Exploiting DNA Ligase III addiction of multiple myeloma by flavonoid Rhamnetin

The Schrödinger Suite version 2018 was employed as a computational tool for carrying out all molecular modeling simulations in this study (Schrödinger LLC, New York-USA, 2018).

Starting from the three-dimensional structure of the human LIG3, deposited in the Protein Data Bank (PDB) with the PDB code 3L2P [6], our molecular modeling simulations [30] were carried out. This is the only crystallographic structure of the LIG3 catalytic region, comprising the DBD (DNA binding domain), Ntase (nucleotidyltransferase), and OBD domains (central origin-binding domain) (ΔZnF-LigIIIβ protein) in complex with a 22-mer nicked DNA substrate. The LIG3 structure was optimized by means of the Protein Preparation Wizard tool (Protein Preparation Wizard; Epik, Schrödinger, LLC, New York, NY, 2021) implemented in Maestro, using OLPS_2005 as the force field. Hydrogen atoms were added, missing side chains were built using the Prime module, and side chains protonation states at pH 7.4 were assigned [Schrödinger. Protein Preparation Wizard; Schrödinger, LLC.: New York, NY, USA, 2018., Schrödinger. Prime; Schrödinger, LLC.: New York, NY, USA, 2018.]. The domain containing PCG was selected for further computational studies and it was submitted to 10,000 iterations of fully energy minimization adopting the Polake-Ribiere Conjugated Gradient (PRCG) algorithm and the “all atoms” notation of the OPLS_2005 force field [31], as implemented in the MacroModel suite [Schrödinger Release 2018–1, MacroModel, Schrödinger, LLC, New York, NY, 2018.]. Moreover, the implicit solvation model GB/SA water was adopted with the aim to consider solvent effects and the optimization step was performed up to the derivative convergence criterion of 0.05 kcal Å-1•mol-1.

For the Virtual Screening (VS) studies, a comprehensive database of about 493 polyphenolic derivatives contained in foods, named Phenol Explorer, was considered. In particular, a complete set of polyphenols included in the above-mentioned database (ver. 3.6) was downloaded [https://​doi.​org/​10.​1093/​database/​bap024, https://​doi.​org/​10.​1093/​database/​bas031, J. A. Rothwell, J. Pérez-Jiménez, V. Neveu, A. Medina-Ramon, N. M’Hiri, P. Garcia Lobato, C. Manach, K. Knox, R. Eisner, D. Wishart, A. Scalbert, Database, 2013.; https://​doi.​org/​10.​1002/​minf.​201501040] and prepared considering the ionization states at physiological pH (pH: 7.4) using the LigPrep platform of Maestro [ LigPrep, Schrödinger, LLC, New York, NY, 2018.] Since it is estimated that approximately 50% of drug candidates’ failures in the clinical stage are due to unfavorable ADME properties, pharmacokinetic properties should also be treated as an important filtering step [32]. Thus, in our study, this selection was performed before the screening. ADME properties were predicted for all the polyphenolic derivatives, by means of the QikProp platform (ver. 3.5) [QikProp, Schrödinger LLC, New York, NY (USA), 2018], and only those compounds with the best ADME profile were retained.

The VS was performed applying the SP protocol of the Glide (v. 7.8) software [Glide, Schrödinger, LLC, New York, NY, (2018).] and 10 poses per ligand were generated. A grid box of 27,000 Å3 was built by considering a receptor van der Waals scaling of 1.0 and by centering on the two basic residues (Lys323 and Arg327) in the DBD of LIG3. In fact, the groove containing Lys323 and Arg327 contributes to the interdomain interactions between the ZnF and DBD that allow the binding of one DNA end while the catalytic core binds a second DNA end to promote blunt end joining.

Given the absolute lack of a known compound able to target this portion of LIG3, we opted for an arbitrary cut-off equal to -6.50 kcal/mol to filter the scored compounds according to their G-score value. Thus, we selected 5 compounds and, given their very close structural similarity, only the first best hit was purchased for the following biological tests.

Peripheral blood mononuclear cells (PBMCs) and primary cells from MM patient bone marrow aspirates, following informed consent and University Magna Graecia (Catanzaro, Italy) IRB approval, were isolated using Ficoll–Hypaque density gradient sedimentation as reported previously [33]. MM patients’ cells were separated from bone marrow samples by antibody-mediated selection using anti-CD138 magnetic-activated cell separation microbeads (Miltenyi Biotec). The purity of immunoselected cells was assessed by flow-cytometry analysis using a phycoerythrin-conjugated CD138 monoclonal antibody by standard procedures. CD138 + cells from MM patients pt#1, pt#2 and pt#3 were cultured in RPMI-1640 medium (Gibco®, Life Technologies) supplemented with 20% fetal bovine serum (Lonza Group Ltd.) and 1% penicillin/streptomycin (Gibco®, Life Technologies). MM cell lines (HMCLs). KMS-26 were kindly provided by Dr. Giovanni Tonon (University of San Raffaele Scientific Institute, Milan, Italy). AMO1, NCI-H929, U266 were purchased from DSMZ (Braunschweig, Germany). AMO1 bortezomib-resistant (ABZB) were kindly provided by Dr. Christoph Driessen (Eberhand Karls University, Tübingen, Germany). MM cell lines were cultured in RPMI1640 (Gibco, Life Technologies) supplemented with 10% FBS (Lonza Group). HS-5 human stromal cell line (purchased from ATCC, CRL-11882TM) was cultured in DMEM supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin [34]. Co-culture experiments were performed in 6 well plate at a density of 2,5 × 10 5 cells/ ml in 1:1 HS-5 /MM cells ratio [35, 36].

Silencer™ select MYC (clone IDs: s9129) were purchased from Invitrogen™ (Thermo Scientific) and were used at 100 nmol/L final concentration. MM cells were transfected using Neon Transfection System (Invitrogen™) (2 pulse at 1.050 V, 30 ms).

Analysis of cell cycle was performed by Propidium Iodide flow cytometry assay (BD Pharmingen), according to manufacturer’s instructions. Flow cytometry analysis was performed by Attune NxT Flow cytometer (Thermo Fisher Scientific).

DSB repair assays was performed as previously described [37, 38]. Briefly, EJ2-GFP plasmid (#44,025, Addgene) were linearized with I-SceI (Thermo Scientific) digestion and transfected into 1 × 10⁶ cells at a ratio of 1 μg per well. 48 h after RHM treatment, the numbers of GFP + cells were determined by flow cytometry (Attune NxT, Thermo Fisher Scientific). For each experiment, FACS analyzed a minimum of 20,000 cells.

Male CB-17 severe combined immunodeficient (SCID) mice (6 to 8 weeks old; Harlan Laboratories, Inc., Indianapolis) were housed and monitored in our Animal Research Facility. Experimental procedures and protocols had been approved by the Magna Graecia University IRB and conducted according to protocols approved by the National Directorate of Veterinary Services (Italy). Mice were subcutaneously inoculated with 5 × 10⁶ H929 cells and treatment started when palpable tumors became detectable (100–200 mm³). Tumor sizes were measured as described [40, 41], and the investigator was blinded to group allocation.

Each experiment was performed at least three times and values are reported as means ± SD. Comparisons between groups were made with Student’s t-test, while statistical significance of differences among multiple groups was determined by GraphPad software (www.​graphpad.​com). Graphs were obtained using Graphpad Prism version 6.0. p value of less than 0.05 was accepted as statistically significant. The synergistic index was determined as previously described [42].

LIG3 shows a unique feature on the surface of the DBD, which distinguishes it from LIG1 and LIG4 (Additional file 1: Figure S1A–D). Specifically, a positively charged groove (PCG) is located at the interface of two halves of the DBD between two parallel helices (α1 and α9), adjacent to the DNA binding surface. For these reasons, the domain containing this positively charged groove was selected for further studies of structure-based drug discovery [43, 44]. In the aim to identify a safe Alt-NHEJ targeting strategy, a database containing several polyphenol structures was selected. Indeed, polyphenols are compounds ubiquitously expressed in plants that have been associated with multiple benefits for human health, such as anti-inflammatory, antimicrobial, antiviral, anticancer, and immune-modulatory effects. Thus, the Phenol-Explorer database, the first comprehensive collection of polyphenols contented in foods, was selected for follow-up screening approaches. Initially, the 493 compounds of the Phenol-Explorer database were assessed for Lipinski's Rule of 5 (RO5) and ADMET properties. In fact, the most important aim of drug discovery is the selection of bioactive compounds characterized by drug-likeness properties. Thus, only 152 compounds were able to fulfill the RO5. Afterward, these compounds were screened against the human LIG3 using a structure-based virtual screening (SBVS) approach. Filtering the compounds according to their G-score value, 5 best hits were selected (Table 1) (Fig. 1A). Since all 5 hits are structurally related, we focused only on the most promising one, which is RHM, a monomethoxy flavone. By analyzing the best docking pose of RHM, we observed that it fitted well within the characteristic groove of LIG3 (Fig. 1B), with the chromone ring facing Lys323 and Arg327, the two most important positively charged residues involved in the interdomain interactions between the Zinc Finger (ZnF) domain and DBD. In particular, the chromone ring established a π-cation interaction with Arg327 and an H-bond with Asn320. Moreover, the binding mode of RHM is further stabilized by the formation of two H-bonds with the side chain of Asp188. Finally, we observed several good hydrophobic contacts, involving Cys183, Ala184, Ala187 and Met335.

To validate the findings obtained from the computational analysis, Saturation Transfer Difference (STD) -NMR experiments were performed to confirm the RHM binding to LIG3 recombinant protein.

From the chemical point of view, RHM [2-(3,4-dihydroxyphenyl)-3,5-dihydroxy-7-methoxy-4 h-chromen-4-one, also known as 7-methoxyquercetin] is a member of a class of compounds known as flavonols. These compounds contain a flavone (2-phenyl-1-benzopyran-4-one) backbone carrying a hydroxyl group at the 3-position. RHM is characterized by ¹H-NMR spectroscopy and the spectrum is reported in Fig. 1C (top). STD-NMR analysis of ligand-target complex was based on the nuclear Overhauser effect that allows the observation of the ligand resonance signals [45, 46]. The term binding epitope is frequently used in the STD-NMR literature to characterize the hydrogens of the ligand that are closer to the protein upon binding and became visible in STD spectra.

STD-NMR analysis of RHM in the presence of LIG3 (Fig. 1C, bottom) showed a specific binding between the ligand and the protein. The resonance of all protons of the molecule have been observed. The resonance of 3,4 dihydroxyphenyl group is shown with blue circles, the two protons of chromenone moiety in red, and the methoxyl group in green. All protons of the molecule are involved in the binding and the quantitative analysis of STD data (short protein–ligand distances produce a strong intensity of the corresponding STD signals) showed that the chromenone ring is the moiety with the strongest interaction (as suggested by docking calculation, this is the part that fits well into the binding pocket). A lower intensity can be detected for the protons of the dihydroxyphenyl ring. The comparison with the docking model suggests that the interaction for this moiety is mediated by the formation of hydrogen bonds with the hydroxyl groups, the presence of D₂O in the buffer for NMR analysis causes the exchangeable protons to be replaced by ²H, which is inefficient for transferring saturation. Hence, ligand protons close to protein exchangeable protons receive less saturation and the contacts mediated by polar groups will show a relative low STD signal. The binding epitope by STD spectrum is comparable to the binding mode suggested by the computational model. Furthermore, the contribution of exchangeable protons (showing very low intensities in the 1D spectrum, Fig. 1C top) by OH groups become visible between 7.1 and 7.3 ppm, indicating that they are engaged in a hydrogen bond with the protein (and therefore they are less prone to exchange with the solvent). Finally, to verify the specificity of the binding, STD experiment was performed in presence of unfolded LIG3 protein. Notably, no RHM binding to LIG3 was observed in this condition (data not shown), suggesting the absence of nonspecific interaction.

Taken together, these data demonstrate that RHM binds directly to LIG3.

After the demonstration of RHM binding to LIG3, the effect on cell proliferation and survival of this flavonoid was investigated in a model of disease such as MM, which is highly dependent on LIG3-driven DNA repair [6].

In vitro, RHM exerted a significant anti-proliferative activity in 3 out of 4 MM cell lines (Fig. 2A), according to LIG3 expression and Alt-NHEJ activity. Indeed, while the half maximal inhibitory concentration (IC50) of ≈ 5 μM was reported in NCI-H929, AMO1, and KMS26, significant resistance to RHM was observed in U266 cells, which express the lowest LIG3 protein level and Alt-NHEJ repair (Fig. 2B, C). Consistently, by analyzing protein levels of c-NHEJ-dependent LIG4 (Additional file 1: Figure S2A), an opposite pattern of expression to LIG3 and an inverse correlation with RHM sensitivity were found, further suggesting that RHM effects on MM cell viability are driven by LIG3-repair dependency. To characterize the induction of cell death, an Annexin V staining was used. RHM induced apoptotic cell death in AMO1 and NCI-H929 cells, (Fig. 2D, Additional file 1: Figure S2B), which was abolished by pan-caspase inhibitor Z-VAD-fmk (Additional file 1: Figure S2C), while no effects were observed on U266 cells (Additional file 1: Figure S2D). Clonogenicity of treated MM cells was also inhibited (Fig. 2E, Additional file 1: Figure S2E), and Haematoxylin and Eosin (H&E) staining showed chromatin fragmentation of treated cells (Additional file 1: Figure S2F). Importantly, RHM impaired the viability of primary MM plasma cells co-cultured with stromal cells (Fig. 2F), thus overcoming the pro-survival role of the MM-microenvironment. Consistently with tumor-selective killing activity, RHM did not affect viability of normal peripheral blood mononuclear cells (PBMCs) from healthy donors (n = 7), thus suggesting a good safety profile (Fig. 2G).

Since LIG3 plays a critical role in DNA DSBs repair, the effect of RHM on DNA damage response was investigated. Importantly, RHM induced a relevant increase in DNA DSBs with a relevant activation of DDR and apoptosis signaling, as demonstrated by increased phosphorylation of CHK1, CHK2, H2AX, occurring together with PARP1 and caspase-3 cleavage (Fig. 3A, B). Cell cycle analysis revealed also G2-arrest following RHM treatment, which was abrogated by caffeine (Additional file 1: Figure S3A), thus suggesting that checkpoint activation induced by RHM treatment depends on DDR signaling [7].

To investigate the molecular mechanisms underpinning the increase of DNA damage, functional analysis was carried out. First, AMO1 cell lysates were fractionated into nuclear-soluble and chromatin-bound fractions after RHM treatment, to evaluate interference on LIG3 nick-sensing activity. Notably, a decrease of LIG3 chromatin binding, as compared to nuclear-soluble fraction was observed (Fig. 3C), suggesting that RHM affects LIG3 recognition of DSBs impairing their repair. Next, the effects on mtDNA metabolism were investigated, given the exclusive role exerted by LIG3 in mitochondria DSB repair. Importantly, RHM treatment reduced mtDNA content (Fig. 3D left), as result of un-repaired mtDNA damage. Consistently with mtDNA function impairment by RHM, a reduction of mitochondrial mass and an increase of reactive oxygen species production (Fig. 3D right, Additional file 1: Figure S3B) were also observed. Finally, a validated plasmid-rejoining assay as a measure of Alt-NHEJ repair was performed. Notably, MM cells treated with RHM showed joining reduction as compared to the vehicle alone (Fig. 3E). Consistently with down-modulation of Alt-NHEJ repair pathway, SNP-array analysis confirmed an overall reduction of copy-number variations (CNVs) acquisition in RHM-treated cells (Fig. 3F), Additional file 1: Figure S3C), further indicating that RHM impairs LIG3-driven error DNA repair.

To investigate the translational relevance of our in vitro findings, in vivo anti-MM activity of RHM was evaluated in NOD-SCID mice bearing sub-cutaneous NCI-H929 xenografts. RHM intraperitoneal (i.p.) daily treatment (400 µg\Kg\day) resulted in a significant tumor-growth inhibition (Fig. 4A), which translated into survival prolongation of treated mice (Fig. 4B). Moreover, Immunohistochemistry (IHC) analysis showed decreased Ki67 expression, tumor regression, and necrotic pattern by H&E staining, thus confirming in vivo antitumor activity of RHM (Fig. 4C). Consistent with in vitro data, an increased expression of DDR and apoptosis markers in tumors retrieved from RHM-treated animals were found, such as phosphorylation of H2AX and Casp-3 and PARP1 cleavage (Fig. 4D).

Hyper-activation of the Alt-NHEJ pathway was been associated with drug resistance in several malignancies [23, 47], including in MM wherein higher LIG3 and PARP1 expression and Alt-NHEJ activity are associated with Bortezomib resistance [15, 16]. On these premises, the activity of RHM on the survival of Bortezomib-resistant cells (Additional file 1: Figure S4A) was then investigated. Notably, RHM exerted higher anti-proliferative activity in AMO1 Bortezomib-resistant (ABZB) cells as compared to their isogenic counterpart AMO1 (Fig. 5A).

To investigate the molecular mechanisms underlying the higher ABZB-sensitivity to RHM and considering the crucial role exerted by MYC in the induction of PARP-mediated Alt-NHEJ [16, 48], a potential role of MYC on LIG3 expression was investigated. Notably, analysis of MM patients’ dataset (GSE24080), disclosed a significant positive correlation between MYC and LIG3 mRNA expression (Additional file 1: Figure S4B). Consistently, a bio-informatic screening (cistrome.org) showed significant enrichment of MYC binding consensus sequences in the LIG3 promoter (Fig. 5B). To wet validate the in-silico data, MYC binding to the LIG3 promoter was first confirmed in AMO1 and ABZB by Chip analysis (Fig. 5C), according to MYC expression (Additional file 1: Figure S4C). Next, we demonstrated that MYC knockdown led to down-regulation of LIG3 protein expression and promoter activity and led to impaired RHM sensitivity (Fig. 5D, E), further suggesting MYC as a master regulator of LIG3-driven genomic instability and as a potential predictor of response to Alt-NHEJ inhibitors.