A ribonucleotide reductase inhibitor with deoxyribonucleoside‐reversible cytotoxicity

Abstract

Ribonucleotide Reductase (RNR) is the sole enzyme that catalyzes the reduction of ribonucleotides into deoxyribonucleotides. Even though RNR is a recognized target for antiproliferative molecules, and the main target of the approved drug hydroxyurea, few new leads targeted to this enzyme have been developed. We have evaluated a recently identified set of RNR inhibitors with respect to inhibition of the human enzyme and cellular toxicity. One compound, NSC73735, is particularly interesting; it is specific for leukemia cells and is the first identified compound that hinders oligomerization of the mammalian large RNR subunit. Similar to hydroxyurea, it caused a disruption of the cell cycle distribution of cultured HL‐60 cells. In contrast to hydroxyurea, the disruption was reversible, indicating higher specificity. NSC73735 thus defines a potential lead candidate for RNR‐targeted anticancer drugs, as well as a chemical probe with better selectivity for RNR inhibition than hydroxyurea.

Highlights

• 12 bioactive inhibitors of human RNR were identified.

• The NSC73735 compound interacts with RNR and has specificity for leukemia cells.

• NSC73735 inhibits cell proliferation by cell cycle arrest.

• The induced cell cycle arrest is rescued by addition of DNA precursors.

• NSC73735 has higher potency and specificity compared to RNR inhibitor HU.

1. Introduction

Ribonucleotide Reductase (RNR) is an essential enzyme present in all free living organisms as well as in some double stranded DNA viruses. RNR catalyzes the reduction of ribonucleoside diphosphates into deoxyribonucleoside diphosphates and is rate‐limiting for dNTP synthesis. The mammalian enzyme consists of two subunits, one large substrate binding subunit denoted R1 (α, gene id RRM1), and one small radical generating subunit denoted R2 (β, gene id RRM2). There is also an alternative form of the small subunit, the p53R2 isoform (gene id RRM2B). RNR activity is mainly present in actively dividing cells that need to synthesize DNA, whereas terminally differentiated cells that have stopped dividing has much lower RNR activity. The activity of RNR is regulated with respect to the cell cycle with a peak at S‐phase, which is achieved by a distinct increase in the expression level of the R2 subunit (Chabes and Thelander, 2000; Eriksson et al., 1984). In contrast, the level of the R1 subunit is more constant throughout the cell cycle (Engström et al., 1985). In addition to regulation at the expression level, the RNR activity is controlled by a sophisticated allosteric mechanism to keep the dNTP pool in balance (Hofer et al., 2012), as unbalanced dNTP pools are mutagenic (Mathews, 2006). The relative ratios between the dNTPs are controlled by the allosteric specificity site (the s‐site), where binding of ATP/dATP induces CDP/UDP reduction and dTTP and dGTP induce GDP and ADP, respectively (Hofer et al., 2012). The human RNR has an additional allosteric activity site (the a‐site), an N‐terminal ATP cone domain (Aravind et al., 2000), which controls overall activity and hence the absolute dNTP concentrations; when ATP is bound to the a‐site the enzyme is active, and when dATP is bound the enzyme activity is shut off (Hofer et al., 2012). In all studied eukaryotic RNRs, binding of ATP or dATP to the a‐site results in hexamerization of the R1 subunit into α₆ complexes (Ando et al., 2016; Crona et al., 2013; Fairman et al., 2011; Kashlan et al., 2002; Rofougaran et al., 2006). The dATP‐inhibited α₆ complex forms a ring structure that binds the R2 dimer (β₂) in such a way that the crucial electron transport chain between the R1 and R2 subunits is disrupted (Fairman et al., 2011). Less is in known about the ATP‐induced α₆ complex, but mutagenesis studies of the Saccharomyces cerevisiae RNR indicated that it is structurally different from the dATP‐induced complex (Fairman et al., 2011). Furthermore, a recent study of the human RNR indicates that the ATP‐induced α₆ complex is less stable than the dATP complex as the ATP‐induced complex changes conformation upon β₂ binding, and higher order filamentous structures were also seen at high ATP concentrations (Ando et al., 2016).

The alternative small subunit, p53R2, is induced by p53 and is therefore associated with DNA repair (Guittet et al., 2001; Tanaka et al., 2000) and is crucial for mitochondrial DNA synthesis (Bourdon et al., 2007; Guittet et al., 2001; Pontarin et al., 2012; Tanaka et al., 2000).

All three RNR subunits, but particularly R2, have been found to be overexpressed in many cancer tissues and in some instances the expression levels of RNR subunits can serve as prognostic markers (Aye et al., 2015, 2011, 2011, 2005, 2006, 2011, 2013, 2008, 2007). RNR is also pharmaceutically relevant as the main target of the anticancer drugs hydroxyurea (HU) and gemcitabine. In addition, RNR has been the focus of many new drug discovery efforts, both in the field of cancer and antibiotics, e.g. in clinical trials involving RNR gene silencing (Davis et al., 2010; Jin et al., 2010; Sridhar et al., 2011), in clinical studies of triapine as an anticancer drug targeted to RNR (Nutting et al., 2009; Traynor et al., 2010), and as a target for novel antibiotics (Tholander and Sjöberg, 2012). In a recent study, Faiz Ahmad et al. identified a set of bioactive RNR inhibitors via in silico screening (Ahmad et al., 2015). In several cases, drugs and drug candidates affects the oligomer state of RNR, for example clofarabine, cladribine, fludarabine and phthalimide could increase the amount of α₆ complex (Ahmad et al., 2015; Aye and Stubbe, 2011; Wisitpitthaya et al., 2016) and gemcitabine was found to induce α₆β₆ complex formation (Wang et al., 2009). Undoubtedly, RNR defines an interesting drug target with proven clinical relevance.

Tholander & Sjöberg previously identified a set of compounds active as inhibitors of the Pseudomonas aeruginosa class I RNR (Tholander and Sjöberg, 2012), of which only a few also possessed activity against P. aeruginosa cells. In contrast, according to available data for bioactivity in the NCI‐60 cell line panel, most of the compounds are active in various human tumor cell lines. Here we show that these compounds are potent inhibitors of human RNR. One of the compounds, NSC73735, has particularly interesting properties suggesting a potential for lead development or usefulness as a chemical probe. Data mining shows that NSC73735 exhibit a bioactivity profile that correlates with that of hydroxyurea, and also with the expression of RNR genes. A combination of thermal shift experiments and quaternary structure determinations indicate that NSC73735 binds to the R1 subunit and interferes with oligomerization. In addition flow cytometry experiments show that the cytotoxicity of the compound causes a decrease in dNTP levels and a disruption of the cell cycle that is reversible by addition of deoxyribonucleosides, which together suggest that RNR is a cellular target.

2. Materials and methods

2.1. Materials

HU, deoxyribonucleosides, Triton X‐100, propidium iodide (PI), and all standard chemicals were from Sigma–Aldrich, Sweden. Glutamine and RPMI medium were from Lonza, RNase A from Macherey‐Nagel, and fetal calf serum (FCS) from Life Technologies. NSC73735 (redoxal, 2‐[(4‐{4‐[(2‐carboxyphenyl)amino]‐3‐methoxyphenyl}‐2‐methoxyphenyl)amino]benzoic acid) were from the Drug synthesis and Chemistry Branch, Developmental Therapeutics Program, Division of Cancer Treatment and Diagnosis, National Cancer Institute, USA, and previously found to inhibit a bacterial Ribonucleotide Reductase by a PCR‐based HTS method (Tholander and Sjöberg, 2012). Custom gene synthesis and subcloning into expression vectors were performed by Epoch Lifescience Inc., USA.

2.2. Cloning, protein expression and protein purification

The coding sequences of human R1 and R2 were codon optimized, custom synthesized and cloned into pET24 vectors. A 6×His‐tag was added to the N‐termini of all protein constructs. The proteins were expressed in Escherichia coli BL21 (DE3) in LB media (for R2) or in TB media (for R1). Overnight pre‐cultures were diluted 20 times into 1.5 L of fresh media and cultivated at 37 °C until the OD₆₀₀ reached approximately 0.5, at which point IPTG was added (250 μM for R2 and 50 μM for R1) to induce protein expression. Cultivation was continued at 15 °C overnight and the cultures then harvested by centrifugation at 4000 × g for 15 min. The collected cells were then resuspended in a buffer solution (50 mM Tris, 500 mM NaCl, 10 mM Imidazole, pH 7.5, and with the addition of complete protease inhibitor tablets from Roche according to the manufacturer's instruction) and lysed by sonication. Cell debris was removed by centrifugation at 10.000 × g for 15 min. All extraction steps were performed with the samples kept chilled.

Extracts of R2 were filtered (0.45 nm) and loaded onto a Ni‐chelate affinity resin (Qiagen) using an automated liquid chromatography system. Weakly bound protein was washed away (buffer: 20 mM Tris–HCl, 500 mM NaCl, and 10 mM Imidazole, pH 7.5) and the protein eluted with a buffer containing 20 mM Tris–HCl, 500 mM NaCl and 500 mM Imidazole, pH 7.5). Fractions containing the proteins of interest were pooled and concentrated (Sartorius Vivaspin, cutoff 30.000 kDa) to a final concentration of 30 mg/ml as determined by the Bradford method. Protein purity was estimated with SDS‐PAGE using a Phast gel system (Pharmacia) with 10–15% gradient gels.

For protein R1, the lysate cleared from cell debris was mixed with solid ammonium sulfate to a concentration of 40% of saturation, stirred for 45 min at 4 °C, and then centrifuged at 15.000 × g for 30 min to collect precipitated protein. The ammonium sulfate pellet was dissolved in buffer (20 mM Tris, 500 mM NaCl, and 10 mM imidazole, pH 7.5), filtered (0.45 nm) and loaded onto a Ni‐chelate affinity column (Qiagen, Sweden) using an automated liquid chromatography system. Weakly bound protein was washed away (buffer: 20 mM Tris–HCl, 500 mM NaCl, and 10 mM imidazole, pH 7.5) and the protein eluted with a buffer containing 25 mM Tris–HCl, 150 mM NaCl and 500 mM imidazole, pH 7.5. Fractions containing the R1 protein was then pooled, desalted on a HiPrep 26/10 desalting resin (GE Healthcare, Sweden) and the concentration determined by the method of Bradford. Protein purity was estimated as described for protein R2.

2.3. Reconstitution of the iron center and the tyrosyl radical content of subunits R2 and p53R2

To reconstitute the iron and radical content of the purified R2 subunit, a small volume (typically 6 μl) from an acidic solution of Fe²⁺ ions (5 mM (NH₄)₂Fe(SO₄)₂ in 10 mM HCl) was mixed with the Tris–HCl‐buffered protein solution (typically 100 μl of 50 μM protein) to give a molecular ratio of approximately 3:1 between iron and protein. The solution was incubated for 10 min at 24 °C and the protein was then used in enzyme activity assays or stored at −80 °C. Tyrosyl radical content was verified by UV absorption spectroscopy (Petersson et al., 1980).

2.4. Enzymatic dose–response assay and determination of IC50 values

To determine IC₅₀ values, a series of activity assays with different concentrations of inhibitors was prepared. The reaction mixtures (50 μl) contained 0.4 μM R2, 0.1 μM R1, 3 mM ATP, 10 mM Magnesium acetate, 0.1 mM FeCl₃, 1.5 mM CHAPS, 15 mM DTT, different concentrations of inhibitor (see results), 100 μM CDP, and 25 mM Tris–HCl buffer adjusted to pH 7.5. CDP was added last (after a pre‐incubation period of 30 min to account for slow active compounds) in a volume of 20 μl to start ribonucleotide reduction. After 60 min the reactions were quenched by boiling. The amount of dCDP formed in the reactions was determined with boronate affinity chromatography and liquid scintillation counting (Shewach, 1992). IC₅₀ values were determined by fitting a four‐parameter dose–response model (B + (B − T)/(1 + 10^{(log[IC50] − log[I])h)}, where B is the lower plateau, T is the top plateau, and h is the slope) to the data by nonlinear regression.

2.5. Thermal shift assay (TSA)

The TSA assay was performed using differential static light scattering (DSLS) on a Stargazer‐384 (Harbinger Biotechnology and Engineering Corporation, Canada) instrument in 384‐well optical bottom plates (Nunc, USA). The assay mixtures (50 μl) contained 0.9–2 μM human R1 or 4 μM human R2 protein in 50 mM HEPES pH 7.5, 100 mM KCl, 12 mM MgAc, 1.5 mM CHAPS, 5 mM TCEP, and 100 μM test compound dissolved in DMSO. The R2 mixture was complemented with 0.4 M Guanidine‐HCl. After the final addition of compounds to be analyzed, the assay mixtures were covered with 40 μl mineral oil and plates centrifuged at 3000 rpm for 5 min in a plate centrifuge (Hettich universal 320, Germany). Plates were heated at 1 °C/min in the range of 25–80 °C with images captured every 30 s. Using the provided software (Harbinger Biotechnology and Engineer corporation), light scattering intensities from the images were plotted as a function of temperature and the aggregation temperature (T_agg) calculated. ΔT_agg represents the difference between the aggregation temperature of the protein with (compound in DMSO) and without (only DMSO) the potential ligand. The data shown represent the mean and standard deviation of 2–4 samples.

2.6. Quaternary structure determination

GEMMA (Gas‐phase electrophoretic mobility macromolecule analysis) quaternary structure analysis was essentially performed as described previously (Rofougaran et al., 2008). The R1 subunit was included in the concentration of 0.05–0.1 mg/ml and the R2 subunit at 0.02 mg/ml in a buffer containing 20 mM ammonium acetate, 0.005% Tween 20, 500 μM ATP + 500 μM Mg²⁺ ions, or 50 μM dATP + 50 μM Mg²⁺ ions and 100 μM NSC73735, as indicated in Figure 1 and Suppl. Figure S2. A 2.5% DMSO control was included to validate that the effects seen were not due to the DMSO that NSC73735 was dissolved in (final concentration 2.5%). GEMMA analysis was performed at reduced pressure (1.4–2 psi) to minimize interference from nucleotides, Mg²⁺ and DMSO. Each sample was scanned 6 times, and a density of 0.58 g/cm³ was used for conversion from diameter to molecular mass.

Figure 1.

Quaternary structure of the R1 subunit in the presence of NSC73735. (A) GEMMA raw counts results with 0.05 mg/ml R1, 500 μM magnesium acetate and 500 μM ATP (black upper trace), in the presence of 2.5% DMSO (blue middle trace) or 100 μM NSC73735 in DMSO (red lower trace). (B) GEMMA results with 0.1 mg/ml R1, 0.02 mg/ml R2, 500 μM magnesium acetate and 500 μM ATP (black upper trace) and in the presence of 100 μM NSC73735 in DMSO (red lower trace). The numbers represent the molecular mass of the particles in kDa after conversion from their determined diameter. The non‐labeled peak in the beginning of some of the traces represents particles formed from nucleotides or other non‐volatile components in the sample. The traces are spaced 500 raw counts apart to allow visualization in the same graph.

2.7. Data‐base mining

All compounds were from NCI's compound collections and have been tested in the NCI‐60 panel of cell lines. To deduce relevant data for compound activities along with RNR gene expression data throughout the NCI‐60 cell lines, the cell miner analysis tool (http://discover.nci.nih.gov/cellminer/home.do) (Reinhold et al., 2012; Shankavaram et al., 2009) was used.

2.8. Analysis of drug likeness

To provide estimation on compound solubility and permeability, standard measures of drug and lead likeness were calculated using Instant JChem 6.2.1, 2014 (http://www.chemaxon.com). The following criteria were analyzed: The Lipinski rule of five (Lipinski et al., 2001) (molecular weight ≥500, number of H‐donors ≤5, and number of H‐acceptors ≤10), the Ghose drug likeness‐filter (Ghose et al., 1999) (160 ≥ molecular weight ≤ 480, 20 ≥ number of atoms ≤ 70, −0.4 ≥ logP ≥ 5.6, and 40 ≤ refractivity ≤ 130), The Veber filter for oral bioavailability (Veber et al., 2002) (number of rotatable bonds ≤12 and polar surface area ≤140 Å²), Chemaxon's filter for bioavailability (six criteria met of: molecular weight ≤500, logP ≤ 5, number of H‐donors ≤5, number of H‐acceptors ≤10, number of rotatable bonds ≤10, polar surface area ≤200 Å², and number of fused aromatic rings ≤5), and Chemaxon's filter for lead likeness (molweight ≤450, −4 ≤ logD_pH=7.4 ≤ 4, number of rings ≤4, number of rotatable bonds ≤10, number of H‐donors ≤5, and number of H‐acceptors ≤8).

2.9. Cell culture

Human promyelocytic leukemia cells (HL‐60) were cultured in RPMI medium supplemented with 10% (v/v) fetal bovine serum and 1% (w/v) glutamine, in a humidified incubator at 37 °C, 5% CO₂.

2.10. Cell viability

For viability measurements, cells were kept for 28 h in FCS‐free medium prior to start of treatment. 1 × 10⁶ cells/ml were then seeded in 24 well plates in complete medium and treated with 1, 5, 10 and 50 μM of NSC73735. 0.5% DMSO was added to the control cells (an equivalent concentration to the 10 μM NSC73735 treated cells). Controls and 10 μM treatments were repeated in triplicates. Cells were counted in the EVE cell counter (NanoEnTek, Seoul, Korea) at 24, 48 and 72 h, using the trypan blue exclusion method as an indication of cell viability.

2.11. Cell cycle analysis by flow cytometry

HL‐60 cells were cultured as described above. On the day of treatment, 0.7 million cells/ml were plated and treated with a deoxyribonucleoside mix (thymidine, deoxyadenosine, deoxyguanosine and deoxycytidine, 10 μM of each), NSC73735 (10 μM), HU (10 or 100 μM), or nucleoside and drug combined. Cells were harvested at 24 and 48 h, resuspended in 1 ml of PBS, permeabilized and fixed with 2 ml of ethanol and kept in the fridge until acquisition. On the day of the analysis, the cells were washed with PBS, and stained with PI in a final concentration of 0,05 mg/mL, including 0.1 mg/mL RNase A and 0,008% Triton X‐100. Cellular DNA content was determined using the BD LSR II flow cytometer operated via the FACSDiva software (BD Biosciences). Cell cycle analysis was performed with the FlowJo vX.0.7 software (Tree Star, Inc.). SSC‐A vs FSC‐A gate was applied to exclude cellular debris, followed by a PE‐H vs PE‐A gating for exclusion of doublets.

2.12. Determination of dNTP pools

HL‐60 cells were synchronized by FCS starvation for 28 h prior to treatment. The cells were then transferred to complete medium and treated with 10 μM NSC73735 or DMSO for 18 h. 20 million DMSO‐ or compound‐treated cells (in duplicates) were harvested by centrifugation (3 min at 1000 × g) and washed with 2 ml ice cold PBS. After removal of PBS, 1 ml extraction solution (65 volume units of acetonitrile, 35 volume units of water, and 100 mM formic acid) and 100 nmol dITP were added to the cells. After incubation for 20 min on ice with mixing for 10 s every second minute, cellular debris was removed by centrifugation (20.000 × g for 10 min). The resulting solution was evaporated to dryness under a stream of argon. The dried material was dissolved in 250 μl HPLC buffer A (see below).

For HPLC analysis, a Waters Symmetry C18 column (150 × 4.6 mm, 3.5 μm pore size) was used. 100 μl samples were injected and eluted at 1 ml/min with a stepwise gradient of 0–50% B in 18 min, 50% B in 20 min and 100% B in 10 min (buffer A: 10% methanol in 50 mM potassium phosphate buffer, pH 7.0, supplemented with 10 mM tributyl ammonium hydroxide (TBAH); buffer B: 30% methanol in 50 mM potassium phosphate buffer, pH 7.0, supplemented with 10 mM TBAH). Compound identification was achieved by comparison with injected standards. Relative quantification was obtained by peak height measurements in the chromatogram (UV absorbance at 271 nm) in relation to the peak for dITP.

3. Results

3.1. RNR inhibitors with activity in human cell lines

The previously identified inhibitors of P. aeruginosa RNR (Tholander and Sjöberg, 2012) are from the NCI diversity set II and have been tested for growth inhibition of human cell lines within the NCI‐60 tumor cell line anticancer drug screen project (Shoemaker, 2006). A rich source of relevant biological data is thus available (Reinhold et al., 2012). Twelve of the inhibitors had statistically reliable data for their bioactivity profiles in the NCI‐60 cell lines. These compounds exhibited selective activity throughout the cell line panel with both resistant and sensitive cell lines (Suppl. Figure S1). Interestingly, many leukemia cell lines were sensitive to the compounds, as well as to HU. The structures of these 12 compounds are presented in Table 1.

Table 1.

Properties of the bioactive RNR inhibitors.

NCS id.	Chemical structure	Drug likenessa	IC50b	R1 ΔTaggc	R2 ΔTaggc
NSC36758		+/+/+/+/+	<[E]	−23 ± 37	−56 ± 0.5
NSC40273		+/+/+/−/+	0.04 ± 0.0004	−0.79 ± 0.7	−1.9 ± 0.8
NSC45384		−/−/−/−/−	0.4 ± 0.003	−5.9 ± 12	−3.2 ± 2
NSC73735		−/+/+/−/+	3.1 ± 0.02	4.0 ± 0.01	−14 ± 0.1
NSC94945		+/+/+/+/+	<[E]	−0.99 ± 2	−8.0 ± 2
NSC102742		+/+/+/+/+	0.61 ± 0.004	−4.4 ± 0.008	−12 ± 3
NSC228155		+/+/+/+/+	1.9 ± 0.0001	0.44 ± 7	−5.9 ± 0.8
NSC278631		+/+/+/+/+	1.5 ± 0.0007	−1.6 ± 4	−10 ± 2
NSC632536		+/+/+/+/+	0.73 ± 0.001	−0.84 ± 0.5	−1.5 ± 1.0
NSC641396		+/+/+/+/+	1.0 ± 0.005	−6.6 ± 0.5	−6.1 ± 0.7
NSC645330		+/+/+/+/+	1.8 ± 0.002	−5.5 ± 0.6	−7.2 ± 0.5
NSC661221		+/+/+/+/+	4.7 ± 0.01	−5.2 ± 0.9	−5.4 ± 1

^aDrug likeness defined as fulfillment (+) or violation (−) of the Ghose, Veber, Lipinski, Chemaxon bioavailability, and Chemaxon lead likeness filters, see methods section. Structure drawing and analysis was performed with Instant JChem 6.2.1, 2014 (http://www.chemaxon.com).

^bIC50 values in μM were determined by in vitro activity assays and fitting a four parameter logistic model to the data by nonlinear regression. >[E] denotes that 100% inhibition was observed at the lowest tested concentration (10 nM), which was lower than the enzyme concentration of the assay. In comparison, the IC50 for HU was determined to be 63.6 μM ± 0.007. Values ± standard errors are shown.

^cΔTagg represents the difference between the aggregation temperature between protein with and without compound in Celsius degrees, as determined by the TSA assay. Values ± standard deviations are shown.

3.2. Correlations between the bioactivity profile of HU, new RNR inhibitors, and RNR gene expression

CellMiner was used to deduce correlation coefficients between the bioactivity profiles of the identified RNR inhibitors in the NCI‐60 panel of cell lines and the RNR inhibitors HU and Gemzar (Table 2). In addition, the table also contains correlation coefficients between the bioactivity profiles and the level of RNR gene expression (Reinhold et al., 2012). The transcript expression data available in Cellminer is compiled from 5 different microarray platforms (Reinhold et al., 2012). Two compounds, NSC73735 and NSC632536, and the two analogues NSC641396 and NSC645330, exhibited high statistically significant correlations >0.5 with HU‐related cell line sensitivities, suggesting a similar mode of action on the cellular level (Reinhold et al., 2012). NSC73735 was the only compound for which cellular sensitivity correlated with RNR gene expression. Several other compounds correlated to a lower extent with the bioactivity profile of HU, but not with RNR gene expression.

Table 2.

Pearsons correlation coefficients.

3.3. Drug likeness

With the exception of NSC45384, most compounds fulfilled the majority of the analyzed drug likeness criteria (Table 1); the Ghose, Veber, and Lipinski rules, and also the Chemaxon rules for bioavailability and lead likeness. Thus, in terms of these measures many of the compounds are possible lead compounds. Notably, these measures do not take the inherent toxicological properties of the compounds into account. Strictly applying such a toxicology filter would probably disqualify many of the compounds for further drug development. However, antineoplastic compounds rarely meet such criteria since their mode of action involves toxic mechanisms to achieve cell killing.

3.4. Dose‐response experiments

The 12 RNR inhibitors with statistically reliable data for bioactivity were analyzed for their potency to the human RNR enzyme. IC₅₀ values for the inhibitors were determined and found to be in the range of 40 nM–4.7 μM (Table 1), which are in level with the values observed for the bacterial enzyme (Tholander and Sjöberg, 2012). In comparison, IC₅₀ for the known RNR inhibitor HU was determined to 64 μM. With consideration to assay conditions, some of these values are in level with the enzyme concentration or lower. For NSC94945 and NSC36758, 100% inhibition was observed at a concentration (10 nM) well below the enzyme concentration used in the assay (0.1 μM). Since the lowest theoretical IC₅₀ value for a reversible inhibitor is half the enzyme concentration, very potent binders or other mechanisms than reversible inhibition may be involved for those compounds (see discussion).

3.5. Thermal shift assay results

Thermal stabilization of proteins in the presence of potential ligands is indicative of direct protein‐ligand interaction. Here a light scattering based thermal shift assay (TSA) (Senisterra et al., 2006) was used to detect binding of the compounds to the RNR R1 and R2 proteins (Table 1). The only significantly stabilizing compound was NSC73725 that caused a shift in the thermal aggregation temperature (ΔT_agg 4.0 ± 0.011 °C) of the R1 protein, at a concentration of 100 μM compound. In comparison, the known RNR ligand ATP stabilized the R1 protein to a lower degree (ΔT_agg 1.3 ± 0.36 °C) at a concentration of 5 mM. None of the other bioactive inhibitors gave a significant stabilization of R1. For R2, none of the inhibitors was found to give a significant increase in the thermal stability. Several compounds caused a decreased stability of either or both R1 and R2 at the tested compound concentration, please see the discussion.

3.6. Quaternary protein structure determination

We used GEMMA to study if NSC73735 binding interfered with the allosteric regulation‐related oligomerization of RNR. Indeed, the ATP‐induced hexamer formation (Figure 1A, black upper trace) is abolished in the presence of NSC73735 (Figure 1A, red lower trace). As a consequence, no induced α₆β₂ complex was formed in the presence of NSC73735 (Figure 1B). Similarly, no dATP‐induced α₆ complex was formed in presence of NSC73735 (Suppl. Figure S2A). The oligomeric state of R1 in the presence of only DMSO at the same concentration as in samples with NSC73735 showed that the effects were not due to the solvent (Figure 1A, blue middle trace). The GEMMA experiments were performed with the mouse R1 and R2 proteins that have 97% and 91% sequence identity with their human counterparts. Human and mouse RNR showed the same trends in GEMMA (Suppl. Figure S2), but because mouse RNR gave better signals and lower backgrounds, this enzyme was used instead of human RNR to get more reliable data.

3.7. Growth inhibition by NSC73735 of HL‐60 cells

Low micromolar potency, physical interaction and complex interfering urged us to progress to cell studies of the NSC73735 inhibitor. In the NCI‐60 screen, human promyelocytic cells HL‐60 showed both sensitivity towards NSC73735 and HU (Suppl. Figure S1) and were chosen for the following studies. Cells were treated with 10 μM NSC73735 and cell viability was determined after 24, 48 and 72 h with trypan blue exclusion as viability indicator (Figure 2A). The number of living control cells increased substantially after 24 h, whereas the number of living cells treated with NSC73735 did not increase at all over time. Notably, when calculating the percentage of viable cells with respect to the total number of cells (Figure 2B) we observed no difference between untreated and treated cells after 24 h, even though the definite number of live cells after treatment was significantly lower, suggesting that treated cells were alive but unable to proliferate. At 48 h of treatment the percentage of viable cells started to decrease, and this effect became even more evident after 72 h, indicating that the arrested cells gradually died after prolonged treatment. Titration of NSC73735 showed that the compound affected cell proliferation and viability in a concentration dependent manner (Suppl. Figure S3).

Figure 2.

Cell viability in presence of NSC73735. (A) HL‐60 cells were treated with 10 μM of NSC73735 and compared to untreated cells. Cells were counted after 24, 48 and 72 h, using trypan blue exclusion as an indication of viability. (B) Living cell count at each time point with % of cell viability ([viable cells/total cell count]*100). Student's t‐test: **p < 0.01, ***p < 0.001, ns: non‐significant.

3.8. Cell cycle analysis

Flow cytometry was used to further analyze a potential cell cycle arrest as the underlying mechanism to the NSC73735 triggered HL‐60 cell growth inhibition. Treatment of HL‐60 cells with 10 μM of NSC73735 reduced the number of cells in the S and G2 phases and increased the number of cells in the G1 phase (Figure 3). Because RNR is required for deoxyribonucleotide synthesis, the addition of deoxyribonucleosides to the growth medium could in principle counteract the cytotoxic effects of a potential RNR inhibitor. Indeed, co‐treatment with 10 μM of each deoxyribonucleoside (thymidine, deoxyadenosine, deoxyguanosine and deoxycytidine) rescued the effect of NSC73735, as can be seen by almost perfectly restored cell cycle distribution (Figure 3). The effect of HU was much weaker (Suppl. Figure S4), and higher concentrations (100 μM) were needed to get a similar effect on the cell cycle distribution as with 10 μM NSC73725 (Figure 3). In contrast to the treatment with NSC73735, the effect of HU could not be counteracted by addition of deoxyribonucleosides, indicating that HU also has non‐RNR related effects. The maximum DMSO concentration used in the experiments did not affect the cell cycle profile (Suppl. Figure S4).

Figure 3.

Cell cycle profile analysis of NSC73735 treated cells by flow cytometry. HL‐60 cells were treated as indicated for 24 h and the cell cycle profile analyzed by flow cytometry after PI staining. Graphs are representative from two independent experiments. Sub‐G, G1, S and G2 fractions are indicated as percentage of the total number of singlets. dNs, deoxyribonucleosides.

3.9. dNTP pool analysis

Nucleotides were extracted from synchronized HL‐60 cells, cultured in the presence or absence of 10 μM NSC73735 for 18 h. All four dNTP levels exhibited lower relative amounts in the compound‐treated cells compared to DMSO‐treated controls (Figure 4). dCTP was detectable in the control cells but below the detection limit in the drug‐treated cells.

Figure 4.

Relative dNTP levels in NSC73735‐treated cells. dNTPs were extracted from NSC73735‐ and DMSO‐treated HL‐60 cells and their relative amounts determined. The levels in the control samples were defined as 100%. Note that dCTP was detectable in the control cells but below the detection limit in the drug treated cells.

4. Discussion

RNR is a recognized but underexplored target for antiproliferative drugs. In this study we set out to explore the potential of a set of the compounds (Tholander and Sjöberg, 2012) from the NCI diversity II compound library as inhibitors of the human RNR. A majority of compounds in the NCI diversity sets have been tested for growth inhibition in the NCI‐60 panel of cell lines and the data together with for example gene expression profiles are available via CellMiner (Reinhold et al., 2012, 2014, 2009). The bioactivity profile and the cell line selectivity are naturally first major measures of the lead potential of a particular compound. In addition, the mode of action of a certain compound can be evaluated by comparing cell line sensitivity with the expression level of the hypothesized target gene, and by looking for correlations in biological activities with a compound with a known mode of action (Reinhold et al., 2012).

Using CellMiner we found that 12 of our RNR inhibitors possessed growth inhibitory activity in many different human cell lines. Notably, the bioactivity of these compounds is selective and the compounds are not generally cytotoxic to all types of cell lines tested; while some cell lines exhibit resistance, others exhibit sensitivity towards the inhibitors. Since a potential lead compound should possess cellular selectivity for cancer cells this is an interesting property of the compounds. A general trend that emerges from the data is that leukemia cell lines are sensitivity to the compounds (with the exception of NSC45384), and that many lung cancer cell lines are resistant. Among cell lines with other tissues of origin, the response to the compounds are more variable.

The compounds NSC73735, NSC632536 and the three analogues NSC641396, NSC645330 and NSC661221, clearly has an above average activity against leukemia cell and also exhibit selectivity for this type of cell line. The responses of cell lines to these compounds, together with available cell line data (Section 3.2.), make them particularly interesting and suggest that they target RNR in vivo. This is supported by the fact that the bioactivity profile of these compounds in the NCI‐60 panel of cell lines exhibits a statistically significant positive correlation with the bioactivity profile of HU, a well‐known RNR inhibitor (Suppl. Figure S1 & Table 2). For NSC73735, there is also a positive correlation between the transcript levels of R1 or R2 genes and the bioactivity of the compound (Table 2).

We used recombinant human RNR to determine the IC₅₀ values of the 12 bioactive RNR inhibitors. The inhibitors were found to inhibit human RNR with IC₅₀ values between 40 nM and 4.7 μM (Table 1). Two compounds caused 100% inhibition at a concentration (10 nM) well below the enzyme concentration used in the assay (0.1 μM). This suggest that other mechanisms than reversible inhibition is involved, such as DTT driven redox‐cycling inactivation of the enzyme, where DTT repeatedly regenerates the reduced form of the inhibitor.

In addition to its correlation between RNR gene expression and bioactivity, NSC73735 was particularly interesting as it was the only compound that induced a shift in the thermal aggregation of the R1 protein. A positive shift is a strong indication of a direct binding of the molecule to the protein. In line with this, we found that NSC73735 hinders the oligomerization of the R1 subunit into hexamers (Figure 4). This is in contrast to the previously known oligomerization modifying inhibitors that instead stabilized the α₆ (Ahmad et al., 2015; Aye and Stubbe, 2011; Wisitpitthaya et al., 2016) or α₆β₆ complexes (Wang et al., 2009). In the Thermal shift assay, the majority of the bioactive compounds caused a decreased stability of the RNR proteins. Destabilization of a truncated version of the R1 protein in the presence of the nucleoside analog azathioprine has previously been observed (Egeblad et al., 2012) and we have observed that the absence of the R2 di‐iron cofactor decreases the aggregation temperature ∼3.6 °C compared to holo human R2 protein. Even though compounds might cause destabilization by interfering with subunit oligomerization and cofactor binding, unspecific destabilization cannot be ruled out in these cases.

Because NSC73735 exhibited several properties suggesting a selective inhibition of RNR, we investigated the effects of this compound on cultured HL‐60 cells and also analyzed its effect on the cell cycle distribution by flow cytometry. In terms of growth inhibition, we observed that NSC73735 caused an apparent arrest of cultured HL‐60 cells and a gradually induction of cell death after prolonged (>24 h) exposure. We also saw that cells treated with NSC73735 exhibited lower levels of dNTPs compared to DMSO‐treated control cells. To further assess if RNR was the cellular target we included combination treatments with our compound and deoxyribonucleosides. Previous studies indicate that nucleosides are taken up by HL‐60 cells mainly via equilibrative nucleoside transporters (ENT) (Baldwin et al., 1999; Ritzel et al., 2001; Tang et al., 2012; Yamauchi et al., 2014). Because RNR is required for deoxyribonucleotide synthesis, addition of deoxyribonucleosides was used to counteract the toxic effects of the compound. Indeed, the combined treatment of NSC73735 and deoxyribonucleosides rescued the cells from the effect on the cell cycle observed upon treatment with NSC73735 alone. In contrast, the toxic effects of HU, used at a concentration giving a similar effect (10‐fold higher) on the cell cycle distribution as NSC73735, could not be reverted by co‐treatment with deoxyribonucleosides.

HU is considered to mainly act by inhibition of RNR and is often used as a chemical probe and reference inhibitor in studies pertaining to RNR and dNTP pool levels (Saban and Bujak, 2009). HU inactivates RNR by quenching the tyrosyl radical on the R2 subunit, but probably has many other cellular effects due to its inherent metal‐chelating properties, formation of cytotoxic reactive nitrogen species (RNS) and reactive oxygen species (ROS), like nitric oxide (Heo et al., 2014) and hydroxyl radicals (Davies et al., 2009). Notably, the dNTP pool depletion caused by HU is not easily reversed by addition of exogenous deoxyribonucleosides to cultured cells (which would be expected for a clean inhibition of RNR), and attempts in this regard has given different outcomes (Adams and Lindsay, 1967; Lagergren and Reichard, 1987; Plagemann and Erbe, 1974; Scott and Forsdyke, 1980; Snyder, 1984; Young et al., 1967), thus further suggesting that HU not only acts by inhibition of RNR.

Interestingly, NCS73735 has previously been shown to target dihydroorotate dehydrogenase (DHODH) (Cleaveland et al., 1995; Knecht and Loffler, 2000), a key enzyme in the synthesis of pyrimidine nucleotides. Inhibition of RNR by NSC73735 might therefore augment the nucleotide depleting effect caused by DHODH inhibition.

5. Conclusions

We have identified a set of new inhibitors of the human RNR. Several of these inhibitors possess biological activity in cell line growth assays, and have selectivity for certain cell line subtypes. The compounds NSC73735, NSC632536 and the three analogues NSC641396, NSC645330 and NSC661221 are particularly interesting with distinct activity against leukemia cell lines and with properties suggesting that RNR is being targeted in the cell. In addition, the compounds meet standard criteria for drug and lead likeness, as well as criteria for bioavailability. For NSC73735, a positive effect on protein stability, interference with protein complex formation, a lowering effect on the dNTP pools, and a deoxyribonucleoside‐reversible disruption of the cell cycle distribution of cultured HL‐60 cells suggests that RNR is a cellular target. This compound is thus a potential lead candidate for RNR‐targeted anticancer drugs, but can also serve as a pharmaceutical probe with better specificity for RNR inhibition than HU.

Supporting information

The following are the supplementary data related to this article:

Supplementary Figure S1 Normalized bioactivity of the identified RNR inhibitors in the NCI‐60 cell lines. Cell line names are given to the left and the graphs are grouped by color according to the tissue of origin: Breast (BR, dark blue), Central Nerve System (CNS, brown), Colon (CO, orange), Leukemia (LE, light green), Melanoma (dark green), Non‐small Cell Lung (LC, light blue), Ovarian (OV, wine red), Prostate (PR, yellow), Renal (RE, red). Compound names are given above each panel, for comparison, the sensitivity of the cell lines to HU is shown in the right most panels. The data was derived using the CellMiner database (http://discover.nci.nih.gov/cellminer/home.do) with the novel RNR inhibitors as search criteria. The scale on the X‐axis represents the Z‐score of the given compound in each cell line, where a positive value indicates a cell line with higher sensitivity towards to the indicated compound in relation to the mean effect of the compound on all cell lines (the zero value), whereas a negative value indicates higher resistance than average. The average zero values (IC50/IG50) in μM are 0.8 (NSC36758), 3.6 (NSC278631), 0.7 (NSC45384), 9.8 (NSC73735), 3.6 (NSC94945), 1.4 (NSC102742), 4.3 (NSC228155), 35.5 (NSC40273), 24.6 (NSC632536), 9.1 (NSC641396), 6.3 (NSC645330), 2.8 (NSC661221), 660.7 (HU).

Supplemental Figure S2 Comparison of quaternary structure of the mouse and human R1 proteins. GEMMA results with 0.05 mg/ml R1 alone (black upper trace) in the presence of 50 μM dATP and 50 μM magnesium acetate (blue middle trace) and with the additional presence of 100 μM NSC73735 in DMSO (red lower trace). Results from the mouse R1 protein are shown in A and the human R1 in B. R1 hexamer formation is clearly disturbed for both the mouse and human R1 proteins. The numbers represent the molecular mass of the particles in kDa after conversion from their determined diameter. The non‐labeled peaks in the beginning of some of the traces represents particles formed from nucleotides or other non‐volatile components in the sample. The traces are spaced 200–500 raw counts apart to allow visualization in the same graph.

Supplemental Figure S3 Cell viability in presence of different concentrations of NSC73735. HL‐60 cells were treated with the indicated NSC73735 concentrations. Cells were counted at indicated times, using trypan blue exclusion as an indication of viability. Living cell count for each concentration at different time points (A). % of cell viability ([viable cells/total cell count]*100) calculated for the different times and concentrations used (B).

Supplemental Figure S4 FACS analysis of low HU‐ and DMSO‐treated cells. Cells were treated with 10 uM of HU or 0.7% DMSO for control purposes.

Supplementary data 1.

1.1.

Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.molonc.2016.07.008.

Crona Mikael, Codó Paula, Jonna Venkateswara Rao, Hofer Anders, Fernandes Aristi P., Tholander Fredrik, (2016), A ribonucleotide reductase inhibitor with deoxyribonucleoside‐reversible cytotoxicity, Molecular Oncology, 10, doi: 10.1016/j.molonc.2016.07.008.