Mechanistic Insights into Harmine-Mediated Inhibition of Human DNA Methyltransferases and Prostate Cancer Cell Growth

Abstract

Mammalian DNA methyltransferases (DNMTs), including DNMT1, DNMT3A, and DNMT3B, are key DNA methylation enzymes and play important roles in gene expression regulation. Dysregulation of DNMTs is linked to various diseases and carcinogenesis, and therefore except for the two approved anticancer azanucleoside drugs, various non-nucleoside DNMT inhibitors have been identified and reported. However, the underlying mechanisms for the inhibitory activity of these non-nucleoside inhibitors still remain largely unknown. Here, we systematically tested and compared the inhibition activities of five non-nucleoside inhibitors toward the three human DNMTs. We found that harmine and nanaomycin A blocked the methyltransferase activity of DNMT3A and DNMT3B more efficiently than resveratrol, EGCG, and RG108. We further determined the crystal structure of harmine in complex with the catalytic domain of the DNMT3B-DNMT3L tetramer revealing that harmine binds at the adenine cavity of the SAM-binding pocket in DNMT3B. Our kinetics assays confirm that harmine competes with SAM to competitively inhibit DNMT3B-3L activity with a K_i of 6.6 μM. Cell-based studies further show that harmine treatment inhibits castration-resistant prostate cancer cell (CRPC) proliferation with an IC₅₀ of ∼14 μM. The CPRC cells treated with harmine resulted in reactivating silenced hypermethylated genes compared to the untreated cells, and harmine cooperated with an androgen antagonist, bicalutamide, to effectively inhibit the proliferation of CRPC cells. Our study thus reveals, for the first time, the inhibitory mechanism of harmine on DNMTs and highlights new strategies for developing novel DNMT inhibitors for cancer treatment.

Introduction

DNA methylation is one of the key epigenetic modifications that affects the outcome of gene expression without altering the underlying DNA sequence.¹ Mammalian DNA methylation occurs at cytosine residues via two de novo DNA methyltransferases, DNMT3A and DNMT3B, and is maintained by DNMT1.² These three DNMTs contain a similar catalytic methyltransferase domain (CD) which can transfer a methyl group from the cofactor S-adenosyl methionine (SAM) to cytosines at the C5 position.³ Emerging evidence indicates that dysregulation of DNMTs by overexpression, mutations, and deletion is linked to various diseases and tumorigenesis.^4,5 DNMT1, DNMT3A, and DNMT3B may be overexpressed in colon, prostate, breast, liver, colorectal, and leukemic cancers.⁶ These tumor cells generally exhibit localized hypermethylation in the promoter regions of tumor-suppressor genes that silence their expression.⁷ Since overexpression of DNMTs and hypermethylation of tumor suppressor genes are well established as key players in cancer, DNMTs have been identified as promising therapeutic targets for cancer treatment.

Two DNMT inhibitors, azacitidine and decitabine, have been approved by the FDA (U.S. Food and Drug Administration) and EMA (European Medicines Agency) to treat hematological cancers, including acute myeloid leukemia (AML), myelodysplastic syndromes (MDS), and chronic myelomonocytic leukemia (CMML).⁸ These two azanucleoside compounds are cytidine analogues that are incorporated into DNA (and RNA for azacitidine) strands during DNA replication in the S phase of rapidly proliferating cells. Incorporation of azanucleoside into DNA (and RNA) induces a DNA damage response, resulting in degradation of the trapped DNMT, because the modified cytidine forms a covalent bond with the conserved cysteine residue in the active site of DNMTs.⁹ After suppression of aberrantly overexpressed DNMTs, DNA demethylation patterns at the promoter regions of tumor suppressor genes can be restored, leading to normal cell growth and differentiation. Despite these two DNMT inhibitors being the most widely used epigenetic drugs, their application for oncological diseases is restricted by their severe toxicity, poor bioavailability, and low specificity.¹⁰ Thus, developing novel compounds that noncovalently bind to DNMTs with greater specificity and lower toxicity remains an important medical unmet need.

Various new classes of non-nucleoside inhibitors against DNMT catalytic activity have been developed using high-throughput virtual screening and enzymatic screening methods.^11,12 Applying a docking-based virtual screening approach against DNMT1, a group of non-nucleoside compounds were isolated from the National Cancer Institute (NCI) database, including RG108 and nanaomycin A.¹³⁻¹⁵ After enzymatic assays, the results showed that these compounds and their derivatives could inhibit DNMTs and reduced the methylation levels of some gene promoters in cancer cells.¹⁶ Novel non-nucleoside DNMT inhibitors have also been identified by means of high-throughput enzymatic screening methods using chemical libraries of 114,¹⁷ 1120¹⁸ and 180 000¹⁹ compounds. Natural compounds and analogues identified in those studies included EGCG and harmine, representing promising compounds for epigenetic drug development.²⁰ Recently, new groups of DNMT1 inhibitors have been identified based on molecular modeling studies, including quinoline-based derivatives.²¹⁻²⁵ Moreover, the crystal structure of DNMT1 in complex with a DNA-intercalating compound GSK3685032 was reported, revealing how it penetrates into hemimethylated DNA between two CpG base pairs, competing the active-site loop.^26,27 However, the molecular mechanism of a large number of non-nucleoside DNMT inhibitors still remains elusive, greatly hampering the development of epigenetic drugs that target DNMT enzymes.

In this study, we selected five reportedly effective non-nucleoside inhibitors (harmine, nanaomycin A, resveratrol, EGCG, and RG108) toward DNMT1, DNMT3A, or DNMT3B to examine systematically their effects on inhibiting all of the three human DNMTs. Harmine was originally isolated from the medicinal herb Peganum harmala, which exhibits a diverse range of pharmacological properties, including antitumor, anti-inflammatory, antioxidant, neuroprotective, antidepressive, antiviral, and antimicrobial activities.^28,29 A library screening identified harmine as an inhibitor of murine DNMT3a,¹⁸ with an IC₅₀ of 2.3 μM. It was also shown to down-regulate DNMT1 expression in a leukemia cell line, leading to hypomethylation and reactivation of the p15 tumor suppressor gene promoter.³⁰ Nanaomycin A is a quinone antibiotic isolated from Streptomyces(31) that inhibits DNMT3B (IC₅₀ = 0.5 μM) by reactivating transcription of the RASSF1A tumor suppressor gene, and it was found to inhibit cell growth in three different cancer cell lines, i.e., HCT116 (colon), A549 (lung), and HL60 (bone marrow).³² EGCG is one of the most studied natural polyphenols in green tea, exhibiting inhibitory activity on human DNMT1 (IC₅₀ = 0.47 μM).^33,34 It can reactivate tumor suppressor genes, reducing DNMT3B expression and inducing growth inhibition and apoptosis of HeLa cells.^35,36 Resveratrol is another natural polyphenol from plants that has been identified as a potential DNMT inhibitor based on its structural resemblance to NSC14778, a compound previously identified as a DNMT inhibitor by virtual screening.³⁷ Resveratrol was found to inhibit DNMT3B (IC₅₀ = 65 μM) and DNMT3A (IC₅₀ = 105 μM), but it had no inhibitory activity on DNMT1.³⁷ Resveratrol has been shown to reduce expression levels of DNMT3B and DNMT1 in breast cancer and to reactivate silenced tumor suppressor genes that may reverse breast cancer progression and metastasis.^38,39

Here, we show that four of these five non-nucleoside compounds—harmine, nanaomycin A, EGCG, and resveratrol—not only inhibit human DNMT1 but also DNMT3A and DNMT3B, yet to different extents. Of these inhibitors, harmine and nanaomycin A exhibit the greatest inhibitory activities against DNMT3A and DNMT3B. The crystal structure of the catalytic domain of DNMT3B–DNMT3L in complex with harmine reveals that harmine is bound at the adenine cavity in the SAM-binding site of DNMT3B. Through kinetic analysis, we confirm that harmine outcompetes cofactor SAM for binding, thereby inhibiting the methyltransferase activity of DNMT3B. Moreover, our cell-based study shows that harmine not only inhibits cell growth and the methylation levels of the RASSF1A tumor suppressor gene promoter, but it also reactivates a number of hypermethylated silenced genes in prostate cancer cells. Our study thus reveals the inhibitory mechanism by which a non-nucleoside DNMT inhibitor operates and provides a framework for rational design of non-nucleoside inhibitors as epigenetic drugs.

Results

Inhibition of DNMT1, DNMT3A, and DNMT3B by Non-Nucleoside Inhibitors

To investigate the inhibitory effects of selected non-nucleoside DNMT inhibitors, we expressed the catalytic domain of human DNMT1 (residues 646–1616), DNMT3A (residues 627–912), and DNMT3B (residues 571–853) in BL21-type bacterial cells (Figure 1A). We also expressed the catalytic-like domain of DNMT3L (residues 179–379) since it may form a heterotetrameric complex with DNMT3A and DNMT3B, respectively, to stabilize their conformation and enhance their activity.⁴⁰⁻⁴² We reconstituted the tetrameric DNMT3A-3L and DNMT3B-3L complexes and purified the truncated DNMT1, DNMT3A-3L, and DNMT3B-3L variants by chromatographic methods (see the elution profiles from the size-exclusion chromatography in Figure 1B and SDS-PAGE in Figure 1C). The molecular weight determined by size exclusion chromatography-coupled multiangle light scattering (SEC-MALS) for DNMT1 was 111.9 kDa (theoretical monomeric MW: 109.6 kDa); for DNMT3A-3L and DNMT3B-3L, they were 103.2 kDa (theoretical tetrameric MW: 113.3 kDa) and 104.9 kDa (theoretical tetrameric MW: 112.5 kDa), respectively (Figure 1D–F).

Figure 1.

Expression and purification of the catalytic domains of human DNMT1, DNMT3A-3L, and DNMT3B-3L. (A) Domain arrangements of human DNMT1, DNMT3A, DNMT3B, and DNMT3L. Apart from the catalytic domain, DNMT1 contains DMAP (DNA-methyltransferase-associated protein), RFTS (replication focus targeting sequence), and BAH (bromo-adjacent homology) domains, and both DNMT3A and DNMT3B harbor PWWP (proline–tryptophan–tryptophan–proline) and ADD (ATRX-DNMT3-DNMT3L) domains. DNMT3L contains an inactive catalytic-like domain (CLD) and an ADD domain. (B) Elution profiles of DNMT1, DNMT3A-3L, and DNMT3B-3L via size-exclusion chromatography (HiLoad 16/60 Superdex 200, GE Healthcare). Protein markers: A, aldolase, 158 kDa; B, conalbumin, 75 kDa; C, ovalbumin, 44 kDa; D, carbonic anhydrate, 29 kDa. Oligomeric states of DNMT1, DNMT3A-3L, or DNMT3B-3L are indicated. (C) SDS-PAGE analysis of purified DNMT1, DNMT3A-3L, and DNMT3B-3L reveals high homogeneity for each recombinant protein. M, molecular weight marker. (D–F) The molecular weights of DNMT1 (111.9 kDa), DNMT3A-3L (103.2 kDa), and DNMT3B-3L (104.9 kDa) were determined by size exclusion chromatography-coupled multiangle light scattering (SEC-MALS).

Next, we measured the inhibitory activities of the five selected non-nucleoside inhibitors—harmine, nanaomycin A, resveratrol, EGCG, and RG108—against the three human DNMTs. Methyltransferase activities of the truncated DNMT1, DNMT3A-3L, and DNMT3B-3L variants were measured using a MTase-Glo methyltransferase assay kit (Promega) to monitor the formation of the S-adenosyl homocysteine (SAH) reaction products. The calculated half maximal inhibitory concentration (IC₅₀) values were derived from the dose-dependent inhibition by each inhibitor. Notably, harmine, nanaomycin A, resveratrol, and EGCG were capable of inhibiting the methyltransferase activities of DNMT1, DNMT3A, and DNMT3B (Figure 2A–D). However, no significant inhibitory effect was observed for RG108 (Figure 2E), consistent with several studies reporting a weak activity of RG108 toward DNMT1 (IC₅₀ value of 390 μM) and murine DNMT3A (IC50 value > 500 μM).^17,43,44

Figure 2.

Inhibition of DNMT methyltransferase activities by non-nucleoside inhibitors. Dose response curves for the methyltransferase activities of DNMT1, DNMT3A-3L, and DNMT3B-3L reveal IC₅₀ values in the presence of 0.3–400 μM of (A) harmine, (B) nanaomycin A, (C) resveratrol, (D) EGCG, and (E) RG-108. Methyltransferase activities of truncated DNMT1, DNMT3A-3L, and DNMT3B-3L were measured using an MTase-Glo methyltransferase assay kit (Promega). Chemical structures of each inhibitor are shown in the plots. Data represent mean ± SD from three independent experiments.

Among the four effective inhibitors, harmine and nanaomycin A presented the strongest inhibitory effects, with IC₅₀ values of ∼13 μM or less for DNMT3A-3L and DNMT3B-3L (harmine: IC₅₀ = 9.38 ± 2.1 μM for DNMT3A-3L, IC₅₀ = 12.46 ± 2.3 μM for DNMT3B-3L; nanaomycin A: IC₅₀ = 9.78 ± 2.64 μM for DNMT3A-3L, IC₅₀ = 5.09 ± 0.28 μM for DNMT3B-3L; Figure 2A and B). Since nanaomycin A inhibited both DNMT3A and DNMT3B, it is not a specific DNMT3B inhibitor, as reported previously.³² Resveratrol and EGCG both exhibited low inhibitory activities, with IC₅₀ values of 40–200 μM against the three DNMTs. Both of these latter ones displayed similar inhibitory activities on DNMT3B-3L (IC₅₀ of 44.95 ± 5.27 μM and 54.05 ± 8.39 μM, respectively; Figure 2C and D). Resveratrol displayed stronger inhibition of DNMT3B-3L than DNMT3A-3L and DNMT1, consistent with the data published previously.⁴⁵ In summary, our analysis indicates that harmine, nanaomycin A, resveratrol, and EGCG all exhibit inhibitory activities against human DNMTs, with harmine and nanaomycin A more effectively blocking the activities of DNMT3A and DNMT3B than resveratrol or EGCG.

Harmine Directly Binds to DNMT3B-3L

To elucidate the underlying inhibitory mechanism of the selected DNMT inhibitors, we screened the cocrystallization conditions for human DNMT3B-3L with each inhibitor or soaked the DNMT3B-3L crystals with each inhibitor. However, we only succeeded in cocrystallizing harmine with DNMT3B-3L, with the cocrystals diffracting X-rays to a resolution of 3 to 4 Å. The orthorhombic crystal of the human DNMT3B-3L-harmine complex belonged to the space group C222₁, with cell dimensions isomorphous to the structure of the DNMT3B-3L-SAH complex reported previously (PDB code 6KDP). Unexpectedly, the refined protein structure revealed an “apo-form” of DNMT3B-3L without the cofactor SAH, and it did not show harmine binding at the active site (data not shown). To probe the effects of different harmine concentrations, we performed cocrystallization of DNMT3B-3L with different molar ratios of harmine and then subjected the crystals to X-ray diffraction. We observed that the electron density of SAH at SAM-binding sites gradually eroded with increasing concentrations of harmine, with the density signal for SAH almost disappearing at a high harmine concentration (Figure S1). This result is consistent with a previous report in which the bacterial DNMT1 homologue M.HhaI in complex with RG108 also presented an empty SAM-binding pocket.⁴⁵

To examine if harmine disrupts the DNMT3B-3L structure to inactivate its activity, we measured the circular dichroism (CD) spectra of the DNMT3B-3L complex in the absence or presence of harmine. The resulting CD spectra revealed that the DNMT3B-3L complex alone exhibits an α/β-type folding pattern, with K2D3 software⁴⁶ estimating 32.5 ± 0.8% α-helix and 16.7 ± 1.6% β-strand. However, CD spectra remained relatively unchanged upon incubating the DNMT3B-3L complex with harmine (31.3 ± 1.0% α-helix, 19.5 ± 0.8% β-strand; Figure 3A). Thus, harmine binding does not appear to alter folding of the DNMT3B-3L complex.

Figure 3.

Harmine directly binds DNMT3B-3L. (A) Representative circular dichroism spectra of the DNMT3B-3L complex in the presence or absence of 1 μM harmine (blue and red, respectively) and harmine alone (black). The CD spectra were recorded at 25 °C with 0.1 μM DNMT3B-3L complex in CD buffer (20 mM phosphate buffer, pH 7.4) from 260 to 190 nm. The inset shows predicted secondary structure compositions (%) as mean ± SD derived from three replicates. (B) Representative intrinsic tryptophan fluorescence spectra of DNMT3B-3L recorded from 300 to 450 nm at an excitation wavelength of 280 nm by fluorescence spectroscopy. Fluorescence intensities declined from 267 to 232 au with increasing harmine concentrations. Numbers indicate the maximum fluorescence wavelength and corresponding fluorescence intensity for the mock (black) and maximum harmine concentration (red) titration traces. The arrowhead indicates the maximum fluorescence wavelength of harmine at 418 nm. (C) Changes in fluorescence intensity plotted against harmine concentration (0 to 80 μM). The data from three independent experiments were fitted to obtain the dissociation constant (K_d) between DNMT3B/3L and harmine. (D) Representative isothermal titration calorimetry (ITC) profile obtained from titration of harmine into DNMT3B-3L (blue) or into the buffer (green). Upper panel: raw ITC data. Lower panel: best fitting of the processed results. The inset shows thermodynamic parameters as mean ± SD derived from three replicates. An * marks the data excluded from data fitting.

Next, we investigated if harmine binds directly to DNMT3B-3L. To do so, we adopted tryptophan fluorescence spectroscopy to detect any changes in the tryptophan microenvironment upon harmine binding. The heterotetrameric DNMT3B-3L complex harbors 24 tryptophan residues: five in DNMT3B (Trp639, Trp694, Trp736, Trp801, and Trp834) and seven in DNMT3L (Trp185, Trp235, Trp258, Trp282, Trp323, Trp335, and Trp359). Notably, Trp834 is located near the active site in DNMT3B, so its fluorescence signal might be affected by harmine if it binds near the active site. Our tryptophan fluorescence spectroscopy of the DNMT3B-3L complex revealed an absorption maximum at 341 nm in the absence of harmine, with an excitation wavelength of 280 nm. Titration of the DNMT3B-3L complex with harmine not only resulted in a 2 nm red shift (longer wavelengths) of the emission fluorescence peak but also reduced the intensity of the emission fluorescence in a harmine-concentration-dependent manner (Figure 3B), indicating that tryptophan residues were more exposed to the hydrophilic environment in the presence of harmine. We plotted changes in the emitted tryptophan fluorescence intensity in response to harmine concentration and determined the binding dissociation constant (K_d) to be 3.48 ± 1.95 μM (Figure 3C). To corroborate this result, we further measured the binding affinity between the DNMT3B-3L complex and harmine by isothermal titration calorimetry (ITC) which resulted in a K_d of 3.35 ± 0.30 μM (Figure 3D). The binding reaction was spontaneous at 25 °C with an exergonic Gibbs energy (ΔG = −31.27 ± 0.26 kJ/mol). The thermodynamic profile (ΔG < 0, ΔH > 0, and – TΔS < 0) suggests that hydrophobic effects mainly contribute to the entropically driven binding.⁴⁷ Together, these data suggest that harmine directly binds DNMT3B-3L with a micromolar affinity.

Crystal Structure of the DNMT3B-3L–Harmine Complex

To obtain the structure of human DNMT3B-3L in complex with harmine, we preincubated DNMT3B-3L overnight with a low concentration of harmine and then removed harmine and/or SAH by means of two rounds of PD-10 desalting. This procedure produced an apo-form of DNMT3B-3L with an empty SAM-binding pocket (Figure S1). The cocrystals were then grown by mixing the apo-DNMT3B-3L eluted from the desalting column with a high concentration of harmine in a 1-to-10 ratio (45 μM of DNMT3B-3L and 450 μM of harmine) using the same aforementioned cocrystallization conditions.

The DNMT3B-3L–harmine cocrystal diffracted X-rays to a resolution of 3.08 Å, exhibiting an identical space group and unit cell dimensions to those of the previously reported structure of the DNMT3B-3L–SAH complex (PDB code 6KDP). The crystal structure (PDB code 7X9D; Figure 4A) was refined to 3.08 Å resolution with R-factor and R-free values of 0.2127 and 0.2446, respectively (Table 1). After the first refinement cycle, we observed a strong electron density at the 2.5σ cutoff level in the |F_o| – |F_c| difference Fourier map located at the SAM-binding pocket, which fit well with a harmine molecule, as revealed using the LigandFit mode in PHENIX. In the crystal structure of the DNMT3B-3L–harmine complex, harmine interacts with the amino acids of DNMT3B via hydrogen bonding and van der Waals interactions (Figure 4C and E). The N2 atom of harmine forms a hydrogen bond with the DNMT3B backbone nitrogen atom of Val628. Harmine N9 also forms a hydrogen bond with the DNMT3B side-chain carboxylate oxygen (Oε2) of Glu605. The oxygen of the harmine methoxy group forms a hydrogen bond with the nitrogen (Nε) of the guanidinium group of Arg832. Additionally, harmine makes van der Waals interactions with several amino acid residues on DNMT3B, including Phe581, Ser604, Val606, Asn626, Asp627, Pro650, and Leu671. Moreover, all harmine atoms, except for C6 and C7, form van der Waals interactions with DNMT3B, which shows that harmine binds to the pocket with a high ligand efficiency.

Figure 4.

Crystal structure of the DNMT3B-3L–harmine complex reveals harmine binds at the SAM-binding pocket. (A) Overall crystal structure of the DNMT3B-3L–harmine complex: DNMT3B in green, DNMT3L in blue, and harmine represented by the orange spheres. Two harmine molecules are bound in the SAM-binding pockets of dimeric DNMT3B. Inset shows a close-up view of the interactions between harmine and DNMT3B. Green dotted lines indicate hydrogen bonds. (B) The cofactor SAM-binding pocket in the DNMT3B-3L–SAH complex (PDB code 6KDP). SAH is bound in the SAM-binding pocket. (C) Harmine binds at the SAM-binding pocket of the DNMT3B-3L–harmine complex (this study, PDB code 7X9D). Harmine is displayed as a stick model fitted in the 2F_o – F_c omit map (contoured at 1 σ). (D) 2D schematic depicting the interactions between DNMT3B and SAH in the crystal structure of the DNMT3B-3L–SAH complex. (E) 2D schematic depicting the interactions between DNMT3B and harmine in the crystal structure of the DNMT3B-3L–harmine complex. The interactions were generated by LigPlot⁺.⁴⁹ Residues linking SAH or harmine via hydrogen bonds are shown as green dashed lines. Residues that provide van der Waals interactions with SAH or harmine are labeled with red eyelash symbols.

Table 1. X-ray Diffraction and Crystal Structure Refinement Statistics for the DNMT3B-3L-DNA-Harmine Complex.

data collection
PDB entry	7X9D
space group	C2221
unit cell dimensions a, b, c (Å)	72.02, 236.16, 230.65
wavelength (Å)	1.00
resolution range (Å)	29.9–3.1 (3.2–3.1)
unique reflections	35,934
total reflections	185,378
I/σa	19.11 (2.20)
Rmergea,b(%)	0.08 (0.53)
completenessa (%)	97.7 (91.0)
redundancya	5.2 (4.6)
CC1/2a	0.986 (0.939)

refinement statistics
resolution (Å)	29.7–3.1 (3.3–3.1)
Rwork (%)/Rfree (%)c	21.27/24.46

RMSD
bonds (Å)	0.002
angles (deg)	0.55
mean B factor (Å2)	62.8
protein	62.8
harmine	72.0

Ramachandran plot (%)
favored	95.99
allowed	4.01
outliers	0.00

^aValues in parentheses are for the highest resolution shell.

^bR_merge = Σ_hΣ_i|I_h,_i – I_h|/Σ_hΣ_iI_h,_i, where I_h is the mean intensity of the i observations of symmetry related reflections of h.

^cR_work/R_free = Σ|F_obs – F_calcd|/ΣF_obs, where F_calcd is the calculated protein structure factor from the atomic model (R_free was calculated with 5% of the reflections selected).

Notably, binding of harmine to DNMT3B induces side-chain conformational changes for several residues in the SAM-binding pocket, including Ser604, Glu605, Val606, Asn626, Asp627, Val628, Leu671, Arg832, Ser833, and Trp834 (Figure S2), which coordinate SAH by forming hydrogen bonds or providing hydrophobic or polar interactions in the SAH-bound DNMT3B-3L structure. Significantly, the side chain orientation of Asp627 is different in the SAH- and harmine-bound DNMT3B-3L structures. Asp627 plays an important role in coordinating the adenine moiety of SAM cofactor,⁴⁸ and it is a conserved residue among DNMTs (Figure S3). In the SAH-bound DNMT3B-3L structure, the Asp627 side-chain is oriented into the adenine-binding cavity and forms a hydrogen bond with the N6 atom of adenine to hold the adenine moiety. In contrast, in the harmine-bound DNMT3B-3L structure, the Asp627 side-chain does not interact with harmine and is oriented away from the adenine cavity (Figure S2B,C). Thus, the altered side-chain orientation of Asp627 in the SAH- and harmine-bound DNMT3B-3L structures indicates that binding of harmine slightly alters the behavior of residues within the SAM-binding pocket.

Harmine Competes with SAM to Inhibit DNMT3B

To verify that harmine competitively binds at the cofactor SAM binding site, we performed inhibition kinetics assays to measure the enzymatic activities of DNMT3B-3L at different SAM concentrations in the presence of the four inhibitors, i.e., harmine, nanaomycin A, resveratrol, and EGCG (0 to 50 μM) by means of MTase-Glo assays. The initial reaction rate in terms of SAH production (μmol/min) was derived from the increase in illuminance signal (Figure 5A–D). Lineweaver–Burk double reciprocal plots showed that harmine, nanaomycin A, resveratrol, and EGEC are indeed competitive inhibitors as the lines converged at the y axis with the same y intercept, revealing a constant Vmax and an increased Km upon inhibitor binding (Figure 5E–H). The inhibition constants (K_i) for the four inhibitors, as determined after nonlinear regression curve fitting, are consistent with the IC₅₀ values estimated from our enzymatic assays (Figure 2). In terms of inhibiting DNMT3B-3L, nanaomycin A displayed the lowest K_i of 4.71 ± 0.98 μM, followed by harmine (K_i 6.61 ± 1.74 μM), with resveratrol and EGCG exhibiting similar K_i values of 23.63 ± 4.17 μM and 30.92 ± 8.14 μM, respectively (Figure 5A–D). Thus, our inhibition kinetic assays not only confirm the inhibitory activities of these non-nucleoside inhibitors but also illustrate that harmine is a competitive inhibitor that competes with SAM to inhibit the activity of the DNMT3B-3L complex. This result corroborates our crystal structure showing that harmine binds at the SAM-binding site, enabling it to competitively inhibit the methyltransferase activity of DNMT3B.

Figure 5.

Kinetics assays of non-nucleoside inhibitors in inhibition of DNMT3B-3L. Michaelis–Menten plots of the enzymatic activity of DNMT3B-3L in the presence of 0–5 μM SAM and 0–50 μM (A) harmine, (B) nanaomycin A, (C) resveratrol, and (D) EGCG. Calculated inhibition constants (K_i) are indicated in the respective panels. Lineweaver–Burk plots for DNMT3B-3L in the presence of (E) harmine, (F) nanaomycin A, (G) resveratrol, and (H) EGCG. Data represent mean ± SD from three independent experiments.

Harmine Reactivates Hypermethylated Silenced Genes in Prostate Cancer Cells

DNA methylation plays an important role in tumor invasion, growth, and metastasis of prostate cancer (PCa). Previous studies have reported that DNMT activity, as well as the protein levels of DNMT1, DNMT3A, and DNMT3B, are all significantly higher in PCa tissue than benign prostatic hyperplasia (BPH) tissue.⁵⁰ Moreover, as found for patient tissues, DNMT activity and expression levels of DNMT1, DNMT3A, and DNMT3B were higher in PCa cell lines compared to their non-neoplastic counterparts.⁵⁰ In metastatic castration-resistant prostate cancer (CRPC) patients, 22% of tumors exhibited a CpG methylator phenotype (CMP) with a hypermethylated genome.⁵¹ To test the effect of harmine on the viability of PCa cells, first we treated a commonly used CRPC cell line (C4–2⁵²) with various harmine dosages, and monitored its effect on the cell proliferation by means of MTT assays. We found that harmine exerted a dose-dependent inhibitory effect on C4–2 cell viability (IC₅₀ of ∼14 μM; Figure 6A), revealing that harmine effectively inhibits CRPC cell growth. However, we could not fully exclude the possibility that harmine might affect the enzymatic activity of succinate dehydrogenase and reduced the cell viability due to an impact on cell metabolism. To further confirm the harmine-induced cancer cell death activity, we also treated C4–2 cells with 0, 25, 50, and 100 μM of harmine and found that treatment of cells with 50 and 100 μM harmine increased the levels of cleaved caspase-3 and decreased Bcl-2 protein levels (Figure S4A) and that harmine treatment induced cell apoptosis after Annexin V and PI staining as shown in the flow cytometry analysis (Figure S4B). Harmine-induced cancer cell apoptosis was also observed in melanoma,⁵³ colon,⁵⁴ and breast cancer.⁵⁵ Together these results show that harmine induces C4–2 cell death at concentrations above 50 μM.

Figure 6.

Harmine inhibits cell growth and reactivates hypermethylated silenced genes in prostate cancer cells. (A) Cytotoxicity effect of harmine on C4–2 cells. Cultured C4–2 cells were treated with harmine (1.56–100 μM) and then cell viability was monitored by MTT assays. Data represent mean ± SD of three independent experiments and were fitted using GraphPad Prism version 7.0 software. (B) Quantitative real-time PCR analysis of prostate-cancer-related hypermethylated genes in DU145 cells treated with or without harmine (upper panel) and RG108 (lower panel). Real-time PCR was performed using primers specific for AR, GSTP1, RASSF1A, APC, SSBP2, HIC1, MGMT, EDNRB, PTGS2, and 18S rRNA. Transcript levels were normalized to 18S rRNA levels. Fold changes in mRNA expression are shown as the ratio of transcript levels in the harmine- or RG108-treated cells relative to the untreated cells. Results are expressed as the means ± SD of three separate experiments. (C) Bisulfite sequencing of the RASSF1A promoter region (from −220 to +162 containing 32 CpG sites) in untreated and harmine-treated DU145 cells. The DNA sequence and CpG sites in the RASSF1A promoter region are shown in the bottom panel. The reduced methylation levels in percentage of harmine-treated cells to untreated cells for the 32 CpG sites were estimated by the reduced methylation percentage in the harmine-treated cells divided by the methylation percentage in the untreated cells (see Table S2). The transcription start site is indicated by a black arrowhead. (D) Harmine cooperates with bicalutamide to inhibit the growth of C4–2 cells. Growth rates of C4–2 cells treated over 8 days with vehicle, bicalutamide (10 μM), harmine (7 μM), or a combination of bicalutamide and harmine are shown. Data are plotted as mean ± SD of three independent experiments and fitted using GraphPad Prism version 7.0 software. ****P ≤ 0.0001, ***P ≤ 0.001, **P ≤ 0.01, *P ≤ 0.05.

Next, to provide mechanistic insights into the cytotoxicity of harmine, we assessed transcription of several hypermethylated genes in DU145 cells treated with or without harmine, including genes participating in DNA damage repair (GSTP1, MGMT),⁵⁶⁻⁶⁰ cell adhesion (ENDRB, APC),⁶¹⁻⁶⁵ signal transduction (RASSF1A),^64,66−70 hormone responses (AR),⁷¹⁻⁷³ inflammation (PTGS2),^74,75 and cell development (HIC1).⁷⁶ We found that the mRNA levels of these hypermethylated genes all increased significantly in cells after the treatment of harmine (Figure 6B, upper panel and Table S1). On the contrary, as described previously for the RASSF1A gene,³² treatment of RG108 did not cause increased expression of the hypermethylated genes (Figure 6B, lower panel). Furthermore, harmine or RG108 treatment did not result in significant changes of the mRNA levels of DNMT3A and DNMT3B genes (Figure S5), suggesting that harmine did not affect DNMT protein levels and may directly target the enzymatic activity of DNMT3A or DNMT3B.

Reactivation of the hypermethylated genes by harmine may result from demethylation in their gene promoters arising from inhibition of DNMTs. For this reason, we further investigated the promoter methylation level of the RASSF1A tumor suppressor gene in untreated and harmine-treated (10 μM) DU145 cells. Bisulfite sequencing using two sets of primers that covered respectively 20 and 12 CpG sites in the RASSF1A promoter regions revealed reduced methylation in these CpG sites in the harmine-treated cells (shaded by green and yellow in Figure 6C and Table S2). Unlike the demethylation caused by nanaomycin A that was restricted to a few CpG sites,³² harmine caused a dispersed demethylation effect that 11 out of 32 CpG sites in the promoter region of RASSF1A had reduced methylation levels in a range of 2.06% to 16.64% (middle panel in Figure 6C). Together, these results indicate that harmine effectively inhibits prostate cancer cell growth that is likely attributable to transcriptional reactivation of silenced hypermethylated genes.

Harmine Cooperates with Bicalutamide to Inhibit Prostate Cancer Growth

Next, we evaluated if harmine could work cooperatively with a widely used prostate cancer drug, bicalutamide, to enhance the efficacy of this androgen-deprivation therapy. Bicalutamide is an antiandrogen drug that works cooperatively with the AKT inhibitor AZD5363⁷⁷ and an antihelminthic drug to inhibit tumor growth, and it is used to treat patients with nonmetastatic or metastatic CRPC.⁷⁸⁻⁸¹ We assessed C4–2 cell viability in the presence of harmine with or without bicalutamide. Treatment with bicalutamide (10 μM) or harmine (7 μM) alone reduced the number of viable cells by 17% and 40%, respectively, after 6–8 days of cell growth. Significantly, the combination of the two compounds greatly reduced cell proliferation by 65% (Figure 6D). Harmine and bicalutamide exhibited a synergistic effect when cells were treated with a high (7.0 μM) or low (3.5 μM) concentration of harmine, defined by the Bliss independence model (Figure 6D, Figures S6 and S7). Thus, targeting DNMTs with harmine significantly inhibits CRPC cell growth, and this approach can be potentially combined with bicalutamide to improve the antiandrogen therapy against prostate cancer.

Discussion

Based on compelling biochemical, structural, and enzymatic evidence, here we report the molecular mechanism underlying how a non-nucleotide inhibitor, harmine, inhibits the methyltransferase activity of human DNMT3B. Indeed, we demonstrate that four non-nucleoside inhibitors—harmine, nanaomycin A, resveratrol, and EGCG—all can inhibit human DNMT1, DNMT3A-3L, and DNMT3B-3L to varying extents, although harmine and nanaomycin A displayed the most significant inhibitory activity on DNMT3A-3L and DNMT3B-3L (IC₅₀ ≤ 13 μM). Our crystal structure of the DNMT3B-3L–harmine complex further reveals that harmine binds at the adenine cavity in the SAM-binding pocket. Enzyme kinetics assays consistently showed that harmine, as well as the other three non-nucleoside inhibitors, compete with SAM for methyltransferase binding to inhibit DNMT3B-3L. Thus, all four non-nucleoside inhibitors presumably bind at the SAM-binding pocket of the DNMT3B-3L complex to inhibit its methyltransferase activity. It was reported that sinefungin, a 5′-aminoalkyl analog of SAH, inhibits the activities of human DNMT1 (IC₅₀ = 80 μM) and DNMT3A (IC₅₀ = 1.8 μM) and serves as a pan-inhibitor against SAM-dependent methyltransferases.^82,83 Optimization of harmine or other compounds that target the SAM-binding pocket may increase its specificity for a certain type of DNMT and provide therapeutic implications in treating diseases caused by dysregulation of DNA methylation.

Harmine is a natural β-carboline alkaloid, and its wide spectrum of pharmacological effects has been studied extensively, including its neuroprotective, anti-inflammatory, antidiabetic, and antitumor activities.⁸⁴ Harmine is believed to exert its neuroprotective effects by inhibiting a group of tyrosine-(Y)-phosphorylation-regulated kinases (DYRKs) that play vital roles in neurodevelopment, with the strongest inhibition on DYRK1A (IC₅₀ = 33 nM).⁸⁵ Inhibition of DYRK1A may also account for harmine’s antidiabetic effects by promoting β cell proliferation and improving blood glucose metabolism.⁸⁶ Harmine also inhibits monoamine oxidase A (MAO-A; IC₅₀ = 0.38 μM) to affect the brain dopaminergic system,⁸⁷ as well as acetylcholine esterase (AChE; IC₅₀ = 9.05 μM) to ameliorate memory impairment.⁸⁸ Harmine has also been extensively reported as exerting antitumor activities, likely via its inhibition of DYRKs and cyclin-dependent kinases (CDKs), which play critical roles in regulating cell cycle progression. For instance, harmine inhibits the activities of Cdk1/cyclin B, Cdk2/cyclin A, and Cdk5/p25 by binding to their respective ATP-binding pockets (IC₅₀ values of 17, 33, and 20 μM, respectively).⁸⁹ Together, these results support that harmine targets a diversity of proteins involved in important signal transduction and/or cell proliferation processes.

Our analyses have demonstrated that harmine binds at the adenine cavity of the SAM-binding pocket in DNMT3B, implying that harmine could fit into the ATP-binding pocket of protein kinases and in the SAM-binding pocket of methyltransferases. ATP and SAM are the two most prevalent nucleotide cofactors in proteins, carrying out important functions in metabolic and regulatory pathways.^48,90 Thus, the potential protein targets of harmine are likely more plentiful than those reported to date. Accordingly, the wide spectrum of physiological effects induced by harmine could arise from its inhibition of multiple protein targets. Here, we have revealed how harmine is bound at the adenine cavity of the SAM-binding pocket in DNMT3B to inhibit the enzymatic activity of this latter one and, consequently, to limit prostate cancer cell proliferation. Our study provides a solid foundation for understanding the various physiological effects induced by harmine and for designing new harmine-derivated inhibitors that may target a specific protein(s) for biomedical applications.

Methods

Chemicals

RG108 (≥98% by HPLC; Catalog No. R8279; Batch No. 0000026100), EGCG (≥98% by HPLC; Catalog No. E4143; Lot No. SLBZ2865), and resveratrol (≥98% by HPLC; Catalog No. R5010; Lot No. SLBV8562) were purchased from Sigma-Aldrich. Nanaomycin A (≥98% by HPLC; Catalog No. A8191; Batch No. 5) was purchased from APExBIO. Harmine HCl (≥99% by HPLC; Catalog No. S3817; Batch No. S381703) and bicalutamide (≥99% by HPLC; Catalog No. S1190; Batch No. S119006) were purchased from Sellekchem.

Protein Expression and Purification

The cDNAs encoding human DNMT1 (hDNMT1), human DNMT3B (hDNMT3B), and human DNMT3A (hDNMT3A) were purchased from Addgene. The cDNA fragments encoding DNMT1 (residues 646–1616), as well as the catalytic domain of DNMT3A (residues 627–912), DNMT3B (residues 571–853), and DNMT3L (residues 179–379), were amplified by polymerase chain reaction (PCR). The DNA fragments encoding DNMT1 (646–1616) were inserted into pSOL Expression Vectors (Lucigen), whereas the cDNAs encoding DNMT3A, DNMT3B, and DNMT3L were cloned into a modified pET28a(+) expression vector expressing an N-terminal 6xHis-SUMO tag and a tobacco etch virus (TEV) cleavage site or an N-terminal 6xHis-tag and a tobacco etch virus (TEV) cleavage site to generate the pSol-tev-DNMT1 (646-1616), pET28a(+)-tev-3B, pET28a(+)-tev-3L, and pET28a(+)-tev-3A plasmids.

All expression plasmids were transformed into the bacterial strain Rosetta2 (DE3) pLysS. Cells were grown at 37 °C in LB Broth (Miller) medium containing 34 μg/mL chloramphenicol and 50 μg/mL streptomycin. Expression of DNMT1 (646–1616) was induced by the addition of 0.002% Rhamnose, whereas expression of DNMT3A (627–912), DNMT3B (571–853), or DNMT3L (179–379) was induced by 0.4 mM isopropyl-thio-β-d-galactoside (IPTG) until an OD₆₀₀ of 0.6 had been attained. After induction, the cells were grown at 18 °C overnight and harvested by centrifugation at 6000 rpm at 4 °C for 30 min. The pellet at the bottom of the centrifuge tube was collected and redissolved in lysis buffer containing 25 mM HEPES (pH 7.4), 500 mM NaCl, 5% glycerol, 0.5 mM TCEP, and ethylenediaminetetraacetic acid (EDTA)-free protease inhibitor cocktail (Roche, Switzerland). Then, the redissolved soups of DNMT3A or DNMT3B were mixed with the redissolved soup of DNMT3L in a 4-to-1 ratio to generate the DNMT3B-3L and DNMT3A-3L complexes.

The pellet mixture was lysed using a microfluidizer. After centrifugation at 17 000g for 45 min at 4 °C, the supernatant was loaded through a HisTrap FF column (GE Healthcare). After equilibrium with nickel wash buffer containing 50 mM sodium phosphate (pH 7.4), 500 mM NaCl, 5% glycerol, 5 mM β-mercaptoethanol, and 40 mM imidazole, the His-tagged protein was eluted by a nickel elution buffer (with the same constitution as the nickel wash buffer) and up to 500 mM imidazole. The 6xHis-tag was removed by TEV protease during dialysis overnight at 4 °C against a buffer containing 50 mM sodium phosphate (pH 7.4), 100 mM NaCl, 5% glycerol, and 5 mM β-mercaptoethanol. The dialyzed protein sample was loaded into a HisTrap FF column (GE Healthcare), and the flow-through fractions containing the cleaved, untagged protein of interest were collected. To further purify and polish the protein sample, the HiTrap Heparin HP column (GE Healthcare) and a gel filtration column (HiLoad 16/60 Superdex 200, GE Healthcare) were used sequentially. The final purified samples of DNMT1 (646-1616), DNMT3A-3L complex, and DNMT3B-3L complex were solubilized in gel filtration buffer containing 20 mM Tris–HCl (pH 7.4), 200 mM NaCl, 5% glycerol, and 0.5 mM tris(2-carboxylethyl)phosphine (TCEP).

Size Exclusion Chromatography-Coupled Multiangle Light Scattering (SEC-MALS)

Molecular weights of DNMT1, DNMT3A-3L, and DNMT3B-3L were measured by SEC-MALS. Superdex Increase 200 10/300 columns (GE Healthcare) connected to a DAWN HELIOS II-18 angle MALS (Wyatt Technology) detector with wavelength set to 680 nm were equilibrated in 20 mM Tris-HCl, at pH 8.0, 200 mM NaCl, and 1 mM TCEP with a flow rate of 0.2 mL/min using ÄKTA-UPC 900 FPLC system (GE Healthcare). Purified protein samples were centrifuged at 17 000g for 15 min at 4 °C and filtered through a 0.22 μm filter (Millipore). Protein samples (100 μL injection volumes) were injected into Superdex Increase 200 10/300 columns. UV fluorescence, MALS, and refractive index data were recorded and analyzed using ASTRA software (Wyatt Technology).

IC₅₀ Measurement of DNMT Inhibitors

DNMT inhibition was assessed using an MTase-Glo Methyltransferase Assay Kit (Promega) in a white 384-well opti-plate system (PerkinElmer). RG108, EGCG, resveratrol, nanaomycin A, and harmine (400 μM) were diluted, respectively, by means of 2-fold serial dilution in a reaction buffer containing 20 mM Tris–HCl (pH 8.0), 50 mM NaCl, 1 mM EDTA, 3 mM MgCl₂, 1 mM dithiothreitol (DTT), and 0.1 mg mL^–1 bovine serum albumin (BSA). The resulting solutions of varied inhibitor concentrations were then incubated with 0.2 μM DNMT1, DNMT3A-3L, or DNMT3B-3L complex, respectively, for 10 min. To initiate the methyltransferase reaction, 0.5 μM SAM, 2 μM 14-bp DNA substrate (forward: 5′-GGAGGCXGCCTGCT-3′; reverse: 5′-AGCAGGCGGCCTCC-3′, for DNMT1, X was 5-methylcytosine; for DNMT3A-3L and DNMT3B-3L, X was cytosine) and 10× MTase-Glo Reagent (400 mM Tris buffer at pH 8.0, 40 mM MgCl₂, 75 mM NaCl, 125 μM dGTP, 200 μM sodium metaphosphate, 80 μg/mL AMP phosphotransferase, 200 μg/mL SAH hydrolase, 50 μg/mL adenosine kinase, 5 mM DTT) were added to the 5-μL mixture and incubated at 37 °C for 1 h to convert SAH into adenosine and homocysteine by SAH hydrolase, before the adenosine was transformed into ADP by adenosine kinase and AMP phosphotransferase. When the reaction time ended, MTase-Glo Detection Solution (1 mM potassium phosphoenolpyruvate, 8 U/mL pyruvate kinase in Kinase-Glo reagent (Promega)) was added to the reaction sample at RT for 30 min, allowing the ADP to be further converted into ATP that can react with luciferase and oxygen to emit 560 nm bioluminescence. Luminescence signals were monitored using an EnSpire Multimode Plate Reader (PerkinElmer). The IC₅₀ curves were constructed using nonlinear regression analysis according to the “dose–response-inhibition” model in GraphPad Prism version 7.0 software.

Circular Dichroism (CD) Spectroscopy

Far UV CD spectra from 260 to 190 nm were recorded on an AVIV Circular Dichroism Spectrometer (Aviv Biomedical, Inc.). All measurements were carried out in a 1 mm quartz cuvette at 25 °C. The protein samples in a 20 mM phosphate pH 7.4 buffer were set to 0.001 μM with or without preincubation with 0.1 μM harmine. The CD spectra were processed by smoothing and subtraction of the baseline by AVIV built-in software. Ellipticity values (θ) were converted to mean residue ellipticity ([θ] in deg cm² dmol^–1) using

where M is the mean molecular mass of the amino acids, l is the light path length in centimeters, and c is the protein concentration in moles per liter. All experiments were performed in triplicate. Secondary structure compositions of DNMT3B-3L with or without harmine were estimated and analyzed by K2D3.⁴⁶

Intrinsic Tryptophan Fluorescence Spectroscopy

All tryptophan fluorescence spectroscopy experiments were performed using a CARY Eclipse Fluorescence Spectrophotometer (Varian Inc.) and a 3-mL quartz cuvette at 25 °C. The excitation wavelength was fixed at 280 nm, and emission spectra were collected between 300 and 450 nm, with a slit width of 5 nm. To measure the interactions of the DNMT3B-3L complex with harmine or SAM, 1 μM DNMT3B-3L complex was titrated with increasing concentrations (0 to 80 μM) of harmine. Control buffer titrations against the DNMT3B-3L complex were performed for background determination in each experiment. The data were analyzed using nonlinear regression analysis with the “one site-specific binding” model in GraphPad Prism version 7.0 software.

Isothermal Titration Calorimetry (ITC)

Binding of harmine to DNMT3B-3L was measured by ITC with the Nano Isothermal Titration Calorimeter (TA Instruments). Aliquots of 7 μL of 0.52 mM harmine were titrated by injection into protein (0.02 mM in 1.03 mL cell) in a buffer containing 20 mM Tris-HCl, at pH 8.0, 100 mM NaCl, and 0.2 mM TCEP. Background heat from ligand to buffer titrations was subtracted, and the corrected heat from the binding reaction was used to derive values for the stoichiometry (n), dissociation constant (K_d), apparent enthalpy of binding (ΔH), and entropy change (ΔS). All experiments were performed in triplicate. Data were fitted by an independent binding model with NanoAnalyze version 3.12.5.

Protein Crystallization and Structural Determination

Purified human DNMT3B-3L complex (5.0 mg mL^–1) and harmine were mixed in a molar ratio of 1:2.5 and stored at 273 K overnight, and then the mixture was purified twice via PD-10 desalting columns to remove residual harmine and/or SAH. The apo-form of DNMT3B-3L (45 μM) was then incubated with harmine (450 μM), and the DNMT3B-3L-harmine cocrystals were grown by the hanging-drop vapor diffusion method by mixing 1.6 μL of the DNMT3B-3L–harmine complex and 0.8 μL of reservoir solution containing 0.1 M bis-tris propane at pH 7.0 and 2.0 M sodium formate. The DNMT3B-3L–harmine cocrystals appeared after 1 week at 277 K. The cryoprotectant used before crystal mounting was composed of 25% glycerol and the reservoir solution. The X-ray diffraction data sets for the DNMT3B-3L–harmine complexes were collected at TPS beamline 05A and TLS beamline 15A at the National Synchrotron Radiation Research Center, Hsinchu, Taiwan. The diffraction data were processed using HKL2000 software.

The crystal structure of the DNMT3B-3L–harmine complex was refined in PHENIX Phaser using the human DNMT3B-3L complex structure (PDB code 6KDP) as the template. The initial model was rebuilt using the program Coot and involved repeated cycles of fitting amino acid side chains and water molecules according to 2F_o–F_c and F_o–F_c electron density maps and refinement in PHENIX. The data collection and structure refinement statistics are presented in Table 1.

Enzyme Kinetics

Kinetics assays were performed using an MTase-Glo Methyltransferase Assay Kit (Promega) in a white 384-well opti-plate system (PerkinElmer). Each inhibitor was diluted to 5, 10, or 25 μM and then incubated with the 0.2 μM DNMT3B-3L complex for 10 min in a reaction buffer containing 2 μM 14-bp DNA (forward: 5′-GGAGGCCGCCTGCT-3′; reverse: 5′-AGCAGGCGGCCTCC-3′), 20 mM Tris-HCl (pH 8.0), 50 mM NaCl, 1 mM EDTA, 3 mM MgCl₂, 1 mM DTT, and 0.1 mg mL^–1 bovine serum albumin (BSA). Methylation of DNA substrates by DNMT3B began upon adding SAM at various concentrations (0 to 5 μM). After incubation for 30 min, the reaction was stopped by adding 1 μL of 0.5% trifluoroacetic acid (TFA). After incubation with TFA for 5 min, 1 μL of 6× methyltransferase-Glo reagent was added, mixed well, and incubated at RT for 30 min. Then, 6 μL of methyltransferase-Glo detection solution was added and incubated at RT for 30 min. Luminescence signals were recorded using an EnSpire Multimode Plate Reader (PerkinElmer). The inhibition constant (K_i) was calculated using nonlinear regression analysis with the “Enzyme Kinetics- Inhibition” model, and corresponding Lineweaver–Burk plots were transformed using the “Pharmacology and Biochemistry Transforms” model, in GraphPad Prism version 7.0 software.

Prostate Cancer Cell Viability Assay

C4–2 and DU145 cells were originally obtained from the American Type Culture Collection (Rockville, MD, USA). Cells were tested and authenticated by genetic profiling using short tandem repeat analysis. C4–2 and DU145 cells were maintained in RPMI 1640 medium with 10% fetal bovine serum (FBS), 2 mM l-glutamine, 1% sodium pyruvate, and 1% penicillin-streptomycin in a humidified, 5% CO₂ incubator at 37 °C.

C4–2 cells were first seeded into 96-well plates (1 × 10⁴ cells/well). Following 24 h of culture at 37 °C, the cells were starved with a serum-free medium for 48 h and then treated with the indicated concentration of harmine (dissolved in ddH₂O) at 37 °C. After 48 h of treatment, 100 μL of 0.5 mg mL^–1 MTT [3-(4,5-dimethyl-2-thiazolyl)-2,5-diphenyltetrazolium bromide] was added into each well and incubated for 30 min at 37 °C. The resulting oxidized form of MTT, purple-colored formazan, was then dissolved in 100 μL of dimethyl sulfoxide (DMSO) and measured at 540 nm using a microplate reader (SpectraMax Paradigm, Beckman Coulter, CA, USA). Cell viability was calculated as

and the IC₅₀ values were calculated using GraphPad Prism version 7.0 software.

Harmine-Induced C4–2 Cell Apoptosis

After 24 h of starvation, C4–2 cells were treated with harmine (0, 25, 50, and 100 μM) in a serum-free medium for 24 h. For cell lysate preparation, cells were washed by cold PBS twice and lysed using lysis buffer (10 mM Tris-HCl, at pH 7.4, 100 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1 mM NaF, 20 mM Na₄P₂O₇, 2 mM Na₃VO₄, 0.1% SDS, 0.5% sodium deoxycholate, 1% Triton-X 100, 10% glycerol, 1 mM PMSF, 0.02% DTNB, and one tablet of protease inhibitor cocktail). After adding lysis buffer into the cell culture, the lysate was scraped, collected in a microcentrifuge tube, and kept on ice for 15 min. Cell lysates were then centrifuged at 13 000 rpm at 4 °C for 15 min. The supernatants were taken and the protein concentrations were measured using Bio-Rad Protein Assay (Bio-Pad, CA, USA). Equal amounts of proteins in each sample were taken and mixed with 6× protein loading dye with 5% β-mercaptoethanol in new microcentrifuge tubes and boiled for 10 min. Samples were subjected to SDS-PAGE and protein transferring from running gels to nitrocellulose membranes (Whatman, USA) within transfer buffer at 300 mA in a 4 °C refrigerator for 3 h. After transfer, the membranes were blocked with 5% skim milk in TBST at RT for 1 h and then incubated with primary antibodies in blocking buffer shaking overnight in a 4 °C refrigerator. Antibodies used in this study were as follows: anti-Bcl-2 (GeneTex, USA), anti-Caspase-3 Antibody-(Pro and Active) (Novus, USA), and anti-Tubulin (Sigma-Aldrich, USA). The next day, the solutions with primary antibodies were removed, and the membranes were washed with TBST three times, 10 min per wash. The membranes were then incubated with the solution with secondary antibodies conjugated with horseradish peroxidase (Jackson, USA) in blocking buffer at RT for 1 h. The secondary antibody solutions were removed, and the membranes were washed with TBST three times, 10 min per wash. The target proteins on membranes were then visualized using an Enhanced Luminol Reagent Plus (PerkinElmer, USA) and detected by a luminescent image analyzer with an e-BLOT WB/SDS-Pages Touch Image camera (shown in Figure S4A).

For the flow cytometry assays shown in Figure S4B, C4–2 cells were treated with harmine (0, 50, and 100 μM) in a serum-free medium for 24 h after 24 h of starvation. The cells were then washed twice with ice-cold PBS buffer, and the cell pellet was resuspended in the annexin V binding buffer (10 mM HEPES, at pH 7.4, 150 mM NaCl, and 2.5 mM CaCl₂) at a density of 2 × 10⁶ cells/mL and stained with an APC Annexin V Apoptosis Detection Kit with PI (Biolegend, USA) following the manufacturer’s protocol and then subjected to flow cytometry analysis (Thermo fisher, Attune NxT). Fluorescence compensation on the flow cytometer was adjusted to minimize spectral overlap.

RNA Extraction and Quantitative Real-Time PCR Analysis

Total RNA was extracted from DU145 cells treated with different doses of harmine (0 or 10 μM) and RG108 (0 or 10 μM) for 16 h using a TRIzol (Invitrogen, USA) reagent-based RNA purification procedure. The DNase-treated total RNA (5 μg) was subjected to reverse transcription using a Maxima First Strand cDNA Synthesis Kit (Thermo Fisher Scientific Inc.) following the manufacturer’s protocol. For quantitative real-time PCR, a reaction master mix containing the following components was gently mixed: 10 μL of SensiFAST SYBR Hi-ROX Kit (Meridian Bioscience), 0.5 μL of forward primer (10 μM), 0.5 μL of reverse primer (10 μM), 1 μL of template DNA, and 8 μL of ddH₂O. The total volume for each eight-strip PCR tube was 20 μL. All primer sets for measuring gene expression levels in this study are presented in Table S1. A two-step cycling protocol was applied in the thermal cycler (Applied Biosystems, USA). Target gene expression levels in each sample were further normalized to 18S rRNA expression levels. A Student’s t-test was used to evaluate significant differences between two groups.

DNA Extraction and Bisulfite Pyrosequencing

Extraction of genomic DNA from DU145 cells treated with different doses of harmine (0 or 10 μM) for 16 h was performed using the Wizard Genomic DNA Purification Kit (Promega, Madison, WI, USA) and quantified with a NanoDrop ND-1000 (PeqLab, Erlangen, Germany). Bisulfite conversion of genomic DNA (2 μg) was performed using the EZ DNA Methylation-Lightning Kit (Zymo Research, Irvine, CA, USA) following the manufacturer’s protocol. The bisulfite-converted DNA (20 μL) was used for pyrosequencing.

Pyrosequencing was performed on a PyroMark Q24 system (Qiagen) with two sets of primers (sequences are provided in Table S3). Assay 2 covers 11 CpGs in the promoter and 1 CpG in exon 1 of the RASSF1A gene. Assay 1 covers 20 CpGs located upstream of the 12 CpGs covered by Assay 2 (Figure 6C). The PCR was performed using a PyroMark PCR Kit (Qiagen) in a volume of 25 μL containing 12.5 μL of 2 × PyroMark PCR Master Mix, 1.25 μL of each PCR primer (5 μM), 2.5 μL of 10 × CoralLoad Concentrate, 6.5 μL of high purity water, and 1 μL of bisulfite-converted template DNA. The PCR cycling program for both primer sets was composed of an initial Taq activation/DNA denaturation step at 95 °C for 15 min, followed by 50 cycles of denaturation at 95 °C for 30 s, annealing at 58 °C for 30 s, and elongation at 72 °C for 30 s. The program was finished by a final elongation step at 72 °C for 10 min. The PCR products (7 μL) were visualized by gel electrophoresis, and 10 μL of the PCR products were subjected to the sample preparation process for pyrosequencing. DNA was mixed with streptavidin-coated sepharose beads (Amersham Pharmacia), followed by strand separation and washing utilizing the vacuum prep tool (Qiagen). The single-stranded DNA bound to the sepharose beads was mixed with 20 μL of 0.375 μM sequencing primer in annealing buffer and heated to 80 °C for 5 min. For the sequencing reaction PyroMark advanced reagents were used (Qiagen). The sequencing results were analyzed using the Advanced PyroMark software (Qiagen). A control PCR reaction without template DNA (no-template control) was included in the assay. PyroMark assays were carried out two times for accuracy.

Treatment of Prostate Cancer Cells with Harmine and Bicalutamide

Cultured C4–2 cells were seeded in a 6 cm cell culture dish (4 × 10³ cells/dish) and incubated with a vehicle (DMSO), 10 μM bicalutamide (dissolved in DMSO), 3.5 or 7 μM harmine (dissolved in ddH₂O), or a combination of bicalutamide and harmine. Cell proliferation was monitored by MTT assay every 2 days after treatment. Experiments were conducted in triplicate, and the standard deviations are reported. The synergistic effect was analyzed by the Bliss Independence model.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acschembio.3c00065.

• Supplementary Figures and Tables (PDF)

Accession Codes

The coordinates and reflection file of the structure for the DNMT3B-3L-harmine complex have been deposited in Protein Data Bank under accession number 7X9D.

Author Contributions

C.C.C., C.J.L., and C.Y.F. expressed and purified DNMTs and performed the biophysical and enzymatic assays. C.C.C., C.J.L., and W.Z.Y. collected the X-ray diffraction data. C.C.C. determined the crystal structure and analyzed data. H.H.H. and H.Y.L. performed the cell-based experiments. C.J.L., C.C.C., and H.S.Y. wrote the manuscript. H.S.Y. and M.S.L supervised and guided the project.

Author Contributions

^§ These two authors contributed equally to the work.

This work was supported by Academia Sinica (postdoctoral fellowship to C.C.C.) and National Science and Technology Council (MOST 108-2311-B-001-009-MY3 and MOST 111-2311-B001-011-MY3 to H.S.Y.; MOST 108-2320-B-002-024-MY3 and MOST 110-2320-B-002-067-MY3 to M.S.L.), Taiwan, ROC.

The authors declare no competing financial interest.