Exploring the Monoterpene Indole Alkaloid Scaffold for Reversing P-Glycoprotein-Mediated Multidrug Resistance in Cancer

Dregamine (1), a major monoterpene indole alkaloid isolated from Tabernaemontana elegans, was submitted to chemical transformation of the ketone function, yielding 19 azines (3–21) and 11 semicarbazones (22–32) bearing aliphatic or aromatic substituents. Their structures were assigned mainly by 1D and 2D NMR (COSY, HMQC, and HMBC) experiments. Compounds 3–32 were evaluated as multidrug resistance (MDR) reversers through functional and chemosensitivity assays in a human ABCB1-transfected mouse T-lymphoma cell model, overexpressing P-glycoprotein. A significant increase of P-gp inhibitory activity was observed for most derivatives, mainly those containing azine moieties with aromatic substituents. Compounds with trimethoxyphenyl (17) or naphthyl motifs (18, 19) were among the most active, exhibiting strong inhibition at 0.2 µM. Moreover, most of the derivatives showed selective antiproliferative effects toward resistant cells, having a collateral sensitivity effect. In drug combination assays, all compounds showed to interact synergistically with doxorubicin. Selected compounds (12, 17, 18, 20, and 29) were evaluated in the ATPase activity assay, in which all compounds but 12 behaved as inhibitors. To gather further insights on drug–receptor interactions, in silico studies were also addressed. A QSAR model allowed us to deduce that compounds bearing bulky and lipophilic substituents were stronger P-gp inhibitors.


Introduction
One of the main concerns regarding chemotherapy failure in cancer and infectious diseases is multidrug resistance (MDR). This complex phenomenon can be classified as intrinsic or acquired resistance. The former comprises all natural features of an organism or cell that makes them resistant to a certain drug, whereas acquired MDR involves a decrease in the susceptibility to a drug, generally due to a certain genetic modification [1,2]. In cancer cells, one well-known mechanism of acquired resistance results from the overexpression of ATP binding cassette (ABC) transporters, which extrude anticancer drugs using ATP, thus decreasing the intracellular drug concentration below the therapeutic window [3,4]. One of the main ABC transporters is the P-glycoprotein (P-gp/ABCB1) that is able to efflux a wide variety of chemically unrelated compounds. Several attempts have been made to develop effective inhibitors of this transporter. However, none of them have passed clinical trials, mainly due to their considerable toxicity or low in vivo efficacy [5,6]. The lack of detailed and reliable structural information of human P-gp, at a molecular level, has also been considered another hurdle in this field. Indeed, the first crystallographic structure of a mammalian P-gp (murine) only appeared in 2009, reported by Aller et al. [7]. Although the sequence identity to human P-gp is 87%, the first human P-gp structures were only reported in the last 3 years [8,9]. Owing to a large drug-binding pocket containing multiple drug-binding sites (DBSs), the substrate polyspecificity is huge [10]; thus, the search for novel molecules to act as MDR reversal agents that could inhibit the P-gp efflux mechanism and circumvent the drug promiscuity and poly-specific binding nature of P-gp has become emergent.
A different anti-MDR approach, known as collateral sensitivity (CS), focuses on searching compounds that are selectively cytotoxic against MDR cells over the parental ones, named CS agents [11]. In fact, MDR cancer cells overexpressing ABC transporters can be, at the same time, hypersensitive to certain agents. The mechanism behind this concept is not completely explained, although potent and highly selective CS agents have been recognized to have specific properties, namely, the ability to produce reactive oxygen species (ROS); to efflux endogenous substrates of a vital molecule; to take advantage of cells sensitivity, overexpressing ABC-transporters; to modify energy levels; or to alter the biophysical membrane properties [11]. Therefore, CS represents a novel strategy for avoiding ABC-transporters mediated MDR during chemotherapy or re-sensitizing MDR cancer cells and consequently reestablishing drug effectiveness.
Aiming at finding effective compounds for reversing MDR, our group has evaluated several plant-derived compounds with different scaffolds as effective ABC-transporter inhibitors, e.g., [12][13][14], with those containing nitrogen substituents receiving particular attention [13][14][15]. Owing to indole alkaloids privileged scaffold, coupled with high bioactivity, including their ability as anticancer agents [16], our ongoing research has been focused on the generation of small libraries of indole alkaloids, from the African medicinal plant Tabernaemontana elegans, through isolation and further derivatization, to establish structure-activity relationships, concerning their ability for reversing MDR [13][14][15]17]. The monoterpene indole alkaloid epimers dregamine (1) and tabernaemontanine, isolated from T. elegans [13], were previously functionalized at the indole nitrogen, giving rise to several N-alkylated derivatives with a significant enhancement in the P-gp inhibitory activity, when compared to the parental compounds [13]. More precisely, it was found that compounds sharing N-phenethyl motifs strongly inhibited the P-gp efflux activity [13]. Likewise, the indole alkaloid scaffold was functionalized in the carbonyl group. The highest and selective P-gp inhibitory activity was found for compounds with a para-methylbenzylidene moiety, whereas other compounds with different substituents were selective for MRP1 [14]. Recently, using a different anti-MDR approach, we have identified an indole alkaloid derivative as an inhibitor of homologous DNA repair by disrupting the breast cancer susceptibility protein BRCA1 interaction with its binding partner, BRCA1-associated ring domain protein (BARD1), in triple-negative breast and ovarian cancers [17].
Therefore, taking into account the previous encouraging results, the present study aimed at preparing novel derivatives for increasing our pool of analogs and thus establishing new structure-activity relationships for optimizing their structures. Thus, by manipulating the dregamine carbonyl at C-3, through the insertion of new substituents containing nitrogen atoms together with aromatic rings, 30  new compounds were prepared. Their ability to inhibit P-gp drug efflux activity was assessed, by flow cytometry, in human ABCB1-gene transfected mouse T-lymphoma cells. Moreover, the antiproliferative activity and the in vitro interaction between the compounds and the antineoplastic drug doxorubicin were also evaluated. The type of interaction between selected derivatives and P-gp was also addressed through the ATPase activity assay. Molecular docking studies were performed to identify the preferred DBS of the derivatives within the P-gp poly-specific drug-binding pocket. A quantitative structure-activity relationship (QSAR) model was also generated for a better comprehension of which molecular descriptors may affect the biological activity.
The structures of the compounds were elucidated mainly by comparing their 1 H and 13 C NMR data with those of dregamine (1), coupled with two-dimensional NMR experiments (COSY, HMQC, and HMBC). When comparing the NMR data of compounds 3-21 with those of compound 1, the main differences were additional carbon and proton signals owing to the new substituent, such as a downfield triplet observed for compound 3 (δ H 7.86) and a singlet for compounds 4-21 (δ H 8.45-9.55) in the 1 H NMR spectra, which were assigned to the H-1 (-C=N-N=CHR). Regarding the 13 C NMR data, as expected, strong diamagnetic effects were observed at C-3 (δ C 159.5-163.1), when compared with those of 1 (δ C 191.6) as well as a carbon resonance (δ C 150.7-162.9) assignable to a second imine function (-C=N-N=CHR).
Similarly, in compounds 22-32, the semicarbazone structural feature (-C=N-NH-CO-NHR) was easily recognized in the 1 H NMR spectra through a singlet, without correlation in the HMQC spectrum, at δ H 7.58-8.82, assignable to the exchangeable NH-2 protons (-C=N-NH-CO-NHR), whereas the NH-2 (-C=N-NH-CO-NHR) proton signal was observed with different multiplicity and location, depending on each substituent, namely, as a singlet (δ H 5.97, 23; δ H 8.11-8.59, 25-31), doublet (δ H 6.15, 24), or triplet (δ H 6.23-6.27, 22 and 32). The assignment of the exchangeable NH protons signals, which were removed with the addition of deuterium oxide after heating (Supplementary Information), was substantiated by the cross-peaks observed in the 1 H-1 H COSY spectra (compounds 22, 24, and 32) (Supplementary Information) and 2 J C-H and 3 J C-H heterocorrelations observed in the HMBC data between the NH-2 proton and the carbonyl C-1 and the imine carbon C-3. In turn, for the NH-2 proton heterocorrelations with C-1 , C-2 , and C-6 were observed, depending on the substituent. Additionally, the semicarbazone moiety was corroborated in the 13 C NMR spectra by the strong diamagnetic effect at C-3 corresponding to C=N (δ C 143.1-149.1) and the presence of a shielded carbonyl group (δ C 154.1-156.7) due to the monomeric effects of the adjacent nitrogen atoms (-C=N-NH-CO-NHR).

In Vitro Antiproliferative Assay and Collateral Sensitivity Effect
The antiproliferative activity of dregamine (1) and derivatives  was assessed through the thiazolyl blue tetrazolium bromide (MTT) assay on sensitive L5178Y mouse T-lymphoma cells (PAR) and corresponding resistant human ABCB1-gene transfected L5178Y subline (MDR). Non-cancer mouse embryonic fibroblasts (NIH/3T3) were also used. The results were obtained in terms of the concentration of the compound causing 50% inhibition (IC50), as shown in Table 1 With the exception of compound 3, all derivatives were found to have higher antiproliferative activity (IC 50 values ranging between 5.43 ± 0. 38 and 26.40 ± 0.53, PAR cells; 4.28 ± 0. 25 and 12.91 ± 0.35, MDR cells) than the parental compound 1 (IC 50 = 37.21 ± 4.99, PAR cells; 22.97 ± 0.48, MDR cells) on both sensitive and resistant cells. It is noteworthy that all compounds but 19, 20, and 29 were proved to have a stronger antiproliferative effect against the resistant cell line when compared to the parental one. Thus, in order to evaluate their potential collateral sensitivity effect, relative resistance (RR) values were determined as the ratio between the IC 50 of a compound against a resistant subline and the IC 50 against the corresponding parental line. Compounds having an RR < 1 show selectivity against the MDR cells, whereas RR ≤ 0.5 means that the CS effect occurs [20]. As can be observed in Table 1, compounds 5, 7, 9, 11, 13, 14, 23, and 24 exhibited RR ≤ 0.5, pointing out their potential as CS agents.
This set of indole alkaloid derivatives possess metal-chelating properties that may explain, at least partially, their CS effect in P-gp-overexpressing cells. In fact, taking into account that the most promising MDR-selective compounds reported in the literature are metal chelators, it is believed that this metal-chelating ability is responsible for in-creased cytotoxicity against MDR cells [21]. This assumption has been substantiated by the MDR-selective toxicity against P-gp-overexpressing cells of several strong metal-chelating thiosemicarbazones such as triapine [22][23][24]. Moreover, it was demonstrated that they do not act only as simple chelators, removing cellular metals, such as iron, but also as metal-interacting agents [25,26].
In a previous study, we have found that some indole alkaloid derivatives, mostly sharing a new aliphatic azine moiety, showed CS activity in MRP1-overexpressing cancer cells. Furthermore, some of these compounds were able to induce MRP1-mediated glutathione efflux, thus increasing its intracellular depletion [14].
In addition, the antiproliferative activity of the compounds was also evaluated on non-cancerous mouse embryonic fibroblast cells (NIH/3T3), whose values were compared with those of PAR and MDR cells by evaluation of the selective index (SI) values. As it can be seen in Table 1, the SI values indicated that most of the compounds exerted a selective activity toward mouse T-lymphoma cells, mainly against the drug-resistant subline (SI C/B > SI C/A ). The highest selective index values (SI C/A = 9.39; SI C/B = 10.79) were found for the derivative bearing the trimethoxyphenyl substituent (17).

Inhibition of P-Glycoprotein Efflux Activity
The evaluation of compounds' ability for inhibiting P-gp efflux activity was assessed using the rhodamine-123 functional assay by flow cytometry on sensitive mouse T-lymphoma cell line (L5178Y-PAR) and the corresponding human ABCB1-transfected MDR subline (L5178Y-MDR). Fluorescence activity ratio (FAR) values were calculated, measuring the quotient between the intracellular accumulation of rhodamine-123 in resistant and sensitive cancer cells. All compounds were tested at 0.2 and 2 µM, and the corresponding results are summarized in Tables 2 and 3. Verapamil (20 µM), a standard P-gp inhibitor, was used as a positive control. Inhibitory activity was assumed to take place when the FAR value was above 1, whereas the compounds were considered as strong inhibitors if the FAR ratio was higher than 10 [13].  The activity of the compounds was mostly observed in a concentration-dependent manner. When tested at the lowest concentration (0.2 µM), most of the azine derivatives were found to be active, with compounds 12, 17-19 exhibiting strong (FAR > 10) P-gp inhibitory activities (FAR values ranging between 11.57 and 30.74) ( Table 2). At 2 µM, it was found that, with the exception of compound 13, the azines bearing aromatic substituents (4-12, 14-21) strongly inhibited the P-gp efflux activity (FAR values ranging from 15.69 to 128.48), having FAR values significantly higher than verapamil (up to 20-fold at a 10-fold lower concentration), those containing benzyloxybenzene, trimethoxyphenyl, naphthyl, or indolyl moieties (16)(17)(18)(19)(20)(21) being the most active (FAR > 95.5). Conversely, no significant activity was found for the azine derivative with an aliphatic substituent (3).
These results clearly emphasized the relevance of extra aromatic motifs attached to the monoterpene indole alkaloid scaffold, whose contribution to the modulatory activity enhancement may be explained owing to additional electrostatic and π-π interactions between aromatic substituents and amino acid residues in the P-gp drug binding site.
Interestingly, the effect of extra aromatic moieties was still more evident for derivatives 16, 18-21, bearing substituents with more than one aromatic ring, exhibiting remarkable inhibition (FAR > 95.5, at 2 µM).
When comparing FAR values obtained at the lowest concentration (0.2 µM) for the azine derivatives bearing mono-substituted phenyl groups (4-16), compound 12 was the most active, corroborating that the presence of a tertiary nitrogen atom is also an important feature for P-gp modulation [27].

Checkerboard Combination Assay
The in vitro interactions between the compounds and the well-known antitumor drug and P-gp substrate, doxorubicin, were evaluated in a combination chemotherapy model on human ABCB1-transfected mouse T-lymphoma cells. The nature of drug-drug interactions was assessed by determination of the combination index (CI) using the Chou and Talalay method (Figure 1), and therefore evaluated as synergistic (CI < 1), additive (CI = 1), or antagonistic (CI > 1) [28]. in the antiproliferative activity of doxorubicin, in drug combination assays, on human ABCB1-transfected L5178Y mouse lymphoma cells (L5178Y. Combination index (CI) parameter is the mean of three CI values determined based on different drug ratios ± standard deviation (SD), for an inhibitory concentration of 50% (IC 50 ). CI < 0.1: very strong synergism; 0.1 < CI < 0.3: strong synergism; 0.3 < CI < 0.7: synergism; 0.7 < CI < 0.9: moderate to slight synergism [29].
As it can be seen in Figure 1, all compounds exhibited a synergistic behavior (CI < 1) when co-administered with doxorubicin, thus substantiating the results obtained in the transport assay. Among them, compound 4 showed very strong synergism (CI < 0.1), whereas other derivatives (5-7, 9, 10, 12, 15, 18-20, 22, 23, 27-29, and 31) exhibited strong synergism with the anticancer drug. On the other hand, compounds inactive or with weak activity in the transport assay, such as the derivatives with aliphatic substituents (3,(22)(23)(24), also enhanced the cytotoxicity of doxorubicin in a synergistic mode, thus suggesting that a different type of mechanism for re-sensitizing the MDR phenotype may occur.

P-gp ATPase Activity Assay
Once it was established that ATP hydrolysis is directly associated with the P-gp efflux activity [8], the interaction mode between selected compounds (12, 17, 18, 20, and 29) and this transporter protein was further investigated using human P-gp membranes in the ATPase activity assay (P-gp-Glo TM ) [30] to gather insights about their interaction with P-gp activity, namely, as stimulators or inhibitors. This ATPase activity assay is based on the ATP dependence of the light-generating reaction of firefly luciferase. After incubating P-gp with ATP, the reaction is stopped, and the ATP consumption by P-gp is given by a luciferase-generated luminescent signal due to the remaining unmetabolized ATP. Thus, a greater decrease in the signal means a higher P-gp activity [30].
Sodium orthovanadate (Vanadate, Na 3 VO 4 ) is an inhibitor of P-gp ATPase activity; thus, after treating samples with vanadate, no P-gp-dependent ATP consumption is observed [30]. Therefore, the P-gp basal ATP consumption (basal activity) is defined as the difference between the luminescent signal of samples treated with vanadate and untreated. Consequently, the tested compounds were ranked as stimulators or inhibitors by comparing their P-gp ATPase activity with the basal activity. Verapamil (0.5 mM), a known P-gp substrate and activator of the ATPase activity of this transporter, thus causing a decrease in the percentage of luminescence of luciferase when compared with untreated samples (Figure 2), was used as control. The results are represented in Figure 2 as the difference between the luminescence of luciferase observed when treated with the compounds and with vanadate in comparison with the basal activity (100%). The compounds are identified as P-gp substrates if they stimulate its basal activity (>100%) or as modulators when inhibiting basal activity (<100%). Compounds were tested at 25 µM and verapamil at 0.5 mM. Results were calculated as the means ± SD from experiments performed in triplicates.
As it could be observed in Figure 2, at the concentration used, an increase in the luminescent signal in relation to the untreated sample (% basal P-gp activity < 100%) was observed for compounds 17, 18, 20, and 29, thus behaving as inhibitors of P-gp ATPase activity. In this way, they may possibly act by binding to a P-gp allosteric residue, decreasing its efflux activity and consequently inhibiting the ATPase activity, or by directly blocking ATP hydrolysis by binding to the P-gp ATP binding site (non-competitive P-gp inhibitors). Conversely, compound 12 showed a similar behavior of the P-gp substrate verapamil, increasing the ATP consumption by stimulating the P-gp activity, thus suggesting that, at the concentration tested, it acts probably as a P-gp substrate that may, competitively, inhibit the efflux of other P-gp substrates.
2.3. In Silico Studies 2.3.1. Molecular Docking Compounds 1, 3-32 were docked inside the P-gp internal drug-binding pocket [7], and the relative location of top-ranked binding poses was assessed to classify the molecules as modulators (M-site) or substrates (R-and H-sites), as previously reported [10] ( Figure 3A). The 20 best poses were visualized, and the main results are displayed in Table 4, coupled with the experimental FAR values at 2 µM, as performed in our previous study. It should be noticed that most derivatives preferred to bind at M-site, and even those that had their best poses at substrate-binding sites (excluding compounds 10 and 11) either had similar modulator-binding energy or a fewer number of poses compared to those found in the M-site. According to our previous studies, it was proposed that the type, number, and distribution of interactions between a molecule bridging N-terminal and C-terminal P-gp halves (cross-interactions) could impact the P-gp ability for conformational changes and consequently have an influence on the efflux phenomenon [10,31]. Therefore, analysis of the interactions between P-gp and the best pose at the M-site was accomplished to determine the cross-interaction capability (CIc) for each molecule (Supplementary Information of [10]). Then, for a better understanding of the MDR-reversal capability of the derivatives, an overall view including binding energies, total number of interactions, and cross-interaction capabilities was performed, in which the main results are summarized in Table 4. These results showed that compounds 4, 6, 16, and 19 had their best pose at the M-site and exhibited a strong binding affinity with the target coupled with strong cross-interactions with both P-gp halves, and therefore were considered as non-competitive inhibitors.
The compounds with the highest FAR values (16)(17)(18)(19)(20)(21) have shown both moderate to strong affinity for the M-site and CIc capability. In this classification, the compound bearing the trimethoxyphenyl substituent (17) is an outlier owing to its moderate affinity, in which the best M-pose obtained was the fifth, coupled with weak cross-interactions with both P-gp halves. However, although compound 17 is among the compounds having the lowest binding affinity to the M-site, it also had a moderate affinity to the substrate-binding sites, making this compound a prototype for competitive modulation through both sites. Unfortunately, the semicarbazones set (22-32) did not show any direct match between the FAR values and the virtual screening results presented herein and used previously to rationalize the activity in other sets of compounds [31]. In this regard, it should be highlighted that several physicochemical properties are not considered in this model, and therefore the virtual screening results were coupled with QSAR analysis in the next section for a better elucidation between molecular features and the biological activity.

QSAR Modeling
Firstly, an extensive database of molecular descriptors (topological, geometrical, and constitutional) was generated, obtaining hundreds of physicochemical properties for each molecule using E-DRAGON, MOE, and PaDEL programs to isolate each descriptor that individually contributed the most to the molecule's potency. Afterward, FAR values were added to the dataset, and a search for the most significant combination of molecular descriptors in each database was performed using WEKA software. The regression results were split between the two sets of compounds (azines 3-21 and semicarbazones 22-32), and the coefficient of determination values for representative descriptors are shown in Table 5. As it can be observed in Table 5, some physicochemical descriptors related to molecular shape and branching (EEig11x, X5, WTPT-2) and atomic contributions to logP (GCUT_SLOGP_3) or molar refractivity (GCUT_SMR_3) were found to have an influence on P-gp inhibitory activity for the compounds with azine substituents attached to the main scaffold. In contrast, the experimental FAR values for the semicarbazones did not show the same tendency, clearly demonstrating that they behave as a class with completely different outcomes regarding activity and are in agreement with the molecular docking analysis.
In order to allow a better understanding of which structural features in dregamine derivatives  contribute the most to P-gp modulatory activity, a QSAR model was built using the most suitable descriptors from E-DRAGON, MOE, and PaDEL software programs and assessed in the WEKA program through the select attributor tool (see the experimental section). After the reduction of the molecular descriptors, a QSAR model was generated in WEKA using the following multivariate linear regression (LR): The statistical data for the LR model ( Figure 4) can be seen in Table 6, in which an R 2 of 0.937 was obtained, with a mean absolute error (MAE) and root mean squared error (RMSE) of 6.941 and 9.998, respectively. The internal and external validations of the model were performed through the 10-fold cross-validation test and test set methods. The corresponding cross-validation parameter, q 2 , and squared correlation for the test set, R 2 pred , showed values of 0.887 and 0.763, respectively, confirming the reliability of the model.
Regarding the regression coefficient values observed in Equation (1), it is possible to verify that descriptor path/walk 4-Randic shape index (PW4) is the most relevant. This topological descriptor introduced by Randic [32] increases with increased branching in the vertices, showing a positive influence of branching on biological activity. The second most significant descriptor verified in this regression was the bond information content (BIC 3 ) index (neighborhood symmetry of three-order) [33], indicating that the higher the edge's number, the lower the BIC value. Consequently, the negative coefficient of BIC 3 shows that the biological activity is estimated to increase for compounds with more edges. The highest eigenvalue No. 2 of the burden matrix/weighted by atomic van der Waals volumes (BEHv2) [34] also influences the biological activity in a positive manner. Thus, the P-gp inhibition tends to increase with the size of the molecule. Finally, other descriptors that proved to correlate with activity were the second hydrophilic-lipophilic balance (vsurf_HL2) and the octanol/water distribution coefficient at pH 7, calculated as a state average (h_logD). As expected, for more lipophilic compounds, the FAR parameter is estimated to increase, which is in agreement with our previous study [13].  In order to increase the robustness and predictability of the study, different QSAR models using machine-learning methods were applied, maintaining the set of descriptors selected above: artificial neural network (MLPRegressor) and support vector machine (SMOReg). As can be seen in Table 6, both models showed R 2 and q 2 values above 0.7, proving the reliability of the models.

General Experimental Procedures
All solvents were dried according to publish methods and distilled prior to use. Other reagents obtained from commercial suppliers were used without further purification. Lowresolution mass spectrometry was performed in a Triple Quadrupole mass spectrometer (Micromass Quattro Micro API, Waters). NMR spectra were recorded on a Bruker 300 Ultra-Shield instrument ( 1 H 300 MHz, 13 C 75 MHz). 1 H and 13 C NMR chemical shifts are expressed in δ (ppm), referenced to CDCl 3 solvent, with the corresponding proton coupling constants (J) in Hertz (Hz). NMR spectra were assigned using appropriate DEPT, COSY, HSQC, and HMBC sequences. Column chromatography was performed on silica gel (Merck 9385). TLC was performed on precoated Merck silica gel 60 F 254 plates, with visualization under UV light (λ 254 and 366 nm) and by spraying either with Dragendorff's reagent or a solution of H 2 SO 4 -MeOH (1:1), followed by heating.

Test Compounds
Dregamine (1) was isolated from the MeOH extract of Tabernaemontana elegans roots, as previously reported [13].
The hydrazone derivative (2) was obtained from reaction of dregamine (1) (1 equiv.) with hydrazine monohydrate solution 98% (5 equiv.) dissolved in MeOH. The mixture was stirred under reflux overnight. Afterward, the reaction mixture was extracted with EtOAc (3 × 50 mL). After drying the organic layers with Na 2 SO 4 , the solvent was evaporated under vacuum at 40 • C, and the resulting residue was submitted to flash chromatography (silica gel, CH 2 Cl 2 /MeOH, 1:0 to 49:1). The preparation of the new derivatives 3-32 is described below.

Antiproliferative Assays
The antiproliferative effects of all compounds were tested in a range of decreasing concentrations (starting with 100 µM, then two-fold serial dilution), using mouse lymphoma cells as experimental model, in 96-well flat bottomed microtiter plates. Cisplatin (TEVA Pharmaceutical Company, Petah Tikva, Israel) used in cell lines served as positive control. First, the compounds were diluted in 100 µL of medium. The maximum tested concentration of each compound was 100 µM. Then, 6 × 10 3 cells in 100 µL of medium were added to each well, with the exception of the medium control wells. The culture plates were initially incubated at 37 • C for 72 h, and at the end of the incubation period, 20 µL of MTT (thiazolyl blue tetrazolium bromide; Sigma-Aldrich Chemie GmbH, Steinheim) solution of 5 mg/mL in phosphate-buffered saline (PBS) was added to each well and incubated for another 4 h. Then, 100 µL of 10% SDS (sodium dodecyl sulfate, Sigma) solution (10% in 0.01 N HCl) was added to each well, and the plates were further incubated overnight at 37 • C. Cell growth was determined by measuring the optical density (OD) at 540/630 nm with a Multiscan EX ELISA reader (ThermoLabsSystems, Cheshire, WA, USA). The percentage of inhibition of cell growth was determined according to Equation (2). All experiments were performed in triplicate. The results were expressed as the mean IC 50 ± SD, and the IC 50 values were obtained by best-fitting the dose-dependent inhibition curves in GraphPad Prism 5 software [29]. Only data from analysis with R 2 > 0.90 were presented.
The assay was performed according to the previously applied experimental settings [13].
The statistical analysis of data was performed using GraphPad Prism 5 software, applying the two-tailed t-test, and p-values of <0.05 were considered significant.

Rhodamine-123 Accumulation Assay
PAR and MDR mouse T-lymphoma cells were used in a density of 2 × 10 6 cells/mL, resuspended in serum-free McCoy's 5A medium and distributed in 500 µL aliquots. The compounds were pipetted at two concentrations (0.2 or 2 µM), and verapamil (positive control, EGIS Pharmaceuticals PLC, Budapest, Hungary) was applied at 20 µM. DMSO at 2% was also added as solvent control. The samples were incubated for 10 min at room temperature, after which 10 µL of rhodamine-123 (5.2 µM final concentration) was measured to the samples. After 20 min of incubation at 37 • C, the samples were washed twice, resuspended in 1 mL of PBS, and analyzed by flow cytometry (Partec CyFlow Space Instrument, Partec GmbH, Münster, Germany). The resulting histograms were evaluated regarding mean fluorescence intensity (FL-1), standard deviation, both forward scatter (FSC) and side scatter (SSC) parameters, and the peak channel of 20,000 individual cells belonging to the total and the gated populations (Supporting Information). The fluorescence activity ratio was calculated on the basis of the quotient between FL-1 of treated/untreated resistant cell line (ABCB1-transfected mouse lymphoma cells) and the respective treated/untreated sensitive cell line (PAR mouse lymphoma cells), according to Equation (3).

Drug Combination Assay
Doxorubicin (2 mg/mL, Teva Pharmaceuticals, Budapest, Hungary) was serially diluted horizontally in 100 µL as previously described, starting with 8.6 µM. The resistance modifier was subsequently diluted vertically in 50 µL; the starting concentration was determined based on the IC 50 . After resuspending the cells in culture medium, they were distributed into each well in 50 µL containing 6 × 10 3 cells, with the exception of the medium control wells, to a final volume of 200 µL per well. The checkerboard plates were kept for 72 h at 37 • C in a CO 2 incubator, and at the end of the incubation period, the cell growth was determined by MTT staining method, as described earlier. Drug interactions were evaluated using Calcusyn software [36]. Each dose-response curve (for individual agents as well as combinations) was fit to a linear model using the median effect equation in order to obtain the median effect value (corresponding to the IC 50 ) and slope (m) [28,29]. The goodness-of-fit was assessed using the linear correlation coefficient, r, and only data from analysis with r > 0.90 were presented. The extent of interaction between drugs was expressed using the combination index, in which a CI value close to 1 indicates additivity, while a CI < 1 is defined as synergy and a CI > 1 as antagonism.

ATPase Activity Assay
The P-glycoprotein ATPase activity was determined using the Pgp-Glo TM Assay Systems Promega kit according to the manufacturer's instructions [30]. Briefly, 20 µL of recombinant human P-gp membranes (1.25 mg/mL), expressing high levels of human P-gp, were incubated in 20 µL of Pgp-Glo TM assay buffer for 5 min at 37 • C. Compounds were tested at 25 µM sodium orthovanadate (Na 3 VO 4 , 0.25 mM) as an inhibitor control, verapamil as a substrate control (0.5 mM), and DMSO at 2% as solvent control. The reaction was initiated by adding 10 µL of 25 mM MgATP and incubated at 37 • C for 40 min. The ATPase reaction was stopped, and after adding 50 µL of ATP Detection Reagent, the samples were incubated at room temperature for 20 min. The luciferase-generated luminescent signal emitted was measured in a CLARIOstar Plus plate reader (BMG Labtech, UK) at 580 nm. The % of basal P-gp activity was calculated through the ratio between the luminescence measured of the P-gp ATPase activity of each compound and the basal activity according to Equation (4).

Molecular Docking
Compounds were previously drawn, energy minimized (default force-field, adjustment of hydrogens and lone pairs), and exported as mol2 files in MOE v2019.01 program. PDBQT files were created with AutoDockTools v1.5.6rc2 for further use in AutoDock VINA 1.1.2 docking software. The murine P-glycoprotein structure (ID: 3G5U) was obtained from the Protein Data Bank (PDB), in which the small linker sequence was added according to Ferreira et al. [10]. The binding location was defined by a docking box, including the whole internal cavity as defined previously by Ferreira et al. [10]. AutoDock VINA was used to generate docking poses, from which the 20 top-ranked were visualized to determine at which different drug binding sites (M, R or H) the pose was located. The ability to modulate efflux by cross-interacting with both P-gp halves (CIc) and, therefore, hinder conformation changes leading to efflux was performed using the top-ranked docking pose at the M-site (box centered at the M-site with dimensions xyz of 35.25 Å × 30.75 Å × 29.25 Å. [10] The corresponding CIc was calculated by the ratio between the nonbonded interactions at the N-terminal and C-terminal halves (with LIGPLOT). MDR-reversal capability of derivatives and inhibition category was thereafter evaluated considering the total number of contacts, binding energies, binding site, and CIc values.

QSAR Modeling
For each molecule, an extensive database of molecular descriptors (constitutional, topological, and geometrical) was generated from E-DRAGON [37] (1665 descriptors), PaDEL [38] (105 descriptors), and MOE (429 descriptors) software programs. Constitutional descriptors give information about the number of atoms and bonds within each molecule, whereas topological and geometrical descriptors make reference to the composition and spatial arrangement of a certain compound.
Thereafter, experimental FAR values obtained were added to the dataset, and the most relevant combination of molecular descriptors in each database was selected by the CfsSubsetEval attribute evaluator [39] within the WEKA software [40] using the BestFirst algorithm as the search method. In the end, the molecular descriptors were reduced to 21 (17 and 4 for E-DRAGON and MOE subsets, respectively), and the corresponding QSAR models were built in WEKA software [40]. QSAR models from each subset were obtained by univariate linear regression to identify which molecular descriptors perform better. In parallel, accurate QSAR models were also built using the machine-learning methods: a support vector machine by the SMOReg approach, using RegSMOImproved as the learning algorithm [41], and an artificial neural network approach (MLPRegressor), by training a multilayer perceptron model with a single hidden layer using WEKA's optimization class, minimizing the squared error plus a quadratic penalty with the Broyden-Fletcher-Goldfarb-Shanno (BFGS) method [42]. In both cases, the number of descriptors used was reduced as small as possible, leading to models of easier interpretation while keeping a good predictive result. The robustness of the created models was assessed by a k-fold cross-validation correlation coefficient (tenfold, q 2 ) and their predictive power (R 2 pred ) by splitting the dataset into training and test sets (75:25). Other parameters as the mean absolute error (MAE) and root mean squared error (RMSE) were calculated to reinforce the reliability of the model.

Conclusions
As ongoing research on the optimization of monoterpene indole alkaloids as MDRreversers, this work was focused on the generation of new analogs by modifying the ketone group of dregamine (1), yielding 19 azines (3-21) and 11 semicarbazones (22)(23)(24)(25)(26)(27)(28)(29)(30)(31)(32). Among the 30 new derivatives , most showed remarkable enhancement in P-gp inhibitory activity. In the transport assay, the strongest MDR reversal compounds were those having azine substituents attached to the indole alkaloid scaffold, containing trimethoxyphenyl (17) or naphthyl moieties (18,19), being, at a 10-fold lower concentration, up to 20-fold more active than the reference inhibitor verapamil. Moreover, most of the azine derivatives bearing aromatic substituents exhibited, simultaneously, a significant and MDR-selective antiproliferative effect in P-gp-overexpressing cells (5-9, 11, 12, 14-18), thus showing a dual role in reversing P-gp-mediated MDR. The results obtained in the functional assay were substantiated by those found in a combination assay, where all derivatives (3-32) displayed synergistic interactions with doxorubicin. In the ATPase activity assay, it was observed that the selected compounds 17, 18, 20, and 29 showed to behave as inhibitors, whereas compound 12 stimulated the ATPase activity, acting, possibly, as a competitive inhibitor.
In silico studies revealed that despite most compounds having their best pose at the modulator binding site (M-site), only a few azine derivatives (4, 6, 16, and 19) showed to interact strongly with P-gp together with strong cross-interaction capability, acting as non-competitive inhibitors. Conversely, compounds 10, 11, and 17 displayed a high affinity with the substrate-binding sites (R-and H-sites), and the best M-pose showed weak cross-interactions with both P-gp halves and thus was considered as a prototype of competitive inhibitors. A QSAR model was built, and the results showed that compounds having more lipophilic and bulkier substituents may affect, in a positive manner, the P-gp inhibitory activity.