Coumarins from Magydaris pastinacea as inhibitors of the tumour-associated carbonic anhydrases IX and XII: isolation, biological studies and in silico evaluation

Abstract In an in vitro screening for human carbonic anhydrase (hCA) inhibiting agents from higher plants, the petroleum ether and ethyl acetate extracts of Magydaris pastinacea seeds selectively inhibited hCA IX and hCA XII isoforms. The phytochemical investigation of the extracts led to the isolation of ten linear furocoumarins (1–10), four simple coumarins (12–15) and a new angular dihydrofurocoumarin (11). The structures of the isolated compounds were elucidated based on 1 D and 2 D NMR, MS, and ECD data analysis. All isolated compounds were inactive towards the ubiquitous cytosolic isoform hCA I and II (Ki > 10,000 nM) while they were significantly active against the tumour-associated isoforms hCA IX and XII. Umbelliprenin was the most potent coumarin inhibiting hCA XII isoform with a Ki of 5.7 nM. The cytotoxicity of the most interesting compounds on HeLa cancer cells was also investigated.

CAs are ubiquitous metallo-enzymes, which are widely present in prokaryotes and eukaryotes. Sixteen different human (h) CA isoforms (I-XV, VA and VB) are present in mammals. These enzymes catalyse a very simple but essential physiological reaction of the life cycle of many organisms, the reversible hydration of carbon dioxide to bicarbonate and protons. As a consequence, CAs participate in various physiological and pathological processes and the deregulation of CA activity is associated with disorders and diseases such as cancer 9 , glaucoma 10 , epilepsy 11 and obesity 12 . Not surprisingly, hCA inhibitors have been intensively studied and several are in clinical use for diverse pathologies 10,13,14 . However, the systemic administration of hCA inhibitors (hCAIs) causes a wide range of side effects due to the lack of isozyme selectivity that leads to inhibition of the ubiquitous hCA I and II isoforms.
Accordingly, there is a continuous search of isoform-selective CAIs which should not inhibit the off-target CAI and II. The transmembrane hCA IX and XII are well-known tumour-associated isoforms and are overexpressed in many hypoxic tumours showing a restricted expression in normal tissues. In particular, under hypoxic conditions CA IX and XII decrease extracellular pH and promote tumour cell survival and invasion in hypoxic microenvironments 15,16 . For this reason, hCA IX and XII are attractive anticancer drug targets. Few natural coumarins have been evaluated as inhibitors of hCA IX and XII isoforms 8,17,18 but some of them were found active in the nanomolar range 18 .
Continuing our search for biologically active secondary metabolites from Sardinian higher plants 19,20 , we have found that the petroleum ether and ethyl acetate extracts of Magydaris pastinacea seeds exhibited potent inhibitory activity against hCA IX and XII. As a consequence, in order to identify the active compounds, a phytochemical investigation of the extracts was performed. Magydaris pastinacea (Lam.) Paol. (Apiaceae) [syn.: Magydaris tomentosa] is a plant that spontaneously grows in few regions of Mediterranean area such as Sicily, Sardinia, Corse, Baleares and North Africa. Previous studies on the rhizomes of M. pastinacea revealed as main metabolites coumarin and furocoumarin glucosides with inhibitory activity on platelet aggregation 21 . Furthermore, coumarins isolated from the flowers of the same plant showed antibacterial and anticoagulant activities 22 while a more recent work on the petroleum ether extract of flowers reported the cytotoxicity of furocoumarins and simple coumarins 23 .

General experimental procedures
Optical rotations were measured in CHCl 3 or MeOH at 25 C using a Perkin-Elmer 241 polarimeter. Circular dichroism spectra were recorded on a JASCO J-810 spectropolarimeter equipped with a Peltier temperature controller using a 10 mm path-length cell. All measurements were performed in methanol at compound concentration of 300 mM. Each reported spectrum represents the average of 3 scans recorded with 1-nm step resolution. Observed ECD signals were converted to molar ellipticity [H] ¼ deg Â cm 2 Â dmol À1 . UV spectra were recorded on a GBC Cintra 5 spectrophotometer. NMR spectra of all isolated compounds were recorded at 25 C on Unity Inova 500NB high-resolution spectrometer (Agilent Technologies, CA, USA) operating at 500 MHz for 1 H and 100 MHz for 13 C, respectively. Spectra were measured in CDCl 3 and CD 3 OD and referenced against residual non-deuterated solvents. HRESIMS were measured on an Agilent 6520 Time of Flight (TOF) MS instrument. Column chromatography was carried out under TLC monitoring using silica gel (40-63 lm, Merck), and Sephadex LH-20 (25-100 lm, Pharmacia). For vacuum-liquid chromatography (VLC), silica gel (40-63 lm) (Merck) was used. TLC was performed on silica gel 60 F 254 or RP-18 F 254 (Merck). LiChrolut RP-18 (40-63 lm) 500 mg, 3 mL (Merck) solid phase extraction (SPE) cartridges were also used. Semi-preparative HPLC was conducted by means of a Varian 920 LH instrument fitted with an autosampler module with a 1000 mL loop. The peak purities were monitored using a dualwavelength UV detector settled at 254 and 360 nm. The columns were a 250 Â 10 mm Spherisorb silica, particle size 5 lm (Waters) and a 300 7.5 mm Polymeric Reversed Phase (PLRP-S 100 Å), particle size 8 lm (Varian).

Plant material
The seeds of M. pastinacea were collected in July 2017 at Siniscola (Nuoro), Sardinia, Italy. The plant material was identified by Prof. Marco Leonti (University of Cagliari, Department of Biomedical Sciences). A voucher specimen (No. 0485) was deposited in the Herbarium of the Department of Life and Environmental Science, Drug Sciences Section, University of Cagliari.

Extraction and isolation
Air-dried and powdered seeds of M. pastinacea (720 g) were ground and extracted with petroleum ether (3.5 L) by percolation at room temperature to give 77.6 g dried extract. The remaining plant material was then extracted with EtOAc (3 L), giving 42 g dried extract.

Molecular modelling
The ligands were built within the Maestro platform. The most stable conformation has been determined by molecular mechanics conformational analysis performed with Macromodel software version 9.2 25 using the Merck Molecular Force Fields (MMFFs) 26 and GB/SA water implicit solvation model 27 , Polak-Ribier Conjugate Gradient (PRCG) method, 5000 iterations and a convergence criterion of 0.05 kcal/(mol Å). All the other parameters were left as default.
The coordinates for CA enzyme were taken from the RCSB Protein Data Bank 28 (PDB code 4ww8) 29 . The protein was prepared by using the Maestro Protein Preparation Wizard. Original water molecules were removed. Molecular docking studies were performed using the QMPL workflow protocol. Grids were defined around the refined structure by centreing them on the co-crystallized ligand. The other settings were left as default.
In order to better take into account the induced fit phenomena, the most energetically favoured generated complexes were fully optimised with the OPLS2005 force field in GB/SA implicit water 30 . The optimisation process was performed setting 10,000 steps interactions up to the derivative convergence criterion equal to 0.05 kJ/(mol Ã Å). The resulting complexes were considered for the binding modes graphical analysis with Pymol and Maestro.
2.6. Biological activity 2.6.1. Carbonic anhydrase inhibition assay An Applied Photophysics stopped-flow instrument has been used for assaying the CA catalysed CO 2 hydration activity 31 . Phenol red (at a concentration of 0.2 mM) was used as indicator, working at the absorbance maximum of 557 nm, with 20 mM Hepes (pH 7.5) as buffer and 20 mM Na 2 SO 4 (for maintaining constant the ionic strength), following the initial rates of the CA-catalysed CO 2 hydration reaction for a period of 10-100 s. The CO 2 concentrations ranged from 1.7 to 17 mM for the determination of the kinetic parameters and inhibition constants. For each inhibitor, at least six traces of the initial 5-10% of the reaction have been used for determining the initial velocity. The uncatalyzed rates were determined in the same manner and subtracted from the total observed rates. Stock solutions of inhibitor (0.1 mM) were prepared in distilled-deionized water, and dilutions up to 0.01 nM were done thereafter with the assay buffer. Inhibitor and enzyme solutions were preincubated together for 6 h at room temperature prior to assay in order to allow for the formation of the E-I complex. The inhibition constants were obtained by nonlinear least-squares methods using PRISM 3 and the Cheng-Prusoff equation, as reported earlier 32,33 , and represent the mean from at least three different determinations. All CA isoforms were recombinant ones obtained in-house as reported earlier 34 .
2.6.2. Cytotoxic assay 2.6.2.1. Cell culture. Human carcinoma HeLa cell line was obtained from the American Type Culture Collection (ATCC, Rockville, MD). Cells were grown in Dulbecco's modified Eagle's medium (DMEM) with high glucose, supplemented with 10% foetal calf serum (FCS), penicillin (100 units/mL)-streptomycin (100 lg/mL), and 2 mM L-glutamine in a 5% CO 2 incubator at 37 C. Subcultures of the HeLa cells were grown in T-75 culture flasks and passaged with a trypsin-EDTA solution. Cell culture materials were purchased from Invitrogen (Milan, Italy).
The HR-ESIMS of compound 11 showed a molecular ion at m/z 277.1078 (M þ H) þ which is in accordance with the molecular formula C 15 H 16 O 5 (calcd. 277.1076). The 1 H NMR spectrum of compound 11 showed two doublets at 6.23 (H-3, J ¼ 9.5 Hz) and 7.59 (H-4, J ¼ 9.5 Hz) ppm and a singlet at 6.77 ppm characteristic of a coumarin nucleus substituted at C-6, C-7 and C-8 (Table 1). In the high field spectrum region a further singlet at d 3.91 (3H, s) was characteristic of a methoxy group, while two singlets at 1.25 (3H, s) and 1.40 (3H, s) were assigned to two tertiary methyl groups. Finally, the signals at 3.36 (2H, dd, J ¼ 1.5, 9 Hz) and 4.86 (1H, t, J ¼ 9 Hz) could be ascribed to a methylene and a methine group, respectively. The 13 C NMR spectrum highlighted 14 carbons of which those at 161.2, 143.7 and 112.8 ppm were characteristic respectively of C-2, C-3 and C-4 of a coumarin nucleus ( Table 1). The connectivity of each proton with its respective carbon has been identified through HSQC experiment. In the HMBC spectrum, the correlations between the aromatic proton at d 6.77 (1H, s) and the carbons at 152.5, 146.1, 143. 7 and 141.8 ppm located this proton at position 5 of the coumarin system ( Figure 2). This was confirmed by the fact that no cross-peak of this proton with the carbon at 143.7 ppm could be observed if it would be located at C-6, C-7 or C-8. In the same spectrum the cross-peaks between the signal at 1. In the HMBC experiment it could be observed that the methoxyl group at 3.91 ppm was linked to the coumarin nucleus on carbon at 141.8 ppm. This data, together with the correlation between the aromatic proton at C-5 and carbon at 141.8 ppm, allowed us to place the methoxy group at C-6. The absence of further aromatic protons and the presence of only one alcohol group (d 1.58) in the proton spectrum, suggested the cyclisation of the alcohol chain at position 7 of the coumarin ring.
HSQC, HMBC, and ROESY experiments allowed the complete assignment of all signals and the identification of the structure as reported in Figure 2.
The absolute configuration of the (þ)-diidrofurocoumarin 11 has been established studying its Electronic Circular Dicroism (ECD) spectrum ( Figure 3). The positive Cotton effect at 348 nm in the ECD spectrum was ascribed to the n ! p Ã electronic transition of the unsaturated lactone ring, which permitted the assignment of the (2'S) absolute configuration based on the modified octant rule 36,37 . In fact, the 1-hydroxy-1-methylethyl group was located in the upper left (þ) octant. Compound 11 is a previously undescribed diidrofurocoumarin and was named magydarin.
The absolute configuration of the (þ)-heraclenol acetate 24 10 is not reported in the literature and therefore it has been established through the acetylation of the isolated (À)-heraclenol 9, for which the S configuration has been previously assigned 38,39 . The specific optical rotation of compound 9 ([a] D 25 ¼ À7.1) revealed that its absolute configuration is S 40 . Acetylation with acetic anhydride in pyridine of (À)-S-heraclenol yielded S-heraclenol acetate possessing a specific optical rotation [a] D 25 þ10. As consequence, the natural (þ)-heraclenol acetate (10) must have an absolute configuration S.
Compound 4 has been identified as trichoclin by comparison with analytical and spectroscopic data reported in the literature 41 . However, trichoclin contain a double bond whose geometric configuration could not be determined by comparison with the data reported in the literature. In order to assign the Z or E configuration to the double bond, a ROESY spectrum of compound 4 was recorded. The ROESY spectrum revealed a cross-peak between the olefinic proton at d 5.74 (1H, t, J ¼ 7 Hz) and the oxymethylene protons at 5.05 (2H, d, J ¼ 7 Hz) but the correlation between the proton at d 5.74 with the methyl group at 1.87 (3H, s) ppm could not be observed. This is well suited to an E geometry. In fact, a 3 D molecular model of E-trichoclin highlighted that the distance between the olefinic proton and the oxymethylene protons was 3.2 Å, whereas the distance between the same proton and the methyl group was 4 Å. In the case of Z isomer, the distance of 3.9 Å between the olefinic proton and the oxymethylene protons preclude any correlation whereas the short distance (2.3 Å) of the methyl at d 1.87 with methine at 5.74 ppm should be enough to give a cross-peak in the ROESY spectrum. Thus, compound 4 has been identified as E-trichoclin.
Compound 5 has been identified as 5-methoxy-trichoclin by comparison with analytical and spectroscopic data reported in the literature 41 . With respect to trichoclin, compound 5 contained the same chain and an additional methoxyl at position 5. As for E-trichoclin, the geometry of 5-methoxy-trichoclin double bond was determined by ROESY experiments. In the ROESY spectrum the cross peak between the olefinic proton at d 5.73 (1H, t, J ¼ 7 Hz) and the methylene protons at d 4.89 (2H, d, J ¼ 7.5 Hz), but not the methyl group at d 1.87 (3H, s) could be observed. Therefore, compound 5 has been identified as E-5-methoxy-trichoclin.

CA inhibition
The petroleum ether and ethyl acetate extracts together with all the isolated coumarins were tested against four hCA isoenzymes (I, II, IX and XII) ( Table 2). The results showed that both extracts were selective towards the tumour-associated CA IX and CA XII since none was active against CA I and CA II up to the concentration of 100 ng/mL, whereas they were highly potent especially against CA XII. The ethyl acetate extract was more potent compared to the ether extract either against CA IX (K i 1.74 ng/mL) either against CA XII (K i 0.5 ng/mL). The furocoumarins 1-9, 11 and coumarins 12, 15 inhibited both hCA IX and hCA XII in a nanomolar range and they were completely inactive (K i values > 10,000 nM) against hCA I and II, thus showing high selectivity over these isoforms that are considered responsible for the side-effects of CAIs. Interestingly, meranzin (13) and especially the furocoumarin 10, inhibited exclusively the IX isoform of CA with K i ratios XII/IX > 5 and >52, respectively. On the contrary, compound 14 was a selective inhibitor of the XII isoform with a K i ratio IX/XII > 34. The most potent compound was umbelliprenin (15), with a K i value against CA XII of 5.8 nM, comparable to that of acetazolamide (AAZ) and demonstrated high selectivity over the off-target CA I/II isoforms (K i ratios I/XII > 1724). Also S-heraclenol (9) (K i 27.5 nM), osthol (12) (K I 75 nM) and E-5-methoxy-trichoclin (5) (K I 63.5 nM) revealed a strong inhibitory effect towards CA IX (heraclenol) and CA XII (osthol and E-5-methoxy-trichoclin).
The new angular dihydrofurocoumarin, magydarin (11), had submicromolar inhibition of the validated cancer-associated isozymes hCA IX and XII with K i s of 150.9 and 623 nM, respectively.

3.3.
In silico modelling of the interaction of compounds 9, 11, 15 with CA XII Compounds 9, 11 and 15 are characterised by an interesting inhibitory activity towards the CA XII isoform, therefore, their mechanism of action was investigated in more details by means of computational methods. The protocol consisted of docking experiments followed by the energy minimisation of the obtained complexes. Recently, an interesting CA inhibition mechanism was reported for coumarin derivatives 8, 48 . This was also found as a plausible mechanism of action of previously synthesised compounds 49 .
Hence, the coumarin derivatives 15, 11 and 9 were docked to understand if coumarin moiety could be hydrolysed by the Zn 2þ activated water molecule of the enzyme cavity, which acts as a very potent nucleophile. These experiments showed that only compound 15 was able to dock the coumarin portion close enough to the Zn 2þ (Figure 4(a)). Therefore, only this compound could be hydrolysed. Instead, the compounds 11 and 9 binding did not show the right orientation, probably due to the steric hindrance of the furocoumarin moiety (Figure 4(b,c)). Thus, both the open conformations (E/Z) of compound 15 were subjected to docking experiments in order to predict the binding mode of hydrolysed forms. It is possible to see in Figure 5 as both diastereoisomers are stabilised by several hydrogen bonds and p-p interactions in the catalytic site. The hydrophobic portion fold up and is stabilised by internal contacts and interactions with hydrophobic residues in the cavity. The predicted affinity of open compounds was estimated to be better than the closed ones.

Cytotoxicity of compounds 5, 9-12 and 15 towards cancer cells
On basis of the CA inhibition results, the growth inhibitory effect of the most active/interesting coumarins (5, 9-12, 15) on HeLa cancer cell line has been carried out (Table 3). Umbelliprenin (15) showed moderate cytotoxicity against HeLa cells (IC 50 75 lM) and this result is in accord with other studies reporting the low cytotoxicity of the compound towards the breast cancer cell lines MCF and 4T1 50,51 . However, umbelliprenin was effective to inhibit tumour growth, angiogenesis and metastasis in mammary tumour-bearing mice when intraperitoneally administrated 52 . These data were confirmed by Zhang et al. 53 , which provided evidence that umbelliprenin may inhibit the growth, invasion and migration of gastric cancer cells by targeting the Wnt signalling pathway, with little to no harm in the lung, heart and kidney.
The low cytotoxicity of umbelliprenin may be related to its very high liposolubility. This hypothesis was confirmed by the fact that when umbelliprenin was encapsulated in nanoliposomes its antiproliferative effect against 4T1 cells increased by five folds 50 . Also osthol (12) showed only a moderate cytotoxicity with an IC 50 of 98 lM but, as reported in previous works 54,55 , its cytotoxicity greatly depends on the cancer cell line.
Most importantly, CA IX and CA XII are overexpressed in cancer cells under hypoxic condition. Thus, the moderate antiproliferative action showed by the isolated compounds is not surprising since the assay has been carried out under normoxic conditions.

Conclusions
Fifteen coumarins were isolated from the seeds of M. pastinacea. Magydarin (11) is new to the literature, while meranzine (13), xanthotoxin (2), E-trichoclin (4), E-5-methoxy-trichoclin (5), fellopterin (7), (þ)-S-heraclenol acetate (10) and meranzine acetate hydrate (14) were isolated for the first time from this plant. None of the 15 isolated coumarins have ever been evaluated for its inhibitory activity on carbonic anhydrases. All isolated compounds were inactive towards the ubiquitous cytosolic isoform hCA I and II (K i > 10,000 nM) that are considered responsible for the side-effects of CAIs. On the contrary, they were significantly active against the tumour-associated isoforms hCA IX and XII. CA XII was inhibited with single-digit K i by umbelliprenin (15) (5.7 nM) and with K i s spanning in the nanomolar range (63.5-74.5 nM) by compounds 5 and 12. As regards the other tumour-associated isoform hCA IX, it was potently inhibited by S-heraclenol (9) (K i ¼ 27.5 nM) and, to a lesser extent, by the new furocoumarin magydarin (11). Particularly noteworthy is the selectivity of heraclenol acetate (10) towards hCA IX over hCAI, hCAII and hCAXII (SI > 52). As far as we know, umbelliprenin is the most potent natural coumarin CA inhibitor. Molecular docking experiments suggested that the most potent coumarin 15 could be hydrolysed in the enzyme catalytic pocket. Furthermore, docking experiments estimated that the predicted affinity of open compounds was better than that of the closed ones. Overall the computational methods helped to rationalise the strong activity of compound 15 towards the CA-XII isoform and to suggest a plausible mechanism of action that would be further investigated to confirm it. , (e,f) CA -XII-9 and the relative 2D representation of the complexes stabilising interactions with the binding site residues represented with different colour depending on their chemical-physical properties: green, hydrophobic; cyan, polar; violet, positive; red, negative charged residues; grey, metal atoms. Instead, magenta arrows indicate the formation of hydrogen bond between protein and ligand, while grey lines indicate the interaction with the complexed ion.
Future work on these compounds will attempt to synthesise analogues with lower lipophilicity to achieve a better drug-like profile.