The magnitude of the shift and the multitude of bands seen in SDS-PAGE (Figure 1b) indicated that the serine/threonine-rich Sld3 protein (Figure 1c) is multiply phosphorylated after checkpoint activation. We used purified Rad53 to phosphorylate a series of Sld3 fragments (Figure 1c) in vitro. Figure 1d (and Supp Figure 1b) shows that Rad53 predominately phosphorylates Sld3 in two C-terminal fragments. To compare this to phosphorylation in vivo, we expressed and purified a version of Sld3 in which the N- and C-terminal domains were tagged and could be separated by cleavage with TEV protease (Figure 1c top). Figure 1e shows that only the C-terminal domain of Sld3 exhibited a λ phosphatase-sensitive shift in SDS PAGE after HU treatment in vivo. To identify individual Rad53 sites in Sld3 in vitro we phosphorylated arrays of peptides corresponding to the entire Sld3 amino acid sequence attached to a cellulose membrane. Consistent with Figure 1d, most of the phosphorylated peptides occurred within the C-terminal domain of Sld3 (Supp. Figure 1c). Because of the extensive overlap in the peptides on the array (Supp. Figure 2a-d) most sites could be identified unambiguously. All 38 potential serine and threonine phosphorylation sites were mutated to alanine (Figure 1c and Supp. Table 2). Compared to the wild type protein, this allele of Sld3 (sld3-A) shifted only slightly in SDS-PAGE after HU treatment (Figure 1f). A variety of sld3 mutants containing subsets of the 38 sites mutated to alanine all show reduced phosphorylation shift indicating that many or most of the sites contribute to the full phosphorylation shift and Sld3 inhibition in vivo (Supp. Figure 2e,f). The residual shift in sld3-A may be due to additional sites missed in our analysis or may due to be cryptic sites only phosphorylated when the stronger sites in the wild type protein are absent. Yeast strains expressing sld3-A as the sole copy of Sld3 showed no sensitivity to HU or DNA damaging agents and did not exhibit synthetic growth defects with several conditional alleles of essential replication proteins (Supp. Figure 3) arguing that the Sld3-A protein is functional for DNA replication.

These Rad53 sites are primarily in the C-terminal portion of Sld3, where the essential CDK sites (Thr600, Ser622) are found (Figure 1c). Figure 2a shows that, whilst CDK phosphorylation of the C-terminus of Sld3 in vitro allows direct binding to Dpb11 but not to a Dpb11 truncation lacking the first BRCT repeat (ΔN); however, subsequent Rad53 phosphorylation of Sld3 inhibits interaction with Dpb11. Mutation of the strongest Rad53 sites in the C-terminus of Sld3 to aspartate residues (sld3-12D – Figure 1c, Supp. Table 2), to mimic constitutive phosphorylation, also blocks interaction with Dpb11 in vitro (Figure 2b) without blocking CDK phosphorylation (Supp. Figure 4) and is unable to support growth in vivo (Figure 2c). The CDK-dependent interaction between Sld3 and Dpb11 can be bypassed in vivo by direct covalent fusion of these proteins³. Figure 2c shows that fusion of the Sld3-12D mutant protein to Dpb11 restored its ability to support growth. This argues that Sld3 phosphorylation by Rad53 inhibits its ability to interact with Dpb11 in vitro and in vivo. Two additional Rad53 phosphorylation sites (Ser306, Ser310) that contribute to the Rad53 phosphorylation-dependent shift of Sld3 in SDS-PAGE (Supp. Fig. 2e) lie considerably further upstream of the Dpb11 interaction domain. Mutation of these two sites to aspartate in the sld3-12D background also produced a protein (Sld3-14D – Figure 1c) that could not interact with Dpb11 (Figure 2b) and could not support growth (Figure 2c). In contrast to sld3-12D, however, sld3-14D could not support growth after fusion to Dpb11 (Figure 2c). Previous work has shown that Sld3 also interacts with Cdc45 and the GINS subunit Psf1 in a two hybrid assay¹⁶. A mutant protein in which Ser306 and Ser310 were changed to aspartate (Sld3-2D) interacted with both Dpb11 and Psf1 in a two hybrid assay (Figure 2d). However, compared to wild type, Sld3-2D showed a reduced interaction with Cdc45. Consistent with this weakened interaction, over-expression of Cdc45 allowed the Sld3-14D-Dpb11 fusion to support growth (Figure 2e). These results indicate that Sld3 phosphorylation by Rad53 inhibits interactions with both Dpb11 and Cdc45. A mutant of Sld3 in which 34 Rad53 sites were converted to aspartate (sld3-34D – Figure 1c) did not support growth, even when fused to Dpb11 (Figure 2c) in the presence of over-expressed Cdc45 (data not shown). This suggests that additional functions of Sld3 may be inhibited in the fully phosphorylated molecule.

Since Sld3 is essential for replication initiation but not elongation¹⁷ and interactions with Cdc45 and Dpb11 are essential for its function, we hypothesised that phosphorylation of Sld3 by Rad53 may act to prevent late origin firing. Mutation of these sites to alanine might, therefore bypass the ability of Rad53 to block late origin firing. Figure 3a lanes 1-4 (bottom panel) shows, however, that the sld3-A allele by itself was not sufficient to allow accumulation of replication intermediates (RI) from the late origin ARS501 in HU. Since Dbf4 is also a substrate of Rad53 (Figure 1a), we next tested the possibility that Rad53 inhibits both Sld3 and Dbf4. We combined the sld3-A allele with the mcm5-bob1 mutant, which bypasses the requirement for Cdc7-Dbf4 in vivo¹⁸. Similar to sld3-A, the mcm5-bob1 mutant by itself did not allow firing of ARS501 in HU (Figure 3A, lanes 5-8). However, Figure 3a shows that the combination of sld3-A with mcm5-bob1 allowed ARS501 firing in HU (Figure 3a, lanes 9-12). The timing and extent of Rad53 activation as judged by its hyperphosphorylation was similar in these strains (Figure 3b), indicating that late origin firing in the double mutant is not due to a defect in checkpoint activation. Indeed, the extent of Rad53 phosphorylation appears greater in the double mutant, consistent with an increased number of origins being fired and, hence, an increased number of stalled replication forks (Ref ¹⁹ – see also Figure 3d and Supp Figure 6,7).

We used the peptide array approach described above to identify 19 Rad53 phosphorylation sites in Dbf4 (Supp Figure 5a and Supp Table 3). Mutation of all these sites to alanine (dbf4-19A) significantly reduced the shift of Dbf4 in HU (Figure 3c). Combination of dbf4-19A with sld3-A resulted in the activation of several late origins in HU as shown by the accumulation of replication intermediates both by 1D electrophoresis (Figure 3d and Supp. Figure 5b,c) and 2D electrophoresis (Figure 3e). Rad53 activation was similar in the single and double mutant strains (Figure 3d, lower panel).

Strains expressing dbf4-19A, however, were unable to grow at 37°C (Supp. Figure 6a). We employed two approaches to exclude that the deregulation of late origin firing was linked to this temperature sensitivity. Firstly, we identified four serine residues in Dbf4 which, when changed to alanine, generated a mutant (dbf4-4A) that allowed origin firing in HU when combined with sld3-A but was no longer temperature sensitive (Supp. Figure 6b). Secondly, we reasoned that if deregulation of late origin firing was linked to loss of Dbf4 function, then both temperature sensitivity and late origin firing should be recessive; however, if dbf4-19A contributes to late origin firing in HU because it is refractory to Rad53 inhibition, it should be dominant. We introduced the sld3-A and dbf4-19A alleles as second copies into cells containing wild type DBF4 and SLD3. This strain is no longer temperature sensitive (Supp. Figure 6b), and the late origin ARS501 fires in HU (Fig. 3f). Therefore, deregulation of late origin firing in HU is not due to a recessive defect in the Sld3/Dbf4 alanine mutants. This also argues that Rad53 must inhibit the entire cellular pools of Sld3 and Dbf4 to completely block origin firing, which is consistent with the fact that the entire population of both Sld3 and Dbf4 appears phosphorylated in HU (Figure 1b and Figure 3c). Taken together, these results show that Rad53 prevents origin firing redundantly by phosphorylating both Sld3 and Dbf4. As is the case for both rad53Δ and mrc1Δ cells^8,20,21, the order of origin firing in dbf4-19A, sld3-A appears unaffected since the late origins ARS501, ARS603 and ARS1412 fire after the early origins, ARS305, ARS607 and ARS315 in HU (Figure 3d and Supp. Figure 5b) suggesting that the temporal order of origin firing may remain intact.

In contrast to wild type cells, rad53 mutant cells traverse S phase quickly in the presence of MMS²² at least partly because of global derepression of origin firing¹¹. This ‘intra-S checkpoint’ is lost in the dbf4-4A, sld3-A double mutant because its S phase in MMS is at least as fast as that of the rad53Δ sml1Δ strain. Rad53 activation occurred normally in the dbf4-4A, sld3-A double mutant (Figure 4b). Therefore, origin firing appears to be globally derepressed in this double mutant. The dbf4-4A, sld3-A double mutant actually traverses S phase slightly faster than the rad53 mutant in MMS. Rad53 is also required to stabilise forks that have stalled^11-14,19. rad53 mutants do not resume DNA synthesis efficiently after an HU arrest; however, the dbf4-4A, sld3-A double mutant is capable of resuming replication with similar kinetics to the wild type strain (Supp Figure 7a). Consistent with this, the dbf4-4A, sld3-A double mutant retained high viability during this rapid S phase in the presence of MMS (Figure 4c) as well as in the presence of high concentrations of hydroxyurea (Supp. Figure 7b). Under both conditions the rad53 mutant lost viability. This provides additional evidence that the primary role of Rad53 in maintaining cell viability after DNA damage correlates with the protection of replication forks. The checkpoint-dependent block to late origin firing appears to make only a minor contribution to cell viability after genotoxic stress. It will be interesting to determine whether it contributes quantitatively to genome stability.

Checkpoint-dependent inhibition of origin firing is a conserved feature of eukaryotic DNA replication. Here we have shown that the checkpoint inhibits origin firing in budding yeast by interfering with both the CDK- and DDK-dependent activation of origins: by phosphorylation of Sld3, which prevents interaction with Cdc45 and Dpb11, and by phosphorylation of Dbf4. How DDK is inhibited by Dbf4 phosphorylation is presently unclear. Inhibition of late origin firing by Rad53 is analogous to the inhibition of replication during G1 phase (Figure 4d) when initiation is prevented by the inhibition of both CDK and DDK. CDK cannot be inhibited during S phase in budding yeast because it is required to prevent the re-licensing of early-firing origins⁶. In higher eukaryotes, the DNA replication checkpoint may inhibit CDK activity, for example through phosphorylation and degradation of the Cdc25 phosphatases²³, since there are CDK-independent pathways that also prevent re-licensing, such as geminin⁶ and PCNA-dependent Cdt1 degradation²⁴. However, Cdc25 degradation is unlikely to represent the sole mechanism by which origin firing is inhibited in mammalian cells^25,26. It is likely that the checkpoint in higher eukaryotes also targets replication initiation factors directly and Dbf4 has indeed been shown to be a checkpoint substrate in other organisms¹⁵. An orthologue of Sld3, Treslin/ticrr has recently been identified in metazoans^27-29 and it will be interesting to see if they are also targets of checkpoint kinases. With the identification of these key Rad53 substrates in yeast, together with recent work identifying the key CDK and DDK substrates, we now have a detailed understanding of how protein kinases regulate the initiation of DNA replication.

1