Discovery of the Inhibitory Effect of a Phosphatidylinositol Derivative on P-Glycoprotein by Virtual Screening Followed by In Vitro Cellular Studies

P-glycoprotein is capable of effluxing a broad range of cytosolic and membrane penetrating xenobiotic substrates, thus leading to multi-drug resistance and posing a threat for the therapeutic treatment of several diseases, including cancer and central nervous disorders. Herein, a virtual screening campaign followed by experimental validation in Caco-2, MDKCII, and MDKCII mdr1 transfected cell lines has been conducted for the identification of novel phospholipids with P-gp transportation inhibitory activity. Phosphatidylinositol-(1,2-dioctanoyl)-sodium salt (8∶0 PI) was found to significantly inhibit transmembrane P-gp transportation in vitro in a reproducible-, cell line-, and substrate-independent manner. Further tests are needed to determine whether this and other phosphatidylinositols could be co-administered with oral drugs to successfully increase their bioavailability. Moreover, as phosphatidylinositols and phosphoinositides are present in the human diet and are known to play an important role in signal transduction and cell motility, our finding could be of substantial interest for nutrition science as well.


Introduction
Around 5% of the human genes are involved in lipid synthesis and regulation. Not surprisingly, these biomolecules have crucial biological functions: They are the primary component of the cellular membrane, can act as first and second messengers in signal transduction and molecular recognition processes, and serve as energy storage [1]. Moreover, they modulate the activity of transporter membrane proteins, like the (ATP)-binding cassette Pglycoprotein (P-gp). Substrates transported by P-gp are highly diverse, mostly from amphipathic and neutral or weakly basic nature, and including compounds ranging in molecular weight from less than 200 Da to almost 1900 Da [2]. As a result, the transporter is capable of effluxing a broad range of cytosolic xenobiotic substrates, thus affecting the absorption of several types of drugs and eventually leading to multi-drug resistance (MDR) [3,4]. In fact, there is strong evidence of the role of P-gp in the development of diseases and addictions [5][6][7][8], posing a real threat for their therapeutic treatment. During the last decades, inhibition of P-gp has been used as a mechanism to combat MDR in cancer therapy, and many small molecules modulating its activity have been described [9,10]. Examples of such molecules include the calcium channel blocker verapamil, natural products like quinidine and the immunosuppressive agent cyclosporine A, and other surfactants and amphiphilic substances [11][12][13]. Even though the exact molecular mechanisms have yet to be completely elucidated, P-gp inhibition is mainly due to tightly binding and blocking the transmembrane substrate binding pockets or by inhibition of the ATPase activity of the cytoplasmic nucleotide-binding domain [2,14]. Inhibition can also be induced by allosteric modulation with substances binding to non-substrate binding sites within the transmembrane regions or by interaction with the P-gp surrounding cell membrane [15][16][17]. Recently, the modulation of ATPase activity and transport inhibition of several phospholipids, including phosphatidylcholine (PC), -ethanolamine (PE), -glycerol (PG), and -serine (PS) derivatives, has been studied at the cellular level. Two phosphatidylcholine derivatives, namely 8:0 PC and 10:0 PC, have been found to significantly reduce ATPase activity and inhibit membrane transport of P-gp substrates [18]. Phosphatidylinositols and their phosphorylated adducts play a key role in many biological processes, e.g. signal transduction and cytokinesis [19][20][21][22]. However, their role in substrate transportation by P-gp has not yet been studied.
P-glycoprotein consists of two a-helical transmembrane domains (TMDs), and two nucleotide binding domains (NBDs) (Fig. 1). The TMDs contain the substrate-binding sites and the translocation conduit [23]. Different binding sites for xenobiotics or drugs have been described, as well as the possibility of allocating two molecules simultaneously [24,25]. Drug transport by P-gp is driven by a switch between two main conformational states of the NBD dimer: ATP binding to the drug-binding competent state induces a rotation of the NBDs and adoption of a close conformation, whereas ATP hydrolysis leads back to the open conformation of the dimer. The close conformation mediates substrate translocation in the TMDs drug-binding sites, thus triggering the release of the substrate to the extracellular face of the membrane [26]. The elucidation of the 3D structure of mouse P-gp in the drug-binding competent state allows for the virtual screening and rational design of modulators of human P-gp [27]. In the present manuscript, a virtual screening experiment followed by in vitro cellular assays was carried out for the identification of novel phospholipids with ability to modulate transmembrane transportation by P-gp.

Target Generation
Three crystal structures of P-gp in the drug-binding competent conformation from Mus musculus are available so far (PDB IDs: 3G5U, 3G60, and 3G61 [27]). As they show a low structural difference [28], we utilized the best resolved 3G5U apo-P-gp structure as a template to generate a homology model for the human transporter ( Fig. 1, protein model included in PDB format as Supplementary Material), using the software Prime 2.2 (Schrödinger, LLC, New York, USA). Model quality was assessed using the Swiss-Model server [29]. The whole inner intermembrane region of the transporter was subjected to a computational molecular binding analysis. The quality of the model was assessed by performing molecular docking experiments with the known inhibitors Elacridar and Verapamil (XP scores of 210.9 and 29.9, respectively). We further used experimental Caco-2 results to study the prediction capabilities of our docking model, and observed 70% accuracy in the discrimination between active and inactive phospholipid molecules (Table S2).

Library Preparation
We used an in-house instance of the workflow management system Galaxy [30] to generate a ligand library containing 178 distinct phospholipids with chain length $6 of the phosphatidylcholine (PC), -ethanolamine (PE), -glycerol (PG), -inositol (PI), -serine (PS) and phosphatidic acid (PA) compound classes by performing substructure searches (query scaffolds shown in Table  S1) on the Dictionary of Natural Products [31], ChEMBL [32], and ZINC [33] libraries. Those compounds were duly processed using LigPrep 2.4 (Schrödinger, LLC, New York, USA) at pH 6.562.0, for the sake of consistency with the buffered experimental validation process.

Statistical Analysis of the in vitro Assays
We applied One-Way ANOVA combined with Tukey and Levene's test using the OriginPro 8.1G (Origin Lab) software to determine significant effects of lipid formulations compared to positive (verapamil) or negative control samples (buffer). Further details on the statistical methods along with the experimental setting have been already described elsewhere [18].

Results and Discussion
Herein, we performed a virtual screening campaign for the discovery of novel phospholipids with cell transport activity (see Methods). Among the possible mechanisms described for P-gp inhibition [34], we focused on molecules that would either sterically block, allosterically modulate, or compete with other substrates of interest; therefore we took into consideration the TMDs of the transporter. Figure 1 depicts the homology model built for human P-gp. The transmembrane region of the transporter, subjected to the molecular docking experiment, shows a very high sequence identity. Regions of the protein exposed to the intra-and extracellular spaces present a lower conservation. This model exhibited 70% accuracy in the discrimination between active and inactive phospholipids in docking studies (Table S2). Thus, a total of 178 distinct molecules of the PA, PC, PE, PG, PI, and PS compound classes were virtually tested on the substratebinding domain of the transporter. Interestingly, phospholipids of the previously unreported PI class very significantly accumulated on the top 10% of the screening results (see Table 1), suggesting a preference of the transporter to interact with phosphatidylinositol derivatives. From those 18 top-ranked molecules, PA derivatives accumulated significantly but to a lesser extent. Docking poses of the other phosphatidyl classes did not accumulate significantly within the top scoring range (XP scores from 212 to 215). These findings led to the selection of two purchasable candidate compounds of each of the PI and PA classes for further experimental in vitro testing, namely 8:0 PI, 6:0 PIP 1 , 18:0/ 18:1 PA, and 14:0 PA (Fig. 2).
The binding mode obtained for compound 8:0 PI in the substrate binding pocket of human P-gp indicates that its polar head group entangles in hydrogen-bond interactions with Asn721, Gln725, and Gln990 (Fig. S2). These polar residues define a positively charged region on the protein surface (Fig. S3, coloured in blue) that surrounds the acidic and inositol moieties of the lipid, while the aliphatic fatty acid chains remain in lipophilic cavities of the pocket (mainly coloured in white or pale). Indeed, most of the pocket is of hydrophobic nature, accounting for the promiscuous behaviour of the transporter [2]. Notably, the analysis of the location of the phosphate group in the rest of the top-ranked 18 molecules allowed for the identification of a favoured region for phosphate positioning in a cavity defined by the TM7 and TM12 a-helixes, in the asymmetric protein's substrate pocket (Fig. S3, [27]). The phosphate group of 8:0 PI precisely locates in the identified favoured region.

Characterisation of the Lipid Formulations by DLS and Cryo-TEM
Depending on their structure, phospholipids can occur in a wide range of aggregation states when dispersed in an aqueous medium: dissolved as monomers, aggregated to small disordered structures, or as clearly defined vesicles, i.e. liposomes [18]. A key factor is the chain length of the fatty acid residues, since chains shorter than 6-8 carbon atoms hamper vesicle formation due to the decreased hydrophobic interactions within the lipid bilayer. Another influence is exerted by the polar head group: Charged structures induce strong repulsive forces between lipid monomers, leading to unstable aggregates, if any.
This circumstance could be also confirmed with Cryo-Transmission Electron Microscopy (Cryo-TEM) for the derivatives tested in the current study (Fig. 3). The long-chained 18:0/ 18:1 PA formed vesicles in demineralised water, but of a very broad size distribution (Table S3). Given that the tested compounds were formulated as sodium-salts, the negatively charged phosphate group might have prevented intensive vesicle formation due to intermolecular electrostatic repulsive forces. In the case of PI derivatives, presenting much shorter carbon chains in addition to the negatively charged head group, the liposomal aggregation is rather unlikely. Indeed, this hypothesis is in agreement with Dynamic Light Scattering (DLS) measurements (Table S3), as the PDI values exceeded the limit of 0.8, therefore impeding meaningful size determinations. Figure 3b verifies that a 8:0 PI ( = PtdIns 8:0-Na) formulation of 1.0 mM in demineralised water contained neither vesicles nor aggregates. Thus, we concluded that 8:0 PI occurred as a solubilized monomer.

Transport Studies across Caco-2 Monolayers
We studied the interference of our candidate compounds with the transport capabilities of the P-gp substrate digoxin, by measuring the differential flux of the transport marker across a cell layer, due to significant changes in secretion or absorption ( Fig. 4 and Table S4). As transport medium we chose isotonic hank's balanced salt solution (HBSS) pH 6.5. Digoxin transport (1 mM; partially 3 H-labelled) in untreated human colon carcinoma (Caco-2) cell layers was distinctly asymmetric, the secretory flux exceeding roughly by 12-fold the absorptive one (control). This accounted for the distinct expression of the efflux transporter, especially as the substrate flux was nearly equalized after pre- incubation with the established P-gp inhibitor verapamil. Caco-2 cells are a well-established biopharmaceutical tool to investigate absorption processes across the human intestinal mucosa; therefore, these transport studies are of high importance to predict oral bioavailability enhancement due to P-gp inhibition [35,36].
In the case of phosphatidic acid derivatives (Fig. 4a), no alteration of digoxin transport was observed after pre-incubation with 14:0 PA at 0.1 mM. Increase of the lipid amount led to a reduction of the apparent permeability (P app ) ratio value by 30%. However, this increase in concentration provoked a drop to 63% in transepithelial electrical resistance (TEER), indicating a disruption of the cell monolayer. The unsaturated PA derivative, 18:0/18:1 PA, did not show any potency in inhibition of digoxin transport, even at a high concentration of 0.3 mM. These observations do not agree with the in silico predictions. However, docking studies were conducted with the respective lipid monomer and any aggregation, such as vesicle formation or other possible polymeric interferences, was not taken into account. Since both PA derivatives present long, mainly saturated fatty acid chains, they tend to form lipid aggregates, as depicted in Fig. 3a [37]. Previous studies have shown that phospholipids might interact with P-gp preferentially in their monomeric state [18], thus compounds with a low monomeric concentration might not be able to cause observable transporter inhibition. Figure 4b features the modulation produced in the digoxin transportation by human P-gp in Caco-2 studies after preincubation with our candidate phosphatidylinositol derivatives. Interestingly, 8:0 PI ( = PtdIns 8:0-Na) induced a significant (45%) reduction of P app ratio at 0.1 mM combined with a TEER value within the acceptable limit. A 2-fold increase in lipid concentration promoted a stronger inhibition of transportation, but at the expense of destroying the cell integrity nearly completely. No effect on transportation was observed after pre-incubation with 6:0 PIP 1 .   The striking and singular efficacy of 8:0 PI might be related to its caprylic moiety, also found in 8:0 PC, which was recently identified as well as active in inhibiting P-gp transportation [18]. Moreover, this short fatty acid chain increases the critical micelle concentration of the lipid, leading to a comparably high monomer amount in the formulation (Fig. 3b). Another key aspect of the compound's activity could be explained by analysing its binding pose (Fig. S2). Axial hydroxyl groups at positions C 2 and C 3 of the myo-inositol moiety of 8:0 PI are involved in hydrogen-bond interactions with Gln725, accounting for specific and unique interactions that would increase ligand affinity. The location of the compound upon binding, as shown in Figure 1, indicates that 8:0 PI inhibits P-gp apical transportation by tightly binding and blocking its TMDs.

Calcein Accumulation Assay (CAA)
After obtaining indications for P-gp modulation activity for 8:0 PI in Caco-2 studies, it was further subjected to CAA, which detects the P-gp-dependent intracellular accumulation of the fluorescent dye calcein. This assay was performed in parallel in a parental wild type (wt) Madin Darby canine kidney (MDCK) II cell line with basal P-gp expression and its corresponding P-gpoverexpressing strain, which is stably transfected with human mdr1. In this test, the degree of P-gp-inhibition directly correlates with the entrapped amount of calcein and, thus, the intensity of intracellular fluorescence. Given that P-gp expression level is the unique difference between both MDCKII strains, different increases in fluorescence intensity in the mdr1 transfected and wt clearly indicate true P-gp modulation [38]. Figure 5 summarizes the results of CAA using HBSS pH 6.5 as assay buffer. The fluorescence intensities of each cell line treated solely with the calcein-derivative were set to 100% ( = control). With the other intensity values referring to this control, an enhancement due to P-gp inhibition became visible. Incubation with the positive control verapamil (0.5 mM) led to an increase in intracellular fluorescence by 1.5-fold in wt cells and 7-fold in mdr1. This difference in effect verified the true impact of Verapamil on the P-gp transporter. The same effect can be observed for 8:0 PI ( = PtdIns 8:0-Na), inducing a 3.8-fold increase in calcein accumulation in the overexpressing strain, but only an 1.8-fold enhancement in wt.
These results indicate that 8:0 PI exerts a reproducible-, cell line-, and substrate-independent P-gp inhibition.

Conclusions
In the present manuscript we have conducted a virtual screening campaign for the identification of novel phospholipids capable of inhibiting cellular membrane transportation by P-gp. Compounds of the previously unstudied phosphatidylinositol class were identified as potential binders by molecular docking analysis ( Table 1). The experimental validation of 4 candidate compounds, namely 18:0/18:1 PA, 14:0 PA, 6:0 PIP 1 , and 8:0 PI, led to the discovery of the latter as an active modulator of P-gp transportation. 8:0 PI exerted a significant transportation inhibition of the substrate digoxin in Caco-2 cells, which was subsequently unequivocally correlated to P-gp modulation in comparative MDCKII CAA studies ( Fig. 4 and 5). Moreover, this compound showed a high tendency to exist as a monomer in solution (Fig. 3b), which has been described as a favoured state for P-gp inhibition [18]. Even though several possible mechanisms have been formulated for the inhibition of this transporter, our hypothesis -based on the assumptions of our theoretical model-is that 8:0 PI modulates transportation either by steric blockade of the substratebinding domain, by allosteric modulation or by competition with other transported substrates (Fig. 1, [34]).
All in all, the identification of compounds that inhibit P-gp transportation is particularly interesting for their implications in combating multi-drug resistance. In fact, the co-administration of P-gp inhibitors along with P-gp substrate drugs might be a promising option in pharmacological formulation development in Figure 5. Calcein Accumulation Assay results. Relative intracellular calcein fluorescence signal in MDCKII mdr1 and wild type (wt) cells (in grey and black, respectively) after pre-incubation with verapamil or 8:0 PI ( = PtdIns 8:0-Na) in HBSS with respect to the control (100% RF; calceinacetoxymethylester in HBSS); n = 6; results are given as mean+standard deviation (*: significance level: 0.05 compared to control; **: significance level: 0.01 compared to control; ***: significance level: 10 23 compared to control). doi:10.1371/journal.pone.0060679.g005 order to overcome drug resistance. Moreover, and due to the important role of phosphatidylinositols and phosphoinositides in many biological processes, such as cell motility and signalling [19][20][21][22], the unexpected discovery of the P-gp inhibitory properties of 8:0 PI may account for a natural mechanism of regulation of cellular membrane function and xenobiotic efflux. Currently, several biological functions of exogenous PI are known, including inhibition of vascular endothelial growth factor (VEGF)-induced angiogenesis, inhibition of low concentrations of amyloid b protein induced degeneration, and promotion of cholesterol transport and excretion eventually inducing anti-obesity response in mice [39][40][41][42][43]. As phospholipids are a common constituent of human dietfor example, phosphoinositols are found in significant amounts in soybean-, it becomes important to further characterise and analyse their role in transmembrane transportation, and study their possible effects in in vivo models for combating MDR.  Table S1 for details). For reference purposes, the binding pose of 8:0 PI is shown in thin purple sticks. Protein surface coloured by electrostatic potential: positive in blue, neutral in white, and negative in red.