitylation. In our recent structure of the human replisome bound to a replication fork, cryo-EM density that we attributed to the excluded strand was positioned in the channel between the ZnF domains of MCM3 and MCM5, consistent with previous structures of Drosophila and budding yeast CMG,,. To further validate our assignment of the excluded strand, we identified a subset of particles lacking CUL2LRR1 from our dataset of replisomes assembled without an excluded strand (Extended Data Figs. 8, 9g). In the resulting density map, the MCM N-tier was identical to our previous map of replication fork-associated CMG, apart from a single region of density, extending from the fork junction between the ZnF domains of MCM3 and MCM5 (Fig. 4a, Extended Data Fig. 10o). This density was present only in the complex associated with the replication fork, thus confirming that it is contributed by the excluded DNA strand.

14Crucially, Fig. 4b shows that the presence of the excluded strand between the ZnF domains of MCM3 and MCM5 sterically blocks the engagement of the Dia2 and LRR1 LRR domains with MCM. As the LRR–MCM interaction is essential for ubiquitylation of CMG and, in turn, replisome disassembly, the occlusion of this interface by the excluded strand provides an elegant and universal explanation for the regulation of replisome disassembly across yeasts and metazoa. Notably, the LRR domains of Dia2 and LRR1 are not demonstrably homologous in sequence or structure. Thus, we propose that the binding of Dia2LRR and LRR1LRR across the exit channel of the excluded strand reflects convergent evolution, probably indicative of a stringent evolutionary pressure to accurately regulate replisome disassembly, and thereby safeguard replication forks. This evolutionary constraint is not evident in parts of the replisome disassembly machinery that do not contribute to the regulation of CMG ubiquitylation. For example, the Dia2 tetratricopeptide repeat domain binds yeast Mrc1 and Ctf4, whereas the LRR1 pleckstrin homology domain binds human TIMELESS.

15On the basis of our results, we propose the model summarized in Fig. 4c. Ubiquitylation of the MCM double hexamer is blocked by the occlusion of the LRR binding site at the inter-hexamer interface (Extended Data Fig. 11a). This occlusion probably also suppresses ubiquitylation during the conversion of MCM double hexamers into pairs of active CMG helicases, before the lagging-strand template is excluded. Once bidirectional replication forks are established and elongation begins, the spooling of the excluded DNA strand between the ZnF domains of MCM3 and MCM5 sterically blocks LRR engagement on MCM. It is possible that the binding of proteins to the excluded strand may help to block LRR–MCM engagement. However, ubiquitylation of CMG is inhibited at reconstituted budding yeast replication forks in the absence of the lagging-strand machinery (Extended Data Fig. 11b, c), consistent with the excluded DNA alone being sufficient to suppress SCFDia2 during elongation. In principle, the binding of yeast Dia2 to Mrc1 and Ctf4, and human LRR1 to TIMELESS and AND-1, could still occur at replication forks, even when the LRR–MCM interaction is blocked by the excluded strand. Accordingly, Mrc1–Ctf4 can support SCFDia2-13A recruitment to reconstituted replisomes (Extended Data Fig. 11d, e). Critically, however, the essentiality of the LRR–MCM interaction for ubiquitylation of CMG will restrict replisome disassembly to termination, independent of the timing of E3 ligase recruitment, and irrespective of whether a replication fork terminates via fork convergence, or at a telomere.

16Finally, we note that if the excluded strand is ever mispositioned, for example, during replication fork stalling or reversal, replisome disassembly could be triggered, due to premature LRR–MCM engagement. As such, the regulatory mechanism that we describe here may have implications for the stability of the replication fork under conditions of replication stress.

17Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

18Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-021-04145-3.