Quinolin-2-one-pyrimidine Hybrids Acting as Potent Reversal Agents on the P-gp-mediated Multidrug Resistance: Insights into Structure-activity Relationships and the Binding Mode

P-gp-associated multidrug resistance (MDR) is a major impediment to the success of chemotherapy. With the aim of nding non-toxic and effective P-gp inhibitors, we investigated a panel of quinolin-2-one-pyrimidine hybrids. Among the active compounds, two of them signicantly increased intracellular doxorubicin and rhodamine 123 accumulation by inhibiting the eux mediated by P-gp and restored doxorubicin toxicity at nanomolar range. Structure-activity relationships showed that the number of methoxy groups, an optimal length of the molecule in its extended conformation, and at least one exible methylene group bridging the quinolinone moiety favored the inhibitory potency of P-gp. The best compounds showed a similar binding pattern and interactions to those of doxorubicin and tariquidar, as revealed by MD and hybrid QM/MM simulations performed with the recent experimental structure of P-gp co-crystallized with paclitaxel. Analysis of the molecular interactions stabilizing the different molecular complexes determined by MD and QTAIM showed that binding to key residues from TMH 4–7 and 12 is required for inhibition. of Dox resistance in Lucena 1 cells were evaluated as previously described 5,8,29 . Briey, 5 x 10 4 Lucena 1 and K562 cells were seeded in 96-well plates containing RPMI-1640 medium with the anticancer drug Dox (3.4–431 or 0.3–34.5 µM, respectively) in the absence or presence of 0.02 to 1.25 µM of compounds 5c and 7b. Negative controls contained 0.5% v/v DMSO while verapamil was used as standard inhibitor. The cells were incubated at 37°C at 5% CO 2 for 48 h and the same protocol described above was followed. IC 50 values of Dox were then obtained. The FR values were calculated as the ratio of the IC 50 obtained with Dox / IC 50 value for Dox in combination with the different concentrations of each tested compound 8,29 . most effective 5c and 7b on PBMC by MTT assay 39 . PBMC were isolated from heparinized blood of healthy volunteers by density centrifugation using Ficoll-Hypaque separation. the Catholic of Ethics Board and informed consents were obtained from all donors. Briey, 1 × 10 5 PBMC/well were incubated with the tested compounds at concentrations ranging from 2.5 to 20 µM in the presence of 10 𝜇 g/mL PHA, or with 1% DMSO as control. After 48 h incubation, MTT was added and the same protocol described above was followed to calculate the IC 50 values. An erythrocyte hemolysis assay was carried out following previous procedures 8 with the compounds assayed at 2.5 to 20 µM. IV) 50 ns of simulation in the NTP ensemble, at 1 atm and 300 K. Procedures III-IV were repeated in two or three independent trajectories using the Andersen thermostat and barostat 44 . In the reasonably equilibrated system, the density uctuated slightly around 1.019 g/mL. Due to the transmembrane nature of P-gp, a 50.0 kcal/Å 2 harmonic restraint was kept for the backbone atoms. Electrostatic interactions were computed using the Particle Mesh Ewald (PME) method with a cutoff of 10 Å 45 . Bonds involving hydrogen atoms were constrained using the SHAKE algorithm, allowing for an integration time step of 0.002 ps. The integration was made using the pmemd.CUDA module of the AMBER18 program 43 with the auxiliary force eld GAFF for the ligands and ff14SB for the protein 46 . The trajectories were analyzed using standard AMBER cpptraj analysis tools. The free energy calculations were made using the mmpbsa module of AMBER 18 by applying Poisson-Boltzman (PB) and Generalized Born (GB) models 47 . The energy analyses were made for the last 8–12 ns of simulation as the average over at least two independent trajectories. The frames were sampled once each 5–10 saved frames (saving 1 frame every 10 ps, thus ensuring that the energy self-correlation is small enough). The clustering analyses for visualization of representative conformations and rendering of some gures were made using Chimera 1.14 48 . Most graphic rendering was prepared using VMD 1.9.3 49 .


Introduction
Cancer is a complex disease which is among the major causes of death worldwide 1 . Chemotherapy plays a key role among the strategies to treat cancer 2 in particular for the treatment of hematological malignancies 3 . Despite its therapeutic advantages, the success of this treatment is impeded by the development of resistance to conventional chemotherapeutic drugs 4 . which is a major negative factor for the survival of patients 2 . One primary mechanism involved in this phenomenon is the over-expression of P-glycoprotein (P-gp), an ATP binding cassette (ABC) transmembrane protein that pumps a great variety of anticancer drugs out from the cell 5 . This e ux leads to a diminished intracellular concentration and a consequent insensitivity, which is known as multidrug resistance (MDR) 6 . The increased expression of P-gp is decisive in the resistance of different types of cancer 7 , in particular in chronic myelogenous leukemia with more than a half of the patients unresponsive to Vinca alkaloids and anthracyclines due to its presence 8 . Structurally, P-gp consists of two homologous transmembrane domains (TMD), each formed by six transmembrane α-helices (TMHs) and two cytosolic nucleotide-binding domains (NBDs) 9 .
Considerable attempts have been made to nd P-gp-interacting compounds, but clinical trials with several modulators have failed for diverse reasons 5,10 . It is thus a matter of great concern to develop novel chemical entities able to tackle P-gp-mediated outward transport and thus overcome MDR in cancer cells.
The importance of the 6,7-dimethoxy-1,2,3,4-tetrahydroisoquinoline moiety in the highly effective third-generation P-gp inhibitors, tariquidar, which also bears an isoquinoline group, elacridar, HM30181, WK-X-34 11,12 and their derivatives, some containing aromatic amide 2,12 , as well as the presence of an aminopyrimidine moiety in the P-gp inhibitor, imatinib 13 , and the promising MDR reversal activity of the multi-substituted or fused pyrimidine scaffold 14 , encouraged us to search for potential P-gp inhibitors in a library of new quinolin-2-one-pyrimidine hybrids 15 .
The present work investigated their ability to inhibit P-gp functionality and to restore sensitivity to the chemotherapeutic drug, doxorubicin (Dox), in multi-resistant Lucena 1 leukemia cells. The various interactions that accommodate ligands to certain sites of P-gp can be explained in depth by understanding the stereo-electronic intricacies involved in their binding modes. With this aim, the possible molecular complexes were evaluated using combined molecular modeling techniques.
Inhibitory effects on P-gp-mediated outward transport by the target compounds, cytotoxicity and SAR studies.
Flow cytometry ndings showed that compounds 2, 2a-c, 3a-c, 4, 5, 8, 8a, 9 and 9a ( Fig. 1), tested at 20 µM, were ineffective (p > 0.05) for increasing the intracellular accumulation of the uorescent substrate Dox. However, compounds 3, 4a-c, 5a-c, 6, 6a-c, 7, 7a-c, 10, 10a and 11 ( Fig. 1) at 20 µM signi cantly enhanced Dox retention with uorescence intensity ratio (FIR) values ranging from 1.21 to 2.87, by e ciently blocking the e ux function of P-gp (Table 1). We therefore further evaluated these compounds at serial dilutions. As the results show in Table 1, compounds 7a, 10, 10a and 11 were able to induce a higher accumulation of Dox than the negative control, as from 1.25 µM (p < 0.05) followed by compounds 4c, 5b, 6c and 7c which showed minimum effective concentrations (MECs) of 0.62 µM (p < 0.05). Compounds 5c and 7b, were the most potent, as they still exhibited effectiveness at 0.31 µM (p < 0.05 and 0.001, respectively) ( Table 1 and Fig. 2A and B). The e cacy of compounds 5c and 7b was similar to that observed with the classical P-gp inhibitor, verapamil, at all the concentrations assayed (p > 0.05), except at 0.31 µM (p < 0.05 and 0.01, respectively). The intracellular uorescence-associated values in Lucena 1 cells treated with both compounds at 20 µM were similar to those observed in treated and untreated sensitive K562 cells (p > 0.05), which in turn showed no differences in their intracellular Dox uorescence (p > 0.05) (Fig. 2C). These results suggested a complete reversal effect on P-gp-mediated MDR and a selective inhibition on P-gp function. Table 1 Inhibitory effect of target compounds on P-gp transport activity in Lucena 1 and K562 leukemia cells  19,20 .
Rhodamine 123 (Rho123) is a uorescent probe e ciently e uxed by P-gp, which binds to the transporter at sites distinct from those of Dox 5 .
Therefore, the ability of compounds 5c and 7b to enhance Rho123-associated intracellular uorescence was further explored by ow cytometry. As shown in Fig. 3A  It has been previously described that the number of hydrogen bond acceptor methoxy groups on terminal phenyl rings is favorable for P-gp inhibitory activity [21][22][23] . Re ecting this, compound 5c, bearing two OCH 3 groups, displayed pronounced activity, followed by 5b with one methoxy and then compound 5a lacking this substituent (MECs = 0.31, 0.62 and 2.50 µM, respectively). The same trend was observed with compounds 7c, 7b and 7a (MECs = 0.62, 0.31 and 1.25 µM, respectively) ( Table 1). It is important to highlight that the instability of the dimethoxy compound 7c may have slightly masked its anti-P-gp effect, therefore showing a higher minimum effective concentration (MEC) value than that expected in comparison to compound 7b, feathered with only one OCH 3 . Increasing the number of methoxy substituents also resulted in improved activity in the chlorinated compounds 4c and 4b compared to compound 4a (MECs = 0.62, 2.5 and 20 µM, respectively) ( Table 1). The same was observed with the chloride compounds 6c and 6b with respect to compound 6a (MECs = 0.62, 2.5 and 5 µM, respectively). The presence of the p-chlorophenyl group was detrimental to the inhibitory effect of compounds 4a-b (MECs = 20 and 2.5 µM, respectively) and 6a-b (MECs = 5 and 2.5 µM, respectively) in comparison to their respective analogues 5a-b and 7a-b which instead bear a naphtha-2-yl group. Interestingly, the presence of two OCH 3 groups counteracts the adverse effect exerted by the chlorobenzene substituent, as was observed when compounds 4b and 6b (MECs = 2.5 µM) were compared with their respective analogues 4c and 6c, showing MEC values of 0.62 µM, the latter with similar activity to that of the closely related compounds 5c and 7c, respectively (Table 1). It has been previously reported that compounds featuring chlorine at ortho-, metaand para-positions, also bearing isoquinolines with two methoxy groups, exhibited similar P-gp modulatory activity to that observed with verapamil 11 . The results obtained clearly indicated that the methoxy substituents signi cantly in uenced P-gp inhibition. The replacement of the p-phenylenediamine in compounds 4a-c and 5a-c for a carbonyl group as in the respective compounds 2a-c and 3a-c prevented the anti-P-gp effect, regardless of the presence of OCH 3 or chloride substituents ( Table 1). The clear loss of activity of these compounds was further revealed by docking and molecular dynamics (MD) simulations.
Reversal activity on doxorubicin toxicity of the most promising compounds 5c and 7b.
To validate the assumption that increased accumulation of Dox is associated with a concomitant enhancement in its cytotoxicity, compounds 5c and 7b were co-administered with this chemotherapeutic drug and the effect on cell viability was determined. Consistent with the accumulation of Dox, compounds 5c and 7b at 1.25 µM substantially decreased the half-maximal inhibitory concentration (IC 50 (Fig. 4).

Molecular modelling.
With the aim of gaining insight into the way that ligands interact with P-gp, molecular modeling was performed. The protocol for this analysis was divided into three interrelated parts with increasing levels of detail. First, in the docking protocol, the interior of the whole transmembrane region was scanned, making no assumptions about the localization of a particular binding site. The main binding sites were determined for the experimental reference chemotherapeutic drug, Dox, the known reference inhibitors tariquidar and verapamil and the subject compounds shown in Fig. 1. Next, detailed analyses were performed of the binding dynamics of the selected species with poor, mild or high activity, and an estimate of each free energy of binding was calculated by means of classical MD simulations. The trajectories obtained were also analyzed to identify the main contributions to these by each residue. Finally, each interaction was discussed in more detail in the light of the QTAIM analyses of hybrid QM/MM calculations.
The primary binding region for the experimental reference chemotherapeutic drug, Dox, was found to be overlapped with the region of the bulkier cytotoxic agent paclitaxel (Taxol), which was co-crystallized in the novel P-gp structure (Fig. 5) 24 . The docking results started to shed light on the differences in the activities of the panel of assayed compounds, even for those having similar structures but sharp differences in Dox accumulation pro les. Most of the mild and all the powerful inhibitors fell inside the site that was also shared by Dox and Taxol (Fig. 5), involving strong interactions with the aromatic residues, mainly from TMH 4, 5 and 6 from one homologous half and with those from TMH 7 and 12 from the other half of the transporter. Figure 6A shows a superimposition of the most stable docked poses of active compounds 4b, 5bc, 6a-b, 7a-c, 10 and 11 along with the top reference inhibitor, tariquidar and Dox. A gross common feature for all these poses in this binding pocket is that they all fold into a distorted "U"-shape. The aromatic edges of each compound accommodated as the branches of the "U" with an aliphatic/conjugated exible bridge as its camber.
This exible intermediate portion, which also itself interacts with TMH 12, allowed for the turn or camber. In contrast, the compounds depicted in Fig. 6B are those that were too short or too long and not exible enough to accommodate as the more powerful compounds did at the site of Fig. 6A. None of them were active. These molecules can either t into just one of the branches of the "U", as observed with compounds 5, 8a or 9a, or fall into a different pocket where there is no interaction with residues from TMH 6 and 12, as observed with compounds 2, 2b, 2c, 3a-c, 8 or 9. The relevance of such structural determinants may be clearly illustrated by comparing the inactive compound 3c with one of the strongest inhibitors, compound 5c, both with very similar chemical structures, including the ortho-dimethoxy substitution in the quinolinone ring. The only difference between both entities is that the exible linker to the quinolinone, the group -NHCH 2 -, is replaced by a longer and less exible carbonyl-vinyl (-C(O)-CH CH-) ( Fig. 1). In addition to the two main groups of compounds shown in Fig. 6A and B, compound 3 with moderate activity (MEC = 10 µM, Table 1) was found to t partially into the pattern of Fig. 6A, also touching TMH 11, a contact that is not shared by any of the other ligands ( Supplementary Fig. S3, online).
While the docking did reveal some clear trends in the type of interaction and sites of relevance, MD simulations provided a more detailed understanding of the nature, strength, and persistence of interactions with the key residues. The inactive compounds 5 and 8, with the patterns shown in Fig. 6B, were selected for the studies, as well as the active compounds 3, 5a, 5c and 7b, the respective activity of which increased from low to mild to the highest ( Table 1). The free energies of binding (ΔG 0 b ) of these representative compounds, estimated by MD, were compared to those of the substrates and of the two reference inhibitors, verapamil, used as a positive control, and tariquidar, one of the most potent in vitro inhibitors known 2 , used as a positive control for the in silico studies. Although one substrate and tariquidar and verapamil were previously simulated 8, 25 , for comparison purposes, these simulations were repeated from scratch with the novel PDB structure and using exactly the same simulation conditions as for the subject ligands. As shown in Table 2, compounds 5 and 8 were found to be unable to compete with either Rho123 or Dox, added to the fact that compound 8 went to another site of lower a nity. The ΔG 0 b values of compounds 3, 5a, 5c and 7b (Table 2) reproduced well their inhibitory properties ( Table 1). The most promising structures, 5c and 7b, showed similar energies to that of verapamil although not as favorable as tariquidar ( Table 2). The latter comparison was not surprising, since tariquidar is known to have a low nanomolar activity 26 , despite that its high toxicity 27 , in contrast to our subject compounds, and other issues 28 prevent it being clinically suitable. Table 2 Free energies of binding from the MMPBSA analyses of the MD simulations. An analysis of the dynamic interaction of the potent compound 7b throughout the simulation revealed that it persistently shared the region and the contacts of the known therapeutic P-gp substrates, Dox and Taxol. In contrast, the inactive compound 8, besides its reduced a nity in terms of ΔG 0 b , stayed in a different region and had greater mobility as shown in Fig. 7, where four representative snapshots of each MD trajectory (i.e. the conformations most visited at the simulation temperature) are superimposed for Dox, Taxol and compounds 7b and 8. A similar behavior was observed for compounds 5a and 5c. Figure 8A-C shows the ΔG 0 b components decomposition in terms of per residue energy for compound 5c, compared to the substrate Dox and tariquidar. Most of the contacts of 5c were shared with the top reference inhibitor.
More important is the fact that most of these contacts, such as Ala229 and Trp232 from TMH 4, Phe303 and Ile306 from TMH 5, Ile340, Phe343 and Gln347 from TMH 6, and Phe983 and Met986 from TMH 12, among others, were proposed to play a key role in the binding to P-gp on both computational and experimental bases 5,8,25,29 . Compound 5c ( Fig. 8C), as well as compound 7b ( Supplementary Fig. S4, online) showed important peaks (i.e. persistent contacts) at residues from both homologous halves (TMHs 1-6 and 7-12), in particular from TMH 6 and 12, both connecting the transmembrane domain (TMD) to the nucleotide binding domains. As depicted in Fig. 8D, and as previously observed in the docking poses, the inactive compound 5 shared most of the contacts present in TMH 5 and 6 with compounds 5c (Fig. 8C) and 7b ( Supplementary Fig. S4, online) but did not interact with those from TMH 12 or with any other from the second homologous half. This energy analysis also revealed a certain importance of contact with TMH 7 (mainly with Gln725), which is observed with compounds 5a, 5c and 7b as well as with tariquidar, but not with Dox or the less active compound 3 or the inactive compound 5 (Figs. 8 and Supplementary S2, online). Indeed, compounds 3 and 5 contact with neither TMH 7 nor TMH 12; the latter only contacted with the second half with a mild interaction with Gln946 from TMH 11 (the decomposition pro le for compounds 3, 5a, 8 and verapamil are shown in Supplementary Fig. S4, online).
The MD simulations also revealed that the number of hydrogen bond acceptors in the molecule would not be as important as the other features discussed so far. Indeed, even though tariquidar is the species with the most H-bond acceptors, it just had a total average of about one H-bond during the simulation time; the same was observed for compounds 5c and 7b. On the other hand, the compound which kept the highest number of persistent H-bonds during the simulation was 3, one of the poorest of the set (see Supplementary Fig. S5, online for a comparison of the H-bond analyses over the last 10 ns of simulations for 3, 5c, 7b and tariquidar). The main interactions with the residues bearing the major peaks in Fig. 8 are hydrophobic in nature and are analyzed in more detail with QM/MM calculations.
Charge density analysis of complexes.
QTAIM calculations are very useful tools for evaluating accurately and in detail the molecular interactions which stabilize ligand-receptor complexes 30 and can be used successfully in different biological systems [31][32][33][34][35] . In order to analyze the molecular interactions stabilizing the complexes of the target molecules and P-gp more quantitatively and with more detail, QM/MM calculations were performed, choosing the most representative complexes of the series. Therefore, the complexes of Dox, tariquidar and compounds 3, 5a, 5c, 7b and 8 with P-gp were further studied by Charge Density Analysis in the context of the quantum theory of atoms in molecules (QTAIM) framework. Reduced models of complexes were constructed, containing receptor residues within the range of noncovalent interactions from ligand atoms, as reported in the experimental section. The QTAIM topological analysis was performed by mapping the gradient vector eld onto the pre-computed charge density (i.e., Δ (r)) of reduced models, thus giving rise to the topological elements of the charge density, i.e., the bond critical points (BCPs) and the bond paths (BPs) that connect the interacting atoms. Figure 9 shows the charge density values obtained for the complexes in function of the different TMH domains of P-gp, and the numerical data is shown in Supplementary Table S1, online. It is interesting to note that the complexes with Dox and the inhibitors tariquidar, 5a, 5c and 7b presented a very similar binding pattern, with all the signi cant interactions located in residues from TMH 4-6 and 12, the most important of which were with TMH 6 and 12. It is important to highlight that tariquidar as well as compounds 5c and 7b, and to a lesser extent compound 5a, showed signi cant interactions with TMH 7. Weaker interactions were also seen with residues located at TMH 1 and TMH 10, while no interactions with TMH 2 were observed in any complex. One exception to this motif was the case of compound 5c, which showed somewhat signi cant interactions with amino acids from TMH 10. Regarding the mildly effective compound 3, the pattern of interactions was different from that observed for the active compounds since it showed the strongest interactions with TMH 6 but lacked interactions with TMH 12. In addition, it was the only compound displaying interactions with TMH 11. The inactive compound 8 did not establish interactions with the residues of either TMH 6 or TMH 7, although it interacts with TMH 12, but with different residues to those which compounds 5c and 7b contacted. Likewise, 3 was the only compound that established interactions with TMH 8 and TMH 9. Therefore, the low or null activity of compounds 3 and 8 may be explained by their binding in widely different sites.
Besides the bene ts of QTAIM as a tool for describing molecular interactions through visual inspection of the charge density in molecular graphs, it has been shown that there is also a direct relationship between the charge density value in the BCP interaction and the complex interaction energy 36 . Therefore, one might expect this relationship to still hold for biomolecular complexes, i.e., the sum of charge density values from all the intermolecular BCPs should be related to the complex stability. Such a relationship concerning a particular system is important since it enables the contributions of individual functional groups to the overall anchoring strength of the ligand within the binding pocket to be measured. In order to clarify the interactions from the point of view of the target ligands, the molecules were arbitrarily divided into three portions (moiety 1, linker and moiety 2), as shown in Fig. 10. Figure 11 shows the total of interactions established according to the three portions of each molecule. Compounds 3 and 8 were not included because these interacted in a different site and, therefore, comparing their interactions with active compounds would be futile. All the complexes showed a similar pattern of interactions and indicated that the three portions of the ligands establish signi cant interactions, stabilizing the complexes. In other words, it appears that the presence of the three portions of the ligands is necessary for the activity. As expected, the interactions observed for the complex with tariquidar were the strongest while interactions obtained for the complex of 5a were the weakest, while the interactions obtained for Dox, 5c and 7b complexes gave intermediate values. These results fully match both the experimental data and the classical MD estimations. Figure 12 shows the molecular interactions between Dox and P-gp. As can be seen, Dox mainly interacts with residues from TMH 6 (Phe343, Ser344 and Gln347) and TMH 12 (Met986 and Gln990). Atoms belonging to the naphthoquinone ring (moiety 1) are stacked between the aromatic residues Trp232, Phe303 and Phe343 (TMH 4, TMH 5 and TMH 6 respectively), establishing several π-stacking type interactions.
These interactions certainly contributed to the overall anchoring of this ligand in the active site as it favorably oriented the rest of the molecule to interact with TMH 6 and TMH 12. Particularly, OH groups and the hydroxyacetyl substituent from the linker were hydrogen-bonded to neighboring polar residues such as Ser196 (TMH 3), Ser344, Gln347 (TMH 6) and Gln990 (TMH 12). Moreover, moiety 2 signi cantly interacted with Met986 and Gln990 and also established hydrophobic interactions with Leu65 (TMH 1) and Ile340 (TMH 6). On the other hand, the larger size and greater exibility of tariquidar compared to Dox prevented the latter from establishing the same interactions. In the case of tariquidar, interactions with TMH 1, TMH 5 and TMH 7 were signi cantly increased while those with TMH 3, TMH 10 and TMH 12 were reduced or even not present at all (Fig. 9).
Although the quinoline ring was still anchored over the aromatic residue Trp232, unlike the moiety 1 of Dox for which the main stabilizing interactions were with TMH 4, TMH 5 and TMH 6, the modeling of tariquidar suggested that this portion interacted especially with TMH 5 (Fig. 13). As observed in this Figure, the quinoline ring formed several hydrophobic interactions with Leu236 from TMH 4, and with Ile299 Phe303 and Ile306 from TMH 5. In addition, several interactions of the linker were formed with hydrophobic and aromatic residues from various TMH. The presence of two aromatic rings and two methoxy substituents in the linker also increased the possibility of hydrophobic interactions. The strongest interactions of this moiety were with Leu65 (TMH 1), Tyr307 (TMH 5), Ile340 and Phe343 (TMH 6), Phe728 (TMH 7) and Phe983 (TMH 12), while a single H-bond was also observed between Gln347 (TMH 6) and the amide carbonyl oxygen bound to moiety 1.
This result supports the idea that the biological effects of the compounds studied here may not be mainly related to the possibility of forming H-bonds but depend on other structural features. Regarding moiety 2, while Dox was almost completely bounded through strong interactions to residues from TMH 12, tariquidar displayed a different pattern. This formed several interactions with TMH 5 (Phe303 and Tyr307), TMH 7 (Leu724) as well as with TMH 12 (Ala987, Gln990 and Val991). Thus, charge density value was signi cantly higher in the tariquidar complex.
In addition, although interactions with TMH 12 were weaker compared to Dox, the charge density regarding TMH 12 remained high as tariquidar establishes a greater number of interactions. Active compound 5c showed some similarities with both Dox and tariquidar, as well as some differences. It must be pointed out that moiety 1 is not located in the same region discussed for the other compounds. In fact, moiety 1 of compound 5c did not interact with Trp232, which could explain in part the low ρ(r) value observed for the naphthyl ring (see blue zones of Fig. 10). Nevertheless, strong interactions were observed between moiety 1 and TMH 5 and TMH 6, and in particular with Phe303, which resemble those observed for Dox. In addition, even though the conformation adopted by 5c prevented it from effectively interacting with Trp232, strong π-stacking type interactions with Phe343 (TMH 6) were formed. The anchoring loss from moiety 1 is compensated by the signi cant interactions established by the linker. As can be seen in Fig. 14, residues Gln347 (TMH 6) and Glu875 (TMH 10) were H-bonded to both NH groups of the compound 5c linker moiety. Also, hydrophobic interactions with Ala229 (TMH 4) and C-H···π interactions between the pyrimidine ring, located at the linker zone, and the aromatic Trp232 were formed. In addition, the linker established CH 2 ···SCH 3 and Ar-H···NH 2 interactions with Met986 and Gln990, respectively, both from TMH 12. Finally, moiety 2 interacted with TMH 12 (i.e., Ala987 and Gln990) in a similar way to that discussed for tariquidar. However, Fig. 14 also shows that a signi cant interaction occurred with TMH 7, especially with Gln725, which is H-bonded to methoxy groups from compound 5c.
Toxicity assessment of most active compounds. As the most potent molecules, the cytotoxicity was determined of compounds 5c and 7b against normal peripheral blood mononuclear cells (PBMC), in an effort to evaluate their safety pro le. The results showed that compounds 5c and 7b were devoid of cytotoxicity, with 21% and 12% of inhibition on proliferation at 20 µM, respectively. In addition, compounds 5c and 7b did not alter the erythrocyte membrane at concentrations lower than 10 and 20 µM, respectively. The results show that these compounds may be proposed as safe candidates for developing novel agents to circumvent MDR by P-gp inhibition.

Conclusions
In this study, 2-substituted amino-4-arylpyrimidines and their quinoline hybrids were tested for their inhibitory properties on P-gp function.
Among the 31 compounds tested, 18 proved to be effective, with compounds 5c and 7b being the most potent to enhance Dox and Rho123 intracellular accumulation by blocking their extruding from cells. Subsequently, these compounds possess outstanding reversal activity of the P-gp-mediated Dox resistance with nanomolar potency. No in uence on proliferation of PBMC nor alterations in erythrocyte membranes were observed with the most effective compounds 5c and 7b, thus suggesting these molecules as promising starting points for chemosensitizer agents.
Results obtained from the preliminary SAR study mainly suggest that both the presence of methoxy groups and their number were favorable for P-gp activity. Important restrictions on the length and exibility of the labeled linker region were found to play a critical role. At least one aliphatic sp 3 carbon of a methylene group seems to be required to enable folding between the moieties labeled 1 and 2. This feature was discussed as an example by comparing the almost identical compounds 3c and 5c, inactive and highly effective, respectively. The number of H-bond acceptor groups per se played no role unless they positively contributed to anchoring the inhibitor in certain poses located in speci c sites and interacting with important residues.
Another interesting contribution of this work is the insight into details of certain structural aspects, which are essential for understanding the formation of the complex ligand/P-gp. Combined MM/QM/QTAIM calculations enabled us to describe the molecular interactions that stabilize the different complexes of the ligands with the protein and to determine which portion of the compounds should be changed in order to increase their a nity with P-gp. The molecular modeling study indicated that the compounds reported here bound in the same region of the active site of P-gp as that of tariquidar and Dox; however, our simulations indicated that these molecules were arranged spatially in a slightly different way. Therefore, these inhibitors bound to some of the same amino acids as tariquidar but also to different ones.

Chemistry.
Most chemicals and solvents were used pure or dried when necessary. All starting materials were weighed and handled in air at room temperature.
Thin layer chromatography (TLC) monitoring was performed on 0.2 mm pre-coated aluminum plates of silica gel (Merck 60 F 254 ) and spots were visualized by UV irradiation (254 nm) to determine the progress of the reaction. If necessary, the compounds were separated by means of column chromatography on ash silica gel 60 (40-63 µm, Merck, Darmstadt, Germany). HPLC puri cation was further performed to ensure the > 95% purity of each compound, as determined by HPLC analysis (see details in Supplementary Information, online). High-resolution mass spectra (HRMS) analysis was performed for identity check. The characterization of synthesized compounds was performed measuring the melting points (Brastead Electrothermal 9100 melting point apparatus) and all were tted within ± 2 ºC of difference with respect to those reported 15 . 1 H spectra were recorded in a Bruker Avance 400 spectrometer at 400 MHz at 298 K, using CDCl 3 and DMSO-d 6 as solvents, except for compounds 2a-c and 3a-c, which were recorded in DMSO-d 6 at 393 K, and tetramethylsilane (0 ppm) as the internal reference, or the residual hydrogen of such solvents.
Compounds were prepared from the quinoline-3-carbaldehydes Ia-c and derivatives 2-3 with an excess of aqueous KOH by heating to re ux in MeOH for 3 days (see Supplementary Fig. S2, online).
Material for biological studies. Cell lines and cell culture.
The sensitive chronic myelogenous leukemia cell line K562 and its resistant derivative, Lucena 1, were cultured in supplemented RPMI-1640 medium (Invitrogen Life Technologies, Carlsbad, CA, USA) with 10% fetal bovine serum, 2 mM L-glutamine, 100 U/mL penicillin and 100 µg/mL streptomycin (Invitrogen Life Technologies) at 37°C in a 5% CO 2 humidi ed atmosphere. Lucena 1 selectively overexpressed P-gp 37 and its MDR was maintained by once weekly exposure to the anticancer drug, Dox at 60 nM till 4 days before the experiments 29 . Dox displayed a weak toxic effect against Lucena 1 cells, which showed 35.8-fold resistance to this drug compared to its parental K562 (IC 50 to Dox = 22.97 ± 5.41 and 0.64 ± 0.13 µM, respectively). The Dox and Rho123 uorescence intensity was 2.31 and 5.87-fold lower, respectively, in Lucena 1 with respect to K562. Cells used in the experiments were in the logarithmic growth phase, with a viability above 90% determined by trypan blue exclusion.
The intracellular accumulation of the uorescent probes, Dox and rhodamine 123 (Rho123), was determined by monitoring its uorescence intensity using ow cytometry (96-well plate format), as previously reported 8 . The inhibition on the outward transport of these P-gp substrates results in an increased medium uorescence intensity (MFI). Both cell lines, Lucena 1 and K562 at a density of 5x10 4 cells/well, were seeded in 96-well culture plates containing RPMI-1640 medium and exposed to the target compounds dissolved in DMSO at the non-toxic concentration of 20 µM. Those compounds showing inhibition on P-gp at the maximum tested concentration were further evaluated at serial dilutions to determinate their MECs. Negative controls were performed with DMSO 0.5% v/v, while verapamil, a known P-gp inhibitor, was used as positive control. The plates were then incubated in a 5% CO 2 incubator for 1 h at 37°C. Afterwards, 5 µM Dox was added to each well. The other uorescent speci c substrate of P-gp, Rho123, however, was added at 500 ng/mL to additionally measure the ability of the most promising compounds to block its e ux. Subsequently, cells from both assays were washed twice with cold PBS, and 15,000 events from each sample were analyzed to determine the cell-associated uorescence. Dead cells were eliminated by means of forward-and side-scatter parameters.
The MFI of retained intracellular Dox and Rho123 were estimated using a 488 nm laser and 585/42nm or 530/30 nm bandpass lters, respectively, and was analyzed through Flowjo software (Tree Star, Inc. Ashland, OR). The blocking of the P-gp transport function was evaluated by the parameter FIR, calculated as the ratio of the MFI of Dox or Rho123 with the addition of the likely inhibitor to the MFI of Dox or Rho123 alone.
The viability of Lucena1 and K562 cells treated with the active compounds 3, 4a-c, 5a-c, 6, 6a-c, 7, 7a-c, 10, 10a and 11 was determined by the MTT colorimetric assay as previously described 3,5 . Both cell lines, at a nal density of 5 × 10 4 cells/well, were incubated with 0.16 to 20 µM of each tested compound previously dissolved in DMSO (0.5% v/v as nal concentration since at this no cytotoxicity was observed) in a 5% CO 2 humidi ed atmosphere at 37°C for 48 h. Then 0.5 mg/ml MTT in PBS was added. After additional incubation for 4 h, the formazan crystals obtained were dissolved in DMSO and the absorbance at 595 nm was read on an iMark micro-plate reader (Bio-Rad, USA). IC 50 values for cytotoxicity were then calculated as reported 38 using nonlinear regression by GraphPad Prism 7 (Graphpad Software, Inc., CA, USA).
Based on the results obtained in the accumulation assays, the effects of the most effective compounds 5c and 7b on reversing Dox resistance in Lucena 1 cells were evaluated as previously described 5,8,29 . Brie y, 5 x 10 4 Lucena 1 and K562 cells were seeded in 96-well plates containing RPMI-1640 medium with the anticancer drug Dox (3.4-431 or 0.3-34.5 µM, respectively) in the absence or presence of 0.02 to 1.25 µM of compounds 5c and 7b. Negative controls contained 0.5% v/v DMSO while verapamil was used as standard inhibitor. The cells were incubated at 37°C at 5% CO 2 for 48 h and the same protocol described above was followed. IC 50  Cytotoxicity on peripheral blood mononuclear cells and hemolysis assay.
The toxic effect of the most effective compounds 5c and 7b on PBMC was determined by MTT assay 39 . PBMC were isolated from heparinized blood of healthy volunteers by density centrifugation using Ficoll-Hypaque separation. Ethical approval was provided by the Catholic University of Córdoba Research Ethics Board and informed consents were obtained from all donors. Brie y, 1 × 10 5 PBMC/well were incubated with the tested compounds at concentrations ranging from 2.5 to 20 µM in the presence of 10 g/mL PHA, or with 1% DMSO as control. After 48 h incubation, MTT was added and the same protocol described above was followed to calculate the IC 50 values. An erythrocyte hemolysis assay was carried out following previous procedures 8 with the compounds assayed at 2.5 to 20 µM.

Statistical analyses.
The results are expressed as mean ± SE. The statistical analysis was performed using both paired and unpaired one-tailed Student's t test (GraphPad Prism 7.0). P-values ≤ 0.05 were considered as statistically signi cant. All experiments were repeated independently at least three times.

Molecular Modelling.
Docking and molecular dynamics.
The structures of the reference substrate (Dox), the substrate co-crystallized in the experimental P-gp structure (Taxol), the reference inhibitors tariquidar and verapamil, and the subject compounds listed in Fig. 1, were obtained by performing a conformational search (when relevant) and a full geometry optimization at the semiempirical PM6 level of theory. The structures were characterized as minima by diagonalizing the Hessian matrix and ensuring the absence of negative eigenvalues using the Gaussian 16 (Rev. A03) package 40 . Two different docking protocols were used: 1) the Autodock 4.2.6 package 41 was used by precomputing a grid inside the whole TMD, as illustrated in Fig. 5A. The procedure has already been described and applied to a different family of compounds 5,29 . Brie y, 4000 runs of Lamarckian genetic algorithm were performed for each ligand; the relatively large number of runs was in order to take into account the large size of the docking region. The population was set at 150 individuals, up to 105 generations with 1 survivor per generation and a limit of 6 x 10 6 energy evaluations and the remaining algorithm control parameters set to program defaults. The cluster analysis was made with 2.5 Å of RMSD 41 . 2) Similarly, within the same box, the second docking protocol was based on Autodok Vina 1.1.2-5 42 , collecting the rst 10 lowest poses or those within 3 kcal/mol above the lowest, setting the 'exhaustiveness' parameter to 64 (default = 8) and repeating the simulations at least three times.
The recent structure of the human P-gp co-crystallized with Taxol (PDB entry 6QEX) 42 was used. Since an experimental P-gp/chemotherapeutic complex (though at relatively low resolution) is available, the Taxol molecule was removed from the site and blind docking was assayed for both Autodock 4.2 and Vina, the latter giving the best approach. Both protocols yielded similar results. However, since the latter performed better when challenged to reproduce the experimental pose of Taxol in the X-ray structure, the discussion will be limited to the Vina results.
The most stable docked structures were used as starting geometries for the MD simulations. The ligand/protein complexes were prepared using the AMBER18 43 leap and antechamber facilities for the parametrization of the inhibitors or substrates. The charges were obtained with the -bcc option, calling the internal AM1 of the sqm module of AMBER 43 . The general setup for the MD simulations was as follows: I) 250 steepest descent minimization steps of the whole system, keeping the protein tightly restrained and embedded into a box of TIP3P water molecules with a minimum distance of 10 Å to each wall, and Cl-counter-ions to reach electro-neutrality as required. II) 6500 conjugate gradient minimization steps of the whole system. III) 100 ps slowly heating in the NTV ensemble with the protein positionally restrained in the backbone. IV) 50 ns of simulation in the NTP ensemble, at 1 atm and 300 K. Procedures III-IV were repeated in two or three independent trajectories using the Andersen thermostat and barostat 44 . In the reasonably equilibrated system, the density uctuated slightly around 1.019 g/mL. Due to the transmembrane nature of P-gp, a 50.0 kcal/Å 2 harmonic restraint was kept for the backbone atoms. Electrostatic interactions were computed using the Particle Mesh Ewald (PME) method with a cutoff of 10 Å 45 . Bonds involving hydrogen atoms were constrained using the SHAKE algorithm, allowing for an integration time step of 0.002 ps. The integration was made using the pmemd.CUDA module of the AMBER18 program 43 with the auxiliary force eld GAFF for the ligands and ff14SB for the protein 46 . The trajectories were analyzed using standard AMBER cpptraj analysis tools. The free energy calculations were made using the mmpbsa module of AMBER 18 by applying Poisson-Boltzman (PB) and Generalized Born (GB) models 47 . The energy analyses were made for the last 8-12 ns of simulation as the average over at least two independent trajectories. The frames were sampled once each 5-10 saved frames (saving 1 frame every 10 ps, thus ensuring that the energy self-correlation is small enough). The clustering analyses for visualization of representative conformations and rendering of some gures were made using Chimera 1.14 48 . Most graphic rendering was prepared using VMD 1.9.3 49 .
Topological analysis of electron density.
To carry out QTAIM analysis, MD trajectories were rst clustered based on the root mean square deviation (RMSD) of the heavy atoms. The representative structures from the most populated clusters were selected. Afterwards, a reduced 3D model was constructed for each compound studied, selecting only those residues that directly interact with the ligands. All amino acids found within a radius of 5 Å of distance from each ligand atom were included. The wave function of the reduced models generated at the M062X/6-31G(d) level of theory were computed with the Gaussian16 40 package and were subjected to a Quantum Theory Atoms In Molecules (QTAIM) analysis 50 using the Multiwfn software 51 . Molecular graphs were depicted with Pymol 52 . QTAIM calculations were performed in order to determine the ρ(r) values at the bond critical points (BCPs) established between each atom of the ligand and a particular atom from the backbone or side chain of an amino acid of the receptor. In order to obtain the total ρ(r) value of the interaction, we performed the sum of the ρ(r) values of each BCP between the amino acid of either the complex or a particular TMH and each atom from the inhibitor. Inhibition on P-gp doxorubicin (Dox) outward transport by different concentrations of compounds 5c (A) and 7b (B), determined by accumulation assay in Lucena 1 and (C) at 20 μM of both compounds in Lucena 1 and K562 cells. Doxorubicin-associated intracellular uorescence signi cantly increased in Lucena 1 by treatments with selected compounds but not in K562. Signi cant differences relative to the respective negative control were determined by using a paired one-tailed Student's t test (*** p < 0.001, ** p < 0.01, * p < 0.05).

Figure 3
Inhibition on P-gp rhodamine 123 (Rho123) outward transport by different concentrations of compounds 5c (A) and 7b (B) determined by accumulation assay in Lucena 1 and (C) at 20 μM of both compounds in Lucena 1 and K562 cells. Rhodamine 123 associated intracellular uorescence signi cantly increased in Lucena 1 by treatments with selected compounds but not in K562. Signi cant differences relative to the respective negative control were determined by using a paired one-tailed Student's t test (** p < 0.01, * p < 0.05).

Figure 4
Dose-response curves for restoring sensitivity in Lucena 1 treated with doxorubicin (Dox) as an antineoplastic drug in the absence or presence of (A) compound 5c and (B) compound 7b. In Lucena 1 cells, doxorubicin toxicity was signi cantly increased when compounds 5c and 7b were co-administered from 0.04 and 0.08 µM, respectively (p < 0.05). The same assay was performed in K562 to discard additional effects other than P-gp. Values are expressed as means ± SE of at least three independent experiments.  Structures of doxorubicin, tariquidar and compounds 5c, 7b and 5a partitioned into three portions: moiety 1 (blue), linker (green) and moiety 2 (light blue).

Figure 12
Molecular graph showing the non-covalent interaction between P-gp amino acid residues (green) and doxorubicin (magenta). Topological elements of the charge density associated with intermolecular interactions are depicted with yellow lines (Bond Paths, BPs) and small red spheres (Bond Critical Points, BCPs).