**2. Inhibitors of DNA Glycosylases: General Considerations**

The example of PARP inhibitors highlights the concept of synthetic lethality, which underlies most of the attempts to develop BER inhibitors into practically useful drugs. Two conditions must be satisfied for such compounds to be effective. First, the target cells must experience genotoxic stress, induced either directly by DNA-damaging factors or indirectly through some kind of metabolic stress (nucleotide pool imbalance, oxidative stress, etc.). Second, if DNA damage caused by this type of stress can be repaired bypassing BER, the bypass must be blocked by mutations or another drug used in combination with a BER inhibitor. The second requirement is often satisfied in cancers, where mutations in DNA repair genes are usually among driver mutations. Genotoxic treatment may also be tuned to produce lesions repaired predominantly by BER (such as uracil accumulated through antimetabolite treatment, uracil analogs used as drugs or prodrugs, or oxidized purines appearing through MTH1 inhibition or induced by photodynamic therapy), in which case BER impairment alone could be sufficient to effect considerable cytotoxicity.

Two considerations are crucial when assessing the cytotoxic potential of DNA glycosylase inhibitors. First, unlike the enzymes underlying common BER steps, such as break signaling by PARPs or AP site cleavage by AP endonucleases, DNA glycosylases are specific for damaged bases, and their inhibition will affect only a subset of BER reactions. This in fact may be advantageous for fine-tuning or selection of concurrently used DNA damaging agents, many of which produce specific primary lesions rather than AP sites of strand breaks [16]. Second, DNA glycosylases are often ambivalent with respect to cell-killing effects of DNA damage (as discussed below in sections about specific types of lesions and their repair): they may either counteract the damage by repairing the induced lesions or potentiate the damage by converting damaged bases to AP sites or strand breaks, which are generally more cytotoxic. Thus, the inhibition of DNA glycosylases is not always warranted for inducing synthetic lethality in cancer cells or bacteria. It is always desirable to validate a particular DNA glycosylase as a drug target by knockout or knockdown approaches in an appropriate cell line or pathogen.

The inhibitors discussed in the remaining parts of this paper are mostly small-molecule compounds. Almost all DNA glycosylases are inhibited to a certain degree by non-specific single- or double-stranded DNA, competing for binding with substrate DNA [17–21], and a number of modified nucleotides tightly bound but not cleaved when incorporated into oligonucleotides have been described [22–28]. Moreover, binding and inhibition of DNA glycosylases by polyanions such as heparin [29–32] likely stems from the ability of these enzymes to bind nucleic acids. Minor-groove ligands of various chemical nature also interfere with DNA glycosylase binding [33,34]. Despite the obvious importance of such interactions for the biological functions of DNA glycosylases, delivery and targeting problems thus far prevent the therapeutic use of oligonucleotides and other macromolecular polyanions as mass action-driven inhibitors of intracellular enzymes. However, one strategy known for a while and recently applied to DNA glycosylases is the use of prodrugs that are metabolized to nucleoside triphosphate analogs and incorporated into DNA [35,36]. For example, 10 -cyano-20 -deoxyuridine triphosphate is a good substrate for DNA polymerases and, when incorporated into DNA, inhibits *E. coli* uracil-DNA glycosylase (Ung) and human UNG, displaying nanomolar *K*<sup>i</sup> values [37]. Interestingly, some lesions, such as 2-deoxyribonolactone and 5-hydroxy-5-methylhydantoin [27,38,39], demonstrate an intrinsic ability to trap bifunctional DNA glycosylases covalently on DNA, reminiscent of the PARP-trapping potential of cytotoxic PARP inhibitors. Thus, the development of nucleotides that can be incorporated into DNA and trap DNA glycosylases may be an interesting direction of the glycosylase inhibitor design.

## **3. Uracil in DNA: Synergism of Glycosylase Inhibitors and Antimetabolites**

Antimetabolites, the class of drugs interfering with nucleotide metabolism pathways and thereby with DNA or RNA synthesis, are one of the staples of therapeutic interventions against cancer and bacterial and protozoan infections and are especially useful in combination therapy [40,41]. Many clinically used antimetabolites, such as antifolates, interfere with thymine biosynthesis and cause the accumulation of uracil (or its analogs) in genomic DNA [42,43]. The repair of drug-induced genomic uracil is a double-edge sword: while it protects cells from the effect of this non-canonical nucleobase at low levels of substitution, extensive uracil buildup and excision are toxic and may be the primary reason of cell death after exposure to antifolates [44,45]. Therefore, the inhibition of uracil repair may have different consequences depending on the level of DNA modification and, possibly, on the nature of the modification (if different from uracil).

Human cells possess four DNA glycosylases capable of excising uracil from DNA. However, for three of them (TDG, SMUG1, and MBD4), uracil either is not the main substrate or is removed only from specific contexts (for example, methylated CpG dinucleotides). The main enzyme responsible for uracil repair, UNG, is among the most important factors limiting the efficiency of antifolates and fludarabine, whose action is based on the accumulation of uracil in genomic DNA [46–48]. *UNG* knockdown in human prostate cancer cell lines increases their sensitivity to H2O<sup>2</sup> and doxorubicin [49]. Non-small cell carcinoma and lung adenocarcinoma cells develop spontaneous resistance to pemetrexed, a folic acid analog inhibiting dihydrofolate reductase, thymidylate synthase, and glycinamide ribonucleotide formyltransferase, due to a significant increase in the level of UNG, and the suppression of *UNG* expression returns the sensitivity to normal [50,51]. Some uracil analogs that are accumulated in DNA (such as 5-fluorouracil) are more toxic for cells when SMUG1, rather than UNG, is downregulated [52,53]. However, due to the structural similarity between UNG, TDG, and SMUG1, low-molecular-weight inhibitors will most likely be active against all three enzymes; therefore, the nature of the glycosylase that removes uracil during treatment with antimetabolites is not of primary importance.

UNG inhibitors in combination with genotoxic stress effectively suppress the growth of *Plasmodium falciparum* [54], *Trypanosoma brucei* [55,56], and *Trypanosoma cruzi* [57,58], which makes BER a promising target for drug intervention in protozoan infections. Importantly, some inhibitors preferentially suppress the activity of UNG from infections agents but have little effect on the human enzyme (Table 2).

Low-molecular-weight UNG inhibitors are still at a preclinical stage. The literature describes three main classes of such inhibitors. All of them are competitive, mechanism-based and mimic certain features of the transition state of the UNG-catalyzed reaction [6,79]. Free uracil and its analogs inhibit UNG enzymes from various sources, presenting submillimolar to millimolar IC<sup>50</sup> [59,80–85], so successful inhibitors required extensive modification of the base to ensure tight binding. Historically, the first class of compounds active against human *Plasmodium* and herpes simplex type 1 UNGs was 6-(*p*-alkylanilino)uracils, of which 6-(*p*-n-octylanilino)uracil showed the strongest affinity, with IC<sup>50</sup> ~8 µM for the viral enzyme [54,59,86–88] (Table 2). Bipartite inhibitors, structurally similar to the 6-substituted derivatives, consist of a uracil base or its analog linked to a phenolic or benzoic fragment [60–62,89] (Table 2). In the structures of bipartite inhibitors bound to UNG, the uracil part occupies the uracil-binding pocket of the enzyme, while the aromatic fragment lies in the DNA-binding groove [61,62] (Table 3). Finally, triskelion inhibitors contain three functional groups at the ends of a branched linker: either one analog of uracil and two aromatic fragments, or two analogs of uracil and one aromatic fragment [63] (Table 2). Interestingly, gentamicin, a clinically used aminoglycoside antibiotic, was reported to inhibit *E. coli* Ung [64,65] (Table 2). Although the reported IC<sup>50</sup> value was quite high (0.4–1.5 mM), this effect may reflect interactions between Ung and the sugar part of DNA [21,90] and suggests another possible direction for inhibitor development.


*Int. J. Mol. Sci.* **2020**, *21*, 3118


*Int. J. Mol. Sci.* **2020**, *21*, 3118


*Int. J. Mol. Sci.* **2020**, *21*, 3118

**Table 2.** *Cont*.


283

SU0383



**Table 3.** Known structures of DNA glycosylases bound to their inhibitors.

Thymine–DNA glycosylase (TDG) has recently been validated as a possible drug target in melanoma: its knockdown causes cell cycle arrest, senescence, and cell death in melanoma cell lines but not in normal cells and prevents tumor growth in a xenograft model [91]. Screening of several mid-scale compound libraries yielded about 40 inhibitors with a variety of structures and IC<sup>50</sup> >10 µM [91].

Uracil-DNA glycosylases from poxviruses (D4 according to vaccinia virus naming convention) provide a unique drug target. While they possess quite efficient uracil-removing activity, the main role of these enzymes is not in DNA repair, but in viral replication. D4 binds the A20 protein to form a processivity factor for poxviral DNA polymerases [92,93]. Deletion of the D4 gene causes a sharp drop in the ability of vaccinia virus to replicate in cells [94–96]. Polycyclic aromatic compounds that disrupt the D4/A20 binding interface are considered a promising class of antiviral drugs active against poxviruses [66,97,98] (Table 2).

Finally, a natural Ung inhibitor, the Ugi protein, is produced by PBS1 and PBS2 bacteriophages [99]. Although Ugi is not considered a therapeutically promising candidate, it has recently found an unexpected use in cell technologies involving CRISPR/Cas9 genome editing. A new generation of Cas9-based tools employs base editors, in which a Cas9 targeting module is fused with a cytosine deaminase to generate C→T mutations [100,101]. Repair by UNG counteracts uracil-mediated targeted mutagenesis, so co-expression of Ugi is commonly used to increase the efficiency of this gene editing procedure [100–102].

## **4. Oxidative Damage Repair: Key to Antibiotic Resistance?**

It has been shown that BER is necessary for the survival of certain pathogenic and opportunistic bacteria (*Mycobacterium*, *Neisseria*, *Pseudomonas*, *Salmonella*) under conditions of genotoxic stress caused either by drugs or by the body's immune response [103–105]. Recently, it was discovered that oxidative stress significantly contributes to the death of bacteria exposed to antibiotics of several classes. Topoisomerase inhibitors, β-lactam antibiotics, membrane-permeabilizing agents, and aminoglycosides induce the generation of hydroxyl radicals in several divergent bacterial species through an iron-dependent Fenton reaction, increasing the lethality of these drugs [106–109], whereas reducing agents such as H2S or NO protect bacteria from a wide range of antibiotics [110,111]. Possible reasons for the enhanced cell death include translation errors due to oxidative RNA damage [112], oxidation of the nucleotide pool followed by massive chromosome breakage at the sites of damaged nucleotide incorporation [113–115], and direct DNA damage by reactive antibiotic molecules, their metabolites, or reactive oxygen species [116,117]. Based on these observations, the systems of antioxidant defense and oxidative damage repair in bacteria are now regarded as promising targets for sensitization towards bactericidal antibiotics, which, if successful, can be a breakthrough in the current antibiotic resistance crisis.

In bacteria, several DNA glycosylases are responsible for oxidative damage repair. In *E. coli*, the best-studied enzymatic system, termed the "GO system" (for Guanine Oxidation), involves three enzymes: Fpg (MutM), MutY, and MutT, which have complementary functions in countering the mutagenic effect of 8-oxoguanine (oxoG) [118–120]. OxoG is an abundant DNA lesion that easily forms stable oxoG(*syn*):A(*anti*) Hoogsteen-type pairs, leading to characteristic G:C→T:A transversions [121,122]. Fpg is a DNA glycosylase that excises oxoG from pairs with C; such pairs appear when G is directly oxidized in DNA or when oxodGMP is incorporated opposite C during replication [123,124]. If oxoG remains in DNA and directs dAMP misincorporation, the excision of oxoG by Fpg would lead to a G→T transversion. To safeguard the cell from this mutagenic route, Fpg does not cleave oxoG:A mispairs, which are recognized by MutY, and A is excised instead of oxoG [125]. If repair DNA polymerases insert the correct dCMP opposite oxoG, the second round of repair is carried out by Fpg; otherwise, dAMP is inserted, and MutY-initiated repair is reinitiated. The third enzyme of the system, MutT, hydrolyses oxodGTP and oxoGTP to monophosphates to prevent oxoG incorporation from the oxidized nucleotide pool [126,127]. *E. coli* also possesses a homolog of Fpg, endonuclease VIII (Nei), which is not considered part of the GO system and preferentially excises oxidized pyrimidines with little opposite-base preference, although it has some activity against oxoG in vitro and prevents G:C→T:A mutations in the absence of Fpg [128–131]. Finally, endonuclease III (Nth) also removes a wide variety of oxidized pyrimidine bases [132,133].

Although the GO system has been extensively characterized in *E. coli*, little is known about its functions and the properties of its components in pathogenic bacterial species. Fpg proteins from *Salmonella enterica* [134], *Neisseria meningitidis* [135], and *Corynebacterium pseudotuberculosis* [136] have been cloned and subjected to limited biochemical characterization, which showed essentially Fpg-like properties. Several Fpg homologs from *Mycobacterium tuberculosis* and *Mycobacterium smegmatis* were characterized and found to have divergent substrate specificities resembling either *E. coli* Fpg or Nei [137–140]. For MutY, limited enzyme characterization has been done for proteins from *N. meningitidis* [141], *Helicobacter pylori* [142], and *C. pseudotuberculosis* [143,144]. The presence of a fully functional GO system with its characteristic antimutator pattern has been confirmed in vivo in their native bacterial cells for *Pseudomonas aeruginosa* [145,146], *N. meningitides* [104,135,147], *M. smegmatis* [137,148,149], and *Staphylococcus aureus* [150]. Fpg was found to be functional in vivo in *S. enterica* [134], and MutY in *H. pylori* [142].

The available information about the relevance of the GO system for the pathogenicity of bacteria shows its dual role. On one hand, it seems that this line of defense indeed assists successful primary infection. MutY deficiency has been shown to compromise mouse stomach colonization by *H. pylori* [142]. Successful macrophage infection by *Brucella abortus* requires intact *fpg* and *mutY* genes [151], and *M. tuberculosis* Fpg and Nei are required for lung colonization in a rhesus macaque model [152]. Hypervirulent *Neisseria* isolates maintain functional *fpg* and *mutY* despite having a general mutator phenotype [147]. On the other hand, hypermutability associated with GO system inactivation sometimes provides the variance for selection of highly virulent or drug-resistant strains [153–156]. This underscores the importance of a thorough characterization of the GO system for a given pathogen to assess it as a possible drug combination target.

Human homologs of Fpg and Nei (NEIL1, NEIL2, and NEIL3) are significantly different from the bacterial proteins in their sequence and structure, making realistic the development of small ligands selectively targeting the bacterial enzymes. MutY and Nth appear to be less selective targets.

Few specific inhibitors of the bacterial GO system have been reported. Fpg is weakly inhibited by free damaged base analogs such as 2,6-diamino-4-oxo-5-(*N*-methylformamido)pyrimidine, 5-nitroso-2,4,6-triaminopyridine, and 5-nitroso-4,6-diamino-2-oxopyridine [67,157]. The base analog with the strongest ability to suppress Fpg activity (IC<sup>50</sup> ~10–100 µM) is 2-thioxanthine (Table 2); however, it is not a true inhibitor but rather a reagent that oxidizes cysteines in the zinc finger in Fpg, Nei, and NEIL2 [67,68]. MutY from *C. pseudotuberculosis* was reported to be sensitive to suramin, an antiprotozoan and anthelmintic drug that acts intracellularly [32] (Table 2); however, the relevant drug uptake pathways are likely absent in bacteria. Interestingly, Fpg is strongly inhibited by Cibacron Blue F3GA, a dye bearing structural resemblance to suramin [69] (Table 2). Several papers described the inactivation of Fpg, Nth, OGG1, NEIL1, and NEIL2 by NO-producing agents and suggested damage to redox-sensitive groups in the enzyme molecules [158–162]. However, the mechanisms of this reaction remain unclear, since the known redox-sensitive groups in these glycosylases are different or absent altogether. A screen of a natural product library revealed several inhibitors of *M. tuberculosis* Nei2, the best of which, norlobaric acid, has *K*<sup>i</sup> = 74 nM [70] (Table 2).
