A new class of disordered elements controls DNA replication through initiator self-assembly

The initiation of DNA replication in metazoans occurs at thousands of chromosomal sites known as origins. At each origin, the Origin Recognition Complex (ORC), Cdc6, and Cdt1 co-assemble to load the Mcm2-7 replicative helicase onto chromatin. Current replication models envisage a linear arrangement of isolated origins functioning autonomously; the extent of inter-origin organization and communication is unknown. Here, we report that the replication initiation machinery of D. melanogaster unexpectedly undergoes liquid-liquid phase separation (LLPS) upon binding DNA in vitro. We find that ORC, Cdc6, and Cdt1 contain intrinsically disordered regions (IDRs) that drive LLPS and constitute a new class of phase separating elements. Initiator IDRs are shown to regulate multiple functions, including chromosome recruitment, initiator-specific co-assembly, and Mcm2-7 loading. These data help explain how CDK activity controls replication initiation and suggest that replication programs are subject to higher-order levels of inter-origin organization.


INTRODUCTION
The appropriate spatiotemporal regulation of DNA replication is essential to genetic integrity and cell proliferation. In eukaryotes, the initiation of DNA replication requires the coordinated action of three proteinaceous factors -the Origin Recognition Complex (ORC), Cdc6, and Cdt1which co-assemble on DNA origins in the G1 phase of the cell cycle to catalyze loading of the Mcm2-7 replicative helicase onto chromatin. Once activated during the transition to S phase, Mcm2-7 helps promote origin melting, replisome assembly, and translocation of the replication fork.
ORC consists of a heterohexameric complex comprising the subunits Orc1-6. Five ORC subunits (Orc1-5), as well as Cdc6 and the six subunits of Mcm2-7, possess an ATPases Associated with diverse cellular Activities (AAA+) domain; Orc6 and Cdt1 are the only non-AAA+ proteins used for Mcm2-7 loading. Despite sharing a high overall degree of conservation across eukaryotes, certain aspects of initiator subunit sequence and function have nonetheless diversified during evolution. For example, S. pombe Orc4 contains a unique domain not found in other ORCs that endows the protein with a preference for A/T rich regions of DNA (Chuang and Kelly, 1999;Lee et al., 2001). Conversely, S. cerevisiae ORC is able to recognize specific origin sequences (Bell and Stillman, 1992;Li et al., 2018) whereas origin specification for metazoan ORCs appears more contextual (Remus et al., 2004;Vashee et al., 2003). S. cerevisiae Cdt1 additionally possesses a catalytically-inactive dioxygenase domain at its N-terminus that is necessary for yeast viability and Mcm2-7 loading (Frigola et al., 2017;Takara and Bell, 2011); this fold is absent in S. pombe and metazoan Cdt1s.
Much of our current understanding of initiator mechanism derives from reconstitution studies using budding yeast initiation factors. These efforts have helped define an orchestrated set of ATP-dependent molecular exchanges between ORC, Cdc6, and Cdt1 that culminate in the loading of two copies of Mcm2-7 onto origin DNA in the form of a stable double-hexamer (Duzdevich et al., 2015;Evrin et al., 2009;Remus et al., 2009;Ticau et al., 2015). One aspect of initiation that has been difficult to probe in vitro, however, is the potential for interactions between initiation factors associated with different origins. Evidence for mesoscale coordination of origin activity derives from multiple sources. For example, genomic studies aimed at defining sites of ORC binding and their relation to replication origins have revealed evidence of origin clustering across chromosomes at nucleosome-free regions (Cayrou et al., 2011;Miotto et al., 2016;Vaughn et al., 1990). Cellular patterns of ORC localization are also strikingly non-uniform, often presenting as concentrated foci on chromatin whether it be in D. melanogaster follicle cells (Austin et al., 1999) or human cells in tissue culture (Lidonnici et al., 2004;Prasanth et al., 2010;Shen et al., 2010). Outside of initiation, there exists a well-characterized phenomenon of origin interference that has been taken as evidence of inter-origin communication (Cayrou et al., 2011); the clustering of co-replicating regions into so-called Replication Factories or replication domains (RD) similarly has been observed in both fixed and live cells (Cook, 1999;Xiang et al., 2018). Finally, the replication timing profile of origins, which fire asynchronously throughout Sphase, partition to specific chromosome territories (Cremer and Cremer, 2001;Gilbert et al., 2 2005;Pope et al., 2014). It is currently unclear whether the clustering patterns observed for ORC binding sites, origin communication, and the timing of replication domains is coincidence or dependent upon an as-yet-discovered set of physiological properties of replication factors.
Present-day views of mesoscale organization within cells typically invoke processes such as protein gradients or membrane compartmentalization. Recently, a rapidly expanding body of work has begun to recognize protein/protein and protein/RNA liquid-liquid phase separation (LLPS) as playing critical functions in generating membraneless pseudo-organelles and colocalized bodies (Boeynaems et al., 2018). Both nuclear and cytoplasmic liquid phase condensates (also known as biomolecular condensates) have been observed and implicated in a panoply of functions, including cellular signaling (Li et al., 2012;Su et al., 2016), centrosome assembly (Woodruff et al., 2017), and chromatin/heterochromatin assembly and maintenance (Larson et al., 2017;Strom et al., 2017) (reviewed in (Maeshima et al., 2016)). A functional role for biological condensates is only beginning to emerge; such entities may help sequester and/or co-localize certain factors to modulate biochemical output and response (Shin and Brangwynne, 2017). Intrinsically disordered amino acid regions (IDRs) are often found in proteins that phase separate and can underpin multivalent interactions that help drive phase separation (Li et al., 2012). Many phase-separating IDRs are additionally enriched for certain amino acids and are hence referred to as low-complexity domains (LCD) (Hennig et al., 2015;Kato et al., 2012).
Although multiple nuclear events are now being scrutinized through the lens of phase separation, how these processes integrate and communicate with other LLPS or dispersivestate pathways is currently unknown.
Here we report that the Drosophila melanogaster replication initiation factors ORC, Cdc6, and Cdt1 possess N-terminal IDRs that facilitate DNA-dependent liquid-liquid phase separation in vitro. Bioinformatic analyses reveal that these initiator IDRs possess a high-complexity sequence signature that is preserved in metazoan homologs, but that is distinct from other condensate-promoting IDRs in other cellular pathways. Biochemical studies show that these IDRs drive the selective co-assembly and enrichment of Drosophila initiation factors into liquid phases in the presence of DNA, while simultaneously excluding non-partner proteins that also phase separate. Although Drosophila Mcm2-7 does not appear to phase separate on its own, initiation factors can recruit the complex into condensates in a DNA-, ATP-, and ORC/Cdc6/ Cdt1-dependent manner. In addition, cellular studies establish that the Orc1 IDR is critical for its recruitment to mitotic chromosomes. Collectively, our observations not only reveal a new class of high-complexity sequences that undergo phase separation, but also provide a model for how these elements promote physical interactions between replication initiation factors that could help promote long-range chromosomal communication observed in replication programs.

RESULTS
D. melanogaster Cdt1 undergoes DNA-dependent liquid-liquid phase separation (LLPS). The architecture of D. melanogaster Cdt1 (DmCdt1) is markedly different from its budding yeast counterpart (Fig 1A). Sequence analysis reveals that the N-terminal sequence of DmCdt1 is predicted to be an extended intrinsically disordered region (IDR) (Fig 1A). Although the sequence of this region is not conserved per se, an N-terminal IDR is present in other metazoan Cdt1s (Supplemental Table 1).
The N-terminal IDR of metazoan Cdt1 contains multiple conserved short linear motifs (SLiMs) (Davey et al., 2012) necessary for Cdt1 regulation, such as a PCNA interacting peptide (PIP) box and kinase consensus sequences (Pozo and Cook, 2016). However, known SLiMs only account for a small fraction (<25%) of the total length of the metazoan Cdt1 N-terminal IDR, so we asked whether this domain might have more general functional significance. The IDR of DmCdt1 has a predicted pI that is relatively basic (pI = 10.2); this feature, combined with the presence of tandem WH domains in the protein, suggested that DmCdt1 might bind DNA. We found that agarose beads coupled with 60bp double-stranded DNA (dsDNA) were able to efficiently pull down DmCdt1, whereas no such interaction was observed for ScCdt1 (Fig 1B).
Quantitative analysis of dsDNA binding using a fluorescence anisotropy-based assay also revealed DNA binding by DmCdt1 (Kd = 83 ± 17 nM) but not ScCdt1 (Fig 1C). Mouse and S. pombe Cdt1 have likewise been shown to bind DNA in vitro (Houchens et al., 2008;Yanagi et al., 2002).
Electrophoretic mobility assays confirmed DNA binding by DmCdt1 (Kd ≈ 100 nM) but also showed that the complex is highly heterogenous (Fig 1D). Indeed, elevated concentrations of the protein (>190 nM) resulted in a well-shifted species, suggestive of increasingly higher-order assemblies; however, size-exclusion chromatography showed that purified DmCdt1 is monodisperse in solution (Supplemental Fig 1). Interestingly, in the course of conducting these studies, we found that upon mixing 20 µM DmCdt1 with stoichiometric amounts of a 60bp duplex oligo, the solution became visibly turbid, and that this turbidity was reversible by the addition of salt (400 mM KCl) (Fig 1E). Inspection of the turbid solution by differential interference contrast (DIC) microscopy unexpectedly revealed the presence of phase-separated droplets up to 4 µm in diameter (Fig 1F). These droplets were fully absent from the DNA alone sample and were significantly reduced in both number and size in the DmCdt1 alone sample.
Together, these data demonstrate not only that DmCdt1 binds DNA, but that DNA-binding in turn induces the protein to phase separate.

DNA-dependent phase separation of DmCdt1 requires the N-terminal IDR and occurs at
physiologic concentrations. The DNA-dependency of Cdt1 phase separation predicts that droplets should be enriched for both protein and nucleic acid. To test this idea, DmCdt1 was expressed and purified with an N-terminal enhanced Green Fluorescent Protein (eGFP) tag (eGFP-Cdt1) (Fig 2A). A Cy5-labeled duplex oligonucleotide (Cy5-dsDNA) was then mixed with an equimolar amount of protein (5:5 µM) and imaged by two-color fluorescence imaging ( Fig   2B). When eGFP-Cdt1 was mixed with Cy5-dsDNA, droplets up to 3 µm in diameter appeared, all containing both protein and nucleic acid. Samples with eGFP-Cdt1 alone did not show such droplets, nor were they observed for samples with Cy5-dsDNA only. These data demonstrate that phase separation by Cdt1 is driven, at least in part, by dsDNA-induced coacervation, and further show that fluorescently-labeled oligonucleotides can serve as a proxy for labeled protein when assessing DNA-dependent protein phase separation by fluorescence microscopy.
Given the known importance of protein IDRs in facilitating phase separation (Mitrea and Kriwacki, 2016), we hypothesized that the IDR of Cdt1 might likewise be critical for the condensates we observed. We therefore purified an N-terminal IDR deletion of Cdt1 (Cdt1 ΔIDR ) and determined whether, upon mixing with DNA, a condensed phase could form (Fig 2C). No droplets were observed when Cdt1 ΔIDR was mixed with Cy5-dsDNA, even at elevated levels (10 µM protein, 10 µM DNA). To confirm these data with unlabeled DNA, we devised a simple depletion assay (Fig 2D). In this approach, protein is incubated with or without DNA, after which the denser, phase-separated material is pelleted by centrifugation, and the degree to which protein is depleted from the supernatant is assessed by SDS-PAGE. In agreement with the microscopy data, no depletion of full-length Cdt1 was observed relative to the load control in the absence of DNA (Fig 2E), indicating that no phase separation occurred. Moreover, the fulllength Cdt1 signal was fully lost from the supernatant in the presence of DNA, indicating that a near complete partitioning of Cdt1 into the condensed (pelleted) phase took place. When the DmCdt1 IDR was removed, the protein was again fully retained in the supernatant, regardless of whether DNA was present or not. Notably, a construct containing only the Cdt1 N-terminal IDR residues (Cdt1 IDR ) bound DNA (Kd = 158 ± 32 nM) and underwent DNA-dependent phase separation on its own ( Supplemental Fig 2A-C), demonstrating that the Cdt1 IDR is necessary and sufficient for condensation and that it directly interacts with dsDNA. Budding yeast Cdt1, which neither has an IDR (Supplemental Table 1) nor binds DNA (Fig 1), did not exhibit phase separation behavior in vitro (Supplemental Fig 2D-E). However, all other metazoan Cdt1 homologs possess an N-terminal IDR; tests with human Cdt1 revealed that, similar to DmCdt1, it likewise can undergo DNA-induced liquid phase condensation (Supplemental Fig 2F-G).
Collectively, these data demonstrate the essential role of metazoan Cdt1 IDRs in facilitating DNA-promoted phase separation.
During active replication in the early D. melanogaster embryo (0-2 hr), Cdt1 concentration peaks at approximately 70 nM (Supplemental Fig 2H-J). Given the low concentration of endogenous Cdt1, it was unclear whether our biochemical experiments, which were conducted at micromolar protein concentrations, would accurately reflect Cdt1 behavior in vivo. To address this question, we used fluorescence microscopy to generate a phase diagram for DmCdt1, titrating protein and salt concentration, and assaying for the presence or absence of a condensed phase ( Fig   2F). Droplets were observed down to the lowest concentration of Cdt1 tested (62.5 nM) when salt concentrations were set at or below physiological levels ([KGlutamate] = 150 mM).
Increasing salt concentration to twice physiological levels (300 mM) shifted the concentration of protein necessary to see phase separation to 500 nM. When the concentration of potassium glutamate was increased to 600 mM no phase separation was observed for any concentration of Cdt1 tested. The ability of Cdt1 to phase separate at nanomolar concentrations was confirmed by the depletion assay, wherein stoichiometric mixtures of DmCdt1 and duplex DNA were prepared from 800 nM to 50 nM, and assessed for DNA-dependent depletion of protein in the supernatant (Fig 2G). DNA-coupled loss of Cdt1 signal was seen at all concentrations tested. Together, these data demonstrate that DmCdt1 undergoes DNA-dependent phase separation at physiological protein and salt concentrations.
ORC and Cdc6 also partition into DNA-dependent liquid phases. Given the role of the DmCdt1 N-terminal IDR in promoting phase separation, we next examined whether other replication initiation factors might possess similar disordered regions of analogous function. Using DISOPRED (Jones and Cozzetto, 2015), we calculated the percent of predicted disordered residues for each Drosophila protein required for initiating replication and ranked them according to low (0-10% predicted unstructured residues), moderate (10-30%), or high disorder (30-100%) (Fig 3A). In addition to Cdt1, Cdc6 and two subunits of ORC (Orc1 and Orc2) were found to possess a high level of disordered content as a proportion of their total polypeptide chain length. Conversely, the Mcm2-7 subunits, though possessing unstructured regions, contained less than 30% predicted disordered sequence overall. Similar patterns were seen for other metazoan replication initiation factors, but not S. cerevisiae proteins (Supplemental Table   1).
For proteins classified with a high fraction of disordered content (> 30%), we next assessed the location of the disordered regions relative to known protein domains (Supplemental Fig 3A). In all cases, the longest regions of uninterrupted disorder reside N-terminal to the bulk of the folded domain content of the chain (i.e., upstream of the Cdt1 WH domains and of the AAA+ and WH domains of Orc1, Orc2, and Cdc6). Cdc6 possesses the shortest N-terminal IDR (246 amino acids) and Orc1 the longest (362 residues). There also exist shorter regions of disorder that serve as linker sequences between tandem globular domains, such as between the AAA+ and WH domain of Cdc6 (26 amino acid IDR) or the two WH domains of Cdt1 (152 amino acid IDR).
The realization that Cdc6 and two subunits of ORC possess long N-terminal IDRs suggested that these factors, like Cdt1, might also undergo phase separation in a DNA-dependent fashion, and that this property might promote their functional integration within a single condensed phase. This idea was first tested by examining the ability of recombinant DmORC to phase separate using the depletion assay (Fig 3B, lanes 1-3). Similar to the behavior of DmCdt1 when it forms condensates, ORC was found to be depleted from the supernatant in the presence, but not in the absence, of DNA. The inclusion of ATP did not have an appreciable effect on phase separation by ORC (Fig 3B, lanes 4-5). Fluorescence microscopy with Cy5-dsDNA confirmed that the DNA-dependent depletion of ORC from the supernatant was due to phase separation (Fig 3C, no droplets are observed for DNA alone). Assaying ORC phase separation over a range of protein concentrations (50 nM to 800 nM) consistently revealed ORC depletion when DNA was present, but not when it was absent (Supplemental Fig 3B). Thus, like Cdt1, ORC is able to undergo LLPS at physiological concentrations when DNA is present. We next investigated whether ORC's ability to undergo phase separation could be targeted to the IDR of a specific subunit, or whether the IDRs of both Orc1 and Orc2 are required for this behavior. In terms of length, the Orc1 IDR, at 362 amino acids, is approximately 100 residues longer than the Orc2 IDR (Fig 3D). Comparisons of sequence composition between Orc1 and within 3% for each amino acid category. Interestingly, Orc1 and Orc2 IDR amino acids are near equally distributed across hydrophobic (36/35%), hydrophilic (30/29%), and charged (32/33%) classes; this preponderance of hydrophobic residues is somewhat unexpected given the predicted unstructured nature of these sequences. Glycine, which is grouped within the hydrophobic class, is often enriched in protein unstructured regions, yet for both Orc1 and Orc2, glycine content is lower than expected for such a region (Brune et al., 2018).
Although the Orc1 and Orc2 IDRs are highly similar in terms of amino acid types, they do show a marked difference in their isoelectric points (pI): the Orc1 IDR (pI = 10.1) is enriched for positively charged residues (20% positive and 12% negative), whereas the Orc2 IDR (pI = 6.0) is weakly enriched for negatively charged residues (17% negative and 16% positive).
Speculating that this divergence in net charge might be an important key for facilitating LLPS by ORC, we constructed two mutant DmORC complexes, one with the Orc1 IDR deleted (ORC 1ΔIDR ) and the other lacking the Orc2 IDR (ORC 2ΔIDR ). Phase separation for both constructs was then assessed by the depletion assay and fluorescence imaging with Cy5-dsDNA.
ORC 1ΔIDR showed no evidence for phase separation in either assay, whereas the construct lacking the Orc2 IDR exhibited wild-type behavior in both assays (Fig 3E-F). These findings demonstrate that Drosophila ORC phase separates using interactions that require the Nterminal IDR of Orc1 but not Orc2. Given the predicted lack of structure in the Orc1 IDR we were surprised to discover that this region is relatively well-conserved across the Drosophila genus (67% identity, Supplemental Fig 4) and that the composition, length, and pI of this region is conserved across the metazoan phyla (Supplemental Table 1). Although S. cerevisiae Orc1 possesses an N-terminal IDR, this region is shorter than observed in metazoa and also has an acidic pI (Supplemental Table 1). Consistent with this observation, yeast ORC does not phase separate (Supplemental Fig 3C).
We next asked whether Cdc6 can also undergo DNA-dependent LLPS. Similar to budding yeast Cdc6 (Feng et al., 2000), DmCdc6 was found to interact with DNA in an ATP-independent manner by dsDNA-coupled agarose bead pull-down assays (Supplemental Fig 3D). Upon testing whether Cdc6 phase separates in the presence of DNA using the depletion assay ( Fig   3G), we observed a reproducible DNA-dependent depletion of Cdc6 from the supernatant (Fig   3G lanes 1-3); however, in contrast to ORC, this depletion was inhibited by the presence of ATP (Fig 3G lanes 4-5). A recent report that ATP can function as a hydrotrope to enhance intracellular protein solubility may provide an explanation for its effect on Cdc6 phase separation (Patel et al., 2017). Repeating the assay at higher concentrations of DmCdc6 shows that unlike DmCdt1, which appeared to fully partition into the concentrated phase at higher concentrations (Fig 2E), Cdc6 was only partially depleted from the supernatant (Fig 3H). When examined over a range of protein concentrations, Cdc6 consistently showed phase separating behavior, although the protein never completely partitioned into the pellet (Supplemental Fig 3E).
Fluorescence microscopy with Cy5-dsDNA confirmed that Cdc6 forms condensates in these 8 conditions (Fig 3I), while tests with a Cdc6 mutant lacking the N-terminal IDR (Cdc6 ΔIDR ) ( Fig   3J) showed that this construct has no capacity to phase separate (Fig 3J-K). Collectively, these studies confirm that Drosophila Cdc6 can form DNA-dependent condensates in a manner dependent on its N-terminal IDR, although its ability to do so is less robust than observed for Cdt1 and ORC.
Initiator IDRs co-assemble and recruit MCM2-7. The independent ability of ORC, Cdc6, and Cdt1 to phase separate, as well as their reliance on the same structural feature for this activity (an N-terminal IDR), suggested that these regions might allow initiators to co-partition into a single condensed phase. To test this idea, condensates were prepared by mixing ORC with Cy5-dsDNA, after which RFP-tagged DmCdc6 (tRFP-Cdc6) and eGFP-Cdt1 were sequentially added. Both tRFP-Cdc6 and eGFP-Cdt1 showed marked enrichment within ORC droplets ( Fig   4A). Moreover, although phase separation by Cdc6 was partially inhibited by ATP, all three initiator components can co-associate in a liquid phase in the presence of ATP (Fig 4A). Thus, while ATP appears to impede phase separation by Cdc6 on its own, the presence of the other initiators overcomes this barrier and allows the full suite of helicase loading factors to co-localize into a highly concentrated protein/nucleic acid-rich phase.
Initiator IDRs have a distinct amino acid sequence signature compared to other phaseseparating factors such as human FUS and EWS, or C. elegans Laf1 (Supplemental Fig 5A-E). Generally speaking, initiator IDRs can be characterized as having: 1) a high degree of sequence complexity, and approximately equal representation of charged, hydrophilic, and hydrophobic residues, 2) a deficiency in aromatic amino acids and glycine, and 3) a net positive charge. This sequence pattern suggested that initiator IDRs might serve as specificity determinants for forming condensates. This hypothesis was tested by assessing the recruitment of eGFP-tagged human FUS (eGFP-FUS) into preformed ORC/Cy5-dsDNA droplets. As opposed to Cdt1 and Cdc6, which readily concentrate within ORC/DNA droplets (Fig 4A), no enrichment was observed for eGFP-HsFUS (Supplemental Fig 5F-I). Thus, initiator condensates show a degree of specificity that allows for exclusion of other proteins with noncongruent IDRs.
Given the liquid phase-forming properties of ORC, Cdc6, and Cdt1 -along with their enzymatically coordinated role in loading the Mcm2-7 replicative helicase onto DNA -we asked whether Drosophila Mcm2-7 (DmMcm2-7) might undergo LLPS as well. We first addressed this question using a modified form of the depletion assay where we assessed partitioning of FLAG-tagged Mcm2-7 to both the dilute and condensed phases by Western blotting (Fig 4B). No evidence for phase separation by Mcm2-7 alone was observed under these conditions either in the presence or absence of ATP (Fig 4C-D). Fluorescence microscopy with FLAG-Mcm2-7 and Cy5-dsDNA likewise showed no evidence of droplet formation (Fig 4E).
Although specific MCM subunits contain predicted disordered domains (Fig 3A), these regions are in all cases shorter than the N-terminal IDRs of ORC, Cdc6 and Cdt1, and have a different amino acid composition (Supplemental Fig 6). The IDRs of Mcm2 and Mcm3, for example, have an acidic pI (pI < 6), and although Mcm4 contains an N-terminal IDR with a basic pI (pI = 11.0), this region has fewer charged residues compared to the other initiator IDRs.
Mcm2-7 recruitment to and stable loading on DNA requires the formation of a pre-Replication Complex (pre-RC) containing ORC, Cdc6, and Cdt1 (Bowers et al., 2004;Evrin et al., 2013;Kang et al., 2014). We therefore asked whether DmMcm2-7 might be recruited into condensed phases in either the presence of the initiators alone, or also with ATP, which is required for loading. Depletion reactions were set up that contained Mcm2-7 and plasmid DNA, and supplemented with either ATP or Drosophila ORC, Cdc6, and Cdt1 ("OCC"), or with both ATP and the OCC (Fig 4F). No appreciable pelleting of Mcm2-7 alone was observed in either the presence or absence of ATP, or with just the OCC; however, when ATP was included with the OCC, an ~8-fold enrichment of Mcm2-7 was observed in the pelleted phase. These data demonstrate that although DmMcm2-7 does not phase separate on its own, it can be specifically recruited into condensates when initiation factors and ATP are present.
Because it was possible that some non-specific, but nevertheless ATP-dependent, feature of the condensates might adventitiously pull Mcm2-7 complexes into the pelleted phase, we next investigated whether ORC, Cdc6, and Cdt1 are all required for partitioning. DmMcm2-7 depletion reactions were repeated with both ATP and DNA, except that this time individual initiators were withheld (Fig 4G). Whereas partitioning of Mcm2-7 into the condensed phase is seen with the intact OCC, no enrichment of the helicase complex is observed when any one of the initiation proteins is omitted from the reaction (Fig 4G). Collectively, these data demonstrate that Mcm2-7 does not phase separate on its own, but that it can be selectively enriched in condensed phases containing all of the factors required for helicase loading (ORC, Cdc6, and Cdt1) and ATP.

The association of ORC with chromatin in vivo is not dependent on its ATPase function but does
require the Orc1 N-terminal IDR. The data presented thus far show that D. melanogaster ORC, utilizing the Orc1 N-terminal IDR, can partition with DNA into phase condensates regardless of whether ATP is present or not. However, previous work has demonstrated that the tight binding of DNA by yeast and metazoan ORC is dependent upon ATP binding to its Orc1 subunit (Bell and Stillman, 1992;Bleichert et al., 2018;Chesnokov et al., 2001;Giordano-Coltart et al., 2005).
This dependency likely reflects the ability of ORC to encircle DNA within its central channel (Bleichert et al., 2018;Li et al., 2018;Yuan et al., 2017). A discrete patch within the DmOrc1 IDR that is close to the protein's ATPase domain has been demonstrated as necessary but not sufficient for the ATP-dependent binding of DNA by ORC in vitro (Bleichert et al., 2018).
However, whether ATP-independent DNA interactions, such as those participating in LLPS, might be important for the recruitment of ORC to chromatin in cells has not been established.
To first assess the need for ATP-binding by ORC to promote chromatin association in vivo, we examined the dynamics of this interaction in live D. melanogaster embryos. A transgenic fly line expressing eGFP-tagged Orc1 under the control of its endogenous promoter was made, and the functional integrity of this transgene confirmed by its ability to rescue a null allele (Supplemental Table 2). In parallel, we constructed a mutant Orc1 transgene, eGFP-Orc1 WalkerAB , that contains mutations in both its Walker A (K604A) and Walker B (D684A/E685A) ATPase motifs; prior biochemical studies have shown that these changes interfere with ATP binding and hydrolysis, respectively, in both budding yeast and flies (Chesnokov et al., 2001;Klemm and Bell, 2001;Klemm et al., 1997). Genetic crosses revealed that eGFP-Orc1 WalkerAB was unable to rescue the null allele (Supplemental Table 2). In principle, the incapacitated function of the eGFP-Orc1 WalkerAB transgene could arise either from an inability to localize to chromatin, or from downstream replication initiation events that require ATP turnover. To test whether the Orc1 WalkerAB construct showed altered chromatin recruitment, we directly visualized the dynamics of eGFP-Orc1 chromatin association in embryos with RFP-tagged histone H2A by lattice light-sheet microscopy (Chen et al., 2014). Other aspects of metazoan ORC are also thought to play a role with how it associates with chromosomes. These include recruitment to histone H4K20me2 marks through an N-terminal Bromo-Adjacent Homology (BAH) domain found in Orc1 (Kuo et al., 2012;Noguchi et al., 2006), and the direct binding of DNA by the Orc6 TFIIB domains (Balasov et al., 2007;Liu et al., 2011).
To more rapidly sort through different Orc1 mutants and their role in chromatin association, we turned to D. melanogaster S2 tissue culture cells. A series of eGFP-labeled Orc1 constructs to be used for co-expression analysis were constructed, along with a mCherry-tagged variant of histone H2A to provide a benchmark for timing events of interest. To confirm that the S2 cell culture system recapitulates the behavior of ORC seen in embryos, the dynamics of wild-type Orc1 in mitotic cells were imaged first. Consistent with previous observations, Orc1 loaded onto chromosomes beginning in anaphase (2.5 ± 0.2-fold enrichment; Fig 5E,I), and mutations within the Orc1 Walker A and B motifs had no effect on the observed recruitment (2.3 ± 0.4-fold enrichment, Fig 5F,I). We next tested how deletion of either the Orc1 BAH domain (Orc1 ΔBAH ) or the Orc1 IDR (Orc1 ΔIDR ) altered Orc1 chromatin recruitment in S2 cells (the design of these constructs retained the nuclear localization signal present in the most N-terminal portion of the Orc1 IDR, which directly follows the BAH domain). The Orc1 ΔBAH construct showed wild-type like association with chromosomes (2.6 ± 0.7-fold enrichment; Fig 5G,I), demonstrating that the BAH domain is not required for chromatin association in vivo. By contrast, chromosome recruitment of Orc1 ΔIDR was fully abolished to background levels (1.0 ± 0.2-fold enrichment; Fig   5H-I), such that no increase in ORC signal on chromosomes is observed as cells progress through mitosis. These data demonstrate that ORC is highly enriched on mitotic chromosomes through an ATP-independent mechanism that uniquely requires the Orc1 N-terminal IDR.
Initiator phase separation is regulated by CDK-dependent phosphorylation. Our analysis of Orc1 dynamics in cells shows that the association of the complex with chromatin is regulated and does not occur until just prior to telophase (Fig 5). This result in turn suggests that a cell-cycledependent change in the status of either chromatin or ORC might be responsible for promoting DNA binding. Previous work has shown that the association of human and X. laevis ORC with chromatin is regulated by CDK-dependent phosphorylation (Findeisen et al., 1999;Lee et al., 2012;Li et al., 2004;Rowles et al., 1999). Interestingly, the major sites for CDK action map to the Orc1 IDR and the phosphorylation of these loci interferes with the binding of Drosophila ORC to DNA in vitro (Remus et al., 2005); the binding of D. melanogaster ORC to mitotic chromosomes also requires cessation of CDK activity (Baldinger and Gossen, 2009). Despite these observations, the mechanism by which the phosphorylation of ORC inhibits its ability to bind chromatin has remained unclear.
Since phase separation by replication initiation factors appears coupled to electrostatic interactions between their IDRs and DNA (Supplemental Fig 2), as indicated by the saltsensitivity of phase separation (Fig 2F) and the basic nature of the initiator IDRs (Supplemental Table 1), we hypothesized that initiator phosphorylation might inhibit the ability of these factors to form condensates. An analysis of putative CDK phosphorylation sites shows that Orc1 hosts sixteen potential CDK sites (Fig 6A), all but one of which localizes to the Orc1 IDR. Of these, seven IDR sites represent full CDK consensus sequences, [S/T]PX[K/R], while the others represent the minimal sequence, [S/T]P. Given the abundance of phosphorylation sites within the Orc1 IDR, we asked whether the Cdc6 and Cdt1 IDRs are also enriched with CDK consensus sequences (Fig 6A). Cdc6 and Cdt1 have seven and sixteen putative sites, respectively. For Cdc6, all sites reside within the N-terminal IDR and four of these represent full consensus sequences. For Cdt1, 13/16 sites are within the N-terminal IDR and nine represent the full consensus sequence. Thus, replication initiation factor IDRs, which constitute the regions responsible for phase separation, retain the vast majority of CDK phosphorylation sites.
To test the functional implications of initiator phosphorylation, we modified our depletion assay to include pre-treatment with CDK/Cyclin. Initiation factors were treated for 60 min with recombinant CDK1/CycA expressed and purified from insect cells. After treatment, DNA was added to induce phase separation and the depletion assay was completed as above, with phase separation assessed by protein depletion from the supernatant after centrifugation (Fig 6B).
Reactions were set up that lacked either CDK1/CycA or ATP to control for the independent effect of each reagent. DNA was again seen to robustly stimulate phase separation by ORC, and the addition of either CDK/Cyc or ATP alone had no effect on this behavior; however, treatment with CDK1/CycA and ATP fully inhibited ORC phase separation (Fig 6C). Analysis of the reactions by light microscopy corroborated the results of the depletion assay (Fig 6D), demonstrating that ORC phase separation is inhibited by CDK-dependent phosphorylation. We next asked whether Cdc6 and Cdt1 are similarly regulated. Although Cdc6 phase separation is inhibited in the presence of both CDK1/CycA and ATP (Fig 6E), it is likewise inhibited by ATP alone and thus a specific effect of phosphorylation could not be assessed. By comparison, phase separation by Cdt1 was fully inhibited by treatment with CDK1/CycA and ATP, but not by either CDK1/CycA or ATP alone (Fig 6F). These results demonstrate that CDK/Cyclin phosphorylation can directly inhibit liquid phase condensation by replication initiation factors.

DISCUSSION
We report here the discovery of a novel condensation property for metazoan replication initiator proteins that directly impacts their biochemical and cellular functions. These findings have substantial implications for the mechanisms of DNA replication initiation, and for understanding how replication is managed across a topologically complex chromatin substrate. More generally, our findings also expand the catalogue and types of IDRs that are known to drive protein phase separation and reinforce the notion that such elements can act as a 'sorting code' for distinguishing partner proteins from other factors in the cell.
Recruitment of ORC to chromosomes. Evolution has created multiple mechanisms for recruiting replication initiation factors to chromosomes (reviewed in (Bleichert et al., 2017)). The prevailing view is that recruitment occurs principally through the ability of the origin recognition complex (ORC) to bind origin DNA in an ATP-dependent manner that depends on DNA encirclement within the central channel of the complex (Bleichert et al., 2018;Li et al., 2018;Sun et al., 2013;Yuan et al., 2017). However, we demonstrate here that in the absence of ATP, D. melanogaster ORC can interact with DNA at physiological protein and salt concentrations in vitro (Fig 3), and that the Orc1 ATP-binding and hydrolysis motifs are dispensable for its recruitment to chromatin in vivo, both in the early embryo and in cell culture. (Fig 5). The N-terminal intrinsically disordered region (IDR) of DmOrc1 is shown to be the key element underpinning the ATPindependent association of ORC with chromosomes (Fig 3 and Fig 5). Corroborating these results is previous work that demonstrates a specific role for human and C. elegans Orc1 in facilitating ATP-independent ORC/DNA binding (Giordano-Coltart et al., 2005;Sonneville et al., 2012;Vashee et al., 2003) and a direct role for a short DNA-binding motif within the Orc1 IDR (Bleichert et al., 2018;Kawakami et al., 2015;Li et al., 2018). It remains to be determined whether the Orc1 IDR alone is sufficient for proper origin recognition (as opposed to general chromatin recruitment), or whether the action of this element works in concert with other DNAinteraction surfaces of ORC, such as the Orc1 BAH domain (Kuo et al., 2012) or the TFIIB domains of Orc6 (Liu et al., 2011).
The long-standing observation that DNA binding by ORC is enhanced by ATP (Bell and Stillman, 1992;Chesnokov et al., 2001;Vashee et al., 2003) would at first seem contradictory to the ATPindependent DNA association phenomena reported here. This apparent dichotomy can be reconciled by invoking a two-state model wherein (for metazoan ORC at least) the tight, ATPdependent encirclement of a short segment of DNA occurs after an initial set of weaker interactions takes place between an initiator IDR and a DNA segment or chromatin region (Fig   7, panels 1-2). The format of these dynamic interactions would promote multivalent contacts between short basic motifs within the Orc1 IDR and the DNA backbone, along with IDR-IDR interactions between ORC molecules. These associations in turn would promote the colocalization and enrichment of the complex along chromosomes as seen in vivo in a manner akin to 'wetting,' a known property of condensates. By contrast, during DNA encirclement, specialized and structurally defined DNA binding motifs, such as the initiator-specific motifs (ISMs) that line the ORC central channel, would rely on a conformational state stabilized by ATP binding to clasp around DNA, rendering the complex competent for catalyzing Mcm2-7 loading.
Consistent with this two-step model, we have found that our eGFP-tagged Orc1 transgene loses its ability to complement a null allele of the protein if it carries a ATP binding and hydrolysis mutation (Supplemental Table 2), even though it associates with chromatin in what appears to be a wild-type manner (Fig 5). Although the precise link between ORC's association with chromatin and the specification of a particular locus as a bona fide origin is unclear, it is tempting to speculate that some controlled aspect of DNA encirclement underpins this step.
The IDR-dependent recruitment and co-association of ORC molecules with chromatin is also intriguing in that it could account for mesoscale-levels of organization observed previously for replication, and thus explain non-uniform organization of both origins of replication and ORC binding sites across chromatin (Cayrou et al., 2011;Miotto et al., 2016;Petryk et al., 2016;Vaughn et al., 1990). In this regard, it is interesting to note that replication domains appear to substantially overlap with topologically associated domains (TADs) (Petryk et al., 2016;Pope et al., 2014), chromosomal regions that provide a basic level of chromatin organization and appear to be yet another biological entity with liquid-phase like properties (Nuebler et al., 2018).
Determining how ORC and other DNA-binding factors are distributed within these chromosomal compartments will be crucial for understanding how different replication zones are coordinated across a complex chromatin substrate.
Formation of the Pre-RC and helicase loading. An important property of initiator IDRs is that they allow for both self-self and partner protein interactions (Fig 4A). Thus, as ORC forms local interactions with itself to spread out along chromatin, this condensation is in turn expected to promote the formation of stable subunit-subunit interactions required for helicase loading (such as the proper docking of Cdc6 into the ORC ring, Fig 7). Importantly, this mechanism of coassociation is seen to predominate in vitro at physiological salt and protein concentrations.
Moreover, the selective partitioning of Mcm2-7 into initiator/DNA phases is seen only when all three initiation components (ORC, Cdc6, and Cdt1) and ATP are present. This stringent requirement for the intact OCC and nucleotide is consistent with what is known to be the minimal set of factors necessary to catalyze the loading of Mcm2-7 complexes onto DNA. Since the pre-RC catalyzes multiple rounds of Mcm2-7 deposition through the reiterative binding and release of individual initiation factors , the 'wetting' of a swath of DNA loci by a multitude of initiation factors would help keep those factors that cycle on and off of ORC at a relatively high concentration to promote successive Mcm2-7 loading cycles. Such behavior may be particularly useful in the early stages of embryogenesis and in pluripotent stem cells, where cycling times are short and helicase loading occurs rapidly (Matson et al., 2017).
Initiator IDRs may serve a role beyond simply increasing the efficiency of chromatin association and helicase loading. It has long been recognized that initiators are under stringent cellular control by CDKs (reviewed in (Parker et al., 2017)). It has also been established that a majority of CDK sites map to unstructured regions of Orc1, Cdc6, and Cdt1; however, how phosphorylation at these sites might regulate initiator function has remained unknown. Here, we demonstrate that the formation of initiator condensates is disrupted by CDK action (Fig 6). The inability of the phosphorylated initiator proteins to phase separate with DNA provides a ready explanation for how such post-translational modifications unique to metazoan ORC could lead to the dissociation of these factors from chromatin late in the cell cycle and then associate again either in anaphase or G1 when the Cdk levels are low (Findeisen et al., 1999;Lee et al., 2012;Remus et al., 2005) (Fig 7). In this regard, it is notable that in metazoans Cdt1 inactivation requires the further degradation of the protein through a PCNA-dependent ubiquitination pathway (Arias and Walter, 2006); why this extra level of stringency has evolved remains to be determined. How phosphorylation interferes with phase separation is unknown and a more sophisticated understanding of the biophysical interactions underpinning the condensation reaction will be required. Given the overall clustering of positively charged residues in the IDR's and the DNA dependence of the condensation, charge neutralization of the domain and/or charge repulsion seem likely.
Molecular mechanism of initiator phase separation. Phase transitions in biology are driven by multivalent interactions (Li et al., 2012). For protein LLPS, multivalency is achieved in two nonmutually exclusive ways: 1) by the linear and repetitive arrangement of folded interaction modules, and 2) through extended regions of intrinsic disorder that contain short linear interaction motifs (Boeynaems et al., 2018). The replication factors fall within the latter of the two classes.
A major family of IDRs that mediate phase separation are the so-called low-complexity domain (LCD) proteins, which include factors such as FUS, Laf1, Ddx4 and hnRNPA1. Studies investigating the biophysical mechanisms of phase separation in these and other proteins have suggested a major role for π-π and π-cation contacts (Vernon et al., 2018). Aromatic residues are major mediators of these type of interactions, and have demonstrated functionality in LCD condensates (Chong et al., 2018b;Jiang et al., 2015;Kato et al., 2012;Lin et al., 2017;Nott et al., 2015;Qamar et al., 2018;Wang et al., 2018). Other sequences are also able to participate in π-π/π-cation bonding pairs, such as repetitive RG/RGG motifs (Chong et al., 2018a;Elbaum-Garfinkle et al., 2015). Even for non-LCD IDRs, such as the Nephrin intracellular domain (NICD), aromatic residue-mediated π-π contacts are of primary importance in facilitating phase separation (Pak et al., 2016). Given these data, we were surprised to find that both aromatic residues (< 3% for Orc1, Cdc6, and Cdt1) and glycine (< 4% for Orc1, Cdc6, and Cdt1) are substantially underrepresented in initiator IDR sequences compared to other phase separating LCD proteins (Supplemental Fig 5E). Initiator IDRs are also devoid of simple repetitive motifs.
Outside of π-π/π-cation interactions, electrostatics can be another major driver of protein phase separation. This force is particularly relevant in systems that undergo complex coacervation with nucleic acids (Aumiller Jr. and Keating, 2016), such as seen with the tau protein and RNA (Wegmann et al., 2018;Zhang et al., 2017). The process of complex coacervation through the utilization of RNA as a counterion scaffold is mechanistically analogous to how we believe the replication initiators behave in the presence of DNA. Lines of evidence supporting this model include the demonstration that: 1) all D. melanogaster initiator IDRs have a pI > 9 (Supplemental Table 1), 2) initiator condensation is salt-sensitive (Fig 2) and each initiator has relatively high fraction charged residues (between 32-35%) (Supplemental Fig 5), and 3) although similar in overall amino acid composition, the Orc1 IDR (pI = 10.1) but not the Orc2 IDR (pI = 6.0) is necessary for ORC phase separation (Fig 3). We envision that initiator IDR interactions with DNA overcome interchain electrostatic repulsion between proximally positioned initiator IDRs, thereby promoting intermolecular interactions between the IDRs. Although at present we possess a minimal understanding of the biophysics of initiator LLPS, the relatively equal representation of hydrophobic, hydrophilic, and charged amino acids within initiator IDRs suggests that intermolecular IDR-IDR interactions may be of multiple types (Supplemental Fig   5).
Conservation of initiator IDRs across eukaryotes. The IDRs of D. melanogaster Orc1, Cdc6, and Cdt1 share a similar sequence composition but no identifiable sequence identity. This has led us to conclude that the general sequence features of these IDRs are sufficient to facilitate DNAdependent phase separation. It is notable, however, that the Orc1 IDR amino acid sequences are highly conserved across the Drosophila genus (representing over 40M years of evolution (Drosophila 12 Genomes et al., 2007)) (Supplemental Fig 4), suggesting that there may be presently unrecognized sequence-specific functional information.
All metazoan homologs of Orc1, Cdc6, and Cdt1 that we have analyzed contain a predicted Nterminal IDR longer than one-hundred amino acids (Supplemental Table 1, including human, mouse, D. rerio, X. laevis, and C. elegans). Interestingly, for all species Orc1 contains the longest IDR and Cdc6 the shortest. While suggestive, an IDR alone does not unequivocally determine whether a protein will phase separate. Further classification of eighteen homologous metazoan sequences demonstrate that they contain sequence characteristics similar to D. melanogaster initiator proteins: all possess a basic pI (calculated pIs range from 9.1 to 11.1), a low content of aromatic and glycine residues, and a relatively high fraction charged residues (ranging from 0.22 to 0.39). Based on our current limited understanding of initiator phase separation, these data suggest that the ability of initiators to undergo DNA-dependent LLPS is likely broadly conserved across metazoa. Consistent with this prediction, human Cdt1, like DmCdt1, forms condensates in a DNA-dependent manner (Supplemental Fig 2G). In instances where Mcm2-7 loading has been reported using purified metazoan proteins (e.g., human ), it is tempting to speculate that this is occurring in condensates.
In contrast to the metazoan initiators, our sequence analysis predicts that the S. cerevisiae initiation factors are likely to be incapable of DNA-dependent phase separation (Supplemental Table 1). Indeed, we found ScCdt1, which lacks an N-terminal IDR altogether, unable to either bind DNA or undergo phase separation (Supplemental Fig 2). Additionally, the IDRs of ScOrc1 and ScCdc6 are relatively short (143 and 31 amino acids, respectively) and have an acidic as opposed to a basic pI (pI = 4.7 and 6.2, respectively), and ScORC showed no evidence of phase separation (Supplemental Fig 3C). For fission yeast the predictions from our analysis are less reliable. S. pombe Cdt1 does not have an N-terminal IDR but homologs of Orc1 and Cdc6 each have an approximately one-hundred amino acid N-terminal IDR with a basic pI.
Intersection with emerging roles of cellular phase separation. The present work establishes for the first time that DNA replication, a cellular pathway vital to cell proliferation, is regulated by macromolecular phase separation. These observations in turn provide a provocative link between the replication machinery and their function in at least two other cellular structures recently predicted to be impacted by phase separation. One is heterochromatin: recent work has demonstrated that Heterochromatin Protein 1 (HP1), a major cellular organizer of heterochromatin, undergoes phase separation in vitro and in vivo (Larson et al., 2017;Strom et al., 2017). Interestingly, there is a conserved linkage between the formation of HP1-dependent heterochromatic domains and ORC binding, with the localization of HP1 and ORC to chromatin being interdependent (Huang et al., 1998;Pak et al., 1997;Prasanth et al., 2004Prasanth et al., , 2010. Notably, Orc1 targeting to heterochromatin relies on the Orc1 N-terminal IDR in both human and D. melanogaster model systems (Lidonnici et al., 2004;Pak et al., 1997). It is tempting to speculate that the formation of heterochromatic domains may involve a condensed phase that forms by the coordinated action of ORC and HP1 (and likely many other proteins), and that this activity may have the dual purpose of excluding transcriptional machineries to maintain a silenced state while enriching ORC in an otherwise inaccessible compartment that must nonetheless be replicated. We anticipate that any mechanism(s) underlying the formation and maintenance of heterochromatic domains may be developmentally regulated, since HP1 domains are clearly visible during interphase in the early embryo (Strom et al., 2017), whereas the association of ORC with chromatin appears restricted to mitosis during the same developmental stage (Fig 5). Beyond HP1, ORC phase separation may have a direct role in chromatin compaction, which could underlie reports that ORC plays a role in mitotic chromosome condensation (Pflumm and Botchan, 2001;Prasanth et al., 2004).
A second condensate connection with ORC may involve centrosomes. It has recently emerged that centrosomes form by the phase separation of scaffolding factors, known as the pericentriolar material (PCM). A major PCM scaffolding factor in C. elegans, SPD-5, undergoes spontaneous self-assembly into spherical condensates that selectively recruit centrosome client proteins necessary for the nucleation of microtubules (Woodruff et al., 2015(Woodruff et al., , 2017. Notably, multiple subunits of ORC, including both Orc1 and Cdc6, are targeted to centrosomes where they regulate centrosome copy number (Hemerly et al., 2009;Kim et al., 2015;Prasanth et al., 2004). We speculate that the sequence features which enable initiator phase separation on DNA (e.g. their N-terminal IDRs) may permit their selective partitioning and enrichment on centrosomes. From a mechanistic standpoint, it is interesting to note that the scaffolding protein SPD-5 and its homologs in other metazoans bear a high net negative charge (DmCentrosomin is 30% charged with pI = 5.8), suggesting that initiator recruitment to centrosomes may proceed through a mechanism similar to their condensation on chromatin, that is, by complex coacervation through charge-charge interactions with a poly-anionic scaffold (in this case, protein).
In conclusion, we have demonstrated that replication initiators contain a novel class of IDRs that facilitate phase separation and affect each stage of the initiator functional cycle. This work establishes a new paradigm for understanding metazoan initiation mechanisms and their regulation by CDK activity, and how initiation can be organized at the mesoscale. These findings in turn open up numerous new avenues of cross-disciplinary investigation into how the replication initiation machineries interface with other cellular pathways, such as those involved in chromatin maintenance and establishment.

MATERIALS AND METHODS
Cloning, expression, and purification of DmCdt1. The coding sequence for full-length D.
Fractions containing full-length Cdt1 were then pooled and, for full-length and Cdt1 ΔIDR , but not eGFP-Cdt1, the tag was cleaved with TEV protease at a 1:10 ratio of TEV:protein in an overnight incubation at 4°C. An additional nickel affinity step was then used to remove TEV, uncleaved protein, and the free tag from purified Cdt1. Finally, the purified sample was concentrated to 2 mL and loaded onto a HiPrep 16/60 Sephacryl S-300 HR column (GE Healthcare) equilibrated and run in Sizing Buffer (50 mM HEPES pH 7.5, 300 mM KGlutamate, 10% glycerol, 1 mM BME). Peak fractions were pooled, concentrated in a 10K Amicon Ultra-15 concentrator (Millipore), flash frozen in liquid nitrogen, and stored at -80°C. All DmCdt1 constructs were purified with the same procedure, except for eGFP-Cdt1 for which the TEV cleavage step was omitted.
Cloning, expression, and purification of ScCdt1. The coding sequence for full-length S.
cerevisiae Cdt1 was cloned into the QB3 Macrolab vector 1b for expression with a TEVcleavable N-terminal His6-tag. ScCdt1 was expressed in bacteria as previously described for DmCdt1. Cells from a 1 L expression were resuspended in 40 mL of Lysis Buffer (50 mM Tris pH 7.5, 300 mM KCl, 10% glycerol, 30 mM Imidazole, 200 µM PMSF, 1x cOmplete EDTA-free Protease Inhibitor Cocktail (Sigma-Aldrich), 1 mM BME and 0.1 mg/mL lysozyme) and lysed by sanitation. The lysate was clarified by centrifugation at 30,000 xg for 1 hour and subsequently filtered through an aPES 0.45 µm bottle-top filter unit. The clarified lysate was loaded onto a 5 mL HisTrap HP column and washed with 10 CV of Nickel Wash Buffer (50 mM Tris pH 7.5, 300 mM KCl, 10% glycerol, 30 mM Imidazole, 200 µM PMSF, 1 mM BME) and eluted with 3 CV of Nickel Elution Buffer (50 mM Tris pH 7.5, 300 mM KCl, 10% glycerol, 250 mM Imidazole, 1 mM BME). The His6 tag was then removed with an overnight incubation with TEV protease at a 1:10 (w:w) ratio of TEV:protein. After cleavage, cleaved ScCdt1 was isolated from TEV, the cleaved tag, and uncleared protein by an additional nickel purification step. Finally, the protein was purified by size exclusion chromatography over a HiPrep 16/60 Sephacryl S-300 HR column (GE Healthcare) equilibrated and run in Sizing Buffer (50 mM HEPES pH 7.5, 300 mM KGlutamate, 10% glycerol, 1 mM BME). Peak fractions were pooled, concentrated in a 10K Amicon Ultra-15 concentrator (Millipore), flash frozen in liquid nitrogen, and stored at -80°C.
Cloning, expression, and purification of human Cdt1. The coding sequence for full-length human Cdt1 was cloned into the QB3 Macrolab vector 4C for expression with a tobacco etch virus (TEV)-cleavable C-terminal hexa-histidine (His6)-maltose binding protein (MBP) tag. This vector was transformed into DH10bac cells for generation of bacmid DNA that was transfected (Cellfectin II) and amplified in Sf9 cells. Virus was amplified twice in Sf9 cells and protein expressed from 2 L of High5 cells infected with high-titer virus. Human Cdt1 was purified from High5 cells according to the method described above for DmCdt1 except that lysozyme was omitted from the Lysis Buffer.
Cloning, expression, and purification of DmORC. All ORC constructs were expressed in High5 insect cells from baculoviruses. Baculoviruses were generated by Cellfectin II (ThermoFisher) transfection of Sf9 cells with DH10bac-derived bacmid DNA that, in all cases, encoded a TEVcleavable His6 and MBP tag on Orc1 and Orc4, respectively. Full-length ORC was generated by co-infection of High5 cells with two baculoviruses, one encoding Orc1-5 and the second encoding Orc6. ORC complexes were also produced with deletions in the N-terminal IDR of Orc1 (ORC 1ΔIDR ) and Orc2 (ORC2 ΔIDR ). ORC 1ΔIDR was generated by co-infection of High5 cells with three baculoviruses, one encoding Orc2-5, a second encoding Orc6, and a final virus from which Orc1 ΔIDR was expressed. To generate the deletion, Orc1 was cloned into Macrolab vector 438B and an internal deletion (deletion of amino acids 248-549) generated by around-the-horn mutagenesis. Similarly, ORC 2ΔIDR was generated by co-infection of High5 cells with three baculoviruses, one encoding both Orc1 and Orc3-5, a second encoding Orc6, and a final virus from which Orc2 ΔIDR was expressed. To generate the deletion, an Orc2 N-terminal deletion (amino acid 270-618) was cloned into Macrolab vector 438A. Baculoviruses were amplified for two rounds prior to infection of High5 cells for protein expression.
All ORC constructs were purified according to the same procedure. The cell pellet from 2 L of cells was resuspended in 80 mL of Lysis Buffer (50 mM Tris pH 7.5, 300 mM KCl, 50 mM Imidazole, 10% glycerol, 200 µM PMSF, 1x cOmplete EDTA-free Protease Inhibitor Cocktail (Sigma-Aldrich), 1 mM BME) and lysed by sonication. The lysate was then clarified by centrifugation at 30,000 xg for 1 hour and the supernatant filtered through an aPES 0.45 µm bottle-top filter unit. Subsequently, the lysate was subjected to a 20% ammonium sulfate precipitation for 30 min at 4°C and then centrifuged again for 1 hour at 30,000 xg. The supernatant was then passed over a 5 mL HisTrap HP column (GE Healthcare) and washed with 10 CV of Nickel Wash Buffer (50 mM Tris pH 7.5, 300 mM KCl, 50 mM Imidazole, 10% glycerol, 200 µM PMSF, 1 mM BME) before protein elution with a 6 CV linear gradient from 50-250 mM Imidazole. The sample was then further purified by a second affinity step over an 8 mL amylose column (NEB), where it was washed with 3 CV of Amylose Wash Buffer (50 mM Tris pH 7.5, 300 mM KCl, 10% glycerol, 1 mM BME) and eluted with 2 CV of Amylose Elution Buffer (Amylose Wash Buffer supplemented to 20 mM maltose). The affinity tags from Orc1 and Orc4 were then removed by adding TEV at a 1:10 ratio and incubating overnight at 4°C. Finally, the sample was concentrated and then purified by size exclusion chromatography on a HiPrep 16/60 Sephacryl S-300 HR column (GE Healthcare) equilibrated and run in Sizing Buffer (50 mM HEPES pH 7.5, 300 mM KGlutamate, 10% glycerol, 1 mM BME). Peak fractions were pooled, concentrated in a 30K Amicon Ultra-15 concentrator (Millipore), flash frozen in liquid nitrogen, and stored at -80°C.
Cloning, expression, and purification of DmCdc6. The coding sequence for full-length D. DNA pull-down assays. A random oligonucleotide of sixty basepairs was produced and annealed with a complementary oligonucleotide in Annealing Buffer (50 mM HEPES pH 7.5, 50 mM KCl). This oligonucleotide is referred to throughout the paper as "dsDNA" and has sequence: 5'-GAAGCTAGACTTAGGTGTCATATTGAACCTACTATGCCGAACTAGTTACGAGCTATAAAC -3'. For DNA pull down assays, dsDNA was produced with a 5' biotin label (biotin-dsDNA, IDT).
To generate DNA-coupled agarose beads, High Capacity Streptavidin Agarose Resin (Pierce) was first washed with Annealing Buffer and then 25 µM biotin-dsDNA was added to the beads and incubated for 1 hour at room temperature; control beads were generated by adding Annealing Buffer alone. Following coupling, the beads were washed three times with Annealing Buffer. To assay DNA binding, the beads were first washed three times with Assay Buffer (50 mM HEPES pH 7.5, 150 mM KGlutamate, 10% glycerol, 1 mM BME) and then 10 µM Cdt1 or Cdc6 was added to control and DNA coupled beads (Cdc6 binding was assessed in the presence and absence of 500 µM ATP). The beads were incubated with protein for 1 hour at room temperature and then washed three times with 10x volume of Assay Buffer before resuspending and boiling the beads in Laemmmli sample buffer and assessing bound proteins by SDS-PAGE analysis and Coomassie staining.
Fluorescence polarization DNA-binding assays. Fluorescence polarization readings were taken on a BioTek Synergy using 384-well black bottom plates (Greiner) and detection of Cy5-labeled dsDNA (Cy3-dsDNA). Ten 15 uL reactions were prepared in Assay Buffer (50 mM HEPES pH 7.5, 150 mM KGlutamate, 10% glycerol, 1 mM BME) containing 50 nM Cy3-dsDNA and a serial dilution of Cdt1 from 1 µM to 0.5 nM. The reactions were incubated 10 min at room temperature and then polarization measured relative to a control. The mean and standard deviation of three independent experiments were plotted as a function of protein concentration and the data fit in Prism with a Hill equation to determine the dissociation constant.
Electrophoretic mobility shift assay (EMSA). EMSA gels were imaged on an Odyssey imaging system (LI-COR) through detection of an IRDye800 fluorescent tag appended to the 5' end dsDNA (IRDye800-dsDNA, IDT). Ten 20 µL reactions were prepared in Assay Buffer (50 mM HEPES pH 7.5, 150 mM KGluatmate, 10% glycerol, 1 mM BME) containing 25 nM IRDye800-dsDNA and a serial dilution of Cdt1 from 3 µM to 0.2 nM. The reactions were incubated for 45 min and then 5 L of each sample was run on a 5% native PAGE gel at 100 V for 1 hr.
Assaying liquid-liquid phase separation. Multiple methods were used to assay for phase separation. For the turbidity assays, 20 µM DmCdt1 was combined with 20 µM dsDNA in microcentrifuge tubes; as controls, samples were prepared that contained only DmCdt1 or dsDNA. After mixing, the solution became immediately turbid and the tubes were imaged on an Epson Perfection V700 scanner. We asked whether this was a reversible process by adding concentrated KCl (4 M) to a final concentration of 400 mM to the Cdt1/DNA mixture and again imaging the sample. To determine the source of the turbidity, 7 µL of each sample was spotted on a slide, covered with a coverslip, and imaged on a wide-field microscope equipped with DIC optics at a magnification of 63x with an oil immersion objective. For assaying droplet formation by fluorescence microscopy, samples were prepared by mixing protein with Cy5 or Cy3-labeled dsDNA (Cy5-or Cy3-dsDNA) and incubating 2 min prior to spotting 7 µL on a glass slide and covering with a glass coverslip. Subsequently, the samples were imaged on a Zeiss LSM 710 inverted confocal microscope using a 63x oil immersion objective and appropriate filter sets. For multi-color fluorescent imaging, control samples were imaged to ensure no crosstalk was observed between channels. Multi-color droplet recruitment assays were completed by first preparing droplets with ORC and Cy5-dsDNA. Subsequently, fluorescently-tagged proteins were spiked into each reaction and imaged. Quantitation of protein recruitment to preformed ORC droplets was completed in FIJI (Schindelin et al., 2012). First, regions of interest (ROI) for ORC/ Cy5-dsDNA droplets were generated with the auto-threshold function and then the signal intensity for the eGFP channel was measured within these regions, form which a mean and standard deviation was calculated. Finally, depletion assay samples were prepared by mixing protein with an equimolar amount of dsDNA and incubating the samples for 30 min at room temperature. After incubation, the samples were centrifuged for 10 min at 16,000 xg and the supernatant removed for separation by SDS-PAGE analysis and subsequent Coomassie or Silver staining; in all cases, a protein load control was also assessed that had not been centrifuged.
Disorder and complexity calculations. The per-residue disorder score was calculated using the online DISOPRED3 disorder prediction server (Jones and Cozzetto, 2015) and The centrifugation step was then repeated, the supernatant aspirated, and the pellet resuspended in Protein Buffer (50 mM HEPES pH 7.5, 300 mM KGlutamate, 10% glycerol, 1 mM BME). Finally, the supernatant and/or pelleted material was analyzed by SDS-PAGE analysis and silver stain or blotting (flag-tag on MCM3). Anti-FLAG primary antibody (Sigma-Aldrich F7425) was used at a 1/1,000 dilution and an IRDye800 CW Donkey anti-Rabbit secondary used at 1/10,000. Blots were imaged on a LI-COR Odyssey imager. To assess the contribution of ATP, OCC, or ORC, Cdc6, and Cdt1 separately, the individual reagents were removed from the loading reactions and the experiment completed as described. Blots were quantitated using FIJI (Schindelin et al., 2012).

Drosophila genetics. The full genomic coding region of D. melanogaster Orc1 including 200
basepairs of downstream regulatory sequence and approximately 1000 basepairs of upstream regulatory sequence was amplified by PCR from OregonR genomic DNA and cloned into the pattB vector with an N-terminal eGFP tag (eGFP-Orc1). To generate eGFP-Orc1 WalkerAB , the same procedure was used but PCR was utilized to incorporate two point mutations: K604A (Walker A) and D684A/E685A (Walker B). Transgenes were generated by injection (GenetiVision Corporation) for site specific PhiC31 integration at location P23L68A4 on the third chromosome. Homozygous lines were generated to confirm viability and were then balanced and crossed to flies carrying the Orc1 4739 null allele (Park and Asano, 2008) to determine each transgene's capacity to rescue lethality in flies homozygous for Orc1 4739 .
For imaging of embryos by lattice light-sheet microscopy it was desirable to have not only Orc1 tagged with eGFP, but to also be able to visualize histones. We therefore used a recombination cross between flies homozygous for eGFP-Orc1 or eGFP-Orc1 WalkerAB and flies containing a His2A-RFP transgene, also residing on the second chromosome. Recombinants containing both eGFP-Orc1/eGFP-Orc1 WalkerAB and His2A-RFP were identified by eye color.

Lattice light-sheet imaging of D. melanogaster embryos.
Embryos were collected from collection cages after a 90 minute laying period. Embryos were dechorionated in 100% bleach for 90 s and placed on the surface of a 5mm diameter glass coverslip using a fine haired paintbrush. The coverslip surface was made adhesive by applying a small drop of homemade glue solution (prepared by dissolving a roll of double-sided scotch tape in heptane overnight) (Mir et al., 2018).
Imaging was performed using a custom built Lattice Light-Sheet Microscope as previously described (Chen et al., 2014;Mir et al., 2017Mir et al., , 2018. For all experiments a 40 beam square lattice pattern was used with a minimum and maximum excitation numerical aperture of 0.44 and 0.55 respectively. The lattice pattern was dithered in the x direction over a 5 µm range in 0.1 µm steps over the duration of each exposure to generate a uniform excitation pattern. Zstacks were acquired by synchronously scanning the excitation sheet using a galvo-mirror, and the detection objective using a piezo stage over a 25 µm range in 0.250 µm increments. A 488 nm laser was used to excite eGFP (eGFP-Orc1 or eGFP-Orc1 WalkerAB ) with an exposure time of 50 ms at each slice and a 561 nm laser was to excite RFP (His2A-RFP) with an exposure time of 10 ms at each slice. During Z-stack acquisition an image was acquired for both channels at each z-position before moving to the next. The time interval between each stack was 5.67 s.
The excitation laser powers were set to 375 µW and 335 µW for the 488 nm and 561 nm lines respectively as measured at the back focal plane of the excitation objective. Images were recorded using two Hamamatsu ORCA-Flash 4.0 digital CMOS cameras. A long-pass dichroic (Semrock FF-560) nm was placed between the two cameras to separate emission wavelengths of above and below 560 nm, and a bandpass emission filter was placed in front of each camera (Semrock FF01-525/50 for eGFP and Semrock FF01-593/46 for RFP).
Once an embryo of suitable age was found (typically starting in nuclear cycle 12 as determined by examining the His2A-RFP channel) it was continuously imaged as described above until the completion of the mitosis preceding nuclear cycle 14.
Images of mitotic events, from metaphase to the completion of telophase were analyzed by first segmenting out chromatin using the His2A-RFP channel (Supplemental Movie 3).
Segmentation was performed using a custom script written in MATLAB R2017B. Briefly, first a 3D median filter was applied, followed by 3D Gaussian filtering to each z-stack. A locally adaptive threshold was then calculated, which was then used to generate a binary mask. The binary mask was then filtered by removing regions connected to a border so that only completely imaged nuclei were analyzed, and then filtered for size to remove segmentation errors. A label matrix was then generated from the binary mask for each time point. An found that the local maxima in the eGFP-ORC1 channel intensity corresponded exactly to early telophase in each nucleus (Fig 5) and this peak was then used to align each individual trace for averaging. Data from all cleavage events for each line were averaged together as no difference was found between traces from the 13th and 14th divisions in either the eGFP-ORC1 or eGFP-Orc1 WalkerAB lines (Supplemental Fig 7) . A total of 88 and 100 nuclei were analyzed for the eGFP-Orc1 and eGFP-Orc1 WalkerAB mutants respectively. Image stacks were collected on a Zeiss LSM 710 inverted confocal microscope using a 40x oil immersion objective (NA = 1.4) and filter sets for mCherry and eGFP. Samples were scanned at low magnification (20x) using mCherry-H2A signal to identify mitotic cells (< 1% of cell population). When metaphase cells were identified they were positioned at the center of the field of view and magnification increased (40x objective with 3x digital zoom). To assess Orc1 dynamics a z-stack (1 µm spaced images over the entire cell volume) was collected every 1.5 min from metaphase through cytokinesis. Images were processed using FIJI. First, each stack of slices was summed and a gaussian blur (radius = 1) filter applied. Subsequently, the average background intensity was subtracted from each channel and an ROI generated for each time point by auto-thresholding the mCherry-H2A channel. The eGFP-Orc1 intensity was measured for each ROI and the fold-change in intensity relative to metaphase intensity was calculated for each time point. We report the average and standard deviation of the fold-change in anaphase intensity for at least three individual mitotic cells of each transfected construct.
Phosphorylation assays. The effect of initiator phosphorylation on phase separation was assessed by both the depletion assay and fluorescence microscopy. For depletion assay analysis, phosphorylation reactions were prepared by mixing 1 µM of either ORC, Cdc6, or Cdt1 with 200 nM CDK1/CycA in the presence and absence of 5mM MgOAc and 1 mM ATP in Protein Buffer (50 mM HEPES pH 7.5, 300 mM KGlutamate, 10% glycerol, 1 mM BME). The phosphorylation reactions were incubated for 60 mins at room temperature and then, to set up the depletion assay, the reactions were mixed with dsDNA and incubated for 30 min at room temperature. The final depletion assay reaction conditions include 500 nM initiator, 125 nM CDK1/CycA, 2.5 mM MgOAc, 500 nM ATP, and 500 nM dsDNA in Assay Buffer (50 mM HEPES pH 7.5, 300 mM KGlutamate, 10% glycerol, 1 mM BME). After incubation, the samples were processed as previously described, assaying protein depletion by SDS-PAGE analysis and Silver staining. For assaying the effect of phosphorylation by fluorescence microscopy, protein concentrations were scaled up 4-fold for the phosphorylation reaction (4 µM ORC and 0.8 µM CDK1/CycA); all other reaction conditions are as described. Subsequently, the phosphorylation reactions were diluted 1:1 with equimolar Cy5-dsDNA in Assay Buffer, incubated for 2 min, and 7 µL were spotted on a glass slide, covered with a glass coverslip, and imaged as described previously.

ACKNOWLEDGEMENTS
We thank past and present members of the Berger and Botchan labs for helpful discussion and advice. We also thank Alison