Identification of Selective ATP-Competitive CMG Helicase Inhibitors for Cancer Intervention that Disrupt CMG-Replisome Function

SUMMARY The human CMG helicase (Cdc45-MCM-GINS) is a novel target for anti-cancer therapy due to tumor-specific weaknesses in CMG function induced by oncogenic changes and the need for CMG function during recovery from replicative stresses such as chemotherapy. Here, we developed an orthogonal biochemical screening approach and identified selective CMG inhibitors (CMGi) that inhibit ATPase and helicase activities in an ATP-competitive manner at low micromolar concentrations. Structure-activity information and in silico docking indicate that CMGi occupy ATP binding sites and channels within MCM subunits leading to the ATP clefts, which are likely used for ATP/ADP ingress or egress. CMGi inhibit cell growth and DNA replication using multiple molecular mechanisms. CMGi block helicase assembly steps that require ATP binding/hydrolysis by the MCM complex, specifically MCM ring assembly on DNA and GINS recruitment to DNA-loaded MCM hexamers. During S-phase, inhibition of MCM ATP binding/hydrolysis by CMGi causes a ‘reverse allosteric’ dissociation of Cdc45/GINS from the CMG that destabilizes the replisome and disrupts interactions with Ctf4, Mcm10, and DNA polymerase-α, -δ, -ε, resulting in DNA damage. These novel CMGi are selectively toxic toward tumor cells and define a new class of CMG helicase-targeted anti-cancer compounds with distinct mechanisms of action.


Introduction 71
The replicative CMG helicase is an emerging target for anti-cancer intervention due to 72 exploitable vulnerabilities in cancer cells resulting from oncogene-driven mismanagement of 73 CMG assembly and function 1 . However, to date, no small chemical inhibitors of the human CMG 74 helicase have been identified. The CMG is a multi-subunit enzyme that performs the primary DNA 75 melting and unwinding steps within replisomes during DNA replication in eukaryotic cells 2 . The 76 CMG helicase is composed of Cdc45, a Mini-Chromosome Maintenance (MCM) heterohexameric 77 ATPase core, and the GINS tetramer (Go-Ichi-Ni-San in Japanese for Sld5, Psf1, Psf2, and Psf3) 2, 78 3, 4 . Assembly and activation of the CMG occur in a stepwise manner, with an excess of MCM 79 hexamers loaded onto DNA during G1 phase (called Licensing 5, 6 ), followed by recruitment of 80 Cdc45/GINS near the G1-S transition to a subset of these MCM hexamers 3,4,7,8 . The dynamics 81 of this MCM-CMG conversion process are important for maintaining genome stability in human 82 cells 1 . Unused MCM complexes act as reserves that are converted to CMG helicases during 83 replicative stress to facilitate recovery of DNA replication 1,9,10,11,12 . Reserve MCM complexes 84 also modulate replication fork speeds to prevent DNA damage 13 . 85 Oncogenic changes cause problems with CMG assembly and function 1 . Cyclin E 86 is in agreement with biochemical assays that clorobiocin is a more effective inhibitor of the hCMG 231 than novobiocin. Such specificity likely derives from the noviose sugar group specific to 232 clorobiocin and CA1. However, overall hCMG inhibition by these compounds cannot be 233 determined from analysis of a single cleft, but instead derive from a sum of inhibitor effects on 234 multiple clefts that individually may or may not be required for specific hCMG/hMCM functions. 235 At present we do not know if certain MCM ATPase clefts display preferences for inhibitor 236 binding over other clefts, particularly in cells, as determining this is technologically challenging 237 for an enzyme comprised of six ATPase domains that are not easily separated for analysis. Going 238 forward, we assess hCMG ATPase and helicase inhibition in cells with CA1 from a holo-enzyme 239 perspective (AT P -competitive effects on all clefts combined) rather than a particular ATPase cleft. 240 241

CMGi Inhibit MCM/Cdt1 Assembly on Chromatin 242
We used the immortalized, non-tumor derived human keratinocyte HaCaT cell line to 243 assess effects of CMGi on growth, DNA replication, and CMG helicase assembly and function. 244 HaCaT cells are synchronized in a quiescent state by serum deprivation and released into the cell 245 cycle to study G1 and S-phase events. A cell viability analysis found that, while novobiocin has 246 little effect, CA1 reduces viability with an IC 50 of ~15 µM ( Figure 3A). Since this concentration 247 of CA1 aligns with that required to inhibit in vitro hCMG helicase activity, and novobiocin has 248 little effect, these results support that the hCMG is a major target of CA1 in human cells at these 249 low concentrations. CA1 inhibits DNA replication when added to HaCaT cells in early G1, while 250 novobiocin has little effect ( Figure 3B), consistent with assembly and activity of hCMG complexes 251 being required for progression into S-phase. 252 12 Orc2 and Orc4 affinity for chromatin is not affected by CA1 exposure in middle G1 ( Figure  275 3E), again suggesting that the ATPases of ORC are not a target of CA1. Cdc6 protein is not affected 276 in chromatin association by CA1 when measured using a polyclonal antibody ( Figure 3E). In 277 contrast, analysis with a monoclonal antibody to Cdc6 suggests that one form of Cdc6 increases 278 on chromatin in parallel with MCM elevation and is sensitive to CA1. Cdc6 contains an ATPase 279 domain that is not required for MCM assembly on DNA, but instead for removal of improperly 280 loaded MCMs 29, 30 . While we cannot rule out the possibility that the ATPase site of Cdc6 may be 281 affected by CA1, these prior studies suggest that it is unlikely that this would contribute to the 282 inhibition of MCM assembly we have observed in the presence of CA1. 283

GINS Recruitment to DNA-loaded MCM Complexes is Inhibited by CMGi 285
HaCaT cells treated with CA1 in late-G1 (12 hrs), but not novobiocin, fail to undergo DNA 286 replication ( Figure 3F). Near the G1/S transition (15-18 hrs) there is additional MCM loading onto 287 chromatin, which coincides with GINS and Cdc45 being recruited to loaded MCM hexamers on 288 chromatin ( Figure 3G). Treatment with CA1 during this late-G1 period has only a small effect, if 289 any, on the remainder of MCM loading and does not affect Cdt1 dynamics, suggesting that Mcm2-290 7 ATPase functions in MCM/Cdt1 loading are no longer required at this later time. However, while 291 Cdc45 is not appreciably affected, CA1 inhibits GINS recruitment to MCM hexamers ( Figure 3G) There are other enzymes with ATPase domains that function in MCM/CMG assembly, 303 including Dbf4-Cdc7 (DDK) and Cdk2 2 . We asked whether CA1 affected these and other enzymes 304 in human cells. DDK phosphorylates two sites in Mcm2 (S53 and S139) to facilitate Cdc45 305 recruitment 35 , both of which show no phosphorylation changes after extended exposure to CA1 306 ( Figure 4A). This agrees with our observation that Cdc45 is recruited to DNA-loaded MCM 307 hexamers ( Figure 3G) and indicates that DDK is not a target of CA1. A pan-Cdk inhibitor 308 (AT7519) that efficiently targets Cdk1 (Cdc2), Cdk2, Cdk3, Cdk4, Cdk6, and Cdk9 blocks 309 phosphorylation of Cdk2 targets, including Rb and Cdc6 (S54P) 36 37 , and the Cdk1 target PP1a 38 . 310 However, extended exposure to CA1 has no effect on these substrates ( Figure 4B). We conclude 311 that CA1 does not target these kinases required for MCM/CMG assembly. 312 We next asked if two helicases (BLM and WRN) involved in DNA repair and 313 recombination contribute to MCM/CMG assembly 39 . Reduction of BLM or WRN using siRNA 314 has no effect on MCM, Cdc45, or GINS recruitment to chromatin ( Figure 4C, left), and has no 315 effect on cell growth, indicating that they are not required for unperturbed DNA replication in 316 HaCaT cells ( Figure 4C, right). It is not known if CA1 can inhibit BLM or WRN. Since these data 317 demonstrate that neither helicase is required for MCM/CMG assembly, any effects of CA1 on 318 these helicases does not contribute to the outcomes we have observed. Finally, an in vitro 319 decatenation assay shows that Topo-II is not affected by low doses of CA1 (Supplementary Figure 320 We determined how CMGi exposure affected the dynamics of hCMG and replisome 326 structure during S-phase. Synchronized HaCaT cells were treated with CA1, novobiocin, or 327 DMSO once cells reached early S-phase (18 hrs) and immunoblots were performed assessing 328 chromatin-bound and total protein components of the replication machinery ( Figure 5A). 329 Etoposide has no effect on hCMG activity in vitro (Supplementary Figure S5B), and was included 330 to compare how inhibition of Topo-II affected replication dynamics. Treatment with CA1 and 331 etoposide effectively suppressed DNA replication, while novobiocin/DMSO did not ( Figure 5B), 332 confirming that hCMG and Topo-II activities are required for ongoing DNA replication. MCM 333 association with chromatin/DNA was not affected by any compounds ( Figure 5C, left). However, 334 GINS and Cdc45 chromatin association was notably suppressed by CA1, and not by novobiocin 335 or etoposide. Total protein levels were slightly affected for Psf3 and Cdc45, but not for other 336 subunits ( Figure 5C, right). We conclude from these results that DNA topological issues and DNA 337 replication arrest due to Topo-II inhibition do not disrupt GINS/Cdc45 interactions with hCMG 338 helicases. However, CMGi inhibition of ATP binding and/or hydrolysis by the Mcm2-7 ATPases 339 causes GINS and Cdc45 dissociation from hCMG helicases during ongoing DNA replication. 340 Structural studies have shown that components of the human replisome interact directly 341 with the hCMG helicase, mediated in part through GINS and Cdc45 40,41 . We asked if CMGi-342 induced loss of Cdc45 and GINS from hCMGs resulted in disruption of replisomes in human cells. 343 Treatment of S-phase cells with CA1, but not other compounds, caused a loss of DNA 344 polymerases-a, -d, and -e from chromatin ( Figure 5C). Factors such as Ctf4 and Mcm10, which 345 interact with the hCMG and facilitate DNA polymerase-a function on the lagging strand 40, 41 are 346 also reduced on chromatin by CA1. Consistent with hCMG and replisome disruption, RPA (single-347 stranded binding protein) is also reduced on chromatin. Intriguingly, exposure to etoposide, which 348 stops DNA replication, does not significantly diminish replisome components or RPA on 349 chromatin, except for a small change to DNA polymerases-a and -d ( Figure 5C). We conclude 350 that, while inhibition of Topo-II (and DNA replication) does not have a significant negative effect 351 on replisome integrity, the structural co-integrity of replisomes and hCMGs is dependent on ATP 352 binding and/or hydrolysis by the Mcm2-7 ATPases during ongoing DNA replication in human 353

cells and is disrupted by CMGi. 354
We next determined if CMGi treatment of partially purified hCMG helicases and 355 replisomes from human cells displayed similar outcomes. Nuclear extracts were prepared from 356 HaCaT cells enriched in S-phase and subjected to immunoprecipitation using antibodies to Psf1 357 or Mcm2. Immunoprecipitated complexes were treated directly with CA1 or DMSO, followed by 358 immunoblotting for associated proteins ( Figure 5D). Psf1 associates with Psf2, Psf3, Mcm2, and 359 Mcm6, indicating that hCMG helicases were extracted from cells ( Figure 5D, middle). We could 360 not examine Cdc45 in this experiment due to signal interference with IgG on immunoblots.   shows that gamma-H2AX levels increase after 24 hr exposure to CA1, but not novobiocin. CA1 384 treatment also causes increased cleavage of PARP, indicative of apoptosis ( Figure 6A), and causes 385 an increase in phosphorylated Chk1 ( Figure 6B). We conclude that CMGi-induced loss of hCMG 386 and replisome structural integrity is associated with increased DNA damage, checkpoint signals, 387 and apoptosis. 388 389 Tumor Cells Are Selectively Sensitive to CMGi 390 CMG helicases and reserve MCMs may be weakened in tumor cells due to oncogene-driven 391 changes 1 , which would render tumor cells more sensitive to CMGi relative to non-tumor cells. To 392 test this, we compared three tumor lines to HaCaT cells for their sensitivity to a lower dose of CA1 393 or novobiocin (24 hr exposure, 5 µM) by assessing PARP cleavage. All three tumor lines displayed 394 significant PARP cleavage only in the presence of CA1, while HaCaT showed no effect at this 395 lower dose ( Figure 6C). Viability analyses determined that tumor lines were ~4-10 times more 396 sensitive to CA1 exposure relative to HaCaT, with IC 50 estimates of 1-4 µM ( Figure 6D). These 397 results demonstrate that tumor cells are selectively sensitive to CA1/CMGi exposure, and that the 398 CMG helicase (or MCM complexes) is a novel target for anti-cancer intervention. 399 400 DISCUSSION 401

Mode of hCMG Binding and Inhibition by CMGi 402
We have identified the first small chemical compounds capable of inhibiting ATPase and 403 helicase functions of the human replicative CMG helicase. Clorobiocin and CA1 are effective and 404 selective CMGi at low concentrations, competing with ATP for binding and hydrolysis at one or 405 more ATPase sites within the MCM ring. SAR suggests selectivity of CA1 derives from the 406 composition of the noviose sugar groups. Consistent with the mechanism of binding and ATP-407 competitive inhibition by these compounds toward their bacterial target, gyrase 18, 27 , CA1 and 408 clorobiocin bind channels within MCM subunits leading to particular ATPase clefts. The sugar 409 groups occupy space where adenosine and ribose of ATP normally interact, thereby producing a 410 direct competition for ATP binding. Modeling also suggests that these inhibitors compete with 411 ATP through a steric block to ATP/ADP ingress or egress through channels at the ATPase 412 domains. Thus, these CMGi act like a 'cork in a wine bottle,' but one that also displaces the 'wine' 413 (i.e., ATP). 414 The existing cryo-EM structural data for the hCMG 31 allows only three ATPase sites to be 415 docked by . Since the other three MCM ATPase 416 channels/clefts cannot be docked with CMGi due to steric hindrance, two somewhat mutually 417 exclusive interpretations emerge. In one case, CMGi may never be capable of binding these clefts 418 if the channels remain in a non-permissible conformation. A corollary of this would require that 419 such conformations not restrict ATP/ADP movement through the narrow channels. Alternatively, 420 there may be certain enzyme states not observed in this cryo-EM structure during which channels 421 become wider and accessible to CMGi entry. Such a molecular 'breathing' model would suggest 422 that ATP/ADP movement at each ATPase domain might be controlled by altering the width of 423 these channels, and thus ATP hydrolysis around the MCM ring might be regulated through 424 allosteric ATP accessibility.

Mechanisms of CMGi on ATP-dependent MCM/CMG Functions in Human Cells 439
Mutation of the majority of S. cerevisiae MCM (ScMCM) ATPase clefts results in loss of 440 viability, derived in part from loss of ScMCM loading onto DNA 28,29,30,43  to speculate that cells might possess a soluble factor that is capable of targeting the hCMG ATPases 471 (directly or via post-translational modifications), leading to replisome disruption during intra-S 472 checkpoints or DNA repair at stalled forks, or when replisomes from approaching forks in adjacent 473 replicons meet. CMGi will be a valuable chemical probe to further interrogate these hCMG and 474 replisome functions in mammalian cells. 475 476

CMGi Define a New Class of Anti-Cancer Compounds 477
Common targets in anti-cancer regimens often include the DNA replication and repair 478 machinery. Clearly, the CMG helicase represents another such target, and the CMGi discovered 479 here will inform the development of additional derivatives for use in the anti-cancer arsenal. 480 Aminocoumarins were originally marketed or investigated in clinical trials as anti-bacterial agents, 481 but their use was supplanted by quinolones such as ciprofloxacin 18 . However, aminocoumarins 482 such as these CMGi may offer clinical advantages with novel modes of action in suppressing tumor 483 growth due to targeted inhibition of the CMG and replisomes. More importantly, oncogenic 484 changes mismanage the regulation of reserve MCMs/CMGs, predicting deficiencies in CMG 485 functionality and a tumor-specific vulnerability in the CMG relative to non-tumor cells 1,46 . In 486 support of this, our results show that tumor cells are indeed selectively sensitive to CMGi, and, 487 importantly, that a therapeutic window exists for targeting the CMG helicase with CMGi. 488 Consistent with our findings, the National Cancer Institute has publicly available data 489 showing that many solid tumor cell lines (NCI-60 set) are highly sensitive to growth suppression 490 by the CMGi identified here (CA1, NSC107412; clorobiocin, NSC227186), but are insensitive to 491 novobiocin (NSC2382; https://dtp.cancer.gov/dtpstandard/dwindex/index.jsp). Until now, a 492 cellular target consistent with these differences was not clear. It had been suggested that HSP90 493 might be a target, but the high doses required to inhibit HSP90 (700-1000 µM) and its sensitivity 494 to CA1 and novobiocin suggest it is not the target 47, 48 . The direct relationship between the low 495 doses that inhibit the CMG in biochemical assays, and similar low doses that inhibit the CMG in 496 vivo to arrest cells, strongly suggests that the CMG is a major target of these CMGi. Interestingly, 497 novobiocin can inhibit a different anti-cancer target, DNA Polymerase-theta, which also contains 498 a helicase domain and is involved in DNA repair 49 . The inability of novobiocin to inhibit the CMG 499 demonstrates that these aminocoumarins display target specificity and distinct modes of action. 500 Given the many roles for MCM/CMG complexes in cell cycle regulation, DNA replication, and 501 DNA repair, future CMGi with drug-like qualities and distinct mechanisms of action will provide 502 a novel means for cancer intervention. All original data used in this report will be deposited after acceptance in a publicly available site 516 for examination, managed either through the journal or a third party site. Assistance with data 517 analysis or methodology used in this report can also be obtained by communication with the 518 corresponding author. 519

Declaration of Interests: 520
The authors declare an interest in this research related to a patent submission (Filing Ref# 10110-521 Novobiocin (cat# 46531) and Courmermycin-A1 (cat# C9270) were obtained from Sigma. 534 Etoposide (cat# S1225) and AT7519 (cat# S1524) were obtained from SelleckChem. All stock 535 solutions of inhibitors were stored at -20 o C as 10 mM suspensions in DMSO. 536 537

Cell Viability Assays 538
Cell viability determinations were performed using CellTiter-Glo Assays (Promega; cat# G7572). 539 Cells were seeded in 96-well plates at a density of ~3 x 10 3 cells/well and treated with drugs for 540 72 hr, after which the cells were processed for viability using CellTiter-Glo reagent according to 541 the instructions of the manufacturer. Each drug concentration test was performed using four 542 replicates, and results averaged and plotted on graphs, +/-1 s.d. 543

Immunoblotting and Antibodies 545
Immunoblotting was performed using standard enhanced chemiluminescent (ECL) and 546 polyacrylamide gel techniques. Lysates from equal cell numbers were separated into Triton X- Estimation of hCMG protein amount isolated was determined by comparing to BSA 663 standards. We nominally achieved purification of 5-6 ng/µL hCMG enzyme from three glycerol 664 29 fractions, or ~7.5 fmol hCMG/µL. Across multiple preps this varied from 5-15 fmol hCMG/µL, 665 consistent with that reported by Hurwitz and colleagues 19,20 . The specific ATPase activity of our 666 isolated hCMG enzyme was consistent with the hCMG ATPase activity obtained by Hurwitz and 667 colleagues 19, 20 . This prior hCMG analysis determined that the human helicase hydrolyzes ATP to 668 ADP at a rate of ~80 mol-ADP per minute per mol-hCMG in the presence of 500 µM ATP. Using 669 the ADP-sensing fluorescent-polarization assay (described below), 15 fmol (in 2 µL) of hCMG 670 elicits an ~52% mP change relative to the assay window (see Results), equating to production of 671 ~7-8 µM ADP in 1 hr (80 x 15 fmol x 60 min = 72 pmol ADP in a 10 µL reaction, or 7.2 µM ADP). 672 This activity represents an ~1.5% ADP conversion rate by the hCMG and falls within the reliability 673 range of Z' = 0.6-0.8 for screening in the primary ATPase assay.

Fluorescent-Polarization Measurements with ADP 2 Transcreener Assay 699
The fluorescent-polarization (FP) ADP-sensing assays were performed using the Transcreener® 700 ADP 2 FP assay kit (cat# 3010-1K, Bellbrook Labs). For hCMG inhibitor screening, assays were 701 performed at 37°C for 1 hr in 10 µl reactions in a 384-well plate (cat. #4514, Corning) using 2 µL 702 of hCMG (~15 fmol), 25 mM Hepes-NaOH (pH 7.5), 10 mM NaCl, 0.5 mM ATP, 10 mM 703 magnesium acetate, 1 mM DTT, 0.1 mg/ml BSA, and DNA fork substrates. But assays without 704 DNA substrates can also be performed, since the hCMG does not require DNA to be present to 705 hydrolyze ATP 20 . Selected inhibitors or samples from a chemical library were added into the 706 reactions when conducting the screening or testing inhibitor effectiveness. The NCI Diversity Set 707 VI chemical library was obtained from the National Cancer Institute (Bethesda, MD). 708 The window of sensitivity for the ADP-sensing assays is determined by setting up two 10 709 µL control samples without any added hCMG helicase: a Low-FP mixture (4 nM ADP Alexa 710 Fluor-633 Tracer alone) and a High-FP mixture (4 nM ADP Alexa Fluor-633 Tracer plus patented 711 anti-ADP 2 Antibody). The amount of ADP 2 Antibody used had to be adjusted to account for the 712 use of 500 µM ATP in our hCMG assays, versus 10 µM ATP in enzyme reactions typically 713 assessed with standard kits prepared by BellBrook Labs. The ADP 2 Antibody has a significantly 714 higher affinity for ADP compared to ATP, but since it can bind to ATP to some extent, this must 715 be offset by including more ADP 2 Antibody in our high-ATP assays. This was done according to 716 the manufacturer by performing a titration with increasing ADP 2 Antibody, fixed 4 nM ADP Alexa 717 Fluor-633 Tracer, and 500 µM ATP to determine the ~EC 80-85 for millipol (mP) changes, which 718 determined the optimal ADP 2 Antibody concentration to use as 0.64 mg/ml (Supplementary 719 Figure S2A). Samples are read with a Perkin Elmer Envision II plate reader with optimized Cy5 720 (far-red) FP-compatible mirror and cubes (cat# 2100-8390, Perkin Elmer). The Low-FP sample is 721 the least polarized and gives a low mP reading, while the High-FP sample is the most polarized 722 and gives the highest mP reading. The difference between the Low-FP and High-FP values defines 723 the FP assay window, which is normally in a range of 150-200 mP under ideal conditions for 724 screening purposes. The hCMG helicase hydrolyzes ATP to ADP and decreases the mP reading 725 within this window, with an ideal change (∆mP) in the FP window of at least 50% to be in a 726 readable range for inhibitor screening (as per manufacturer). Potential hCMG chemical inhibitors 727 reverse this effect and cause the mP readings to increase. 728 We performed a titration with increasing and decreasing concentrations of ADP and ATP, 729 respectively, to assess the sensitivity of the FP assay in detecting ADP production 730 (Supplementary Figure S2B). Consistent with the stated manufacturer predictions, the assay can 731 reliably detect 1-3% changes in ADP production (i.e., 5-15 µM ADP production) in starting 732 concentrations of ATP of 500 µM, with a Z' efficiency between 0.6-0.          HaCaT cells were released into the cell cycle and allowed to progress into early S-phase (18 hrs). Cells were then treated with compounds and analyzed 3 hrs later for DNA replication using BrdU labeling. Results are averages from three fields, +/-1s.d. CA1/novobiocin, 15 µM; etoposide, 5 µM; these doses were also used in the following panels. (c) Immunoblots of chromatin-bound or total proteins from synchronized HaCaT cells treated with compounds at 18 hr (early S-phase) and assessed at times indicated. (d) Experimental design (left) for in vitro assessment of CA1 effects on hCMG complexes. A nuclear extract was prepared from synchronized HaCaT (20 hrs after release; middle S-phase) and subjected to immunoprecipitation with anti-Psf1, anti-Mcm2, or IgG control. Samples were separated in half, then treated with DMSO or CA1 (15 µM) for 30 minutes prior to immunoblotting for (co-)precipitated proteins. (e) Experimental design (left) for in vitro assessment of CA1 effects on hCMG and replisome complexes from asynchronous HEK-293T cells stably expressing ectopic Flag-Mcm2. A nuclear extract was prepared and subjected to immunoprecipitation with anti-Flag or IgG as a control. Samples were treated with DMSO or CA1 (15 µM) for 30 minutes and immunoblotted for co-precipitated proteins (or Mcm2).