Mitoxantrone, rhodamine 123, GF120918 (Elacridar), Ko143, hydrocortisone, prednisolone, dexamethasone, ivermectin, lopinavir, hydroxychloroquine, chloroquine and oseltamivir and MTT were purchased from Sigma-Aldrich. Hoechst 33342 and TRIzol were purchased from Invitrogen. All other reagents were commercial products of the highest available purity.

Human HEK293 parental cells (wild-type) and HEK293 cells stably transfected with ABCG2 (HEK293-ABCG2), mouse NIH3T3 parental cells (wild-type) and NIH3T3 cells stably transfected with P-gp (NIH3T3-ABCB1), hamster BHK21 parental cells (wild-type) and BHK21 cells stably transfected with MRP1 (HEK293-ABCC1) were provided by Dr Attilio Di Pietro (IBCP, Lyon, France). All cells were maintained in Dulbecco's Modified Eagle's Medium (DMEM high glucose) supplemented with 10 % fetal bovine serum (FBS), 1 % penicillin/streptomycin, and with 0.75 mg/mL G418 (HEK293-ABCG2), 60 ng/mL colchicine (NIH3T3-ABCB1) or 0.1 mg/mL methotrexate (BHK21-ABCC1) at 37 °C in 5 % CO₂ atmosphere.

The ability of drugs to inhibit the transport function of ABC proteins was evaluated using fluorescent substrates by flow cytometry. Cells were aliquoted at a density of 1.0 x 10⁵ cells/tube. Cells were exposed to fluorescent substrates (Hoechst 33342 at 3 µM, rhodamine 123 at 10 µM, calcein-AM at 0.15 µM and mitoxantrone from 2.5 to 25 µM) with or without drugs at different concentrations, and incubated at 37 °C in 5 % CO₂ for 45 min. The cells were centrifuged (2,000 x g for 5 min) and resuspended with 300 µL of cold phosphate buffer saline (PBS) and kept on ice until flow cytometry analysis. Intracellular substrate fluorescence data were acquired using a FACS Celesta (equipped with three lasers: 355, 405 and 488 nm) or a FACS Calibur (equipped with two lasers: 488 and 635 nm) flow cytometer. At least 10,000 events were collected, and the median fluorescence intensities were used for the calculations. The inhibition percentage was calculated using parental cells or a reference inhibitor to achieve 100 % of inhibition. In all experiments, at least three independent replicates were used and IC₅₀ values were calculated by using GraphPad prism software version 6.01.

Cells were seeded (2.0 x 10⁴ cells/well) into a 96 wells plate and incubated for 24 h to the attachment. The cells viability was evaluated by MTT (3-[4,5-dimethylthiazol-2-yl]-2,5 diphenyl tetrazolium bromide) assay. Cells were treated with increasing concentrations of drugs for 72 h. After this period, the medium was discarded and the cells monolayer was washed with PBS (100 μL), followed by incubation at 37 °C with MTT solution (100 μL of solution 0.5 mg/mL in PBS) for 4 h. Then, the solution was discarded, and the formazan crystals dissolved with 100 μL of ethanol/DMSO (1:1). The absorbance was measured using a microplate reader at 595 nm (Bio-Rad iMark).

The effect of drugs on the binding of a conformational antibody was determined by flow cytometry. HEK293-ABCG2 cells were cultivated at 37 °C in 5 % CO₂ until approximately 90 % of confluence, then detached and separated in tubes with 5 x 10⁵ cells/tube. Cells were centrifuged at 2,000 x g for 5 min and the supernatant was discarded. The cell pellet was suspended in PBS (100 µL) containing 40 µg/mL of BSA. Samples were incubated with inhibitors for 10 minutes at 37 °C. After this period, the primary antibody anti-human ABCG2 clone 5D3 (BD Pharmingen - dilution 1:100) was added to each sample and incubated for 30 min at 37 °C. Cells were centrifuged and the supernatant was discarded. Cells were suspended in PBS (100 µL) and the secondary antibody was added (anti-mouse PE, Abcam - dilution 1:200). The samples were incubated at 37 °C for 30 min, centrifuged, and the cell pellet was suspended in 300 µL of PBS. Data were recorded by flow cytometry. At least 10,000 events were collected.

Total RNA was obtained of HEK293-ABCG2 cells from tissue flasks-25 cm² (at approximately 90 % of confluence) after treatment for 72 h with lopinavir (6.25 µM) and ivermectin (1.56 µM). The total RNA isolation was performed using TRIzol (Invitrogen) protocol according to the manufacturer's instructions. RNA concentration was quantified by absorbance using the NanoDrop™ spectophotometer and the integrity was evaluated by 1 % agarose gel electrophoresis. RNA was stored at -80 °C. Two micrograms of total RNA were reverse transcribed using High Capacity cDNA Reverse Transcription Kit (Applied Biosystems) according to the manufacturer's instructions, and the resulting cDNA was stored at -20 °C. Using cDNAs as the template, quantitative real-time PCR was performed using the SYBR Green PCR Master Mix (Applied Biosystems) in a 7500™ Real-Time PCR Detection System (Applied Biosystems). A dissociation cycle was performed after each run to check for non-specific amplification or contamination. The mRNA expression levels were normalized using the geNorm 3.4 software, and the corresponding housekeeping gene expression levels. Sets of specific primers were designed using Primer designing tool - NCBI and validated through BLAST and BLAT, and their respective sequences are shown in Table 1(Tab. 1) (References in Table 1: Ali et al., 2015[3]; Jacob et al., 2013[24]; Lemma et al., 2016[33]; Potashnikova et al., 2015[44]; Saidova et al., 2018[48]; Sharan et al., 2015[49]; Zhang et al., 2017[73]). Relative expression levels were estimated using the method described by Pfaffl (2001[43]).

Protein was obtained from HEK293-ABCG2 cells from tissue flasks-25 cm² (at approximately 90 % of confluence) after treatment for 72 h with lopinavir (6.25 µM) and ivermectin (1.56 µM) using 200 µL of RIPA+ buffer and 2 µL of 0.5 M EDTA pH 8. Protein quantification was performed by Bradford and 40 µg was separated by gel electrophoresis (8 % polyacrylamide). Proteins were transferred to a PVDF membrane using a semi-dry transference system (GE Healthcare). The membrane was cut in two parts guided by the PageRuler™ Prestained protein ladder. The upper part of the membrane was incubated overnight (4 °C) with ABCG2 antibody (BXP-21, diluted 1:500) in TBST buffer (1.5 M NaCl, 0.5 mM Tris and 0.1 % Tween 20) containing 1 % powdered milk. The lower part of the membrane was incubated overnight (4 °C) with GAPDH antibody (mAB, diluted 1:5000) in TBST buffer containing 1 % powdered milk. The membranes were washed in TBST, and incubated with anti-mouse IgG HRP antibody, diluted 1:2,000 in TBST containing 1 % powdered milk at room temperature for 1 hour. After this incubation, the membranes are washed with TBST. ECL Plus kit substrate (GE Healthcare) was added to the membranes and exposed to a Amersham Hyperfilm (GE Healthcare) for 10 seconds.

ABCG2 structure with mitoxantrone was retrieved from Protein Data Bank (PDB), PDB ID 6VXI, resolution 3.7 Å (Orlando and Liao, 2020[41]). The structure was prepared and minimized by adding hydrogens, adjusting protonation states (pH 7.4) of amino acids, and fixing missing side-chain atoms using Maestro PrepWizard (version 2021.4). The missing loops between K46 and E60 (N-terminal domain or NTD), S302-P327 (NTD) and G354-Y369 (connecting the transmembrane bundle and the NTD) were generated using Prime (Jacobson et al., 2004[25]) and the final structure encompassed from A35 to S655.

Subsequently, the structures had only polar hydrogens maintained and were converted to pdbqt format using the Autodock Tools 1.5.6 (Morris et al., 2009[38]). The ligand structures for ivermectin, lopinavir and mitoxantrone were downloaded from ZINC Database (Sterling and Irwin, 2015[51]), taken into the Avogadro software and subjected to a geometry pre-optimization using the Auto Optimization tool (MMFF94s force field) (Halgren and Nachbar, 1996[21]; Halgren, 1999[20]), followed by visual inspection to ensure that there were no errors. Using Avogadro, the molecule file was prepared for a second geometry optimization in MOPAC2016 (Stewart, 2016[52]), with the semi-empirical quantum PM7 method. At the end of this step, the files were converted to the pdbqt format, also in Autodock Tools, ensuring that all torsions were set to active. For docking, the Autodock Vina 1.2.3 (Eberhardt et al., 2021[14]) was used, in which a grid box was delimited, based on the central region in which the co-crystallized ligand was originally detected. The mitoxantrone + lopinavir (MTX + LPV) and mitoxantrone + ivermectin (MTX + IVT) docking was performed sequentially, using the 6VXI + MTX docked structure for the inhibitors. The atomic coordinates of the gridbox centroid for all docking experiments were defined as: X = -0.181; Y = -0.222; Z = 0.571, with a 40 Å distance on all three axes. The exhaustiveness parameter was set to 75 and the maximum number of results (poses) to twenty, with a maximum allowed variation of 2 kcal/mol from the first to the last conformation. After, the process was performed with all molecules (substrate and inhibitors), resulting poses served as a starting point for molecular dynamics simulations (MD).

The minimized structures were submitted to MD simulation for further refinement, using a previously published protocol (Zattoni et al., 2022[72]). Selected docking poses were further validated by MD simulation, where ligand stability within the proposed pocket and its interactions were evaluated. The MD simulations were carried out using the Desmond engine (Bowers et al., 2006[9]) with the OPLS4 force-field (Lu et al., 2021[34]). The simulated system encompassed the protein-ligand complex, a predefined water model (TIP3P;´Jorgensen et al., 1983[27]) as a solvent, POPC membranes (automatically positioned according to the alpha-helices), and counterions (Na⁺ or Cl^- adjusted to neutralize the overall system charge). The system was treated in an orthorhombic box with periodic boundary conditions specifying the shape and the size of the box as 10x10x13 Å distance from the box edges to any atom of the protein. RESPA integrator time steps of 2 fs for bonded and near, and 6 fs for non-bonded terms far were applied. Short-range coulombic interactions were performed using a time step of 1 fs and a cut-off value of 9.0 Å, whereas long-range coulombic interactions were handled using the Smooth Particle Mesh Ewald (PME) method (Darden et al., 1993[10]). Standard Desmond relaxation protocol was employed. Simulations were run in the NPT ensemble, with a temperature of 310 K (Nosé-Hoover thermostat) and pressure of 1.01325 bar (Martyna-Tobias-Klein barostat). MD trajectories were visualized, and figures were produced using PyMOL v.2.5 (Schrödinger LCC, New York, NY, USA). At least three independent simulations were performed for each ligand, being 200 ns for inhibitors and 500 ns for mitoxantrone.

Protein-ligand interactions were determined using the simulation event analysis pipeline implemented in Maestro (Maestro v2021.4). Distance calculations were performed employing the Maestro event analysis tool (Schrödinger, LLC, New York, NY). The molecular mechanic energies with generalized Born and surface area continuum solvation (MM/GBSA) were calculated with Prime (Jacobson et al., 2004[25]) thermal MM/GBSA script provided by Schrödinger. Each 5^th frame of MD was used for MM/GMBSA calculations. Trajectories were clustered according to the ligand's RMSD values in order to select relevant conformations for discussion and figures, using the trj_cluster.py script provided by Schrödinger. Figures were generated using PyMOL v2.5 (Schrödinger, LLC, New York, NY).

In order to identify ABCG2 inhibitors, a repurposing drug strategy utilizing a cell-based model was used. All the eight drugs were selected based on the fact that they were used as therapeutic alternatives for COVID-19, regardless their efficacy. Hydrocortisone, prednisolone, dexamethasone, ivermectin, lopinavir, hydroxychloroquine, chloroquine and oseltamivir were tested to inhibit the activity of ABCG2 in stably transfected HEK293-ABCG2 cells overexpressing the ABCG2 transporter. This initial screening was performed by flow cytometry using Hoechst 33342 as a fluorescent substrate of ABCG2, and all drugs were essayed at 10 and 100 µM. Only lopinavir and ivermectin inhibited ABCG2 activity (Figure 2A(Fig. 2)). Both drugs produced a mild inhibition effect at 10 µM, about 25 %. However, a complete inhibition (100 %) was observed at 100 µM (Figure 2A and B(Fig. 2)). The ABCG2 inhibition caused by lopinavir (Weiss et al., 2007[69]; Bierman et al., 2010[7]) and ivermectin (Jani et al., 2011[26]) was already reported, and most recently, this inhibition effect was confirmed, showing IC₅₀ values (compound concentrations giving a half-maximal inhibition) of ABCG2 inhibition of 13.1 and 3.1 µM, for lopinavir and ivermectin, respectively (Telbisz et al., 2021[58]). Here, we observed IC₅₀ values of inhibition of 23.4 and 25.5 µM for lopinavir and ivermectin, respectively (Figure 2C and D(Fig. 2)). The differences among the IC₅₀ values could be associated with the substrate used in each study. Telbisz et al have used PhenGreen (PG)-AM as substrate of ABCG2 (Telbisz et al., 2021[58]), while we used Hoechst 33342.

We also investigated the inhibition effect of these eight drugs on P-gp and MRP1 using stably transfected NIH3T3-ABCB1 cells overexpressing P-gp and BHK21-ABCC1 cells overexpressing MRP1, respectively. In this case, rhodamine 123 was used as substrate of P-gp and calcein-AM was used as substrate of MRP1. Lopinavir and ivermectin acted as P-gp (see supplementary information Figure 1) and MRP1 (see supplementary information Figure 2) inhibitors. The inhibition at 10 µM was higher for P-gp than for MRP1. Interestingly, the inhibition caused for both drugs at 10 µM on MRP1 was similar to the observed toward ABCG2 (~ 25 %). Both drugs completely inhibited P-gp and MRP1 at 100 µM. Using calcein-AM (Telbisz et al., 2021[58]) or doxorubicin (Bierman et al., 2010[7]) as substrate, ivermectin and lopinavir were also able to inhibit P-gp and MRP1. It was recently described that hydroxychloroquine at concentrations higher than 10 µM inhibits the P-gp activity, without effect on ABCG2 activity (Weiss et al., 2020[68]). Here, we observed that both chloroquine and hydroxychloroquine inhibited the P-gp activity at 100 µM (see supplementary information Figure 1) without effect on ABCG2 (Figure 2A(Fig. 2)). In general terms, despite the search for specific inhibitors, the identification of dual- and pan-inhibitors is attractive (Zattoni et al., 2022[71]). In fact, the clinic trials failure of specific P-gp inhibitors was partially attributed to the overlap of transported substrates among the ABC transporters (Tamaki et al., 2011[56]; Robey et al., 2018[47]). Taken together, we confirmed that lopinavir and ivermectin act as pan-inhibitors of ABC transporters at 100 µM (Figure 2A(Fig. 2), see supplementary information Figures 1 and 2).

To further investigate the interaction of the eight drugs with ABCG2, a cell viability assay was performed using HEK293 cells and transfected cells overexpressing ABCG2 (HEK292-ABCG2). Many classes of ABCG2 inhibitors also are recognized as ABCG2 substrates, and this effect can be initially investigated by a cell viability assay (Zattoni et al., 2022[71]). However, this approach is only useful for cytotoxic drugs. A lower cytotoxic effect triggered by drugs on cells overexpressing the ABC transporter than on the parental cell line suggests a transport. Since SN38 (the active metabolite of irinotecan) is a substrate of ABCG2, this drug was used as control. As shown in Figure 3(Fig. 3), SN38 was transported by ABCG2, showing a lower cytotoxic effect on cells overexpressing ABCG2. All drugs decreased the cell viability after 72 hours of exposure. Some of these drugs, such as oseltamivir showed a very low cytotoxic effect, decreasing the cell viability only in concentrations higher than 200 µM (Figure 3H(Fig. 3)). In sharp contrast, lopinavir and ivermectin were highly cytotoxic, decreasing the cell viability at 12 and 3 µM, respectively (Figure 3D and E(Fig. 3)). Interestingly, the two drugs that inhibited ABCG2 transport activity also were the most cytotoxic agents.

The cytotoxic effect is an important parameter to be considered during the identification of ABCG2 inhibitors since many potent inhibitors did not follow pre-clinical studies due to their high intrinsic cytotoxicity (Boumendjel et al., 2011[8]; Zattoni et al., 2022[71]). To correlate the inhibition potency and cytotoxicity, the inhibitors were compared based on their therapeutic rate (IG₅₀/IC₅₀), where the IG₅₀ value is the concentration that decreases 50 % cell viability. In this way, compounds that present a high therapeutic rate (TR), such as chromone 6g (MBL-II-141), that showed a TR of 2000 (Valdameri et al., 2012[63]), are considered promising to be explored in pre-clinical models and clinical trials studies (Zattoni et al., 2022[71]). Here, the calculated TR for lopinavir and ivermectin were 0.64 and 0.26, respectively (see supplementary information Table 1). These very low TR values indicate that both compounds were highly cytotoxic and probably the ABCG2 inhibition effect might not be validated in animal models. However, considering a drug repurposing strategy, another parameter that should be considered is the drug plasma concentration. The highest plasma concentrations described for lopinavir and ivermectin were 15 µM (https://www.accessdata.fda.gov/drugsatfda_docs/label/2007/021226s018lbl.pdf) and 53 nM (https://www.accessdata.fda.gov/drugsatfda_docs/label/2009/050742s026lbl.pdf), respectively. Thus, the clinical use of ivermectin as an ABCG2 inhibitor is impracticable, however, the IC₅₀ value of ABCG2 inhibition of lopinavir is in the same order magnitude of the plasma concentrations (µM range). Considering that the most promising ABCG2 inhibitor identified by the drug repurposing strategy is the febuxostat (Miyata et al., 2016[36]; Toyoda et al., 2019[60]), showing a IC₅₀ of 27 nM and plasma concentration of 90 nM, its use in clinical dose possibly inhibits ABCG2 (Toyoda et al., 2019[60]). In accordance, we found that ABCG2 is possibly inhibited by lopinavir in clinical dose.

Regarding the cytotoxic profile of the eight drugs comparing both cell lines, HEK293 and HEK293-ABCG2, no difference was evidenced (Figure 3(Fig. 3)). These results suggest that none of these drugs are recognized as substrates of ABCG2 transporter. Our data further confirmed that hydroxychloroquine and lopinavir are not substrate for ABCG2 (Agarwal et al., 2007[2]; Bierman et al., 2010[7]; Weiss et al., 2020[68]). The absence of transport mediated by the target transporter is a desirable feature of inhibitors, which shows that lopinavir and ivermectin chemical structures promising scaffolds for the rational design of more potent ABCG2 inhibitors.

Although the efflux mediated by P-gp of hydrocortisone (Nakayama et al. 1999[40]), hydroxychloroquine (Weiss et al., 2020[68]), lopinavir (Agarwal et al., 2007[2]), dexamethasone (Ueda et al., 1992[61]), oseltamivir (Morimoto et al., 2008[37]), prednisolone (Karssen et al., 2002[29]) and ivermectin (Didier and Loor 1995[12]) has been described, we decided to evaluate the transport of all drugs by the MTT-based cell viability assay using parental and a cell line overexpressing P-gp. Ivermectin, dexamethasone and oseltamivir behaved as P-gp substrates, whereas hydrocortisone, prednisolone, lopinavir, hydroxychloroquine and chloroquine did not (see supplementary information Figure 3). Using the same approach for MRP1, we found that oseltamivir was the only drug not recognized as substrate. Lopinavir could be considered as strong substrate, whereas ivermectin could be considered as weak substrate (see supplementary information Figure 4). Ivermectin, chloroquine and dexamethasone have already been described as MRP1 substrates (Vezmar and Georges, 1998[66]; Ardelli, 2013[6]; Aberuyi et al., 2021[1]; Rendic, 2021[46]), while prednisolone and lopinavir have been described as non-substrates (Webster and Carlstedt-Duke, 2002[67]; Bierman et al., 2010[7]). In summary, our data confirmed that many drugs that were used for COVID-19 interact with the main ABC transporters involved in MDR, being inhibitors or substrates, and in some cases, the inhibitor is also transported, such as ivermectin on P-gp.

Since lopinavir and ivermectin inhibited ABGC2, and the IC₅₀ value of ABCG2 inhibition caused by lopinavir is in the same order of magnitude of the plasma concentration, the molecular mechanism of inhibition was studied. The interaction of drugs with ABC transporters can be studied by different in vitro approaches, including the use of conformational antibodies (Zattoni et al., 2022[71]). In the case of ABCG2 transporter, the antibody called clone 5D3 recognizes an extracellular epitope of the protein (Taylor et al., 2017[57]). In general, ABCG2 inhibitors induce a conformational change that increases the 5D3 binding, in contrast to ABCG2 substrates, which do not trigger this “5D3 shift” (Telbisz et al., 2012[59]; Zattoni et al., 2022[71]). Lopinavir and ivermectin induced an increase in 5D3 binding, such as the ABCG2 reference inhibitor Ko143 (Figure 4A(Fig. 4)). As shown by the histograms, the 5D3 shift is more pronounced for lopinavir and ivermectin than Ko143 (Figure 4B(Fig. 4)). This result agrees with our previous data that demonstrated that lopinavir and ivermectin are non-transported inhibitors of ABCG2 (Figures 2(Fig. 2) and 3(Fig. 3)).

Drugs that inhibit the transport activity of ABC transporters are called functional inhibitors. This inhibition effect is settled on the direct binding of ligands in a druggable binding pocket of these transporters, commonly located at transmembrane domains (Kowal et al., 2021[32]; Zattoni et al., 2022[71]). However, drugs targeting transcriptional or posttranslational protein levels have been described as promising to overcome the MDR phenotype. Drugs that trigger this effect are considered modulators, to differentiate functional inhibitors. In addition, a dual effect, direct transport inhibition and decreased protein levels are advantageous but poorly observed (Zattoni et al., 2022[71]). Here, the effect of lopinavir and ivermectin on ABCG2 was tested at transcriptional and translational levels, by qPCR and western blot, respectively. Considering a half-life of approximately 60 hours for ABCG2 in different cell lines (Imai et al., 2009[23]; Peng et al., 2010[42]), both assays were performed after 72 hours of drug exposure. In addition, to avoid a bias associated with the cytotoxicity effect, the highest drug concentration that does not decrease the cell viability was used to treat the cells. In this case, 6.25 and 1.56 µM for lopinavir and ivermectin, respectively (Figure 3D and E(Fig. 3)). As shown in Figure 4C(Fig. 4), lopinavir and ivermectin did not modulate the mRNA expression levels of ABCG2. The western blot results also revealed an absence of effect of both drugs on protein expression levels (Figure 4D(Fig. 4)). Together, these data confirm that lopinavir and ivermectin are not modulators of ABCG2 expression levels and should be classified as functional inhibitors of ABCG2.

To get insights into the biochemical mechanism of ABCG2 inhibition, the type of inhibition caused by ivermectin and lopinavir was investigated by varying the concentrations of the inhibitor and the substrate mitoxantrone. This ABCG2 substrate was used instead of Hoechst 33342 because only with mitoxantrone a saturation curve was achieved (Figure 5A and C(Fig. 5)). Both drugs caused a noncompetitive inhibition since an increase of V_max with no effect on K_M value was observed (Figure 5B and D(Fig. 5)). The ABCG2 noncompetitive inhibition was already reported by some compounds, such as stilbene derivatives (Valdameri et al., 2012[64]). Recently, an uncompetitive and mixed-type of inhibition were also described for indenoindole (Guragossian et al., 2021[19]) and porphyrins, respectively (Zattoni et al., 2022[72]).

In order to gain further insights into the noncompetitive inhibition caused by lopinavir and ivermectin, in silico analyses were performed using molecular docking, molecular dynamics and MM/GBSA free binding energy calculations. Initially, molecular docking analysis showed a compatibility of the binding of inhibitors lopinavir and ivermectin in presence of the substrate mitoxantrone, revealing multiple conformations of each ligand co-occupying the drug binding cavity (DBC). The results suggested that the large DBC, located between both subunits of transmembrane helices, can accommodate one molecule of mitoxantrone and one molecule of each inhibitor simultaneously (Figure 6(Fig. 6)). Analysis of the docked complex showed that mitoxantrone occupied an “upper” position, closer to the L554/L555 plug, but ivermectin and lopinavir occupied a “below” position, closer of the cytoplasmic side (Figure 6(Fig. 6)).

To further investigate the stability of the complex ligands-ABCG2 and specific types of interaction and other potential binding modes, molecular dynamics (MD) simulations were performed. The results were similar to those observed by docking (see supplementary information Figure 5). Mitoxantrone acted similarly in the presence or absence of an inhibitor, basically conserving the high frequency of F439/N436 interactions. The binding free energies (ΔG_bind) estimated revealed that the binding of ivermectin or lopinavir does not significantly alter the mitoxantrone binding affinity (Figure 6A(Fig. 6)). Docking results initially showed that the most common interaction between ABCG2 transporter and mitoxantrone involves a π-stacking with F439 (Figure 6(Fig. 6)). MD results confirmed this observation and mitoxantrone binding was mainly favored by π-stacking interactions between the anthracenedione core of mitoxantrone and F439, as well by H-bonds with N436 from both subunits. The lack of interactions between mitoxantrone and ivermectin revealed by docking results was further elucidated by MD, which showed frequent water-mediated H-bonds present between the tetrahydropyran moiety of ivermectin and E446 (Figure 6B and C(Fig. 6)). This amino acid residue also interacted with lopinavir, indicating an important role in the stabilization of these inhibitors inside the drug pocket in presence of mitoxantrone. Together, these results suggest an absence of overlap between inhibitors and mitoxantrone, providing a plausible explanation for the noncompetitive inhibition.

Furthermore, analysis of root-mean-square deviation (RMSD) values via a clustering algorithm showed three states in proximity for each inhibitor, indicating a low variation in terms of mobility inside the drug-binding pocket of ABCG2 (Figure 6C and D(Fig. 6)). These data suggest that binding of a second molecule, in this case, an inhibitor, cannot displace mitoxantrone from the DBC. In accordance, previous works of MD simulations using cryo-EM obtained structures strongly suggests that ligands can alter the protein conformation, allowing a simultaneous binding of substrates and inhibitors (Nagy et al., 2020[39]; Yu et al., 2021[70]).

Previous reports suggest that F439 is critical for the binding and stabilization of substrates by clamping the molecule inside the cavity (Gose et al., 2020[15]), and mutations at position 439 severely impair Hoechst 33342 and pheophorbide a (Gose et al., 2020[15]). Interestingly, the importance of N436 appears to be more specific and dependent on the molecule (Gose et al., 2020[15]; Kowal et al., 2021[32]), is indispensable for mitoxantrone transport (Guragossian et al., 2021[19]). The importance of E446 for ABCG2 function was also confirmed by mutagenesis studies (Khunweeraphong et al., 2017[30]; Szöllősi et al., 2019[55]). Our data suggest that mitoxantrone binds to the F436 region, while the inhibitors were trapped between E446 residues, blocking the proximity of transmembrane domains needed for the catalytic cycle of transport mediated by ABCG2. This spatial configuration allows these inhibitors to act as a wedge, preventing the approach of transmembrane helixes necessary to attain turnover conformations that potentially precede substrate efflux to the extracellular space (Yu et al., 2021[70]), but not avoiding substrate accommodation inside the DBC. However, a complete understanding of the catalytic cycle and drug dislocation from cavity 1 (drug binding pocket) to cavity 2 (above the L555), as well as the conformational changes induced by these inhibitors require additional studies.

We thank Dr Attilio Di Pietro from IBCP (France), Dr Robert W. Robey and Dr Michael Gottesman from NCI/NIH (USA) for having provided the cell lines. This work was supported by PROIND 2020 (UFPR), Fundação Araucária/PPSUS (grant number 2020131000003) and Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) (Finance code 001). The authors would like to thank the Finnish IT center for science (CSC), Ltd. for the generous computational resources. The study was financially supported by the TüCAD2, a program funded by the Federal Ministry of Education and Research (BMBF) and the Baden-Württemberg Ministry of Science as part of the Excellence Strategy of the German Federal and State Governments (TK). In addition, TK was funded by the Fortüne grant initiative under the Excellence Strategy.

The authors declare that they have no conflict of interest.