Synthesis and Biological Evaluation of Phenanthrenes as Cytotoxic Agents with Pharmacophore Modeling and ChemGPS-NP Prediction as Topo II Inhibitors

In a structure-activity relationship (SAR) study, 3-methoxy-1,4-phenanthrenequinones, calanquinone A (6a), denbinobin (6b), 5-OAc-calanquinone A (7a) and 5-OAc-denbinobin (7b), have significantly promising cytotoxicity against various human cancer cell lines (IC50 0.08–1.66 µg/mL). Moreover, we also established a superior pharmacophore model for cytotoxicity (r = 0.931) containing three hydrogen bond acceptors (HBA1, HBA2 and HBA3) and one hydrophobic feature (HYD) against MCF-7 breast cancer cell line. The pharmacophore model indicates that HBA3 is an essential feature for the oxygen atom of 5-OH in 6a–b and for the carbonyl group of 5-OCOCH3 in 7a–b, important for their cytotoxic properties. The SAR for moderately active 5a–b (5-OCH3), and highly active 6a–b and 7a–b, are also elaborated in a spatial aspect model. Further rational design and synthesis of new cytotoxic phenanthrene analogs can be implemented via this model. Additionally, employing a ChemGPS-NP based model for cytotoxicity mode of action (MOA) provides support for a preliminary classification of compounds 6a–b as topoisomerase II inhibitors.


Cytotoxicity
The cytotoxic assay of 11 naturally occurring and 19 synthesized phenanthrenes was carried out on a diverse set of human liver (HepG2 and Hep3B), oral (Ca9-22), lung (A549) and breast (MEA-MB-231 and MCF7) cancer cell lines, and a human fetal lung fibroblast (MRC-5) cell line (Tables 1 and 2). Doxorubicin was used as a positive control and an IC 50 .4 mg/ mL was considered inactive.
Among the 11 naturally occurring compounds [2], calanquinone A (CA-1) (5-OH, 6-OCH 3 ) and calanquinone B (CA-2) (5-OCH 3 , 6-OH) simply have reversed placements of the OH and one OCH 3 group, but CA-2 was much less potent than CA-1 ( Figure 2 and Table 1). The SAR results of CA-1 and CA-2 could possibly be explained by intramolecular hydrogen bonding between C = O (C-4) and OH (C-5) groups in 3-methoxy-1,4-PQs that may be a necessary moiety for cytotoxicity. To set up SAR correlations and identify active phenanthrene analogs, calanquinone A (CA-1) was selected as a lead compound for further studies.
Accordingly, 19 analogues including calanquinone A (6a; CA-1) were synthesized and tested in cytotoxicity assays. As shown in Table 2, calanquinone A (6a) and denbinobin (6b) exhibited significant potency against all cancer cell lines (IC 50 0.08-1.06 mg/mL). PQs 7a and 7b also showed very high potency against five cancer cell lines (IC 50 0.16-1.66 mg/mL), not including HepG2. Conversely and interestingly, PQs 5d and 5e were active only against the HepG2 cancer cell line with IC 50 values of 1.49 and 1.24 mg/mL, respectively. PQs 4a, 4b, 4c and phenanthrene 8a displayed selective activity toward the Ca9-22 cancer cell line with IC 50 values of 2.17, 3.45, 1.90 and 3.91 mg/mL, respectively. The SAR study of cytotoxicity suggested that the skeleton of 1,4-PQ is preferable to that of phenanthrene. To evaluate a potential SAR of the intramolecular hydrogen bond between C-4 and C-5, 3-methoxy-1,4-PQs 5a-b, 6a-b and 7a-b were designed. Compounds 6a and 6b, with OH at C-5 and C = O at C-4, can form an intramolecular hydrogen bond. However, the hydrogen donors of 5a-b and 7a-b have been replaced with OCH 3 and OAc groups, respectively. Among the six 3-methoxy-1,4-PQs, 6a and 6b exhibited the most significant potency, especially 6a (IC 50 0.08-0.89 mg/mL). Compounds 5a and 5b showed marginal activities against all human cancer cell lines. Surprisingly, the new 7a and 7b, with OAc at C-5 and C = O at C-4, were active toward five human cancer cell lines (IC 50 0.16-1.66 mg/mL), but not HepG2. These data also represent the first time we have found this phenomenon in a cytotoxic assay of PQ derivatives. To expand upon the SAR study, all natural and synthesized compounds were used for the 3D pharmacophore model building.

3D Pharmacophore Modeling
To further identify the critical structural features of the phenanthrene analogs, 29 compounds ( Figure 2) were used for pharmacophore modeling with Catalyst HypoGen. In this spatial aspect model, the phenanthrene structures and their cytotoxicity toward MCF-7 cancer cell line showed some interesting information.
The best pharmacophore model was established as a result of thirty runs with various parameters and characterized by a best correlation coefficient (0.931), the lowest total cost value (109.366), the highest cost difference (42.417), and the lowest RMS (0.790) (for details see Tables S1, S2 and Figure S1; Text S1). As shown in Figure 7, four essential features, three hydrogen bond acceptors (HBA1, HBA2 and HBA3) and one hydrophobic feature (HYD) were defined. All mutual distances of the four features can be measured. The distances between HBA1 and HBA2 or HBA3 were found to be 5.13 and 7.40 Å , respectively. The distances  between HBA2 and HBA3 or HYD were found to be 5.95 and 5.92 Å , respectively. The distances between HYD and HBA1 or HBA3 were found to be 3.83 and 4.78 Å , respectively. The distance between HBA2 and HBA3 is especially critical for the MCF-7 cytotoxic effect in this model.
The mappings of the best model with all compounds show fit values ranging from 6.18 to 8.57 (Table S3; Text S1). Calanquinone A (6a) mapped to the best hypothesis model with the fit value of 8.57 reveals significant features in Figure 8A. Obviously, the HBA1 links to the carbonyl group of quinone ring at C-1, HBA2 links to the oxygen atom of the methoxyl group at C-3, HBA3 links to the oxygen atom of the hydroxyl group at C-5, and HYD aligns to the aromatic ring (B-ring). Compound 6b has a similar alignment as 6a, with a high fit value of 7.95 ( Figure 8B and Table S3). As shown in Figures 8C and 8D, 7a and 7b, which are highly toxic to MCF-7 cells, also match against all features of the best hypothesis in which 7a was originally designed to remove the intramolecular hydrogen bond and was previously speculated to be less cytotoxic. The main difference in structure between 7a-b and 6a-b is the carbonyl group of the acetoxyl substituent at C-5. Consequently, the conserved distance between HBA2 and HBA3 explains why 7a and 7b, with OCOCH 3 at C-5 and C = O at C-4 but without the same intramolecular hydrogen bonds as 6a and 6b, can still possess significant cytotoxicity, in contrast to our previous speculation. For less MCF-7 cytotoxic compounds, mismatching one hydrogen bond in the triad and thus disrupting the structures of 5a and 5b results in moderate inhibition, one order higher in mg/mL. The oxygen atom of the methoxyl group at C-6, the carbonyl group at the C-4 position, and the aromatic atom at C-9 of 5a fit into the HBA2, HBA3 and HYD, respectively. However, they do not fit into HBA1 ( Figure 8E). Also, the carbonyl group at C-1, the oxygen atom of the methoxyl group at C-3, and the aromatic atom at C-9 of 5b match against HBA1, HBA2 and HYD features, but are not linked to HBA3 ( Figure 8F).
Thus, HBA1, 2 and 3 complete the triad of the hydrogen acceptor feature and clearly explain the MCF-7 cytotoxic variation of 5a-b, 6a-b and 7a-b. In addition, the hydrophobic feature HYD indicates a pharmacophore anchor for a three-ring core as in phenanthrene and PQ derivatives in the series. The hydrophobic feature links to all compounds and the HBA feature always  links to the carbonyl group at C-1 or C-4 of the quinone ring in all PQs. As a whole, a pharmacophore and explicit SAR were established herein.

ChemGPS-NP Analysis of Calanquinone A (6a) and Denbinobin (6b)
ChemGPS-NP (chemical global positioning system for natural products) is a computational model based on principal component analysis of physical-chemical properties. Such properties can be estimated directly from structure data, and by performing score prediction in the ChemGPS-NP model, this provides a versatile tool for charting and navigating the biologically relevant chemical space [15]. In a previous study ChemGPS-NP has successfully been used to chart a set of known anticancer agents with different cytotoxic mechanisms. The resulting map has been used as a tool to predict the anticancer Mode of Action (MOA) for new and previously unstudied lead compound [16]. As shown in Figure 9, the two most potent cytotoxic compounds, calanquinone A (6a)  and denbinobin (6b), were predicted in the model. Evaluating their resulting position on the chemical space map, it can be concluded that these phenanthrene derivatives do not unambiguously belong to any of the well defined groups representing alkylating agents, antimetabolites, proteasome inhibitions, tyrosine kinase inhibitors, topoisomerase I, and tubulin inhibitors except topoisomerase II inhibitors. The preliminary result of this ChemGPS-NP analysis indicates that calanquinone A (6a) and denbinobin (6b) might be members of a topoisomerase II inhibitor, which however, still remains to be further elucidated.

Topoisomerase II Assay
From the ChemGPS-NP analysis, it seems that the MOA for 3-methoxy-1,4-PQs might processes cytotoxicity as a topoisomerase II inhibitor. In the previous study [17], 1,4-benzoquinone has been found to poison human topoisomerase IIa. According to these results, we chose the most potential calanquinone A (6a) and its moderate compound 5a to test the DNA cleavage assay, in which known etoposide (VP-16) was used as the positive control. As shown in Figure 10, compound 6a showed the inhibition on Topo IIa in the result of the appearance of supercoided DNA instead of the relaxed one at the concentration of 100 mmol/L. Additionally, compound 5a also had the similar effect at higher concentration (200 mmol/L). Moreover, both compounds induced the formation of linear DNA, suggesting that they could possibly trap Topo IIa into DNA cleavage complex. Our data proved PQs inhibit hTopoII in vitro with inducing DNA strand breaks and protein covalently bound to DNA, ultimately leading to cell cycle arrest and death.
A best ligand-based pharmacophore model against the MCF-7 cancer cell line was successfully established. It explains the SAR of 3-methoxy-1,4-PQs 5a-b (5-OCH 3 ), 6a-b (5-OH) and 7a-b (5-OCOCH 3 ) in a spatial aspect model. Highly active 6a, 6b, 7a and 7b possess three hydrogen bond acceptors forming a hydrogen bond triad combined with one hydrophobic group as a pharmacophore that can interact with a potential target. The revealed pharmacophore model provides a bona fide basis for further design and synthesis of promising phenanthrene structures in vitro to study their anti-breast cancer properties. On the basis of ChemGPS-NP prediction and TopoII assay assessment, 1,4-PQs were suggested as the topoisomerase II inhibitors. This is the first time to apply ChemGPS-NP to previously untested cytotoxic compounds for MOA prediction. In the future, ChemGPS-NP could be used to effectively find the most possible MOA in the new drug discovery, as suggested by Rosén and co-workers [16].
Overall, our data demonstrate that PQs could be promising lead compounds for the further development of anti-cancer.

General
Melting points were determined on a YanacoH digital micro melting point apparatus model MP-500D without correction. NMR spectra were recorded on Varian Unity-plus 400 MHz FT-NMR and Varian Mercury-plus 400 MHz FT-NMR instruments. Chemical shift (d) values are in ppm (parts per million) with CDCl 3 as the internal standard, and coupling constants (J) are in Hz.

Cytotoxicity ASSAY
Compounds were tested against human liver (HepG2 and Hep3B), oral (Ca9-22), lung (A549), breast (MEA-MB-231 and MCF7) cancer cell lines, and the human fetal lung fibroblast (MRC-5) cell line using an established colorimetric MTT assay protocol [18]. The absorbance was measured at 550 nm using a microplate reader. The IC 50 is the concentration of agent that reduced cell growth by 50% under the experimental conditions.

3D Pharmacophore Model
The pharmacophore modeling with Catalyst HypoGen was performed via Discovery Studio 2.1 (Accelrys, San Diego, CA, USA) [19]. Twenty-nine phenanthrene derivatives were collected from the natural plant, C. arisanensis, and from chemical synthesis ( Figure 2). Cytotoxicity against MCF-7 cells was determined by the MTT assay and the concentration (mg/mL) of test compound which inhibited 50% of the cancer cells (IC 50 ) was used in the generation of the pharmacophore model. An IC 50 value of .20 mg/mL was defined as 20 mg/mL. All experimental IC 50 values spanned about 2-3 orders of magnitude from 0.09 to 20 mg/mL. The 2D/3D structures of compounds were generated using ChemBioOffice 2008 (Cambridge Scientic Computing, Cambridge, Massachusetts, USA) and then optimized in a Dreiding force field. The conformational ensemble of each compound was generated using the best conformational analysis method based on a CHARMM force field with a 20 kcal/mol energy threshold above the global minimum. A maximum limit of 255 conformations was used to cover maximum conformational space.
The best 3D arrangements of chemical functionalities should explain the activity variations among the 29 compounds. Thirty runs with different parameters were performed for the best pharmacophore hypothesis. Four chemical features, including hydrogen-bond acceptor (HBA), hydrogen-bond donor (HBD), hydrophobic (HYD), and aromatic ring (AR) features, were also tested during the building of pharmacophore hypotheses (Table  S1; Text S1). The best hypotheses were selected via a correlation and a cost analysis in Catalyst HypoGen.
Three costs including the total cost (the sum of weight, error and configuration cost), the null cost and the fixed cost will be evaluated. A total cost that is similar to the fixed cost and far from the null cost indicates statistically significant pharmacophore hypotheses. A difference between the total cost and null cost ranging from 40 to 60 indicates a true correlation of the pharmacophore hypothesis with 75-90% high probability. The true correlation represents ,50% probability when it is less than 40. Generally, the configuration cost should be smaller than 17 in a standard HypoGen model. According to the total cost (109.366), fixed cost (99.558), null cost (151.783), RMS value (0.790), and correlation coefficient (0.931) (Table S2), the best pharmacophore hypothesis, run 22, containing three hydrogen-bond acceptors (HBA1, HBA2, HBA3) and one hydrophobic feature (HYD) was selected (Figure 7).

ChemGPS-NP
The PCA-based model ChemGPS-NP (http://chemgps.bmc. uu.se) is a tool for navigation in biologically relevant chemical space. It has eight principal components (PC; dimensions), derived from 35 molecular descriptors describing physical-chemical properties such as size, shape, polarizability, lipophilicity, polarity, flexibility, rigidity, and hydrogen bond capacity for a reference set of compounds. The ChemGPS-NP descriptors were calculated for compounds 6a and 6b on the basis of their structure information as simplified molecular input line entry specification (SMILES) using the software DRAGON Professional. Compounds 6a and 6b were then mapped onto ChemGPS-NP using interpolation in terms of PCA score together with a reference set of known anticancer agents with previously studied Mode of Action (MOA) (Anticancer Agent Mechanism Database; http://dtp.nci.nih.gov/ docs/cancer/searches/standard_mechanism.html). Principal component and PCA score prediction were calculated employing the software SIMCA-P+, with the training set ChemGPS-NP. Prior to PCA determination, all data were centered and scaled to unit variance [16].

Topoisomeras II Assay
Topoisomerase II assay was performed by using a Topo II Drug Screening Kit (TopoGEN, Inc.). In brief, 0.1 mg of pHOT plasmid DNA was incubated with 2 units of topoisomerase IIa in 20 mL assay buffer at 37uC for 40 min in the presence of tested compounds (6a, 5a) and control drug, etoposide, respectively. 2 mL of 10% SDS and 2.5 mL of 10 mg/mL proteinase K were added into the reaction sample and then incubated for 30 min at 37uC to digest topoisomerase IIa. The samples were mixed with 2 mL of loading buffer and cleaned up by adding an equal volume of phenol:chloroform:isoamyl alcohol (25:24:1) according to the description. The sample was mixed by vortex and centrifuge for 10 sec. An aliquot (10 mL) of the upper aqueous part was analyzed by electrophoresis with 2% agarose gel containing 0.5 mg/mL of ethidium bromide [20].