New Insights on the Mechanism of Quinoline-based DNA Methyltransferase Inhibitors*

Background: 4-Aminoquinoline SGI-1027 and analogs inhibit DNA methylation, which is deregulated in cancers. Results: These compounds induce deviations from Michaelis-Menten equations in DNA competition experiments and interact with DNA. Conclusion: They are competitive inhibitors for the DNA substrate of the DNA methyltransferase and non-competitive for the methyl group donor, S-adenosyl-l-methionine. Significance: These findings suggest a mechanism of inhibition for these 4-aminoquinoline-based DNMT inhibitors.

Initially synthesized as part of a minor-groove binders family of quinolinium bisquaternary salts, SGI-1027 inhibits bacterial DNA methyltransferase M.SssI, human DNMT1, mouse Dnmt3A, and mouse Dnmt3B (9). It is currently the reference compound in several DNMT inhibition assays (10,11) and structure-activity relationship studies (12). Therefore, there is an actual interest in elucidating its molecular mechanism of action.
Two groups performed competition studies on DNMT1 (9,10) and concluded that SGI-1027 was a AdoMet-competitive and DNA non-competitive inhibitor of DNMT1. Here, we studied the mechanism of inhibition of full-length human DNMT1 by SGI-1027 and two analogs that we recently synthesized (compounds 5 and 31 in Fig. 1, described by Valente et al. (13) and Rilova et al. (14), respectively). In contrast to previously reported data (9,10), our findings clearly support a behavior as DNA competitive and AdoMet non-competitive inhibitors. The ability of the compounds to interact with DNA and DNMT1 was investigated to further characterize the mechanism of action using compound 19 (Fig. 1) as a negative control as it did not succeed to inhibit either DNMT1 or human catalytic DNMT3A (DNMT3Acat) (14). Several hypotheses are described, and the differences with the literature are discussed.

EXPERIMENTAL PROCEDURES
General-All commercially available reagents and solvents were purchased from Sigma, and radioactive [methyl-3 H]AdoMet was from PerkinElmer Life Sciences. SGI-1027, compounds 19 and 31, and compound 5 were synthesized as described in Refs. 9, 14, and 13, respectively. 10 mM stock solutions were prepared in DMSO and aliquoted. The compounds were named according to the nomenclature of the respective articles.
DNMT1 Competition Assays-Competition assays on fulllength DNMT1 were realized according to Gros et al. (16). Briefly, the tested compound, biotinylated duplex, [methyl-3 H]AdoMet and DNMT1 were incubated for 2 h in 10 l at 37°C. An aliquot of 8 l was then transferred in a Flashplate TM well containing 190 l of 20 M SAH solution in Tris-HCl. The Flashplate TM was agitated at room temperature for 1 h, washed 3 times with 200 l of 0.05% Tween 20 in Tris-HCl, and read in 200 l of Tris-HCl on TopCount NXT.
In AdoMet competition assays, [methyl-3 H]AdoMet was varied between 0.5 and 15 M at a fixed DNA duplex concentration of 0.6 M. For each AdoMet concentration the tested compound concentration was adjusted between its IC 10 and IC 80 . For each compound concentration, the Michaelis-Menten model was fitted by non-linear regression to the data, and K m app and V m app were calculated according to this model. For Lineweaver-Burk plots, a linear model was fitted by linear regression to the transformed data. Lineweaver-Burk or double-reciprocal plots were only used as a graphic representation to distinguish competitive, non-competitive, and uncompetitive inhibitor. Noteworthy, this graphic representation is known to distort and magnify the errors of the data.
In DNA-competition assays, the DNA duplex concentration was varied between 0.05 and 0.6 M, whereas [methyl-3 H]AdoMet concentration was held at 15 M. For each DNA duplex concentration, the tested compound concentration was adjusted between its IC 10 and IC 80 values. For each compound concentration, the Copeland and Horiuchi (17) non-competitive, uncompetitive, and competitive models were fitted by non-linear regression to the data. For each molecule, the only convergent model at each concentration was the Copeland and Horiuchi (17) competitive model. In the Lineweaver-Burk plots, the double reciprocal of the Copeland and Horiuchi competitive model (17) was fitted by non-linear regression to the transformed data.
In both AdoMet and DNA competition assays, for each substrate concentration, the IC 50 of the tested compound was calculated by non-linear regression fitting with sigmoidal doseresponse (variable slope) with constrained top and bottom at 100 and 0% of inhibition, respectively. All the non-linear and linear regressions were performed on GraphPad Prism 4.03 (GraphPad Software).
T m Assay-DNA thermal denaturation experiments were conducted as described in Mergny and Lacroix (18). Hairpin DNA duplexes hp_2_CG (5Ј-TATATACGTACGGTGTT-TTCACCGTACGTATATA-3Ј) containing 2 CpG sites, hp_ 1_CG (5Ј-TATATACGTACTGTGTTTTCACAGTACGTA-TATA-3Ј) containing 1 CpG site, and hp_0_CG (5Ј-TATATA-TGTACTGTGTTTTCACAGTACATATATA-3Ј) containing no CpG site were used at 2 M in the absence or presence of the inhibitor in the T m assay buffer (100 mM NaCl, lithium cacodylate 20 mM, pH 7.2). The temperature at which 50% of the duplex is denatured, T m , was calculated as previously described (18). Means of at least two experiments with the corresponding S.E. are reported.
DNase I Footprinting-DNase I footprinting experiments were performed essentially as described in Lemster et al. (19) and Racané et al. (20). Briefly, the 117-and 265-bp DNA fragments were obtained from EcoRI and PvuII double digestion of the pBS plasmid (Stratagene, La Jolla, CA). The generated DNA fragments was 3Ј-end-labeled for 30 min at 37°C using 10 units of Klenow enzyme (New England BioLabs) and [␣-32 P]dATP (3000Ci/mmol, PerkinElmer Life Sciences) before isolation on a 6% polyacrylamide gel under native conditions. The radiolabeled 117-and 265-bp DNA fragments were cut off from the gel, crushed, dialyzed overnight against 400 l of elution buffer (10 mM Tris-HCl, pH 8.0, 1 mM EDTA, 100 mM NaCl), and then separated from polyacrylamide gel by filtration through a Millipore 0.22-m membrane followed by ethanol precipitation.
Appropriate concentrations of the various tested compounds were incubated with the 117-or 265-bp radiolabeled DNA fragments for 15 min at 37°C to ensure equilibrium before the addition of 1 unit/l of DNase I in appropriate buffer for 3 min of digestion. The reaction was stopped by ethanol precipitation. The digested DNAs were subsequently dissolved in 4 l of denaturing loading buffer (80% formamide solution containing tracking dyes), heated for 4 min at 90°C, and chilled 4 min on ice before electrophoresis for 90 min at 65 watts on a 8% denaturing polyacrylamide gel in Tris/borate/EDTA buffer. Finally, gels were soaked in 10% acetic acid, transferred to Whatman No. 3MM paper to be dried under vacuum at 80°C, and exposed overnight at room temperature on phosphor-imaging storage screens. The identity of the bases from each DNA fragment was established from comparison of the relative position of the bands to the guanine sequencing standard (G-track) classically obtained using dimethyl sulfate and piperidine treatment of the same DNA fragment.
Differential Scanning Fluorimetry Assay-Experiments were conducted using a CFX384 TM Real-Time System (C1000 Thermal cycler, Bio-Rad CFX Manager 2.0 Software, Bio-Rad). The samples were heated at 0.1°C/s, from 10 to 80°C. The fluorescence intensity was plotted as a function of the temperature. T1 ⁄ 2 was given by the inflection point of the fluorescence curve. ⌬T1 ⁄ 2 was calculated by subtracting the t1 ⁄ 2 in the absence of the compound to the T1 ⁄ 2 in the presence of the compound in the same condition (i.e. in the presence of other partners DNA and/or AdoMet).
The protein was scanned to assess suitability of the method, and the lowest concentration of DNMT1 protein needed to generate a strong signal was determined to be 2.5 M. Com-pound concentrations varied between 5 and 200 M. The DNA duplex used in the enzymatic assays was chosen and added at 5 M. The AdoMet cofactor was added at a final concentration of 5 M or not. The SYPRO orange dye (Invitrogen) was diluted to 1/400th in each sample. Each experiment was repeated for at least two times in duplicate. Means of at least two experiments are displayed with the corresponding S.E.
DNA Duplex and DNMT1 Gel Shift Assays-4 l of compounds or 1% of DMSO, 0.5 M 6-carboxyfluorescein 5Ј-labeled DNA duplex used in the DNMT1 enzymatic assay, and 5 M AdoMet were added in each well and incubated for 30 min at 37°C with or without 2.5 M DNMT1. Samples were loaded on a 0.7% agarose gel and migrated for 50 min at 135 V in 1ϫ Tris/borate/EDTA electrophoresis buffer. Gels were analyzed with the Typhoon trio of GE Healthcare with appropriated filters.

SGI-1027 and Its Analogs Share DNA-competitive and
AdoMet Non-competitive Behavior on DNMT1-We recently developed a new DNMT1 inhibition assay in homogenous phase that allows to study the mechanism of action of potent inhibitors of DNMT1 by carrying out enzymatic studies (16). We decided here to apply it to the study of 4-aminoquinoline SGI-1027, a well described inhibitor of DNMTs, and two analogs that we have recently synthesized, 31 and 5 ( Fig. 1)  (13, 14). Interestingly, the latter is a position isomer of SGI-  Table 1) and of DNMT3Acat (IC 50 of 0.8, 0.9, and 0.7 M, respectively; Fig. 1 and Table 1) (14). The assay that we developed was particularly suitable to carry out competition experiments on DNMT1, which methylates DNA with higher yields than DNMT3Acat. In addition, we dispose of the full-length human DNMT1, containing all the important domains for the interaction of the enzyme with DNA and with the inhibitors (21), whereas we only disposed of the C-terminal domain of DNMT3A (DNMT3Acat). Therefore, full-length DNMT1 was chosen to perform all further competition experiments.
First, we studied the behavior of the inhibitors to compete with the DNA substrate of the enzyme in the DNMT1 enzymatic test. To analyze whether the compounds were DNA competitors or not, we performed non-linear regression with the three models (competitive, non-competitive, and uncompetitive inhibition) described by Copeland and Horiuchi (17). Interestingly, DNA competition studies displayed unexpected results (Fig. 2). Velocity plots against substrate concentration did not follow the Michaelis-Menten behavior and presented a sigmoidal character that was particularly significant at high inhibitor concentrations (Fig. 2, A-C). This phenomenon resulted in a deformation of the Lineweaver-Burk plots (Fig. 2, D-F). Indeed, at high inhibitor concentrations, points no longer displayed a linear behavior but rather an up-turning parabolic character that could be approximated with a quadratic function. Copeland and Horiuchi (17) explained these features as being characteristic of competitive inhibitors that might interact with DNA.
In agreement with these competitive behaviors (22), the measured IC 50 of the compounds increased as the DNA concentration increased (data not shown). However, in the case of substrate-inhibitor interactions, the IC 50 analysis is not sufficient to conclude whether SGI-1027 and its analogs are DNA competitors or not (17). Thus, velocity plots, doublereciprocal plots, and IC 50 against [DNA]/K m DNA plots were analyzed; all suggested a DNA competitive inhibition of DNMT1 ( Fig. 2 and data not shown). Moreover, concerning compound 5, the sigmoid is so stretched to appear as aligned points (Fig. 2B). This is a mathematical limit of the Copeland and Horiuchi (17) model that occurs when nearly all the inhibitor in solution is complexed to the substrate, in our case the DNA, leading to an apparent straight line. This suggested that compound 5 is a stronger DNA binder than 31 and SGI-1027.
In contrast, in AdoMet competition studies, each compound presented a Michaelis-Menten behavior with hyperbolic velocity plots against the substrate concentration (Fig. 3, A-C). Moreover, this was confirmed by the double-reciprocal plots ( Fig. 3D-F). As the K m app remained constant and the 1/V m app increased with increasing inhibitor concentrations (data not shown), SGI-1027 and its analogs resulted in AdoMet noncompetitive inhibitors of DNMT1. In addition, the overall unchanged IC 50 regardless of the AdoMet concentration confirmed this hypothesis (Fig. 3, G-I) (22). Thus, our results showed that SGI-1027 and its analogs display AdoMet non-competitive behaviors in agreement with the Michaelis-Menten model in experimental conditions in which the DNA concentration is constant. On the other hand, the three compounds seemed to be DNA competitors with deviations from the Michaelis-Menten equation at high inhibitor concentrations. As this phenomenon might be characteristic of the inhibitor interacting with the DNA, we next evaluated this feature by using parent compound 19 as a negative control as it inhibits neither DNMT1 nor DNMT3Acat.

SGI-1027
Interacts Weakly with DNA, and Compound 5 Binds Strongly to DNA-To study the interaction between the compounds and DNA, we followed thermal denaturation of short DNA duplexes by UV absorbance in the absence and in the presence of the compounds. Three hairpin duplexes were chosen containing no CpG site (hp_0_CG, T m ϭ 68.0°C), 1 CpG site (hp_1_CG, T m ϭ 71.0°C), and 2 CpG sites (hp_2_CG, T m ϭ 74.5°C). Fig. 4, A-C, and Table 1 report the differences observed in T m (⌬T m ) for the DNA duplexes when in the presence of 19, SGI-1027, 31, and 5. No significant increase in the T m value was observed when the negative control 19 was incubated with any of the DNA duplexes, whereas SGI-1027 showed a slight increase of the T m value when incubated with hp_1_CG (⌬T m ϭ 1.0°C). The data with compounds 31 showed a higher S.E. because the molecule was less soluble in DMSO. It showed a concentration effect on the T m of the duplexes (data not shown), suggesting that the compound interacts weakly with DNA. Derivative 5 resulted to be a strong DNA ligand (⌬T m ϭ 4.0°C, 3.0°C, and 1.5°C for hp_0_CG, hp_1_CG, and hp_2_CG, respectively). The interaction of 5 with the three DNA duplexes

and DNMT3Acat inhibition activity of SGI-1027 and its analogs and differences in the T m values of the duplexes in the presence and in the absence of the compounds (⌬T m )
The concentrations (M) at which 50% of the methylation activity is inhibited (IC 50 ) are reported for DNMT3Acat and DNMT1. For 19, the percentages of inhibition are displayed. The difference in T m (⌬T m in°C) in the presence of 10 M of inhibitors and in their absence is reported for each duplex. Means of at least two experiments with the corresponding standard errors are reported. The means of measured T m are displayed in parentheses (°C).

IC 50
⌬T m (measured T m in°C) at 10 M was also monitored at 375 nm, a wavelength at which the ligand absorbs, allowing the monitoring of specific changes related to the ligand (23). Inverted transitions were observed for the three duplexes with the same temperature dependence as that observed at 260 nm. This inverted transition for duplex hp_2 with compound 5 is shown in Fig. 4D and is representative of experiments with the other duplexes (data not shown). This transition was not observed with SGI-1027 and the other compounds (data not shown). This indicates that, upon duplex melting (as observed at 260 nm), the absorbance properties of 5 are concomitantly modified, strongly supporting that 5 interacts with DNA. Hence, the most potent inhibitor of full-length DNMT1 is also the strongest DNA binder (Table 1). Next, for compounds that interact with DNA, an interesting issue is the selectivity of the compounds toward the DNA sequence. This point could not be addressed upon use of the three hairpin duplexes because they melt at different temperatures. Thus we addressed it by DNase I footprinting analysis.

DNMT3Acat
This technique can show where the compound binds on a DNA fragment of a known sequence. Two DNA sequences were analyzed: a 265-bp DNA fragment containing some AT-rich sites and two CpG sites and a 117-bp DNA fragment with several CpG sites in a less AT-rich environment (Fig. 5).
Both DNase I footprinting gels and densitometric analyses confirmed that SGI-1027 binds weakly to DNA showing some noise (as poor and nonspecific DNA binding) in the differential cleavage pattern, as does compound 19. Compound 31 partially protected the DNA from DNase I digestion at concentrations up to 2.5 M, with a preference for A/T stretches on the 265-bp DNA fragment (Fig. 5, A and C) but shifting to CpG binding (positions 37-38 and 50 -51) on the 117-bp DNA fragment. However, the protection was weak. Noteworthy, for both SGI-1027 and 31, we could not test higher concentrations than the reported ones as it resulted in cleavage inhibition (smearing), which interfered with sequence selectivity analysis.
Compound 5 clearly showed much stronger footprints surrounded by huge cleavage enhancements as evidenced on the gels by strongly digested bands (Fig. 5, A and C) and on the densitometric analyses by positive differential cleavage calculated relatively to control lanes 0 in the absence of the compounds (Fig. 5, B and D). Clearly, in agreement with the T m analysis, 5 is the strongest DNA binder of the series showing binding to CpG dinucleotides (positions 37-38 and 50 -51, Fig. 5D; position 88 -89 evidenced at 10 M, Fig. 5B). The strong DNase I cleavage enhancements observed may reflect some distortion of the DNA helix induced by the binding of 5 (24).
Both DNA duplex thermal denaturation and DNase I footprints showed that 19 does not interact with DNA and that SGI-1027 and 31 display weak interaction. In contrast, 5 is clearly the strongest DNA ligand and the most potent inhibitor of DNMT1. Next, we investigated whether the compounds directly interact with DNMT1 by using thermal shift analysis of the protein unfolding temperature.
Compound 5 Interacts with DNMT1 Only When the DNA Is Present-In the fluorescence-based thermal shift assay (differential scanning fluorimetry), the ability of a molecule to stabilize or destabilize the protein during its thermal unfolding is quantified by its thermal shift (⌬T1 ⁄ 2 ): the difference in the pro- tein unfolding temperature in the presence and absence of the ligand.
The DNMT1 unfolding temperature was determined to be 46.8 Ϯ 0.1°C in absence of any ligand. No variation or ⌬T1 ⁄ 2 of 3.4 Ϯ 0.2°C was observed when DNMT1 was incubated with 5 M AdoMet or 5 M DNA, respectively (data not shown). Finally, in the presence of both DNA and AdoMet, the ⌬T1 ⁄ 2 was 6.1 Ϯ 0.4°C suggesting a cooperation between the two ligands (data not shown). In the presence of 200 M inhibitor 5 (Fig. 6A) no variation was measured in the absence of DNA and AdoMet; in contrast a strong destabilization was observed when the DNA and AdoMet were present in the solution (⌬T1 ⁄ 2 ϭ Ϫ6.6 Ϯ 0.2°C). Thus compound 5 did not directly interact with DNMT1 but needed the formation of the DNMT1-AdoMet-DNA enzymatic complex. This observation was confirmed when each substrate was added separately (Fig. 6B). In fact, a variation in the T1 ⁄ 2 of DNMT1 was only observed when the DNA duplex was present in addition to compound 5. Noteworthy, the destabilization was strongest when AdoMet was also present. In addition, a dose-response curve was obtained for the ⌬T1 ⁄ 2 of the enzymatic complex with the increase in compound 5 concentration (Fig. 6C). Similar results were obtained on DNMT3Acat (data not shown).
The other compounds interacted weakly with DNMT1 alone or in the presence of its two substrates (DNA and AdoMet), with the exception of SGI-1027, which showed a strong destabilization in the presence of the ternary complex (⌬T1 ⁄2 ϭ Ϫ7.1 Ϯ 0.2°C, Fig. 6A). Compounds 19 and SGI-1027 presented a weak destabilizing effect on DNMT1 alone (Fig. 6A) that could be observed also on the ternary complex.
Noteworthy, in agreement with the AdoMet non-competitive behavior observed in the enzymatic studies for SGI-1027, the destabilization of the DNMT1 and of the DNMT1-AdoMet-DNA complexes was little dependent on the AdoMet concentration (Fig. 6D). The slight effect observed mainly arises from the stabilizing effect of increasing concentration of AdoMet on the DNMT1-DNA complex (data not shown).
Compound 5 Destabilizes the DNMT1-DNA Complex-To further explore the destabilization of the enzymatic complex by the compounds, DNMT1 was complexed on a fluorescein-labeled DNA duplex in the absence or in the presence of the different compounds and migrated on an agarose gel (Fig.  7A). Only compound 5 destabilized the DNMT1-DNA complex with the appearance of the free duplex as the concentration of the compound increased. In addition, heat denaturing experiments at 40°C showed that the DNMT1-DNA complex is dissociated faster in the presence of compound 5 (Fig. 7B).

DISCUSSION
Our data indicate that SGI-1027 and its two analogs, 5 and 31, are inhibitors of DNMT3Acat and DNMT1. Compound 5 is the most potent inhibitor of DNMT1 among the three. These compounds behaved as AdoMet non-competi-tive inhibitors of DNMT1 and, rather, displayed non-Michaelian behaviors in DNA competition studies on DNMT1, which is characteristic of DNA competitive inhibition with DNA-inhibitor interactions (17). These results are in disagreement with previous reports describing a AdoMet competitor behavior for SGI-1027 (9, 10). Datta et al. (9) analyzed only parts of the velocity plots against AdoMet concentration; a plateau was not reached in their experimental conditions rendering the interpretation more difficult. Fagan et al. (10) conducted competition experiments on a truncated form of DNMT1 (residues: 621-1600 amino acids) missing the replication focus targeting sequence (RFTS) domain. Importantly, this domain was previously described to be an endogenous DNA competitive inhibitor of DNMT1 (25) and to bind to the DNA binding pocket of DNMT1 (26). Thus, it is not surprising that DNA competition studies on truncated DNMT1 are different from ours carried out on the full-length enzyme. Regarding the deviations from the Michaelis-Menten model observed in our DNA competition experiments, they can also result from other causes than substrate-inhibitor interactions. For example, they can be the consequences of tight binding or time-dependent inhibition. We cannot exclude these other hypotheses. However, we experimentally observed that the compounds interact with DNA and in particular compound 5, supporting this inhibition mechanism. Indeed compound 5, the most potent inhibitor (Table 1) on DNMT1, is a strong DNA binder (Table 1 and Fig.  4), whereas SGI-1027 and 31 interact only weakly. These derivatives present a preference for A/T stretches, G/C-rich stretches, and certain CpG sites in the 117-bp DNA fragment studied here (Fig. 5). Interestingly, the change from the para to the meta bonds from SGI-1027 to 5 increased the DNA binding properties, suggesting a better fitting in the DNA groove (27). Finally, for compound 5, we were able to establish that this molecule only interacts with DNMT when the DNA is present (Fig. 6) and induces a destabilization of the DNMT1-AdoMet- DNA complex (Fig. 7). SGI-1027 was also shown to be able to destabilized the DNMT1 in the presence of DNA and AdoMet by differential scanning fluorimetry (Fig. 6).
In conclusion, both enzymatic assays and biophysical studies indicate that the most potent inhibitor of DNMT1, compound 5, inhibits DNMT by interacting with DNA and destabilizing the enzymatic complex.