Structure-Guided Identification of DNMT3B Inhibitors

Abstract

Methyltransferase 3 beta (DNMT3B) inhibitors that interfere with cancer growth are emerging possibilities for treatment of melanoma. Herein we identify small molecule inhibitors of DNMT3B starting from a homology model based on a DNMT3A crystal structure. Virtual screening by docking led to purchase of 15 compounds, among which 5 were found to inhibit the activity of DNMT3B with IC₅₀ values of 13–72 μM in a fluorogenic assay. Eight analogues of 7, 10, and 12 were purchased to provide 2 more active compounds. Compound 11 is particularly notable as it shows good selectivity with no inhibition of DNMT1 and 22 μM potency toward DNMT3B. Following additional de novo design, exploratory synthesis of 17 analogues of 11 delivered 5 additional inhibitors of DNMT3B with the most potent being 33h with an IC₅₀ of 8.0 μM. This result was well confirmed in an ultrahigh-performance liquid chromatography (UHPLC)-based analytical assay, which yielded an IC₅₀ of 4.8 μM. Structure–activity data are rationalized based on computed structures for DNMT3B complexes.

Over the last three decades, basic cancer research has addressed increasingly the epigenetic, rather than genetic, origins of disease.¹ While specific mutations in DNA sequences have long been known as potential causes of tumorogenesis, a growing body of work shows that errors in gene regulation are common in many cancer cells. One well-studied mechanism of gene regulation is through methylation of DNA facilitated by a family of enzymes called DNA methyltransferases (DNMTs, see Figure 1).²

Figure 1.

DNMT1/DNA complex crystal structure from the Protein Data Bank (PDBID: 3PTA).³ The partially buried SAM methyl donor is shown in blue.

The process involves the transfer of a methyl group from an S-adenosyl methionine (SAM) cofactor to the 5-carbon of cytosine at cytosine-guanine dinucleotides in DNA. This chemical modification inhibits the expression of specific genes, which are required for normal cell development, as has been shown in genetic imprinting, cell differentiation, X chromosome inactivation, and silencing of oncogenes. However, overexpression of DNMTs, especially DNA methyltransferase 3B (DNMT3B), and accompanying aberrant DNA methylation patterns have been found in bladder, kidney, colon, and several other types of cancer cells.⁴ A common observation is that cancer genomes exhibit global hypomethylation leading to the expression of normally silenced genes, and simultaneous local hypermethylation in specific regions named CpG islands, silencing normally expressed genes.⁵ CpG islands are regions in the genome with dense populations of CpG dinucleotides, which appear near a majority of gene promoter regions.

Through increased methylation in CpG islands, tumor suppressor genes have been observed to be silenced in cancer cells, contributing to uncontrolled cell growth.⁶ Furthermore, hypermethylation has been related to the enhanced expression of oncogenes via suppression of miRNA transcription.⁷

Studies have shown that cancer cells deprived of DNMTs are not viable, suggesting that active DNMTs are required for cancer cell proliferation.⁸ Inhibition of DNMT1, the maintenance methyltransferase, is lethal in nearly all cell types yet generally does not drive cancer formation. DNMT3B (and to some extent the closely related DNMT3A) is a driver of a variety of cancer types, e.g., acute myelogenous leukemia, colon cancer, and melanoma.⁹

Due to their important role in carcinogenesis, DNMTs are viewed as attractive targets for drug design efforts.⁹⁻¹¹ Nonselective DNA methyltransferase inhibitors, e.g., decitabine in Figure 2, are standards of care for certain forms of leukemia. These triazene nucleoside analogues are covalently incorporated into DNA and deplete DNMTs by covalently linking them to DNA, frequently resulting in double stranded DNA breaks and cell death. These agents are associated with high toxicity, a limited therapeutic window, and incomplete DNMT inhibition.

Figure 2.

DNA methyltransferase inhibitors.

However, based on knockout phenotypes in mice, DNMT3B inhibition is unlikely to lead to problematic toxicity.¹² DNMT3B is important because (a) it is a de novo DNMT, which is not required for the propagation of epigenetic information during normal cell division,² (b) it is the most commonly overexpressed DNA methyltransferase in cancer cells,⁴ and (c) it is indispensable for cancer cell proliferation.^4,8,12 Furthermore, inhibition of DNMT3B has been linked to re-expression of tumor suppressor genes leading to cancer cell death, though the mechanisms behind this effect are still unresolved.^9,10,13,14

We recently identified key genetic changes in melanoma, one of the deadliest forms of skin cancer, that include Braf mutant protein inhibitor therapy and immunotherapy approaches.¹⁵ We concluded that inactivation of DNMT3B (but not DNMT1 or DNMT3A) markedly reduced melanoma formation in the aggressive Braf/Pten model and in human melanoma cells. Thus, DNMT3B and specific miRNAs play critical roles in melanoma and are therefore promising therapeutic targets.

To date, only nanaomycin A has been reported to be a selective inhibitor for DNMT3B, with an IC₅₀ of 0.5 μM,¹³ while the disalicylic acid derivative NSC14778 was identified as a dual DNMT1 (92 μM) and DNMT3B (17 μM) inhibitor.¹¹ Molecular dynamics (MD) simulations of the DNMT3B/nanaomycin A complex showed that the inhibitor favors binding via several hydrogen bonds with catalytic residues in the presence of the SAM cofactor and could involve a covalent binding mechanism.¹⁶ The anthracycline moiety of nanaomycin A, the relatively low LD₅₀ for mice (28 mg/kg), and the CC₅₀ for Vero cells (0.7 μM) are sufficient to raise concern about the potential of nanaomycin A as an effective, nontoxic therapeutic agent.

Thus, alternative DNMT3B-specific inhibitors are of great interest as chemical probes in experiments designed to elucidate the role of DNMT3B in cancer or as a potential anticancer drug. Herein, we report novel, non-nucleoside DNMT3B inhibitors. Virtual screening by docking was applied, with attention paid to the need for desirable pharmacological properties (Figure 3). To date, a crystal structure of DNMT3B has not been reported. Though a homology model was used in a previous virtual screening study,¹¹ the coordinates were not reported and reinvestigation was deemed desirable. Thus, we constructed a new homology model using Prime¹⁷ and the crystal structure of DNMT3A (PDBID: 2QRV),¹⁸ which is 81% identical and 96% similar in the highly conserved catalytic domain. The resultant model appears similar to the prior one in the vicinity of the active site including the positioning of the catalytic residues, Cys651, Glu697, and Arg832. We then ran a virtual screen using Glide SP¹⁹ with the drug-like subset of the ZINC12 Database, which included more than 10 million compounds (10,637,968)²⁰ with desirable ADME (absorption, distribution, metabolism, and excretion) features. The best-ranked ca. 1000 complexes were visualized, and 248 compounds were selected, redocked, and scored with GOLD²¹ and Glide XP (Figure 3).²²

Figure 3.

Schematic of the computational methodology used here.

Consensus hits were identified, and after additional considerations of potential metabolic lability, synthetic ease of preparing analogues, and novelty, compounds 1–15 in Figure 4 were purchased. As detailed in the SI, a fluorogenic assay for DNMT inhibition was performed using a DNA oligonucleotide with CpG sites as the substrate.²³ After methylation, a methylation-specific endonuclease GlaI cleaves the oligonucleotide resulting in a fluorescent signal. As summarized in Table 1, five (7, 10–13) of the 15 purchased compounds were indeed active against DNMT3B with IC₅₀ values of 13–72 μM. Furthermore, only two compounds, 8 and 12, yielded inhibition of DNMT1. It was possible to then purchase eight analogues of 7, 10, and 12, namely, 16, 17–22, and 23, respectively (Figure 4), which were found by substructure search. Only 16 and 19 showed activity, but with IC₅₀ values of 70–80 μM. Nanaomycin A was also found to be a selective inhibitor of DNMT3B in these assays with an IC₅₀ of 1.5 μM, which is similar to the prior report of 0.5 μM in a related assay using tritiated-SAM.¹³

Figure 4.

Structures of the purchased (1–15) obtained from the virtual screening. Compounds 16, 17–22, and 23, analogues of 7, 10, and 12, respectively, were also purchased and tested.

Table 1. Experimentally Half Maximal Inhibitory Concentration (IC₅₀) Values for DNMT3B and DNMT1, Docking Score, and ZINC ID for the Active Compounds Obtained from the Virtual Screening.

	IC50 (μM)
Compound	DNMT3B	DNMT1	Glide SP (kcal/mol)	ZINC ID
7	72.3	NA	–10.71	31776613
8	NA	28.5	–10.26	21655126
10	66.8	NA	–11.07	04753819
11	22.4	NA	–10.74	11331166
12	13.5	27.2	–10.92	14228671
13	37.2	NA	–10.51	12531405
16	70.4	ND	–8.18	31776635
19	80.0	ND	–8.99	84339111
Nanaomycin A	1.5	NA	–5.37
RG108	0.9

At this point, the diarylpropanamide 11 emerged as the most compelling new inhibitor based on the combination of potency, selectivity, and novelty. The docked structure from XP Glide also seems reasonable, as shown in Figure 5A. There is van der Waals contact near the amide group and Arg832, whose putative role in the catalytic mechanism is to hydrogen-bond to the partially negative cytosine carbonyl group after addition of the thiolate of Cys651 to cytosine C6.¹⁶ There are also hydrogen bonds between the lactam carbonyl group and the hydroxyl group of Thr586 and backbone NH of Trp834, and the 4-pyridinyl N and the backbone NH of Val628. In addition, there is the striking aryl–aryl interaction between the quinazolinone bicycle and the indole of Trp834.

Figure 5.

(A) Docking pose from Glide XP for 11 obtained from the virtual screening for DNMT3B. (B) Substituents R and R′ of 11.

It was decided to explore some analogues of 11 with emphasis on alternatives to the quinazolinone fragment, which is designated R′ in Figure 5B. De novo design was performed with the ligand-growing program BOMB.²⁴ The propanamide core was retained, and BOMB was used to build hundreds of analogues with variation of the terminal groups R and R′. All reasonable conformers of each analogue are generated, and the complexes for each one are built and optimized with OPLS force fields. The optimized structures are scored, and among the most promising molecules, a number were selected for preparation. For R, replacement of the 3-(pyridine-4-yl)-1,2,4-oxadiazole by only 2-phenyl-indole, 2-phenyl-adenine, and 2-phenyl-benzimidazole fragments, which modeled well, was considered.

Three compounds in the indole series, 29a–c, were synthesized via the route summarized in Scheme 1. Reductive amination of the indole carbaldehydes 24a–c provided the corresponded amines 25a–c in near quantitative yield. In parallel, a Sandmeyer reaction was performed on commercially available 26, followed by a C2 regioselective Heck coupling to yield 28. The amines 25a–c were then used in dipeptide couplings with the aryl indole 28 to provide the final compounds 29a–c. These three compounds were found to be inactive in the DNMT3B assay, so attention turned to the benzimidazole and adenine series following the route in Scheme 2. A Pybop coupling provided the tertiary amides 31a–j, which underwent Suzuki coupling to yield the final compounds for both the benzimidazole 32a–32d and adenine 33a–33j series. Though the benzimidazoles were inactive, 5 compounds from the adenine series showed good activity as inhibitors of DNMT3B (Table 2). The most active new compound is 33h, with anIC₅₀ of 8.0 μM. Some predicted properties for the active compounds are also shown in Table 2. The computed aqueous solubilities of 10^–3 to 10^–6 M and octanol/water log P values of 1–3 are all in the normal ranges for oral drugs.²⁴

Scheme 1. Indole Core Series.

Reagents and conditions: (a) CH₃NH₂, NaBH₄, CH₃OH, 23 °C, 94–99%; (b) H₂SO₄ (aq), NaNO₂ (aq), KI, H₂O, 23 °C, 92%; (c) Indole, Pd(OAc)₂, dppm, KOAc, H₂O, 110 °C, 50%; (d) 25, HATU, DIPEA, DMF, 23 °C, 52–77%.

Scheme 2. Benzimidazole and Adenine Core Series.

Reagents and conditions: (a) 25a–25j, PyBOP, DMF, 23 °C, 70–87%; (b) 34 or 35, Pd₂(dba)₃, P(Cy)₃, K₃PO₄ (aq), dioxane, 100 °C, 17%–40%.

Table 2. Enzyme Activities for the Active, Synthesized Compounds and Predicted Properties.

Comp	IC50 (DNMT3B) μM	QPlogSa	QPlogPb	ClogPc
33a	45.0	–5.59	2.76	2.86
33f	28.0	–5.20	2.12	2.29
33g	11.0	–4.10	1.59	1.37
33b	10.0	–3.27	1.81	2.35
33h	8.0	–5.56	1.29	0.85

^aQikProp prediction of aqueous solubility, log S. S in mol/L.

^bQikProp and

^cChemDraw predictions of octanol/water partition coefficient.

To confirm the potency of compound 33h, we subsequently performed a Direct Resolution of One Dalton difference (DRONE) assay described recently by Sasaki et al.²⁵ In this assay an ultrahigh-performance liquid chromatography (UHPLC)-based analytical method is used to directly monitor reaction kinetics of DNA modifying enzymes including cytidine deamination and cytosine methylation.

This assay was previously applied to inactivation of DNMT3B by nanaomycin A and gave an IC₅₀ of 730 nM,²⁵ in good accord with the value of 1.5 μM found here in the fluorogenic assay. When the DRONE assay was performed for compound 33h, an IC₅₀ of 4.8 μM was obtained (Figure 6A), again about a factor of 2 lower than in the fluorogenic assay. Thus, confidence can be expressed in the utility of compound 33h as a basis for further lead optimization in this novel heteroaryl-substituted, tertiary-amide series.

Figure 6.

(A) Inhibition of methyltransferase activity²⁵ by the addition of DNMT3B inhibitor 33h. (B) Docking pose from Glide XP for 33h.

The docked structure for 33h with DNMT3B is illustrated in Figure 6B and helps explain some of the SAR data. The amide carbonyl is hydrogen-bonded to Lys777. The E-configuration for the amide appears to be necessary to properly position the thienopyrimidinone fragment for π-stacking with Trp834. Thus, it is not surprising that the secondary amides 33d and 33e, which should predominantly populate the Z-configuration, are inactive. It is also noted that the adenine N1 in 33h is predicted to be hydrogen-bonded to the backbone NH of Val628, analogously to the interaction with the 4-pyridinyl N in 11 (Figure 5A). The absence of this interaction for the indole 29 and benzimidazole 34 series may be an important factor for their observed lack of activity.

In summary, the desire to discover small molecules that selectively target DNMT3B over DNMT1 led us to virtual screen more than 10 million compounds in conjunction with a DNMT3B homology model. Consensus hits led to discovery of multiple active chemotypes including the novel tertiary amide 11, as bases for lead optimization. A kinetic fluorogenic assay provided IC₅₀ results for 25 purchased compounds including nanaomycin A and RG108. Initial exploration of analogues of 11 led to synthesis of 17 new compounds in three series. The activity of the most potent compound 33h was confirmed in an additional ultrahigh-performance liquid chromatography (UHPLC)-based analytical assay. The SAR data was found to be consistent with computed structures for the complexes of the small molecules with DNMT3B.

Glossary

Abbreviations

DNMT

DNA methyltransferase

DNA

deoxyribonucleic acid

RNA

ribonucleic acid

miRNA

MicroRNAs

SAM

S-adenosyl methionine

HATU

1-[bis(dimethylamino)methylene]-1H-1,2,3-triazolo[4,5-b]pyridinium 3-oxide hexafluorophosphate

DIPEA

N,N-diisopropylethylamine

PyBOP

benzotriazol-1-yl-oxytripyrrolidinophosphonium hexafluorophosphate

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsmedchemlett.0c00011.

• Experimental details for the constructed homology-model, virtual screening, and in vitro characterization of DNMT activity. Full synthetic procedures and spectral characterization data for all intermediates and final compounds. (PDF)

• SMILES strings for the new compounds (XLSX)

• A PDB coordinate file for the complex of 33h and DNMT3B (PDB)

The authors declare no competing financial interest.