Nucleotide binding halts diffusion of the eukaryotic replicative helicase during activation

The eukaryotic replicative helicase CMG centrally orchestrates the replisome and leads the way at the front of replication forks1. Understanding the motion of CMG on the DNA is therefore key to our understanding of DNA replication. In vivo, CMG is assembled and activated through a cell-cycle-regulated mechanism involving 36 polypeptides that has been reconstituted from purified proteins in ensemble biochemical studies2,3. Conversely, single-molecule studies of CMG motion have thus far4–6 relied on pre-formed CMG assembled through an unknown mechanism upon overexpression of individual constituents7,8. Here, we report the first activation at the single-molecule level of CMG fully reconstituted from purified yeast proteins and the quantification of its motion. We observe that CMG can move on DNA in two ways: by unidirectional translocation and by diffusion. We demonstrate that CMG preferentially exhibits unidirectional translocation in the presence of ATP, whereas it preferentially exhibits diffusive motion in the absence of ATP. We also demonstrate that nucleotide binding halts diffusive CMG. Taken together, our findings support a mechanism by which nucleotide binding allows newly assembled CMG to engage with the DNA within its central channel without melting it, halting its diffusion and facilitating the initial DNA melting required to initiate DNA replication.


Introduction 29
Eukaryotic DNA replication is catalyzed by a MDa-sized dynamic protein complex known as the 30 replisome. The replisome is powered by the replicative helicase CMG (Cdc45/Mcm2-7/GINS), 31 which centrally orchestrates the other components and leads the way at the front of replication 32 forks 1 . Understanding the motion of CMG on DNA is therefore crucial to our understanding of 33 how cells successfully replicate DNA. In vivo, loading and activation of CMG on DNA occur in 34 temporally separated fashion. In Saccharomyces cerevisiae in particular, CMG loading occurs at 35 specific sequences known as origins of replication 1 . First, in the G1-phase of the cell cycle, a set 36 of proteins known as 'loading factors' scans the DNA until such origins of replication are located, 37 at which inactive single and double Mcm2-7 hexamers are then loaded onto dsDNA 9-12 . In the 38 subsequent S-phase, double Mcm2-7 hexamers are selectively phosphorylated by the cell cycle-39 regulated Dbf4-dependent kinase (DDK) 13 . Then, a set of proteins known as 'firing factors' 40 facilitates the assembly of full CMG by recruiting the helicase-activating factors Cdc45 and GINS 41 to the phosphorylated Mcm2-7 double hexamers 2 . Upon full assembly, CMG must transition from 42 encircling dsDNA to encircling ssDNA, so that it can unwind dsDNA by steric exclusion of the 43 non-translocation strand 14 . This transition is known as CMG activation and consists of two steps. 44 In the first step, ATP binding allows each CMG in a double hexamer to melt 0.6-0.7 turns of 45 dsDNA within its central channel 3 . In the second step, each CMG extrudes one strand of the double 46 helix from its central channel; this final step requires ATP hydrolysis and the action of the firing 47 factor Mcm10 3 . After having extruded one strand of DNA, each activated sister CMG translocates 48 on ssDNA in a 3'-to-5' direction by hydrolyzing ATP 1,15,16 , allowing the two helicases to bypass 49 and move away from each other 3 , and committing the cell to initiate DNA replication 1 . This entire 50 combination of dual optical trapping and confocal scanning microscopy 21 to image and quantify 74 the motion of fluorescent CMG along DNA molecules held in an optical trap Extended 75 Data Fig.1a,Methods). To this end, we functionalized both ends of a linear 23.6 kb DNA 76 containing a natural ARS1 origin with digoxigenin and desthiobiotin moieties at both ends. We 77 then bound the functionalized DNA to streptavidin-coated magnetic beads and used it to assemble 78 and activate CMG (Methods). In short, we loaded Mcm2-7 hexamers onto the bead-bound DNA, 79 phosphorylated double Mcm2-7 hexamers with DDK and washed the beads with 300 mM KCl. 80 We then assembled and activated CMG for 15 min in the presence of fluorescently labeled 81 Cdc45 LD555 (Extended Data Fig. 1a), which supports unwinding near WT levels (Extended Data 82 Fig. 1b). Following CMG assembly and activation, we washed the beads again with 300 mM KCl 83 to select for fully mature CMG 2,22 , and 'paused' the reaction by removing ATP. DNA:CMG 84 complexes were then eluted from the magnetic beads by competing the desthiobiotin-streptavidin 85 interaction with an excess of free biotin 23 . Following elution, DNA:CMG complexes were tethered 86 between two optically-trapped anti-digoxigenin-coated polystyrene beads, and transferred into a 87 buffer solution containing Mcm10, RPA and either ATP, no nucleotide, or the slowly hydrolyzable 88 ATP analog ATPγS. We then scanned the DNA with a confocal scanning laser and observed 89 fluorescent CMG helicases as diffraction-limited spots on the otherwise unlabeled DNA (Fig. 1a-90 ii). Approximately a third of the trapped DNA molecules contained diffraction-limited fluorescent 91 CMG spots, typically a single one (Fig. 1b). We deduced the number of CMG per diffraction-92 limited spot by counting the photobleaching steps within each spot (Extended Data Fig. 3, 93 Methods). As this showed that most spots contained 1 CMG (Fig. 1c), it followed that most DNA 94 molecules had a total of 1 CMG (Extended Data Fig. 1c), where a priori one might have expected 95 a total closer to 2 or multiples thereof. We cannot attribute our experimentally measured lower 96 number to the labeling efficiency of Cdc45 LD555 , which we measured to be 85 ± 4 %; rather, we 97 attribute it to (a potential combination of) other factors including loss of Cdc45 during the high 98 salt washes and downstream handling, CMG dissociation at nicks 24 on the DNA during the 99 ensemble activation, or CMG diffusing off the ends of the DNA during elution. Furthermore, it 100 was recently shown that each Mcm2-7 in a double hexamer independently matures into CMG 22 . 101 Thus, we cannot discard the possibility that in our system only one of the two Mcm2-7 hexamers 102 is fully matured into CMG. 103 104

Mature CMG is preferentially assembled near origins of replication 105
We first looked at the initial positions of CMG on the DNA. Of note, because we cannot 106 differentiate between the two possible orientations of the DNA in our experiments, we display the 107 initial positions of CMG in plots showing the distance from the center of the DNA 9 . We observe a 108 wide distribution of initial positions with a peak near or at the ARS1 origin (Fig. 1d). Furthermore, 109 spots containing two CMG complexes are less widely distributed around the origin than spots 110 containing one CMG ( Fig. 1e-f). Taken together, these results are consistent with a preferential 111 assembly of sister CMG helicases near the ARS1 origin, followed by the motion of individual 112 activated helicases away from the origin during the 15-min ensemble activation reaction. 113 114

Colocalization of fluorescent Cdc45 and fluorescent Mcm2-7 hexamers is DDK-dependent 115
Salt-resistant Cdc45 is considered a hallmark of mature CMG 2,22 . Nonetheless, if the Cdc45 LD555 116 spots that we observe are part of bona fide CMG, their presence on the DNA should be dependent 117 on DDK 2,13 . To confirm this, we quantified the co-localization of red fluorescently labeled Mcm2-118 7 JF646-Mcm3 with green fluorescently labeled Cdc45 LD555 (shown to jointly support DNA unwinding 119 (Extended Data Fig. 2a)) in the presence and absence of DDK ( Fig. 1g-j). While nearly 20% of 120 Mcm2-7 JF646-Mcm3 spots colocalized with Cdc45 LD555 in the presence of DDK, we observed a ~4-fold 121 decrease in this colocalization in the absence of DDK (Fig. 1h-

Fully reconstituted CMG helicases exhibit two quantitatively distinct motion types 130
We next sought to quantify the motion of CMG in the presence of ATP. For this, we implemented 131 a change-point algorithm (CPA) to fit linear segments through regions of the position-vs.-time 132 plots of individual spots Methods); the slopes of these segments then give us a noise-133 reduced value of the instantaneous velocities of individual fluorescent spots. To calibrate our 134 analysis, we imaged dCas9 LD555 with the same imaging conditions that we used for CMG (Fig. 2a,135 Extended Data Fig. 3); because dCas9 LD555 is static on the DNA, it provides us with a measure of 136 the velocity error in our system. After drift correction, the distribution of instantaneous velocities 137 of fluorescent dCas9 LD555 spots after the CPA fit is centered at 0 bp/s and has a width σ dCas9 that 138 reflects our experimental uncertainty in velocity measurement (Fig. 2a inset). For all CMG motion 139 analysis, we defined a conservative velocity cutoff of 5 × σ dCas9 (= 2.0 bp/s) to categorize 140 fluorescent spots as static or mobile; we considered mobile any fluorescent spot with at least one 141 CPA segment with a slope above this threshold, and all other spots static. 142 Following the approach described above, we determined that ~70% of CMG spots are mobile when 144 imaged in a buffer solution containing RPA, Mcm10 and ATP ( Fig. 2b and 2d, Extended Data Fig.  145 5a). Unexpectedly, when we imaged CMG in a buffer solution containing RPA, Mcm10 and no 146 ATP, we observed that ~40% of CMG spots were also mobile ( Fig. 2c-d, Extended Data Fig. 5d). 147 Nonetheless, we noticed qualitative differences in motion of CMG in the presence and absence of 148 ATP: while CMG seemed to move unidirectionally in the presence of ATP (Fig. 2b,  149 Supplementary Movie 1), it appeared to move in a more random (e.g. diffusive) manner in the 150 absence of ATP (Fig. 2c, Supplementary Movie 2). To quantitatively characterize these two 151 apparently distinct motion types, we employed two independent approaches. First, we looked at 152 the CPA segments of all the traces in each condition, and calculated the probability that 153 consecutive segments have the same direction (Fig. 2e); in the absence of noise, this probability 154 should equal 1 for unidirectional motion, and 0.5 for random motion. As seen in Fig. 2e, our 155 measured probabilities closely match these expected values, providing quantitative underpinning 156 of our initial observations. As an independent approach, we conducted anomalous diffusion 157 analysis of the mobile traces in each condition Extended Data Fig. 5b,5e and 5h). For 158 each individual trace, we calculated the mean-squared displacement (MSD) as a function of the 159 lag time τ, and then fitted the result to the equation MSD(τ) ∝ τ α to extract the anomalous diffusion 160 coefficient α (Methods). The value of α then allowed us to classify each trace into different motion 161 types, as α ≫ 1 for unidirectionally moving molecules, α ≈ 1 for a freely diffusive molecules, and 162 0 < α ≪ 1 for molecules undergoing constrained diffusion (Fig. 2f). This anomalous diffusion 163 analysis confirms that unidirectional motion is most likely when ATP is present (Fig. 2g), whereas 164 diffusive behavior is most likely when ATP is absent (Fig. 2h). 165 We note that we observed a small population of seemingly diffusive CMG spots in the presence 167 of ATP, and a small population of seemingly unidirectionally moving CMG spots in the absence 168 of ATP (Fig. 2g-h). We hypothesized that these subpopulations might have arisen from 169 misclassification of short traces 26 . To test this hypothesis, we simulated two populations of single-170 molecule traces of varying lengths within the range of our experimental data: one population solely 171 consisting of unidirectionally moving traces, and the other population solely consisting of freely 172 diffusive traces (Methods). We then carried out the same anomalous diffusion analysis that we did 173 on the experimental CMG data on both simulated data sets. We observed that the distribution of 174 motion types for the simulated unidirectional traces looked very similar to that of the experimental 175 mobile CMG traces in the presence of ATP (Extended Data Fig. 6a and Fig. 2g), whereas the 176 distribution of motion types for the simulated diffusive traces looked very similar to that of the 177 experimental mobile CMG traces in the absence of ATP (Extended Data Fig. 6b and Fig. 2h). 178 Thus, the results of our simulations suggest that, overall, the mobile traces in the presence of ATP 179 represent unidirectional motion, whereas the mobile traces in the absence of ATP represent 180 diffusive motion. 181 182

Analysis of CMG motor motion 183
Following this identification of two distinct types of CMG mobility, we investigated both in further 184 depth. We first investigated the unidirectional motor motion of CMG, the motion type that powers 185 the replisome. We thus specifically analyzed the velocities of unidirectionally moving CMG spots 186 in the presence of ATP, which yielded a distribution of instantaneous velocities with a peak at ~5 187 bp/s (Fig. 3a), consistent with previous single-molecule studies on pre-formed CMG in the 188 presence of RPA 4 . This distribution has a long tail, reaching up to instantaneous velocities of ~45 189 bp/s. These higher velocities are low in probability, suggesting that CMG can only achieve these 190 high velocities in short time bursts. Consistent with this, when we calculated the time-averaged 191 velocity of each CMG spot and examined the resulting distribution ( Fig. 3a inset), we did not 192 observe such high velocities. The distribution of time-averaged velocities has a peak at ~5 bp/s, 193 which is consistent with ensemble biochemical studies of CMG motion 18 . Notably, we did not 194 observe any noticeable backtracking of CMG (Fig. 2b), which is consistent with previous studies 195 suggesting that RPA prevents CMG backtracking by keeping the lagging strand template out of opposite directions upon their activation. The low probability of these splitting events is to be 204 expected because i) we allow CMG to become activated and translocate on the DNA for 15 min 205 before imaging, and ii) because most DNA molecules contain 1 CMG (Extended Data Fig. 1c). 206 207

Nucleotide binding halts diffusive CMG 208
Our data shows that CMG diffuses on DNA in the absence but not in the presence of ATP (Fig.  209 2b-c and 2g-h), suggesting that ATP is involved in stopping this diffusive motion of CMG. To 210 investigate whether it was the binding or the hydrolysis of ATP that stopped the diffusive motion 211 of CMG, we investigated CMG motion in a buffer solution supplemented with RPA, Mcm10 and 212 the slowly hydrolysable ATP analog ATPγS. When we imaged CMG under these conditions, the 213 vast majority CMG spots were found to be static ( Fig. 4a-b, Extended Data Fig. 5i). Taken together 214 with our data in the presence and absence of ATP ( Fig. 2b-d), our results show that it is the 215 nucleotide binding and not the hydrolysis that halts the diffusive motion of CMG. Furthermore, 216 our results confirm that ATP hydrolysis is required for the unidirectional translocation of CMG 16 . 217

218
Previous biochemical studies showed that ATP binding allows newly formed CMG to melt 0.6-219 0.7 turns of the DNA within its central channel 3 , which was recently confirmed by cryo electron 220 microscopy 25 . Comparing these previous observations with our single-molecule results led us to 221 hypothesize that 1) the diffusive motion that we observed in the absence of ATP corresponded to 222 CMG surrounding dsDNA, and that 2) the halting of such diffusive motion in the presence of 223 ATPγS is due to CMG melting the DNA within its central channel. 224

225
To test whether CMG can diffuse onto dsDNA in the absence of ATP, we developed an ensemble 226 CMG sliding assay Methods). Briefly, we synthesized two 1.4 kb linear DNA constructs 227 biotinylated at one end and containing an ARS1 origin. The non-biotinylated end of the constructs 228 was then either left as a free end or covalently crosslinked to a M.HpaII methyltransferase 27 . 229 Because the crosslinked methyltransferase is too large to fit inside the central channel of CMG 28,29 , 230 it should stop CMG from diffusing off the end of the DNA, which would otherwise be free to 231 diffuse off the free end of the DNA. We then bound both DNA constructs to streptavidin-coated 232 magnetic beads, and assembled CMG in the presence of fluorescent Cdc45 LD555 onto them; 233 importantly, we omit the firing factor Mcm10 from the activation reaction to prevent strand 234 extrusion from the central channel of CMG and ensure that CMG is surrounding dsDNA 3 . After 235 CMG assembly, we incubated the bead-bound DNA in a buffer solution with or without ATPγS, 236 and monitored the amount of fluorescent Cdc45 LD555 present on the DNA over time. As seen in 237 Fig. 4d-e, we detected the fastest decay of Cdc45 LD555 signal in the DNA construct with a free end 238 in the absence of ATPγS, whereas there was only a small decay in the same DNA construct when 239 ATPγS was present. We also observed some decay, albeit smaller in magnitude, in the DNA 240 construct with a capped end in the absence of ATPγS ( Fig. 4d- We report the first single-molecule quantification of the motion of fully reconstituted CMG 262 helicases. To the best of our knowledge, these studies also constitute the first fully in vitro 263 preferentially moves in a diffusive manner (Fig. 2). To explain these motion outcomes, we propose 276 the model summarized in Fig. 4c, which is based on the assumption that our ensemble activation 277 reaction contains a mixture of two populations of CMG: one population encircling dsDNA, and 278 another population encircling ssDNA (Fig. 4c-i). In the first part of our model, we propose that 279 the population of CMG surrounding ssDNA moves unidirectionally in the presence of ATP (as the 280 motor can then hydrolyze ATP to unwind dsDNA), but remains static in the absence of ATP (as 281 the motor is unable to unwind DNA without ATP hydrolysis 3,6,31 ) (Fig. 4c, right half). This part of 282 the model is supported by the fact that we observe very similar instantaneous and average 283 velocities of unidirectionally translocating CMG spots to those of previous single-molecule and 284 ensemble biochemical studies of DNA unwinding 4,5,18 . We consider an alternative scenario in 285 which at least part of the unidirectional translocation that we observe in the presence of ATP 286 corresponds to CMG translocating on dsDNA (Extended Data Fig. 10) -as was postulated to occur 287 for CMG helicases that bypassed each other after the collision of two replication forks or after 288 encountering a flush ss/dsDNA junction 32-34 -to be less likely. Consider, for example, the CMG 289 splitting events that we observe (Fig. 3c). Because CMG translocation on dsDNA has the same 3'-290 to-5' polarity as on ssDNA 34 , one could postulate that the splitting events represent two helicases 291 that surround dsDNA following assembly at a single Mcm2-7 double hexamer, but such helicases 292 would move towards each other (remaining within a single diffraction-limited spot), and not away 293 from each other as our data shows. One could also postulate that the splitting events represent two 294 helicases encircling dsDNA following assembly at distinct Mcm2-7 double hexamers within a 295 single diffraction-limited spot. However, our analysis of Mcm2-7 spots shows such occurrences 296 to be unlikely (Extended Data Fig. 2c). Further studies will be required to assess whether and how 297 the unidirectional motion of CMG differs according to whether it occurs on ssDNA or dsDNA. 298

299
In the second part of our model, we propose that the population of CMG surrounding dsDNA loses 300 its bound ATP when we remove ATP from the buffer following our ensemble activation (Fig. 1a); 301 ATP dissociation then causes CMG to disengage from the DNA and thereby become diffusive. 302 This diffusive population remains so in the absence of nucleotide (giving rise to the diffusive 303 population we observe in the absence of nucleotide); nevertheless, upon ATP re-addition and re-304 binding, CMG is allowed to re-engage with the DNA inside its central channel and thus become 305 static (giving rise to the static population in the presence of ATP). In support of this hypothesis, 306 we showed that CMG can diffuse on dsDNA and that nucleotide binding halts this diffusion (Fig.  307   4a-b). Furthermore, we show that this halting occurs independent of the previously observed DNA 308 melting by CMG upon nucleotide binding 3,25 (Extended Data Fig. 9), suggesting that DNA 309 engagement and DNA melting by CMG need not occur concomitantly. We propose that the 310 presence of ATP in the nucleus prevents newly assembled CMG from diffusing along the DNA, 311 poising the helicase to catalyse the initial melting required to initiate replication. Further studies 312 investigating which of the additional Mcm2-7:DNA contacts within CMG 25 are responsible for 313 halting CMG diffusion upon nucleotide binding will shed further light into our observations. 314

315
Our measured diffusion coefficient of freely diffusive CMG under the ionic strength conditions of 316 this study (250 mM K-glutamate) (Extended Data Fig. 5l) is similar to our previously measured 317 diffusion coefficient of single Mcm2-7 hexamers in higher ionic strength conditions (500 mM 318 NaCl) 9 . This observation suggests that CMG diffuses more freely on the DNA than single Mcm2-319 7 hexamers, which in turn suggests that CMG has fewer contacts with the DNA than Mcm2-7 320 hexamers as recently confirmed by structural studies 25,35 . We also note that single-molecule studies 321 with pre-formed D. melanogaster CMG showed no evidence of CMG diffusion in the presence of 322 ATP 4,5 , in agreement with our observations. However, single-molecule studies with pre-formed S. 323 cerevisiae CMG reported extensive diffusive behavior in the presence of ATP 6 . Further studies 324 will be required to investigate the reasons for this discrepancy, and to probe potential differences 325 between fully reconstituted and pre-formed CMG, such as whether the phosphorylation state of 326 Given the complexity and the number of components required to fully reconstitute CMG assembly 337 and activation, in vitro single-molecule studies quantifying CMG motion have so far relied on pre-338 formed CMG helicases assembled through an unknown mechanism upon overexpression of 339 individual constituents 7,8 . Although these studies have provided us with very important insights 340 into how CMG works, it is unknown whether the assembly mechanism of pre-formed CMG is 341 cell-cycle regulated. It is therefore also unknown whether pre-formed CMG has a similar 342 phosphorylation state to the one it has in vivo. In addition, pre-formed CMG requires an artificial 343 region of ssDNA to bind DNA and thus does not allow us to access the intricacies of CMG 344 assembly and activation. In this study, we report the first single-molecule quantification of the 345 motion of fully reconstituted CMG helicases. We believe that the full reconstitution of CMG 346 makes it more likely that the phosphorylation state of its constituents more closely mimics what 347 happens in vivo; this in turn makes it more likely that fully reconstituted CMG can support all 348 relevant interactions with proteins intrinsic and accessory to the replisome. Furthermore, the 349 reliable assay we developed will allow us and others to address important questions regarding 350 CMG motion, such as the mechanistic role of elongation factors, with unprecedented resolution. 351 Finally, we anticipate that the novel assay we developed, based on the double functionalization of 352 DNA ends with two orthogonal attachment types, will facilitate the study at the single-molecule

Figure 2 | Fully reconstituted CMG exhibits two different motion types. a, Position
vs. time of dCas9 LD555 spots; (inset) distribution of instantaneous velocities coming from the CPA fits of dCas9 LD555 spots; red lines show the instantaneous velocity cutoff (5σdCas9) used to separate CMG spots in b, and c, into static or mobile; CPA fits are not shown for clarity. b-c, Position vs. time plots of CMG spots in the b, presence of ATP c, absence of nucleotide; CPA fits are plotted in black, static traces are shown in light gray. d, Ratio of static CMG traces in the presence of ATP, absence of nucleotide, and static dCas9 traces. e, Frequency of consecutive CPA segments with the same direction for CMG spots in the presence of ATP (blue) or absence of nucleotide (orange); inset diagrams illustrate expected segment directions of a unidirectionally moving spot (top) or a diffusive spot (bottom). f, (left panel) Idealized examples of MSD vs. delay time τ plots with an anomalous coefficients α < 1 (red), α = 1 (yellow) and α > 1 (green); (right panel) diagrams illustrating the types of CMG motion corresponding to each of these three cases: constrained diffusion (α << 1), free diffusion (α ≈ 1) or unidirectional motion (α >>1). g-h, Fraction of mobile CMG traces classified into different motion types in the g, presence of ATP or h, absence of nucleotide.   Fig. 2d, are shown as light bars for comparison. c, model proposed to explain experimental motion results; c-i, proposed two populations of CMG present in the ensemble CMG activation reaction; melted and extruded states correspond to the first and second steps in CMG activation, respectively; c-ii, summary of experimental outcomes in Fig. 2 and Fig. 4a, and proposed explanation of their origins. d, Fluorescent scan of an SDS-PAGE gel showing the amount of Cdc45 LD555 left on linear DNA bound to magnetic beads at one end and containing either a free end or an end capped with a covalently crosslinked methyltransferase. e, Densitometry quantification of the experiment shown in d, showing the average normalized intensity of three replicates together with their standard deviation. Data points are connected by solid lines to guide the eye. Figure 1 | Hybrid ensemble and single-molecule assay and reagent validation. a, SDS-PAGE showing the minimal set of purified proteins required for the reconstitution of CMG assembly and activation; the gels were stained with Coomassie Blue Stain and fluorescently scanned with either a red or a green laser, to show the fluorescently labeled proteins in either color. b, Ensemble unwinding assay showing that Cdc45 LD555 supports DNA unwinding to near WT levels. c, Distribution of total numbers of fluorescent CMG complexes per DNA, obtained by combining the total number of CMG diffraction-limited spots per DNA (Fig. 1b) with the number of CMG complexes within each spot (Fig. 1c). Statistical significance was obtained from a two-sided binomial test (p-value=1.2 × 10 −5 ).