Polycyclic nitrogen heterocycles as potential thymidine phosphorylase inhibitors: synthesis, biological evaluation, and molecular docking study

Abstract New polycyclic heterocycles were synthesised and evaluated as potential inhibitors of thymidine phosphorylase (TP). Inspired by the pharmacophoric pyrimidinedione core of the natural substrate, four series have been designed in order to interact with large empty pockets of the active site: pyrimidoquinoline-2,4-diones (series A), pyrimidinedione linked to a pyrroloquinoline-1,3-diones (series B and C), the polycyclic heterocycle has been replaced by a pyrimidopyridopyrrolidinetetraone (series D). In each series, the tricyclic nitrogen heterocyclic moiety has been synthesised by a one-pot multicomponent reaction. Compared to 7-DX used as control, 2d, 2l, 2p (series A), 28a (series D), and the open intermediate 30 showed modest to good activities. A kinetic study confirmed that the most active compounds 2d, 2p are competitive inhibitors. Molecular docking analysis confirmed the interaction of these new compounds at the active binding site of TP and highlighted a plausible specific interaction in a pocket that had not yet been explored.


Introduction
Angiogenesisthe formation of new blood vessels from pre-existing vasculaturehas been validated as a target for several tumours and has been shown to promote tumour growth and metastasis 1 . Given the complexity of this process, there is a need for new compounds targeting multiple pro-angiogenic factors [2][3][4][5] . In this context, thymidine phosphorylase (TP) seems to be an interesting and understudied therapeutic target.
The physiological role of TP is to catalyse the reversible phosphorolysis of thymidine into thymine and 2-deoxyribose-1-phosphate (2dR1P) that is metabolised into 2-deoxyribose (2dR) ( Figure  1). TP has been shown to be up-regulated in the hypoxic regions of many solid tumours (stomach, pancreas, kidney, oesophagus, breast, ovary, lung, colon, bladder, uterus, kidney … ) 4,6,7 . Usually, this over-expression is highly associated with tumour micro vessel level, infiltration, metastasis and also correlates with the aggressiveness and invasiveness of the cancer 6 . Via 2dR, TP stimulates the secretion of several pro-angiogenic factors such as VEGF, MMP-1, IL-8 by both malignant cells and stromal cells located in the tumour microenvironment 8,9 . Consequently, TP inhibitors seem to be promising agents 10 .
Various crystal structures of Escherichia coli TP and human TP (hTP) complexes with thymine or thymidine analogues have been published [11][12][13][14][15][16] . Briefly, the binding of thymidine is ensured by three hydrogen bonds between the two carbonyls and the NH function with an arginine, a serine and a lysine. These studies revealed a large empty space (pocket 1) facing the C-5 and C-6 of the pyrimidinedione and another (pocket 2) close to the 1-position ( Figure 2).
Various TP inhibitors have been reported in the literature 8,17,18 . For a recent review see Sajid et al. 19 . In order to interact with the thymidine binding site, most of them are pyrimidine-2,4-dione derivatives. 7-Deazaxanthine (7-DX) and 5-chloro-6-[1-(2-iminopyrrolidinyl)methyl]uracil hydrochloride (TPI) have emerged ( Figure  2) 20,21 . Up to now, the most potent inhibitor is TPI (IC 50 hTP ¼ 0.035 mM 8 ), whose anti-angiogenic activity was attested through several in vitro and in vivo models [22][23][24] . TPI has been co-crystallized with hTP 12 . In combination with trifluridine, TPI has been recently approved (TAS-102, Lonsurf V R ) for the treatment of metastatic colorectal cancer 25 . The role of TPI is dual, it prevents the degradation of trifluridine by TP and exerts anti-angiogenic effect.
Only a few polycyclic TP inhibitors have been reported (for a review see Bera et al. 8 ). Crystallographic studies data allowed us to design inhibitors in which the natural ligand feature is linked to a polycyclic aromatic nitrogen heterocycle either by ring annelation or via various linkers.
In the present study, we describe the synthesis and the biological in vitro evaluation as TP inhibitors of four series of new original aromatic derivatives. To elucidate the mechanism of enzyme inhibition of the hits, a brief kinetic study was attempted. In silico molecular docking studies have also been performed to explore the binding site and possible interactions mode of these new derivatives with TP.

Chemistry
In a first time, we have designed three series of new heterocycles as potential TP inhibitors. The general structures are presented in Figure 3. In series A, the pyrimidinedione was annelated to a quinoline giving a rigid tricyclic heterocycle. The substituents on the 5 to 9-positions could modulate the interaction with pocket 1. Compounds of series B and C were more flexible. Thus, a pyrimidinedione substituted by a methylenic chain either on the 1-position (series B) or on the 6-position (series C) was linked to the 2position of a quinolopyrrolidinedione. In a second time, the results of a first docking study allowed us to design compounds of series D in which the quinolopyrrolidinedione has been replaced by a pyrimidopyridopyrrolidinetetraone.
Our strategies for synthesising the tricyclic heterocycles are based on one-pot multicomponent reactions, developed in our group, that involved an aniline, an aldehyde, and a 1,3-dicarbonyl derivative 26-28 .
2.1.1. Pyrimido [4,5-b]quinoline-2,4(1H, 3H)-diones (series A) We have described a very simple and convenient one-pot reaction for the synthesis of pyrimido [4,5-b]quinoline-2,4(1H,3H)-diones involving an aldehyde, an aniline, and barbituric acid 1 as the 1,3dicarbonyl reagent (Table 1) 27 . This straightforward method circumvents the preparation of unstable substituted 2- aminobenzaldehydes that limits the scope of previously described syntheses [29][30][31] . The use of commercially available anilines allowed the facile syntheses of pyrimido [4,5-b]quinolinediones substituted on the 6 to 9-positions with electron donor or electron-withdrawing groups. Access to the 5-substituted derivatives is also possible starting from aliphatic or aromatic aldehydes. We have previously described the synthesis of compounds 2a-d, 2f-h, 2j-k, 2m, 2p-r which have been obtained in 25% to 88% yield 27 . According to this method, six new derivatives (2e, 2i, 2l, 2n, 2o, 2s) have been synthesised in modest-to-good yields (Table 1). It is worth noting that the presence of two electron-withdrawing groups on anilines is allowed but provided compound 2l in low yield (17%). Tetra and pentacyclic derivatives have been obtained respectively using 1-naphthylamine and 1-aminoanthracene (compounds 2n and 2o). In the last case, it was necessary to dilute the reaction mixture (330 mL of AcOH/mmol instead of 100 mL) to avoid the precipitation of the poorly soluble dihydro intermediate. The angular structures of compounds 2n, and 2o have been unambiguously determined by NMR.
The preparation of compound 2s has required another reaction scheme. In the 1,3-dimethyl series, the literature described the synthesis of 5-dimethylaminopyrimidoquinolines using 6-arylamino-1,3-dimethyluracils as starting materials. When the synthesis was performed by nucleophilic substitution on an intermediate 5-thiomethyl heterocycle, the overall yields were very low due to the preparation of the thiomethylheterocycle 32 . The use of phosgene iminium chloride (Viehe's salt) furnished pyrimidoquinolines substituted either on the 5-position by a 5-dimethylamino group (in the presence of triethylamine) or by a chloride (without triethylamine) 33 . The dimethylpyrimidoquinoline 2s has been prepared using this second procedure, so 6-arylaminouracil 5 was obtained by reaction of 6-chlorouracil 3 with 3,4-methylenedioxyaniline 4, under microwave irradiation, according to the method described by Fang et al. 34 . Then, the reaction of the Viehe's salt on 5 gave an iminium intermediate that cyclized into 2s (Scheme  1). Surprisingly, the reaction leads to the dimethylamino compound even in the absence of triethylamine. The low solubility of 2s, which precipitates in the reaction mixture, probably prevents the nucleophilic attack by chloride ions observed in other series.  Although many publications have already reported the synthesis of N 2 -substituted 2H-pyrrolo [3,4-b]quinolines, only three have been covered regarding the N 2 -unsubstituted heterocycle. The reported methods are multistep syntheses. Starting from the corresponding 2-aminoacetophenones, the 9-methyl derivatives were obtained as a guanidinium salt in two steps and 38% overall yield 35 and the 9-phenyl analogue in 5 steps and 12% overall yield 36 . The ultimate step of the third chemical scheme involved the condensation of urea on the quinoline-2,3-dicarboxylic diacid to give the unsubstituted heterocycle 37 .
In a first time, we intended to apply this last method to the synthesis of the 6,7-methylenedioxy derivative 10 using a 6,7methylenedioxyquinoline-2,3-dicarboxylate. In the literature, an analogous dimethyl ester has been already described either by reaction of the 2-aminobenzaldehyde with dimethyl acetylenedicarboxylate and subsequent acidic treatment 38 or by reaction of the methylenedioxyaniline with the dimethyl acetylenedicarboxylate and subsequent treatment with the Vilsmeier reagent 39 . In both cases, the yield of the second step was not specified.
We thought that it would be possible to prepare the diethyl ester 8 in a one-pot reaction. Indeed, 8 was obtained in 40% yield by refluxing in EtOH a mixture of aniline 4, diethyl acetylenedicarboxylate (7) and formaldehyde (6) (Scheme 2). Unfortunately, the condensation of 8 with urea afforded 10 in only 12% yield. Assays with the corresponding diacid were unsuccessful.
Finally, we have synthesised 10 in 39% yield according to an adjustment of our three-component one-pot method replacing the cyclic 1,3-diketone by bromomaleimide (9) which is more easily accessible than the hydroxymaleimide (Scheme 2) 28 . With this example, we have shown that the scope of the one-pot reaction could be extended to 1,3-diketone analogues such as a cyclic b-halogenoketone.
It is well known that pyrimidinediones could be alkylated on the N 1 and/or on the N 3 . As an example, the reaction of thymine with the x-dibromobutane and the x-dibromohexane lead to a mixture of N 1 -substituted derivatives and N 1 ,N 10 -dimers. With the dibromopentyl chain in the same conditions, only the N 1 ,N 3 -disubstituted pyrimidinedione has been isolated 40 . In several papers, the N 1 -alkylation was achieved in two steps via the formation of the 2,4-O-disilyl pyrimidine in order to avoid the formation of side products [41][42][43][44][45][46] . We have prepared compounds 12a,b and 15 according to the described procedure but without isolation of the unstable disilyl intermediates (Scheme 3).
Final compounds 13a and 13b were obtained by alkylation of 10 by 12a and 12b in DMF in the presence of K 2 CO 3 . Unfortunately, in the bromo-uracil series, it was not possible to isolate compound 16b which reacted again with another molecule of 15 to give 17 (Scheme 3).
Access to the 5 0 -bromo derivatives required the use of a benzoyl protecting group on the N 3 position of 5-bromouracil (14) 47 . After alkylation of the N 3 -benzoylbromouracil (18) 48 , the reaction of the resulting compounds 19a,b with 10 and subsequent deprotection of the pyrimidinedione in acidic medium afforded compounds 16a and 16b (Scheme 3). A NOE NMR experiment realised on compound 19a confirmed the N 1 -alkylation of the pyrimidinedione from bromouracil (14).
For the synthesis of series C, it was necessary to protect the 6chloromethyluracil (21) by bis(trimethylsilyl)acetamide (BSA) before the condensation with 10. The reaction of the resulting 5unsubstituted compound 22 with N-halogenosuccinimides lead to the 5 0 -halogeno compounds 23a-c (Scheme 4).
Biological results and molecular modelling have prompted us to synthesise the N 2 -benzyl substituted compound 24 (Table 3) in which the pyrimidinedione moiety of derivative 22 is replaced by a phenyl nucleus. Thus, the reaction of benzyl bromide with 10 in the presence of K 2 CO 3 in DMF furnished 24 in 65% yield. Derivatives of series D have been synthesised from the unsubstituted tricyclic heterocycle 27 (Scheme 5). We have previously described easy access to this heterocyclic skeleton via a new synthon, the 4-formyl-3-hydroxy-2,5-dioxo-2,5-dihydro-1H-pyrrole (26) 28 . The functionalization of 27 was achieved by reaction with two equivalents of the appropriate amine in refluxed DMF, and cyclisation of the resulting diamide using two equivalents of p-toluenesulfonic acid (PTSA) 28 . Four new derivatives 28c, 28e, 28f, and 28h have been synthesised for the present study.
In an attempt to prepare 28i, the reaction of the 6-aminomethyluracil hydrochloride (29)

In vitro thymidine phosphorylase inhibiting activity and molecular docking study
Inhibitory activities of the new heterocyclic derivatives in series A-D have been tested in vitro against recombinant E. coli TP. The adopted protocol is a modification of the original method developed by Khan 51 . TP activity and inhibition assays were based on the phosphorolysis of thymidine to thymine, this conversion results in a significant spectrophotometric decrease. At 290 nm and pH 7.4, thymidine has a higher extinction coefficient than thymine (De ¼ À480 M À1 cm À1 ). The absorbance at 290 nm was measured every 2 min for 30 min in 96-well plates at 25 C.
A preliminary screening has been carried out at 50 mM or at the maximum concentration allowed by the solubility of the tested compound in the incubation medium (aqueous buffer containing 2.5% of DMSO). Results are given in percentage of inhibition (average of three measurements). For the most promising compounds (inhibition > 30% and solubility > 50 mM), IC 50 values have been determined (Tables 2-4). 7-DX, which presents the activity of the same order of magnitude (IC 50 E. coli TP ¼ 40 mM 8 ) was used as a reference.
To help understand structure-activity relationships, the geometry optimised structures of the synthesised compounds were docked into the active site of hTP 12 using GOLD 5.1. In order to validate the docking protocol, an X-ray crystal of TP complexed with TPI was retrieved, and the ligand was redocked into the active site of the enzyme. The predicted binding mode and interaction pattern of TPI generated in our study were found to be in accordance with the reported crystallographic study with an RMS deviation of 0.654 Å 12 .
Finally, a brief kinetic study was performed to elucidate the mechanism of inhibition of the two most interesting derivatives.

Series A
Series A has been designed as a conformationally constrained series in which the pyrimidinedione moiety should interact with the thymine binding site and the other nuclei with pocket 1 facing the C-5 and C-6 of the thymine ( Figure 2). In order to modulate the interactions and/or the solubility, various substituents have been introduced on the 5 to 9-positions.
Seven derivatives exhibited inhibiting activity towards E. coli TP (compounds 2c, 2d, 2e, 2i, 2l, 2m, 2p), three of them (compounds 2d, 2l, 2p) are interesting in terms of activity and solubility as compared to 7-DX (Table 2). This preliminary enzyme inhibition study suggested the following structure-activity relationships for series A compounds: i. A methyl on the benzenic ring was unfavourable except for the 9-methyl derivative 2c that was weakly active. ii. Monosubstitution on the 8-position by an ether group seemed to be interesting in the case of a methoxy group (compound 2d) but the activity decreased with the substituent size (compounds 2e and 2f). The replacement of the oxygen by methylene seemed to be possible (compare compounds 2e-2i). Compounds 2g and 2h were sparingly soluble; consequently, it is not possible to conclude about the effect of the presence of two ether groups. iii. The presence of an electron-withdrawing substituent on the 8-position such as a chlorine atom (compound 2j) or a trifluoromethyl group (compound 2k) was unfavourable. However, the 7,8-dichloro derivatives 2l exhibited an IC 50 in the same order of magnitude as the 8-methoxy derivative 2d. iv. The enzyme seemed to accept the presence of an additional angular benzene ring in compound 2m. It is not possible to conclude concerning the geometrical isomer 2n or the homologue 2o which were less soluble. v. The introduction of methyl on the 5-position of the 8methoxy derivative 2d was possible (compound 2p). In the case of compound 2g, the 5-substituted derivatives remained inactive whatever the substituent sizes (compounds 2q, 2r, 2s).
As expected, preliminary docking study showed that the pyrimidinedione moiety of the series A derivatives interacted with the active site of hTP in a similar binding mode as the 5-chlorouracil  fragment of TPI, that is, by establishing hydrogen bonds with arginine 202, serine 217 and lysine 221 of the a domain and histidine 116 of the a/b domain. As an example, docking of compound 2d superposed on TPI (in green) is presented in Figure 4. Results were the same for the other active derivatives 2p and 2l (Supporting information, Figure S1). The less marked activity of compounds 2c and 2e was also supported by the modelling results that highlighted two major binding modes corresponding in both 180 vertical and 180 horizontal rotations of the tricyclic moieties (Supporting information, Figure S2 for compound 2c as an example). It is worth noting that several substituents on the 8-position seemed to be located in the same area of the active site as showed in Figure 4: methoxy group of compound 2d, benzyl of compound 2i, and phenoxy for one of the conformations of compound 2e.

Series B and C
Series B and C have been designed as more flexible series than series A with the aim of interactions of the pyrimidinedione in the thymidine binding site of TP and of the methylenedioxypyrroloquinoline moiety in pocket 1 (see Figure 2).
In terms of enzymatic inhibition, series B and C did not present really interesting activities in comparison with 7-DX. The unsubstituted derivative 22 and the 5-bromopyrimidinediones 16a, 16b, or 23b seemed to the more interesting but, unfortunately, they are the less soluble. The spacer size has no influence. In the case of a methylene linker (series C), the presence of a halogen on the 5-position was not favourable. More interestingly, the pyrimidinedione did not seem to be essential, it can be replaced by a phenyl nucleus (compound 24) (Table 3).
Surprisingly, docking study showed that in both series, the dioxolane of the pyrroloquinoline moiety is positioned in the same place as the pyrimidinedione of TPI suggesting interactions of the methylenedioxy group with the active site; the pyrimidinedione moiety interacting with a large pocket which has never been explored (superposition of compound 16b and TPI on Figure 5).
However, as presented in Figure 5 for compound 16b (dark green cloud) and compound 23b (white cloud), two differences between series B and C were noticed. On the one hand, the tricyclic moieties were not superposed (180 vertical rotation). On the other hand, the aliphatic chains and pyrimidinediones of the Bseries molecules were positioned in the upper part of the pockets, whereas pyrimidinediones of series C derivatives interacted with the lower part of this pocket. Interestingly, the phenyl nucleus of compound 2i (series A) was also located in this white cloud (Supporting information, Figure S3).

Series D
The B and C series modelling results led us to design the D-series in which the methylenedioxyphenyl rings were replaced by a pyrimidinedione with the aim to optimise interactions with the thymidine binding site of the natural ligand. Furthermore, various substituents have been introduced on the imide nitrogen in an attempt to explore the lower part of the "new" pocket.
In a physicochemical point of view, series D compounds are more soluble than the previous series. However, they were uninteresting in terms of enzymatic inhibition activity except for the N-benzyl derivative 28a (Table 4).
As expected, the tricyclic moiety of compound 28a was well superposed with compound 2i, the benzyl analog in series A. The benzyl nuclei were positioned in the same area but not in the same position ( Figure 6). Surprisingly, the flexible open intermediate derivative 30 was as active as compound 28a. Interestingly, one 6-pyrimidinomethylamino arm was positioned in the same area as previously described compounds (Figure 6), the other one was located at the entry of the pocket.

Mechanism of enzyme inhibition
The synthesised compounds have been designed as TP inhibitors interacting with the thymidine fixation site. However, TP being a two substrates enzyme, it was necessary to verify that new inhibitors do not bind to the phosphate site. This study has been realised with one of the most active and/or the most soluble compounds in each series, that is, compounds 2p, 23b, 28a, and 30. E. coli TP activity was determined in the presence of a saturating concentration of thymidine (equivalent to three times the value of the Michaelis-Menten constant [km] 52 ) and variable amounts of phosphate (2-30 mM). Tested compounds have been used at a concentration corresponding to the IC 50 . For each tested derivative, phosphate concentration did not have any significant influence on the percentage of inhibition, indicating that the selected inhibitors did not bind to the phosphate site.
To explore whether 2d and 2p, the most active compounds which showed the same activities as 7-DX (TP CI 50 : 26-28 mM), acted as a competitive, uncompetitive, or mixed TP inhibitor, a brief kinetic study was performed according to previously described method 51 . The percentage of TP inhibition was determined for different concentrations of the inhibitors and thymidine. Lineweaver-Burk plots showed that all straight lines converged at the same point on the positive side of the y-axis (Figure 7). Consequently, 2d and 2p exhibited competitive inhibition kinetic on TP with thymidine as substrate. This was further supported by the fact that the Km values increased in the presence of 2d and 2p, while the Vmax values did not change significantly as the inhibitor concentration increased.

Cell proliferation assay (MTT)
The two most active compounds against TP (compounds 2d and 2p) were evaluated for growth-inhibiting properties in two cells lines namely human umbilical vein endothelial cells (HUVEC) and epidermal carcinoma cells (A431) which are well known to overexpress epidermal growth factor receptor (EGFR), a pathway involved in the cell proliferation and angiogenesis.  The inhibitory effect on cell proliferation was assessed using MTT assay 53 after 72 h of treatment through dose-response assays performed in the 100 mM to 0.1 mM concentration range. The results were expressed as percentages of growth inhibition.
Compound 2d exhibited better activity against A431 cell line (57% at 50 mM) than against HUVEC cell line (56% at 100 mM) but antiproliferative effect remains low. Compound 2p can be considerate as less cytotoxic against the two cell lines (50% and 31% at 100 mM against respectively A431 and HUVEC cells).

Conclusion
In conclusion, a small library of 38 derivatives has been synthesised and evaluated for its TP inhibiting activity. Around the pharmacophoric pyrimidinedione core of the natural substrate thymidine, four series have been designed in order to interact with wide empty pockets of the active site.
The natural ligand has been annealed to a quinoline in series A (pyrimidoquinoline-2,4-diones), substituted via a methylenic chain by a quinolopyrrolidinedione in the more flexible series B and C, and the polycyclic heterocycle has been replaced by a pyrimidopyridopyrrolidinetetraone in series D. The tricyclic moieties of these new polycyclic nitrogen heterocycles have been synthesised by one-pot multicomponent reactions that involved an aniline, an aldehyde, and a 1,3-dicarbonyl derivative or an analogue.
The biological evaluation identified several structurally distinct TP inhibitors in series A and D. Among them, compounds 2d, 2l, 2p, 28a, and surprisingly the open intermediate 30 showed a modest to good TP inhibition (IC 50 values ranging from 26 to 87 mM) when compared to 7-DX used as a positive control. The two most active compounds 2d and 2p were shown to interact with the thymidine fixation site and to exhibit a competitive mode of inhibition towards TP.
Molecular docking analysis confirmed the interaction of these newly synthesised compounds at the active binding site of TP. Moreover, docking studies highlighted a plausible specific interaction in a wide pocket that had not been yet explored.
For the first time, our study showed that it is possible to inhibit TP with tricyclic heterocycles. It is worth noting that the active compounds have in common a pyrido[2,3-d]pyrimidinedione nucleus and possess a chain interacting with the same part of this pocket. Interestingly, the open intermediate 30 becomes a starting point for a novel series that will be further exploited in order to improve the activity against TP.

Chemistry
General procedures: Commercial reagents were used as received without further purification. Microwave irradiation reaction was performed with an InitiatorTM 2.0 device, Biotage. Reactions were followed with thin-layer chromatography (TLC) (using 0.20 mm silica or alumina gel 60 F 254 aluminium plates, Merck) and visualisation was achieved with UV light (254 and 365 nm). Purifications were achieved through recrystallization or flash chromatography (using 40-63 mM silica, Merck). 1 H and 13 C NMR spectra were recorded on a Bruker AC 300 or 400 spectrometer by using DMSO-d 6 as the solvent and internal standard. Chemical shifts are reported in ppm and coupling constants (J) are given in Hertz. Spin multiplicities are reported as follows: Melting points were measured on a Stuart SMP3 melting point apparatus and are uncorrected. IR spectra were obtained on Perkin-Elmer 1600 spectrophotometer. Elemental analyses were performed at the CNRS Analysis Laboratory, Gif-sur-Yvette, France.

Biologicals
A spectrophotometric assay method was adopted to evaluate in vitro E. coli TP inhibiting activity of all the synthesised compounds. The conversion of thymidine to thymine was recorded at 290 nm. Thymidine was dissolved in buffer (3 mM). DMSO stock solutions of the tested compounds were diluted in buffer, the incubation medium contained 2.5% of DMSO. All experiments were conducted in triplicate.

In vitro thymidine phosphorylase essay
Initially, all synthesised compounds were tested against E. coli TP at 50 mM or at the maximum concentration allowed by their solubility.
The method reported by Khan et al. 51 has been adapted to a screening in 96-well plates (200 mL well, UV-Star V R 96-well microplate, Greiner). Each well contained 160 mL of 50 mM KH 2 PO 4 buffer (pH 7.4), 0.006 U of E. coli TP (T2807-1KU, Sigma Aldrich), and 20 mL of the tested compound solution. The reaction was initiated by the addition of 20 mL thymidine (i.e. a 300 mM final concentration corresponding to the Km 52 ). The decrease in absorbance due to conversion of thymidine to thymine was followed during 30 min (PowerWave HT Microplate Reader, Biotek) at 25 C, with one measure every 2 min and a 30 s shaking before each measure. Data were processed using Biotek KC4 software and the results were expressed in percentage of inhibition at 50 mM or at the maximum concentration allowed by the solubility of the tested compound (Tables 2-4). For the most active and soluble compounds 2d, 2l, 2p, 28a, and 30 (inhibition > 30% and solubility > 50 mM), IC 50 values have been determined.

Phosphate competition study
For this study, the reaction mixture was composed of 160 mL of 10 mM Tris buffer (pH 7.4), 150 mM NaCl, 2 mM EDTA, 0.006 U of E.
coli TP, different concentrations of KH 2 PO 4 (2, 5, 10 and 30 mM). Compounds were tested at a concentration close to their IC 50 . The reaction was initiated by the addition of 20 mL of thymidine (3 mM and 10 mM solutions).

Molecular docking study
Molecular modelling studies were carried out with the GOLD software (version 5.1). The ligands were constructed using the standard fragments of the Sybyl library 6.9.1. Their geometry was optimised using the Tripos force field by assigning partial loads calculated by the Gasteiger-H€ uckel method to a gradient of 0.0001 Kcal/mol/Å. The conformations obtained for each compound were classified by a consensus scoring function and then the conformation groups were observed visually to evaluate their consistency and their complementarity with the active site. Only the most representative conformations of each group were selected.
In order to check the validity of the results, the TPI reference inhibitor was relocated to the hTP binding site (PDB code: 1UOU) following the same protocol as for the inhibitors. The best solution has an RMS on heavy atoms of 0.654 Å with respect to the 1UOU crystallographic structure. Results were visualised by Chimaera and Maestro.
Supplementary data includes docking superposition of compounds 2d, 2p, and 2l in the hTP active site; two docked conformations of compound 2c in the hTP active site; docking superposition of compounds 2i, and 23b in the hTP active site; NMR spectra of the final compounds.