5-Azacytidine–Induced Methyltransferase-DNA Adducts Block DNA Replication In vivo

Numerous chemicals as well as radiation can lead to covalent protein-DNA adducts, but little is known about the consequences of these adducts in vivo. One important chemotherapeutic agent that leads to protein-DNA adducts is 5-azacytidine (aza-C), which was first synthesized in 1964 (1). Since then, aza-C and its derivatives have been approved for myelodysplastic syndrome, and have proven beneficial for patients with acute and chronic myelogenous leukemia (2–4).

Aza-C is a cytidine analogue in which carbon-5 (C5) of the pyrimidine ring is replaced with nitrogen. Normally, a DNA cytosine-C5 methyltransferase (MTase) acts on a cytosine residue in its recognition sequence by covalently binding to C6, and then transferring the methyl group from S-adenosylmethionine to C5; the covalent protein-DNA linkage is then reversed and the enzyme dissociates from the DNA (5). However, aza-C substitution at the target cytosine interferes with the reaction cycle and results in long-lived or irreversible MTase-DNA adducts (6–8). Evidence for adduct formation in vivo has been presented (9, 10).

A major consequence of aza-C treatment is loss of cytosine MTase activity (11–13). Hypomethylation caused by aza-C can lead to alteration of gene expression, replication timing, and decondensation of chromatin (14–18). Significantly, tumor-suppressor genes can be reactivated by aza-C treatment, and synergistic effects have been seen with a histone deacetylase inhibitor (19). In mammalian cells, aza-C treatment also results in defective tRNAs and rRNAs, and inhibits protein synthesis (20). These RNA effects are avoided with the deoxy version, 5-aza-2′-deoxycytidine (aza-dC; ref. 20).

The consequences of covalent MTase-DNA adduct formation can also be important. Embryonic stem cells and transgenic mice with reduced levels of cytosine MTase were more resistant to aza-dC than the wild-type, suggesting that protein-DNA adduct formation mediates aza-dC cytotoxicity (21). In addition, aza-C sensitivity of some tumor cell lines correlates positively with MTase levels (22). Furthermore, Jackson-Grusby et al. (23) detected a high frequency of C:G to G:C transversion mutations at CpG islands in promotor regions after aza-dC treatment, suggesting that MTase-DNA adducts might be involved in mutagenesis. Aza-dC treatment also causes induction of p53 DNA damage response, proposed to be dependent on formation of the MTase-DNA adducts (24). Finally, aza-C–dependent MTase-DNA adducts have been implicated in chromosomal rearrangements, chromosomal breakages, and bone marrow toxicity (20).

Studies in bacterial cells have contributed to our understanding of aza-C cytotoxicity. E. coli cells deficient in the chromosomally encoded DNA cytosine MTase (Dcm) are less sensitive to aza-C than wild-type (25), whereas overproduction of Dcm or M.EcoRII increases aza-C sensitivity (26, 27). In addition, mutations in the recombination genes recA and recBC cause hypersensitivity to aza-C (25–27). Genetic studies of recA and lexA (SOS repressor) mutants argue that the recombination function of RecA is very important for survival after aza-C treatment, and that induction of the SOS response is protective (25, 27). These results imply that aza-C–induced MTase-DNA adducts play a significant role in cytotoxicity, and that some pathway of RecA-dependent recombinational repair is competent to repair these protein-DNA adducts or downstream DNA damage.

Covalent protein-DNA adducts or tightly bound proteins represent a major challenge to the DNA replication machinery. DNA replication forks can be blocked in vivo by replication termination complexes such as Tus-Ter (28) or by drug-stabilized topoisomerase-DNA covalent complexes (29, 30). Here, we report that aza-C–induced MTase-DNA adducts block plasmid pBR322 DNA replication in vivo. We also observed RecA-dependent X structures after replication blockage, implying that blocked forks are processed by RecA. In addition, electron microscopy revealed both theta and rolling circle (RC) replicating plasmids. The formation of X structures and RCs suggest that the stalled replication forks may be prone to breakage. If so, these “collateral” DNA breaks may represent a pathway that contributes to the cytotoxicity and/or genome instability induced by aza-C.

Materials. Aza-C was obtained from Sigma-Aldrich; restriction enzymes were from New England Biolabs; Nytran membranes were from Schleicher & Schuell; Random Primed DNA Labeling Kit was from Roche Applied Science; and radiolabeled nucleotides were from Perkin-Elmer Life Science. Oligonucleotides used for site-directed mutagenesis were synthesized by Sigma-Aldrich and the QuikChange Site-Directed Mutagenesis Kit was from Stratagene. Luria broth contained Bacto tryptone (10 g/L), yeast extract (5 g/L), and sodium chloride (10 g/L).

Plasmids. The M.EcoRII overexpression plasmid with a tetracycline resistance gene, pR215, was kindly provided by Ashok Bhagwat (Department of Chemistry, Wayne State University, Detroit, MI; ref. 31). Monomeric (pR215, pBR322, and pBR322-C1060A) or dimeric (pBR322 dimer) plasmids were gel purified and used for transformation (see below). pBR322-C1060A plasmid, which was generated using the Stratagene QuikChange Mutagenesis Kit, contains a C to A mutation at location 1060; this destroys one of M.EcoRII recognition sites but does not alter translation of the tetracycline resistance gene.

E. coli strains. E. coli strains were AB1157 (his-4 argE3 leuB6 proA2 thr-1 thi-1 rpsL31 galK2 lacY1 ara-14 xyl-5 mtl-1 kdgK51 supE44 tsx-33 rec+) and AB1157 recA. These two strains were transformed with the M.EcoRII overexpression plasmid, pR215. The desired plasmid(s) were introduced into the AB1157 strains by transformation, and colonies with mostly monomeric plasmid (for pR215, pBR322, and pBR322-C1060A) or mostly dimeric plasmid (for pBR322 dimer) were selected based on miniprep results and frozen in aliquots at −80°C. These aliquots were diluted directly into culture medium to ensure the monomeric or dimeric plasmid state for each experiment.

DNA preparation for two-dimensional gel electrophoresis. E. coli strains with the desired plasmid(s) were grown in Luria broth with appropriate antibiotics to maintain selection for the plasmid(s). Aza-C was added to 0.25 mg/mL when the cultures reached an A₅₆₀ of 0.3. After 3-h incubation with or without aza-C, 2-mL samples were collected by centrifugation and frozen in a dry ice/ethanol bath. Cell pellets were resuspended in 500 μL of Triton lysis buffer [50 mmol/L Tris-HCl (pH 7.8), 10 mmol/L EDTA, 1% Triton X-100, and 1.8 mg/mL lysozyme] and incubated at 65°C for 20 min. Proteinase K (0.5 mg/mL) and SDS (0.2%) were added to the samples, which were then incubated at 55°C for 1 h. DNAs were extracted with phenol/chloroform/isoamyl alcohol (25:24:1) followed by dialysis with TE [10 mmol/L Tris-HCl (pH 7.8), 1 mmol/L EDTA] at 4°C overnight.

Two-dimensional gel electrophoresis and Southern hybridization. The first dimension (0.4% agarose) was run in 0.5× Tris-borate EDTA (TBE) buffer at 1 V/cm for 29 h. The desired slices were cut, rotated 90° counterclockwise, and cast within the top of the second-dimension gel (1% agarose with ethidium bromide at 0.3 μg/mL). The second-dimension gel was run at 4.5 V/cm for 16.5 h at 4°C with recirculated 0.5× TBE containing ethidium bromide at 0.3 μg/mL. The gels were analyzed by Southern hybridization with a ³²P-labeled probe that was made using the Random Primed DNA Labeling Kit and a 729-bp ampicillin resistance gene fragment from pBR322 (which does not hybridize with plasmid pR215). Southern blots were visualized by PhosphorImager.

DNA preparation for electron microscopy. DNA samples for electron microscopy analysis were prepared using a modification of the Hirt precipitation method (32, 33). Briefly, cell pellets from 6 mL of culture were resuspended in 0.5 mL SDS lysis buffer [10 mmol/L Tris (pH 7.6), 10 mmol/L EDTA, and 0.6% SDS] and incubated at room temperature for 15 min. NaCl was added to a final concentration of 1.4 mol/L and the lysate was mixed gently and then incubated at 4°C overnight. The lysate was centrifuged at 13,000× g at 4°C for 40 min to remove cell debris and chromosomal material. The supernatant was collected and treated with RNase A (0.05 mg/mL) at 37°C for 1 h, followed by proteinase K (0.5 mg/mL) at 55°C for 1 h. The DNAs were further purified by phenol/chloroform/isoamyl alcohol (25:24:1) extraction and dialyzed against TE at 4°C overnight.

Electron microscopy. Sample preparation for electron microscopy used the denatured cytochrome C method (Kleinschmidt method) in the droplet form. DNA samples were mixed with ammonium acetate (0.25 mol/L) and cytochrome c (Sigma) at 7 μg/mL and placed on a sheet of parafilm for 90 s to 5 min. The surface film was picked up with a parlodion-covered electron microscopy grid and dehydrated in 70% and 90% ethanol, 30 s each, and air dried. The grids were shadowcast with rotation (8°) with Pt/Pd (80%:20%) at 1 × 10⁻⁶ Torr and examined in Tecnai 12 TEM instrument at 40 kV. Images were collected with a 4K × 4K Gatan CCD camera.

Replication fork blockage in vivo after aza-C treatment. To investigate the potential blockage of DNA replication by aza-C–induced MTase-DNA adducts in vivo, we used two-dimensional agarose gel electrophoresis. The first dimension separates DNA based on size with little contribution from shape, whereas the second dimension separates DNA by both size and shape. Replication intermediates of plasmid pBR322 were visualized by Southern hybridization with a specific probe.

Figure 1 shows the expected replication intermediates and predicted two-dimensional gel patterns of pBR322 linearized by HindIII (Fig. 1A) or AlwNI (Fig. 1B; refs. 34, 35). Plasmid pBR322 replicates unidirectionally from position 2535 bp, generating predominantly bubble (B) molecules and double Y molecules (DY) in these restriction digests. We first analyzed DNA isolated from a pBR322-containing strain that has only the endogenous (chromosomally encoded) Dcm. However, we did not detect any accumulation of stalled forks after growth in the presence of aza-C (Fig. 2B and G). The activity of endogenous Dcm protein is weak (26); thus, we reasoned that MTase-DNA adducts formed with Dcm might be insufficient for detecting stalled forks. We therefore generated a strain containing both pBR322 and plasmid pR215, which overexpresses M.EcoRII; M.EcoRII recognizes the same sequence as Dcm (CCWGG; second C is methylated). With overexpression of M.EcoRII, we readily detected three strong spots on the bubble arc (Fig. 2D, and I, closed arrows). These accumulations on the bubble arc were dependent on the presence of pR215 and aza-C treatment (compare the aza-C–treated samples in Fig. 2D and I with the untreated samples in Fig. 2C and H). If aza-C–induced MTase-DNA adducts block DNA replication, then we would expect four classes of bubble molecules (branches near 131, 1,060, 1,443, and 2,501 bp) and two classes of double-Y molecules (branches near 2,622 and 2,635 bp). However, the bubbles near the 2,501 bp site would be very small, and these molecules should therefore migrate very near the linear monomer, which likely prevents their detection. In addition, the two classes of double-Y molecules would be obscured by a strong accumulation of late replicating intermediates in this region of the gel even in the absence of aza-C (compare the double-Y region in Fig. 2C and D).

To determine whether the fork blockage evident on the bubble arc is due to aza-C–induced MTase-DNA adducts, we mutated CCATT to CAATT at the 1,060 bp M.EcoRII recognition site. The mutated plasmid (pBR322-C1060A) was confirmed by restriction enzyme digestion and DNA sequencing (Supplementary Fig. S1; data not shown). The bubble spot consistent with the 1,060 bp site disappeared in both the HindIII and AlwNI digests of the mutated plasmid (Fig. 2E  and J, closed arrows). Therefore, we conclude that aza-C–induced MTase-DNA adducts block DNA replication in vivo.

RecA-dependent formation of X structures. We were surprised to find an accumulation of X structures in the two-dimensional gels, dependent on both M.EcoRII overexpression and aza-C treatment (Fig. 2D and I). We next introduced pR215 and pBR322 into a recA⁻ derivative of AB1157 and analyzed the DNA from aza-C–treated cells. The X structures disappeared from the recA⁻ strain (Fig. 3A  and B, arrow), whereas bubble accumulations were still present. Accumulation of the X structures is therefore RecA dependent.

Wild-type cells transformed with monomeric plasmid still contain a small percentage of dimer plasmids due to RecA-dependent recombination. Thus, the absence of X structures in Fig. 3B could be an indirect consequence of RecA deficiency. Therefore, we created wild-type and recA⁻ strains containing pR215 monomer and pBR322 dimer plasmids. The X structures were again absent from the recA⁻ strain (compare Fig. 3C  with D, arrow). Therefore, accumulation of the X structures is directly dependent on RecA, presumably reflecting either RecA-dependent recombination or RecA-dependent fork regression (see Discussion).

Formation of Y molecules is due to RC replication. Another unexpected pattern from the two-dimensional gels was the accumulation of spots on the Y arc. The Y-spot accumulations were dependent on M.EcoRII overexpression and aza-C treatment (Fig. 2D and I), and much more dramatic with plasmid dimers than monomers (compare spots on the Y arc in Fig. 3C and A). In addition, one of the Y spots disappeared from the mutated pBR322-C1060A plasmid (compare Fig. 2D  with E, open arrows), implying that the Y spots depended on aza-C–induced MTase-DNA adducts.

At least three possible mechanisms can create Y molecules. First, theta replication within a dimer pBR322 creates bubbles if replication is blocked before the first restriction digestion site but Y molecules if replication is blocked after that site (Fig. 4A). According to this interpretation, we would expect a roughly equal intensity for the spots on the Y and bubble arcs from the cells carrying the dimer plasmid. However, we detected much more intense spots on the Y arc. The second possible mechanism is replication blockage within a RC replication structure (Fig. 4B). RC plasmid replication usually does not occur in E. coli because the exonuclease function of RecBCD destroys the RC tail (36, 37). However, the exonuclease activity of RecBCD may be inhibited in aza-C–treated cells. Induction of chromosomal DNA damage, for example, by UV, seems to saturate the exonuclease activity (38). Alternatively, RecBCD translocation might be inhibited by aza-C–induced MTase-DNA adducts on the RC tail. The third possible mechanism for creating Y molecules also involves RC replication, namely blockage of theta replication within the tail of the RCs (Fig. 4C). Blockage of theta replication within the tail would result in a similar set of theta and Y molecules as described above for the first model, but the Y molecules could predominate.

To distinguish between these mechanisms and further analyze aza-C–induced replication blockage, we set out to use electron microscopy to visualize the intermediates. DNA was prepared from aza-C–treated cells that contained dimeric pBR322 and monomeric pR215 plasmid. We used a modification of the Hirt isolation method (32, 33) to separate nonreplicating plasmids and replicating plasmid intermediates from the chromosomal DNA. This procedure precipitates chromosomal DNA, whereas plasmid DNAs remain in the supernatant. We analyzed the efficiency of the Hirt procedure by subjecting the two DNA fractions to both one-dimensional and two-dimensional gels. When analyzed by one-dimensional gel with ethidium bromide staining, roughly 90% to 95% of the chromosomal DNA was precipitated into the pellet, whereas 90% of the plasmid circles remained in the supernatant. Meanwhile, two-dimensional gel analysis of HindIII-digested DNA with Southern blotting showed that ∼65% of the replication intermediates were pelleted, whereas ∼35% remained in the supernatant (Supplementary Fig. S2D and A, respectively).

Although we recovered less than half the replication intermediates in the supernatant, the elimination of chromosomal DNA was critical for clear interpretations of the electron microscopy images. Therefore, we analyzed the DNA from the supernatant by electron microscopy. For each replication intermediate visualized, we measured the lengths of the DNA segments and compared the circle length to the measured length of nonreplicating control pR215 and pBR322 dimer plasmids (see Fig. 5 legend for measurements of molecules in that figure).

We observed many nonreplicating circular pBR322 dimers and several theta structures (Fig. 5A; Table 1). In addition, we found eight unambiguous RCs (Fig. 5C) with a tail length ranging from 1.4 to 2.5 times larger than the attached circle size. Out of these eight RCs, four had a circle consistent with the size of pBR322 dimer, whereas four others had a larger plasmid circle than expected, resembling either trimeric pBR322 or dimeric pR215. Furthermore, we scored five replicating theta forms (Fig. 5B) and four RCs (Fig. 5D) for pR215 monomer plasmid, which reinforces the conclusion that RC replication occurs in the presence of aza-C. We also found a relatively large number of plasmids with small tails (Table 1), which could result from either broken thetas or RCs with short or broken tails. We observed some molecules with apparent X structure in the electron microscopy. However, these are not definitive because they could be either bona fide X structures or two DNA molecules that happen to overlap on the grid. Overall, the electron microscopy data strongly support the hypothesis that aza-C induces a switch from theta to RC in a subset of molecules.

As an additional test for the existence of RC intermediates, we analyzed the Hirt pellet and supernatant fractions by two-dimensional gel analysis without restriction enzyme digestion. RC intermediates produced from a well-characterized in vitro replication system have been shown to migrate in a characteristic “ds eyebrow” arc that emanates from the open circle spot (39). In both the supernatant and pellet, DNA from the aza-C–treated cells (but not the untreated cells) displayed this ds eyebrow arc, providing additional evidence that aza-C induces plasmid RC replication (Supplementary Fig. S2C and F). The aza-C–dependent ds eyebrow emanated from the dimer open circle spot, not from the monomer, indicating that the RC molecules have a dimeric rather than monomeric template circle (consistent with the electron microscopy results above).

The major conclusion of this work is that aza-C–induced MTase-DNA adducts block DNA replication in vivo. We did not detect blockage induced by the endogenous Dcm protein, presumably due to low endogenous Dcm activity (26). However, fork blockage was evident after aza-C treatment of cells carrying the M.EcoRII overexpression plasmid pR215. Previous studies provided evidence for the formation of covalent aza-C–induced MTase-DNA adducts in vivo (9, 10), and our results argue that a major effect of these adducts is the blockage of DNA replication.

In addition to fork blockage within theta intermediates (Fig. 2), we detected strong accumulation of spots on the Y arc, particularly when cells carried dimeric pBR322 plasmid (Fig. 3). These accumulations were also dependent on aza-C–induced MTase-DNA adducts (Fig. 2E), and our results argue that these Y molecules are created by fork blockage in a RC. Using electron microscopy analysis, we found both replicating thetas and sigma-shaped plasmids with tails. Sigma-shaped plasmids would correspond to the Y molecules in the two-dimensional gels after restriction digestion. A number of the sigma molecules in the electron microscopy had tails longer than the length of the attached plasmid circle, demonstrating that RC replication had occurred (Fig. 5; Table 1). We strongly suspect that RCs with even longer tails were preferentially lost in the Hirt precipitation, which eliminates long linear DNA. The simplest explanation for the formation of RCs in the aza-C–treated cells involves breakage of the stalled forks (see below).

We also observed a strong accumulation of RecA-dependent X structures after aza-C treatment of cells with pR215. The blocked forks on the theta and Y arcs appeared as early as 30 min, but X structures only appeared after 60 min of aza-C treatment (data not shown). This result argues that the X structures are somehow generated from molecules with blocked forks. One model is that these X structures are Holliday junctions generated by a recombination reaction. Replication forks stalled by aza-C–induced MTase-DNA adducts may sometimes be processed into strand breaks that can then be used as a substrate for RecA-dependent recombination. Because recombination mutants including recA are hypersensitive to aza-C (26, 27), the Holliday junctions we observe may reflect a key repair pathway for survival after aza-C treatment.

Another model for X structure formation is RecA-dependent replication fork regression of a stalled RC. Robu et al. (40) showed that RecA protein can catalyze fork regression in vitro, and Courcelle et al. (41) presented evidence consistent with RecA-stabilized fork regression in vivo after UV-induced damage. Because of the repeating nature of a RC tail, regression of a RC intermediate can generate an X structure that is exactly twice the plasmid length after restriction digestion; that is, exactly equivalent in structure to a plasmid dimer with a Holliday junction. In summary, the generation of Holliday junctions after aza-C treatment implies that DNA ends are created either by fork breakage or by fork regression.

We propose that RCs are sometimes generated after a unidirectional fork in a theta molecule encounters an aza-C–induced MTase-DNA adduct. Replication fork blockage can allow a branch-specific endonuclease to cleave one arm of a fork (30, 42), and then two general models for RC replication are possible. In one model, RC replication ensues after gap repair/ligation at the broken fork and replication restart at the unbroken fork (Supplementary Fig. S3). The second model involves RecA-dependent recombination triggered by the broken fork. A simple version of this model involves intermolecular recombination of the broken end with a second circle, resulting in a “double-circle RC” with circular plasmids at both ends (Supplementary Fig. S4). In this model, the RC replication itself is activated upon loading of a replisome onto the displacement loop created in the strand invasion reaction. None of the RC molecules that we observed in the electron microscopy had circles at both ends as predicted by this simple recombination model. However, such double-circle RCs could have enormously long intervening linear DNA after RC replication, making them difficult to isolate intact. Further experiments are clearly needed to elucidate the molecular mechanism that triggers RC replication after aza-C treatment.

Several prior studies argued that aza-C–induced MTase-DNA adducts are important in cytotoxicity (see Introduction). Prior studies also suggested that aza-C treatment leads to DNA breaks, likely dependent on formation of the MTase-DNA adducts. Karon and Benedict (43) detected chromatin breaks after treatment of mammalian cells with aza-C, although the mechanism was unexplored. Furthermore, the bacterial SOS response is induced by aza-C (25, 27), and recombination mutants (recA and recBC) are hypersensitive to the analogue (25–27).

As described above, our observation of RC intermediates and Holliday junctions after aza-C treatment strongly suggests that DNA breaks are formed from blocked replication forks. We refer to this kind of DNA breakage as “collateral damage,” because it is an indirect consequence of the primary mechanism of inhibition. We propose that this collateral damage could be a major factor in the cytotoxicity of aza-C treatment and that it could be involved in generation of both DNA damage responses and chromosomal rearrangements induced by aza-C.

Drug-stabilized topoisomerase-DNA complexes are somewhat analogous to aza-C–induced covalent MTase-DNA adducts. A variety of antitumor drugs and antibacterial quinolones have been shown to stabilize covalent topoisomerase-DNA complexes involving either type I or type II topoisomerases, including bacterial DNA gyrase (44, 45). In several cases, these complexes have been shown to block replication forks, and these blocked forks can be cleaved by endonucleases (29, 30, 42, 46). Therefore, fork blockage and collateral DNA damage may be a common mechanism for chemotherapy.

Grant support: NIH grants GM72089 (K.N. Kreuzer) and GM31819 (J.D. Griffith).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Ashok Bhagwat for many useful discussions and for providing plasmid pR215, and Smaranda Willcox in the Griffith laboratory for excellent assistance in the electron microscopy experiments.