Value of quantifying ABC transporters by mass spectrometry and impact on in vitro-to-in vivo prediction of transporter-mediated drug-drug interactions of rivaroxaban

 

Abstract

The pharmacokinetics of numerous therapeutic agents, including the widely utilized anticoagulant rivaroxaban, are profoundly influenced by the activity of ATP-binding cassette (ABC) transporters. Specifically, P-glycoprotein (P-gp), also known as ABCB1, and Breast Cancer Resistance Protein (BCRP), or ABCG2, play critical roles in the absorption, distribution, metabolism, and excretion (ADME) profiles of many drugs. Their involvement in efflux transport can significantly alter the systemic exposure and cellular penetration of substrates like rivaroxaban, thereby predisposing patients to potentially clinically significant drug-drug interactions (DDIs). These interactions can manifest as altered efficacy or increased toxicity of rivaroxaban when co-administered with other medications that modulate the activity or expression of these transporters.

 

Investigations into the “victim” role of rivaroxaban, where its pharmacokinetics are affected by other drugs, and the broader study of transporter-mediated DDIs are routinely conducted using various *in vitro* cellular models. While these models offer a controlled environment for preliminary assessment, a significant challenge in interpreting the results from rivaroxaban efflux transport and DDI studies in cell models lies in the inherent variability and often unquantified abundance of P-gp and BCRP transporters. Discrepancies in transporter expression levels across different cell lines or even within batches of the same cell line can lead to inconsistent and unrepresentative transport data, ultimately hindering accurate prediction of *in vivo* outcomes.

 

Recognizing this critical limitation, the present study was meticulously designed with a primary aim: to develop and validate a highly precise and sensitive quantification method using liquid chromatography-tandem mass spectrometry (LC-MS/MS) capable of assessing the intricate relationship between transporter expression levels and their corresponding functional activities. This comprehensive investigation utilized a selection of well-characterized cellular models, specifically Caco-2ATCC, Caco-2ECACC, MDCK-MDR1, and MDCK-BCRP cell lines, each chosen for its relevance to drug transport studies. The Caco-2 lines mimic intestinal epithelial cells, while the MDCK lines, often genetically engineered, provide high, specific expression of individual transporters, allowing for isolated study of their function.

 

The methodology employed in this study was multi-faceted, ensuring a robust characterization of the transporters. Firstly, the relative and absolute quantities of both P-gp and BCRP transporters within each cell model were precisely determined using the newly developed LC-MS/MS quantification method. This molecular quantification provided a direct measure of the protein levels. Secondly, to further validate and visually confirm the presence and localization of P-gp and BCRP expression, complementary biochemical and imaging techniques were applied, including Western blotting for protein detection and immunofluorescence staining for cellular localization, particularly at the cell membrane. Finally, to assess the functional relevance of the quantified transporters, their intrinsic functional activities were rigorously determined through bidirectional transport experiments. In these assays, the ability of the transporters to efflux a known substrate, rivaroxaban, was measured across the cellular monolayer in both apical-to-basolateral and basolateral-to-apical directions. Furthermore, the half-maximal inhibitory concentrations (IC50s) of two highly specific and well-characterized ABC transporter inhibitors, verapamil for P-gp and ko143 for BCRP, were determined. These IC50 values provide a quantitative measure of the potency with which these inhibitors can block the respective transporter functions.

 

The findings from these rigorous investigations yielded significant insights into the properties of the selected cell models. P-gp and BCRP protein expression was consistently detected at the cell membrane in all relevant models, confirming their appropriate cellular localization for efflux functions. Critically, the expression levels of these transporters were found to be substantially greater in the respective transfected models (MDCK-MDR1 for P-gp and MDCK-BCRP for BCRP), as expected, demonstrating the successful genetic engineering and overexpression. A compelling and pivotal result was the clear correlation observed between the calculated efflux ratios of rivaroxaban and the precisely quantified P-gp and BCRP protein quantities, providing direct evidence that the functional activity of these transporters is indeed proportional to their absolute expression levels. In terms of inhibitory potency, the lowest IC50 values for verapamil were consistently obtained in the MDCK-MDR1 model, indicating its high sensitivity to P-gp inhibition. Similarly, the lowest IC50 values for ko143 were obtained in the MDCK-BCRP model, confirming its potent and selective inhibition of BCRP in this system.

 

In conclusion, this comprehensive study successfully demonstrated that the developed LC-MS/MS method provides a highly accurate and reliable means for quantifying the protein expression levels of P-gp and BCRP efflux transporters in various cellular models. By precisely linking transporter quantity to functional activity, this innovative quantification approach significantly enhances the interpretability of *in vitro* drug transport data. This improved understanding of the relationship between transporter expression and function is crucial for building more robust *in vitro-in vivo* correlations, ultimately leading to more accurate predictions of drug pharmacokinetics and potential drug-drug interactions involving rivaroxaban and other substrate medications in a clinical setting. Such advancements are indispensable for optimizing drug development and ensuring patient safety.

 

Introduction

Rivaroxaban, an oral anticoagulant classified as a direct oral anticoagulant (DOAC), plays a crucial role in preventing and treating various thrombotic conditions. Its pharmacological behavior within the body is significantly modulated by the activity of specific efflux transporters, most notably P-glycoprotein (P-gp), also known as ABCB1, and Breast Cancer Resistance Protein (BCRP), or ABCG2. These transporters are intimately involved in governing the drug’s intestinal absorption, dictating how much of the orally administered dose enters the systemic circulation, and its subsequent elimination from the body via biliary and renal pathways. Consequently, alterations in the activity or expression of these transporters can directly impact the systemic exposure to rivaroxaban.

 

Indeed, clinically significant *in vivo* pharmacokinetic drug-drug interactions (DDIs) have been widely documented when rivaroxaban is co-administered with other medications that potently inhibit cytochrome P450 (CYP) enzymes, particularly CYP3A4, and/or efflux transporters. Notable examples of such strong inhibitors include verapamil, ketoconazole, and ritonavir. These interactions can lead to a marked increase in rivaroxaban exposure, potentially elevating the risk of bleeding complications. ATP-Binding Cassette (ABC) transporters, including P-gp and BCRP, are broadly recognized for their fundamental involvement in drug pharmacokinetics. These integral membrane proteins are strategically positioned on the cell membrane at various crucial physiological barriers throughout the body. Such locations include the intestinal epithelium, which regulates drug absorption; hepatic cells, involved in drug metabolism and biliary excretion; renal tubular cells, responsible for drug elimination in urine; and the placental and blood-brain barriers, which control drug passage to sensitive organs. At these interfaces, ABC transporters function as efflux pumps, actively expelling drugs out of cells. This action effectively limits the intestinal absorption of many drugs, thereby reducing their systemic bioavailability, and simultaneously promotes their elimination from the body via urinary or biliary routes. Given their pervasive influence, the co-administration of drugs that serve as substrates for these transporters with agents that inhibit or induce their activity can precipitate clinically relevant drug-drug interactions, leading to unpredictable and potentially harmful alterations in drug exposure and therapeutic efficacy.

 

To systematically investigate the intricate involvement of P-gp and BCRP transporters in drug pharmacokinetics and to predict potential DDIs, the US Food and Drug Administration (FDA) strongly recommends the utilization of diverse *in vitro* experimental systems, prominently including established cell models. Among these, the Caco-2 cell model has gained widespread acceptance and is extensively employed as a high-fidelity *in vitro* mimic of the human intestinal mucosa. The strong correlation between the permeation of drugs across Caco-2 cell monolayers and their actual human intestinal permeation makes this cell line an invaluable reference standard for studies focused on drug intestinal absorption. Furthermore, the Madin-Darby canine kidney (MDCK) cell line has also been widely reported in drug absorption studies. MDCK cells are increasingly favored due to their more rapid growth rate compared to Caco-2 cells, and crucially, they have been shown to share many common epithelial cell characteristics with Caco-2 cells, validating their utility in such studies. Advancements in cell engineering have led to the generation of modified MDCK cell lines, such as MDCK-MDR1 and MDCK-BCRP, which are created by transfecting the human ABCB1 (encoding P-gp) or ABCG2 (encoding BCRP) gene into parental MDCK cells. These transfected cell models are particularly powerful because they specifically overexpress a single transporter of interest. This targeted overexpression facilitates the precise design of studies aimed at elucidating the isolated involvement of individual transporters in drug disposition pathways, enabling clearer mechanistic insights.

 

However, it is important to acknowledge that despite their utility, the Caco-2 and MDCK cell models exhibit inherent differences in their barrier properties. This is quantitatively illustrated by their transepithelial electrical resistance (TEER) values: Caco-2 cells typically form tighter monolayers with TEER values of approximately 600 Ω·cm^2, whereas MDCK cells form comparatively leakier monolayers with TEER values around 100 Ω·cm^2. Moreover, these models possess distinct characteristics concerning their endogenous protein expression profiles. Transfected MDCK cells, by design, significantly overexpress the target transporters compared to their native counterparts and even compared to Caco-2 cells. Furthermore, the specific expression levels and subcellular localization of endogenous and transfected transporters can vary depending on the specific source of the cell lines used. For instance, Caco-2 cells can be acquired from either the American Type Culture Collection (ATCC) or the European Collection of Authenticated Cell Cultures (ECACC), and it has been well-documented that the mRNA expression levels of various proteins, including ABC and SLC transporters, can differ considerably between these two sources.

 

While *in vitro-in vivo* correlations (IVIVCs) have been successfully established for absorption studies utilizing Caco-2 and MDCK cells, particularly for drugs primarily undergoing passive absorption, a robust and widely accepted IVIVC for accurately predicting *in vivo* transporter-mediated DDIs from *in vitro* model results remains elusive. This gap highlights the ongoing need for further in-depth investigative analyses concerning the experimental protocols and analytical methods employed in *in vitro* assays to establish more reliable IVIVCs for transporter-mediated DDIs. According to current FDA guidelines, specific ratios, such as [I]/IC50 > 0.1 or [I]^2/IC50 > 10 (where [I] represents the clinical unbound intestinal or hepatic concentration of the inhibitor and IC50 is derived from *in vitro* studies), can be used as a preliminary metric for assessing the potential for DDIs. However, a critical review of published IC50 values for several known P-gp inhibitors reveals considerable interlaboratory variations for the same substrate. Such variability significantly limits the clear establishment of consistent and predictive IVIVCs. This observed variability could be partially attributed to numerous factors, including subtle differences in cell culture conditions or varying passage numbers of the cell lines (with higher passage numbers often correlating with increased transporter expression). Nevertheless, it is increasingly understood that this variability is more likely due to the inherent differences in the specific test systems employed across various laboratories.

 

Consequently, obtaining precise quantitative information regarding the expression levels of ABC transporters in these diverse cell models would be immensely valuable. Such accurate quantification would allow for a more correct and nuanced interpretation of the experimental data generated in drug permeability and drug-drug interaction studies, enabling researchers to better account for inter-model variations. Several analytical techniques are available for evaluating protein expression at the cellular level. In recent years, liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) has gained significant prominence and is increasingly employed for the direct evaluation and accurate quantification of membrane transporters in a wide array of biological samples. This advanced technology is becoming more accessible to research laboratories and offers the distinct advantage of enabling the routine and high-throughput quantification of numerous proteins with unparalleled sensitivity and specificity.

 

Therefore, the primary aim of the present study was to comprehensively assess the influence of P-gp and BCRP expression levels on the predictive capacity of *in vitro* cell models regarding rivaroxaban pharmacokinetics and transporter-mediated drug-drug interactions. This was achieved through the innovative development and meticulous validation of a highly reliable and accurate quantification method utilizing LC-MS/MS, all performed in strict accordance with the prevailing FDA guidelines for bioanalytical method validation. This approach sought to provide a direct link between the quantitative presence of transporters and their functional impact, thereby improving the predictive power of *in vitro* models for *in vivo* DDI outcomes.

 

Materials and methods

Chemicals and reagents

All chemical substances and reagents used throughout this study were meticulously sourced to ensure high quality and purity. Specifically, methanol, purified water, acetonitrile (ACN), trifluoroacetic acid (TFA), and formic acid (FA) were acquired from Carlo Erba Reagents, based in Milan, Italy. It is important to note that all these solvents were of LC-MS grade, ensuring minimal impurities that could interfere with sensitive mass spectrometry analyses. Rivaroxaban, the primary drug of interest, and its isotopically labeled internal standard, [13C6]-rivaroxaban, were obtained from Alsachim, located in Illkirch, France. For protein-related assays and quantification, bovine serum albumin (BSA), a common protein standard, AQUA peptides (Absolute Quantification peptides, isotopically labeled synthetic peptides used for absolute protein quantification), a Micro BCA assay kit for protein concentration determination, a Mem-PER Plus membrane protein extraction kit for isolating membrane proteins, a Mass Spec sample prep kit (which included Lys-C enzyme), and Pierce Peptide desalting spin columns were all purchased from ThermoFisher Scientific, based in Massachusetts, USA. Crude heavy-labeled standard peptides, essential for absolute protein quantification by LC-MS/MS, were specifically obtained from JPT Peptides Technologies in Berlin, Germany. Various cell culture components and standard laboratory reagents, including Dulbecco’s modified Eagle’s medium (DMEM), Hank’s balanced salt solution (HBSS), Dulbecco’s phosphate buffered saline without sodium and magnesium (DPBS), HEPES buffer, heat-inactivated fetal bovine serum (FBS), trypsin, nonessential amino acids, penicillin, amphotericin B, streptomycin, and dimethyl sulfoxide (DMSO), along with the specific transporter inhibitors verapamil, ko143, and cyclosporine A, were all procured from Sigma, Saint-Quentin-Fallavier, France. Eagle’s Minimum Essential Medium (EMEM) was obtained from the American Type Culture Collection (ATCC) in Rockville, Maryland, USA.

 

Cell culture

The cell lines critical to this study were carefully maintained and prepared. MDCK II, MDCK-MDR1, and MDCK-BCRP cells were generously provided as gifts by Professor Piet Borst from The Netherlands Cancer Institute, Amsterdam, The Netherlands, recognizing their high quality and specific characteristics. Caco-2ATCC cells (designated HTB-37) were directly purchased from the ATCC in Rockville, Maryland, USA, while Caco-2ECACC cells (designated ECACC 86010202) were acquired from the European Collection of Authenticated Cell Cultures in Salisbury, UK. All cell lines were maintained under strict aseptic conditions in their respective optimal culture media: EMEM was used for Caco-2ECACC cells, and DMEM for all other cell lines. Both media were enriched with 10% heat-inactivated FBS, 1% nonessential amino acids, 100 U/mL penicillin G, 100 µg/mL streptomycin, and 250 ng/mL amphotericin B to support robust cell growth and prevent contamination. The cells were incubated in a controlled environment at 37 degrees Celsius, with 95% relative humidity and a 5% CO2 atmosphere, conditions conducive to optimal mammalian cell proliferation.

 

For the subsequent drug transport assays, cells were meticulously seeded onto insert filters (Falcon®, composed of a translucent PET membrane, with a surface area of 0.3 cm^2 and a 0.4 µm pore size, designed for high-density cell growth). These inserts were placed within 24-well companion plates, obtained from Dominique Dutscher, Strasbourg, France. MDCK II, MDCK-MDR1, and MDCK-BCRP cells were seeded at a density ranging from 1 × 10^5 to 1.5 × 10^5 cells per insert and were cultured for a period of 5 days, with the culture medium being replenished daily to ensure continuous nutrient supply. Caco-2 cells, due to their slower differentiation and monolayer formation, were seeded at a lower density of 1 × 10^4 to 1.5 × 10^4 cells per insert and cultured for a longer duration of 21 days, with medium changes performed every 2–3 days.

 

To confirm the confluence and integrity of the cell monolayers, transepithelial electrical resistance (TEER) was precisely measured in each well both before and after the transport assay using an EVOM resistance meter (World Precision Instruments, Sarasota, USA). This measurement serves as a critical indicator of tight junction formation and monolayer barrier function. The permeability assays were only conducted on inserts that exhibited TEER values above 150 Ω·cm^2 for Caco-2 cells, signifying a well-formed tight monolayer, and above 40 Ω·cm^2 for MDCK II, MDCK-MDR1, and MDCK-BCRP cells, which typically form less resistive monolayers but still adequate for transport studies.

 

Expression and quantification of ABC transporters

Western blotting

To assess the protein expression levels of ABC transporters, confluent cell monolayers were meticulously collected and subsequently lysed in freshly prepared RIPA lysis buffer, a standard solution for extracting cellular proteins. The total protein content within these cell lysates was accurately quantified using the BCA assay kit, ensuring consistent loading across samples. Aliquots containing 50 µg of total protein from each sample were then size-fractionated using a 4–12% Tris glycine acrylamide gel prepared in MOPS buffer, a process known as SDS-PAGE. Following electrophoretic separation, the proteins were efficiently transferred from the gel onto 0.45 µm nitrocellulose membranes, creating a stable platform for antibody detection. These membranes were then subjected to a rigorous immunodetection protocol. They were first incubated overnight at 4 degrees Celsius with primary antibodies diluted in a blocking buffer: 1/100 dilution for P-gp (monoclonal antibody sc-55510; Santa Cruz Biotechnology), 1/250 dilution for BCRP (monoclonal antibody BXP-21; Santa Cruz Biotechnology), and 1/400 dilution for β-actin (monoclonal antibody sc-47778; Santa Cruz Biotechnology), which served as a loading control. After primary antibody incubation, the membranes were thoroughly washed and then probed for 1.5 hours with horseradish peroxidase (HRP)-conjugated secondary antibodies (Santa Cruz Biotechnology) diluted 1/5000 in the blocking buffer. The detection of protein bands was achieved using the electrochemiluminescence (ECL) method, following the manufacturer’s specific protocol, which produces a chemiluminescent signal proportional to the amount of target protein present.

 

Immunofluorescence

Immunofluorescence studies were performed to visually confirm the subcellular localization of the tight junction protein zonula occludens 1 (ZO-1), as well as the P-gp and BCRP transporters, across all utilized cell models. In brief, cells grown on inserts were fixed using either 3.7% paraformaldehyde (PAF) or methanol for 30 minutes, a step crucial for preserving cellular structures. Following fixation, cells were permeabilized with 0.2% TritonX-100, allowing antibodies to access intracellular targets. Non-specific binding sites were then meticulously blocked by incubating the cells in a complete culture medium containing 10% FBS for 30 minutes, minimizing background signal. The insert membranes, now prepared for staining, were carefully excised and incubated with the primary antibodies (sc-55510 for P-gp and BXP-21 for BCRP) at a dilution of 1/100 in the blocking buffer for 1 hour. After appropriate washes, the secondary antibodies, diluted 1/200 in the blocking buffer, were applied and incubated for 1 hour. To visualize cell nuclei and provide a counterstain for context, DAPI was used at a dilution of 1/1000 in the blocking buffer. All stained biomarkers were imaged using an inverted epifluorescence microscope, specifically an Olympus IX81 (Olympus, Tokyo, Japan), which was equipped with CellSens imaging software (Olympus, Munster, Germany), enabling high-resolution image acquisition and analysis of protein localization.

 

Quantification by liquid chromatography – High resolution mass spectrometry

Sample preparation

The preparation of samples for mass spectrometry-based quantification was executed according to a previously established and validated protocol. Briefly, the crucial membrane protein fraction was meticulously extracted from the cells using the Mem-PER Plus membrane protein extraction kit, adhering strictly to the manufacturer’s detailed instructions to ensure efficient and clean isolation of membrane-associated proteins. Following extraction, the protein concentration of the membrane fraction was accurately measured using the Micro BCA protein assay kit, which is essential for normalizing samples and ensuring consistent input for subsequent digestion. For enzymatic digestion, an in-solution tryptic digestion was performed utilizing the Pierce Mass Spec sample prep kit. Specifically, 100 µg of the isolated membrane protein fraction was first subjected to reduction with 10 mM dithiothreitol (DTT) for 45 minutes at 50 degrees Celsius, which breaks disulfide bonds. This was followed by alkylation with 50 mM iodoacetamide (IAA) for 20 minutes at room temperature, with protection from light, to prevent disulfide bond reformation. The reduced and alkylated proteins were then sequentially digested: first with Lys-C (at a 1:100 enzyme-to-protein ratio) for 2 hours at 37 degrees Celsius, and then with trypsin (at a 1:50 enzyme-to-protein ratio) for an extended period of 16 hours at 37 degrees Celsius, strictly following the manufacturer’s guidelines. Each digested sample was then desalted using the Pierce Peptide desalting spin column, following its manufacturer’s protocol, to remove salts and other impurities that could interfere with mass spectrometry analysis, thus ensuring high-quality peptide samples.

 

Relative quantification

Relative quantification of proteins was performed as previously described in reference [27], employing a sophisticated liquid chromatography-mass spectrometry workflow. The prepared peptide samples were analyzed using an Ultimate U3000 liquid chromatography system (Dionex, California, USA) coupled to a mass spectrometer. Peptide separation was achieved on an Easy-spray C18 column (2 µm fully porous particles, 0.075 mm ID × 500 mm) from ThermoFisher Scientific, USA, using a 150-minute gradient elution at a precise flow rate of 300 nL/min. The column temperature was maintained at 35 degrees Celsius, and a sample volume of 1 µL was injected. Mobile phase A consisted of water with 0.1% formic acid, while mobile phase B was an 80% acetonitrile solution containing 0.08% formic acid. Peptides were eluted by a gradient starting from 2.5% to 40% of mobile phase B over 123 minutes, followed by a brief wash at 90% mobile phase B before returning to the initial starting conditions. Following chromatographic separation, peptides were introduced into an Orbitrap mass spectrometer (D30 Orbitrap, with a 5 kV central electrode voltage). A precursor scan of intact peptides was recorded by scanning from m/z 375–1500 with a high resolution of 70,000. Subsequently, the 10 most intense multi-charged precursors detected in the full scan were selected for higher-energy collision dissociation (HCD) analysis in the C-trap, generating fragment ion spectra for peptide identification and quantification. To prevent the repetitive selection of the same peptides for MS/MS analysis, the acquisition program employed a 30-second dynamic exclusion window.

 

Absolute quantification

Absolute quantification was performed using a precise liquid chromatography system coupled to a high-resolution mass spectrometer. Chromatography was conducted using an Ultimate U3000 liquid chromatography system (Dionex, California, USA) operating at a flow rate of 300 nL/min. Peptides were initially loaded onto an Acclaim PepMap 100 C18 trap column (100 µm × 2 cm, C18, 5 µm, 100Å) at a flow rate of 20 µL/min, using water containing 2% acetonitrile and 0.1% trifluoroacetic acid (TFA). Following this trapping step, the peptides were separated on an Easy-spray C18 column (2 µm fully porous particles, 0.075 mm ID × 150 mm) from ThermoFisher Scientific, Massachusetts, USA, utilizing a 45-minute gradient. Mobile phase A consisted of water with 0.1% formic acid (FA), and mobile phase B was an 80% acetonitrile solution containing 0.08% FA. The elution gradient was precisely defined as follows: from 0 to 30 minutes, a linear gradient from 5% to 45% B; from 30 to 31 minutes, a linear gradient from 45% to 90% B; from 31 to 38 minutes, a sustained 90% B; and at 38.5 minutes, a rapid return to the initial conditions, maintained until 45 minutes.

 

Peptides were quantified using parallel reaction monitoring (PRM)-based targeted mass spectrometry in electrospray positive mode, a highly selective and sensitive method for absolute quantification of specific peptides. For each target peptide, the three most intense fragment ions were selected and monitored. The mass spectrometry acquisition method combined a full-scan survey method with a time-scheduled multiplexed PRM method. The full-scan method was performed with a resolution of 17,500, an AGC (Automatic Gain Control) target value of 3 × 10^6, a maximum injection time of 200 ms, and a scan range from 400 to 1200 m/z. The PRM method parameters were set as follows: a resolution of 35,000 (at m/z 200), an AGC target value of 2 × 10^5, a maximum injection time of 200 ms, and a narrow isolation window of 0.7 m/z to ensure high specificity. HCD (Higher-energy C-trap Dissociation) fragmentation was performed with a normalized collision energy of 27.

 

The entire quantitative method was rigorously validated in strict accordance with the FDA recommendations for bioanalytical method validation, ensuring its reliability and robustness. For each target protein (P-gp and BCRP), three proteotypic peptides were carefully selected. Among these, the two most intense peptides were designated as quantification peptides, while the third served as a confirmation peptide to enhance confidence in the identification. For P-gp, the quantification peptides chosen were FDTLVGER (a peptide shared between canine and human P-gp, allowing for quantification across both species) and FYDPLAGK (a peptide specific to human P-gp, enabling species-specific quantification). For BCRP, the quantification peptides were SSLLDVLAAR (shared by canine and human BCRP) and VIQELGLDK (specific to human BCRP). For each of these selected peptides, the three most intense fragment ions were systematically monitored and utilized for quantification. Comprehensive details on the calibration standards and quality control (QC) levels prepared for each peptide, used to establish and validate the assay’s performance, are summarized in Table 1. Additional exhaustive information pertaining to the method validation, including parameters such as linearity, accuracy, and precision, is available in the Supplementary Information.

 

Bidirectional transport assay

Four distinct cell models were employed to comprehensively investigate the bidirectional permeabilities of rivaroxaban. Rivaroxaban was utilized at a concentration of 10 µM, a concentration carefully chosen because it approximates the relevant plasma concentrations observed *in vivo* and has been consistently used in prior transport studies conducted on these same cell models. The transport studies were performed using cell culture inserts, which fundamentally consist of two separate compartments – an apical compartment and a basolateral compartment – precisely separated by a confluent cell monolayer cultured on a porous filter membrane, as meticulously described in previous literature. The assays were conducted under controlled physiological conditions, specifically in HBSS transport buffer supplemented with 10 mM HEPES and 1% DMSO, maintaining a pH of 7.4.

 

Apparent permeabilities (Papp) in both directions (apical to basolateral, A→B, and basolateral to apical, B→A) were rigorously computed using the following standard equation: Papp = (V / (C0 * S)) * (C2 / t), where V represents the volume of the receiver compartment, [C0] is the initial concentration of the compound in the donor compartment, S denotes the surface area of the cell monolayer, [C2] is the final concentration of the compound quantified in the receiver compartment, and t is the total incubation time. To quantitatively assess the functional activity of efflux transporters on drug kinetics across the cell monolayer, efflux ratios (ER) were subsequently computed using the formula: ER = Papp (B→A) / Papp (A→B). The standard deviation of the efflux ratio was calculated according to a formula specifically proposed by Gnoth et al. (2010), ensuring statistical rigor. In accordance with current FDA guidelines, an efflux ratio exceeding a value of 2 is typically considered indicative that the drug under investigation is a substrate for an efflux transporter. All available samples collected from the receiver compartments were precisely analyzed using the established LC-MS/MS method, as detailed in previous sections.

 

Determination of half-maximal inhibitory concentration

The half-maximal inhibitory concentration (IC50) is a crucial pharmacological parameter used as a quantitative measure of an inhibitory drug’s potency. In this study, the IC50 values for the specific P-gp inhibitor verapamil and the BCRP inhibitor ko143 were determined over a wide range of concentrations. For verapamil, concentrations ranged from 0.1 to 1000 µM, while for ko143, concentrations ranged from 0.001 to 25 µM, ensuring full coverage of their dose-response curves. The concentration of rivaroxaban, the transport substrate, was consistently maintained at 10 µM throughout these inhibition experiments. To accurately calculate the IC50 values, the experimental data were fitted using unweighted, nonlinear, least-squares regression modeling of the Efflux Ratio. This curve fitting was performed using the `nls()` function within the R statistical software environment. The specific mathematical model employed for fitting was a sigmoid model, represented by the equation: ER = I0 + ((Imax – I0) / (1 + ([I] / IC50)^h)), where ER represents the observed efflux ratio at a given inhibitor concentration, [I] denotes the concentration of the inhibitor, h is the Hill coefficient (reflecting the steepness of the dose-response curve), I0 represents the initial efflux effect observed in the absence of the inhibitor, and Imax represents the maximal effect observed when the inhibitor is present at saturating concentrations.

 

Data analysis

The acquisition and subsequent analysis of all experimental data were meticulously performed using specialized software packages to ensure precision and integrity. Data acquisition and the preliminary quantitative analysis were conducted using TraceFinder® Clinical software version 4.1 (ThermoFisher Scientific), designed for accurate and high-throughput quantification of analytes. For the more complex relative quantification of proteins, ProteomeDiscoverer® software version 2.2 (ThermoFisher Scientific) was employed, providing sophisticated tools for proteomic data processing. All advanced statistical analyses and the generation of high-quality graphic outputs were performed using the R statistical software (R Foundation for Statistical Computing, Vienna, Austria, accessible at https://www.R-project.org/), leveraging its extensive capabilities for complex statistical modeling and data visualization.

 

Results

Expression of junction proteins and phenotypic markers

The integrity and barrier properties of the cultured cell monolayers are paramount for accurate transport studies. To confirm the tightness of these monolayers, the expression and localization of zonula occludens 1 (ZO-1), a crucial tight junction protein, were investigated using immunofluorescence staining. As visually depicted in Fig. 1, a bright and uniformly distributed signal for ZO-1 was consistently localized at the cell membranes, specifically at the junctions between adjacent cells, across all the cell models investigated. This observation confirmed the successful formation of intact and tight cellular barriers essential for accurate permeability measurements. Complementing the immunofluorescence, protein mass spectrometry analysis was employed to assess the relative expression levels of six key junction proteins in the Caco-2ATCC, Caco-2ECACC, MDCK II, MDCK-MDR1, and MDCK-BCRP cell lines (Fig. 2). This analysis revealed distinct expression patterns consistent with their known barrier properties. Occludin, another vital component of tight junctions, showed a remarkably higher abundance, approximately 10 to 100-fold greater, in the Caco-2 models compared to the MDCK cells. This difference correlates directly with the significantly higher transepithelial electrical resistance (TEER) values typically observed in Caco-2 monolayers, indicative of tighter junctions. Conversely, the abundance of ZO-1 and ZO-2 proteins was found to be higher in the MDCK models than in Caco-2 cells, while ZO-3 exhibited a closely similar abundance across all models. Among the MDCK models themselves, the highest abundance for occludin was specifically observed in MDCK-BCRP cells. Additionally, cadherin-1 and cadherin-17 displayed a 5–10-fold greater abundance in Caco-2 cells compared to MDCK II and MDCK-BCRP cells, with a slightly lower expression observed in MDCK-MDR1 cells.

 

To further validate the cellular identity and differentiation of each model, specific phenotypic markers were employed. Dipeptidyl peptidase-4 (DPP4) was selected as a characteristic intestinal marker for Caco-2 cells, reflecting their enterocyte-like differentiation. For MDCK cells, V-type proton ATPase was chosen as a specific renal marker, consistent with their renal tubule epithelial origin. The quantitative analysis showed that the abundance of dipeptidyl peptidase-4 was approximately 10-fold greater in the Caco-2 models than in the MDCK models, while V-type proton ATPase exhibited a notably higher abundance in MDCK cells (Fig. 3). These positive control findings unequivocally confirmed the relevance and accuracy of the label-free proteomic analysis used for relative quantification of protein expression, providing robust evidence for the distinct epithelial characteristics of each cell line.

 

Expression and quantification of P-gp and BCRP efflux transporters

The core of this study involved a thorough investigation into the expression and precise quantification of the ABC efflux transporters, P-gp and BCRP. Initially, the expression of these transporters was qualitatively assessed by mass spectrometry. For a more precise and quantitative evaluation, absolute quantification of P-gp and BCRP protein amounts in all five cell models was performed using the developed LC-MS/MS method, utilizing two specific quantification peptides for each protein. Prior to the quantitative measurements, the method underwent rigorous validation, demonstrating excellent linearity, accuracy, and precision through the analysis of calibration curves for each of the four chosen peptides (Fig. S1). All calibration curves consistently yielded R^2 values above 0.98, indicative of strong linearity. The FYDPLAGK and VIQELGLDK peptides demonstrated the lowest lower limit of quantification (LLOQ) at 74.25 pM, enabling highly sensitive detection, whereas SSLLDVLAAR had the highest LLOQ at 247.5 pM. The LLOQ value for FDTLVGER was determined to be 148.5 pM. Crucially, the precision of the method was consistently better than 15%, and the accuracy was always within ±15% of the nominal concentration across all quality control (QC) samples (Table S1), confirming the method’s reliability.

 

As anticipated and serving as an internal control for the transfected models, the two peptides specific to human P-gp (FYDPLAGK) and human BCRP (VIQELGLDK) were not detected in the parental MDCK II cells or in the MDCK-BCRP (for human P-gp peptide) or MDCK-MDR1 (for human BCRP peptide) models, confirming the absence of endogenous human transporters. However, these specific human peptides were readily observed in the respective transfected models, MDCK-MDR1 and MDCK-BCRP, validating the successful and specific expression of the human transgenes (Table 2). Quantitative analysis revealed that transfected MDCK-MDR1 cells exhibited a P-gp expression level approximately 5-fold higher than that observed in the Caco-2 models, underscoring their utility for P-gp-specific studies. For BCRP, its expression level in MDCK II and MDCK-MDR1 cells was below the LLOQ, indicating very low or undetectable endogenous expression. Both Caco-2 models showed a BCRP protein amount approximately 2-fold lower than that determined in the transfected MDCK-BCRP cells. Interestingly, Caco-2ECACC cells exhibited a 1.5-fold greater amount of P-gp protein compared to Caco-2ATCC cells, suggesting a significant difference in endogenous P-gp expression between these two commonly used Caco-2 sources, while the amount of BCRP was remarkably similar in both Caco-2 models.

 

These quantitative absolute quantification results were further corroborated by LC-MS/MS relative quantification analyses. Proteomic analyses successfully detected P-gp, BCRP, MRP1, MRP2, MRP3, and MRP4, albeit with varying abundance levels across the different cell models (Fig. 4). P-gp exhibited a strikingly higher abundance, approximately 10-fold greater, in MDCK-MDR1 cells compared to Caco-2 cells, while MDCK II and MDCK-BCRP cells showed a comparatively low abundance of P-gp. The difference in P-gp abundance between the two Caco-2 models was also confirmed, with a greater abundance in Caco-2ECACC cells, consistent with absolute quantification. BCRP showed a closely similar abundance in Caco-2 and MDCK II cells. The highest BCRP abundance was observed in MDCK-BCRP cells, validating its overexpression, while the lowest abundance was detected in MDCK-MDR1 cells, where BCRP is not specifically overexpressed. Regarding other MRP transporters, the lowest abundance for MRP1 was observed in MDCK II cells. MRP1 exhibited a similar abundance across Caco-2ECACC, MDCK-MDR1, and MDCK-BCRP models, but was detected to a slightly greater extent in Caco-2ATCC cells. MRP2 showed the greatest abundance in the two Caco-2 models, being more than 10-fold higher than in the MDCK models, and Caco-2ECACC cells specifically displayed a greater abundance of MRP2 than Caco-2ATCC cells. Conversely, the abundance of MRP3 and MRP4 was approximately 10-fold higher in MDCK models than in Caco-2 models, indicating distinct endogenous transporter profiles.

 

To provide additional orthogonal confirmation of P-gp and BCRP expression, Western blotting was performed (Fig. 5). For P-gp expression, the Western blot data revealed a more intense signal in MDCK-MDR1 cells compared to Caco-2 cells, migrating at an approximate molecular weight of 170 kDa (Fig. 5A), consistent with the expected size of P-gp. No detectable bands were observed in the MDCK-BCRP and MDCK II models, confirming the absence of significant endogenous P-gp or non-specific detection. Within the Caco-2 cells, the band detected in Caco-2ECACC cells was slightly more intense than that observed in Caco-2ATCC cells, mirroring the quantitative LC-MS/MS findings regarding differential P-gp expression between these two Caco-2 sources. BCRP expression was successfully observed in MDCK-BCRP cells and both Caco-2 cell lines (Fig. 5B). The detected bands migrated at a high molecular weight, strongly suggesting that the assay, which was not conducted under reducing conditions, detected the BCRP dimer, a known oligomeric state of the protein. The most intense BCRP signal was consistently detected in the MDCK-BCRP cells, as anticipated due to genetic overexpression, while the signal intensity was found to be similar in both Caco-2ATCC and Caco-2ECACC cells, consistent with LC-MS/MS data for BCRP.

 

Finally, immunofluorescence staining was employed to visually confirm the subcellular localization of P-gp and BCRP, particularly at the cell membrane, which is crucial for their efflux function. P-gp was clearly localized at the cell membrane in both Caco-2 models and in the MDCK-MDR1 model (Fig. 6). In MDCK-MDR1 cells (Fig. 6B), P-gp exhibited strong and homogeneous staining across the entire cell surface, characterized by bright, multiple fluorescent spots densely covering the entire apical membrane of most cells, indicative of high-level overexpression. Caco-2 cells, in contrast, showed a more heterogeneous P-gp signal that was also localized at the cell membrane, predominantly at the junctions between cells, and an apical expression characterized by bright, multiple fluorescent spots observed over some, but not all, cells (Fig. 6D and E), suggesting more heterogeneous endogenous expression. As expected, no specific P-gp signal was observed at the cell membrane of MDCK II and MDCK-BCRP cells (Fig. 6A and C), confirming their lack of specific P-gp overexpression. Similarly, BCRP was consistently localized at the cell membrane across all the cell models where it was expressed (Fig. 7). BCRP exhibited a homogeneous expression pattern characterized by thin and continuous apicolateral membrane staining. A significantly greater immunofluorescence signal for BCRP was consistently observed in the MDCK-BCRP model compared to the other models, further validating the successful and robust overexpression of this transporter.

 

Functionality of P-gp and BCRP efflux transporters

To directly evaluate the functional activity of P-gp and BCRP efflux transporters, comprehensive bidirectional transport assays were meticulously performed. Rivaroxaban, serving as the substrate, was consistently incubated at a fixed concentration of 10 µM, either on the apical or basolateral side of the cell monolayer. These assays were conducted in the presence of increasing concentrations of two highly specific inhibitors: verapamil, known for its potent inhibition of P-gp, and ko143, a highly selective inhibitor of BCRP. The detailed permeability values of rivaroxaban across all the cell models used are thoroughly summarized in Table S2.

 

The baseline efflux ratios of rivaroxaban, determined in the absence of any inhibitor, provided crucial insights into the intrinsic efflux capacity of each cell model. These values were measured at 3.13 for Caco-2ATCC, 5.75 for Caco-2ECACC, 11.5 for MDCK-MDR1, and 7.22 for MDCK-BCRP cells (as graphically represented in Fig. 8 and Fig. 9). The consistent observation of efflux ratios greater than 2 across all models indicates that rivaroxaban is indeed an active substrate for efflux transporters in these systems. Further experiments demonstrated that the bidirectional permeability values of rivaroxaban were significantly and dose-dependently modified as the concentrations of verapamil or ko143 were increased across all models. Specifically, an expected pattern emerged: the apparent apical-to-basolateral permeabilities began to increase, signifying enhanced drug absorption into the basolateral compartment, while concurrently, the apparent basolateral-to-apical permeabilities began to decrease, indicating reduced efflux. This dual effect ultimately resulted in a marked reduction in the calculated efflux ratios, confirming the successful inhibition of transporter activity (as clearly depicted in Fig. 8 and Fig. 9).

 

To quantitatively characterize the inhibitory potency of verapamil and ko143, modeling of the efflux ratios was performed to determine their respective half-maximal inhibitory concentrations (IC50 values) for P-gp and BCRP-mediated transport of rivaroxaban across the four distinct cell models. For verapamil, the most potent inhibition, indicated by the lowest IC50, was notably achieved in the MDCK-MDR1 model, with an IC50 of 6.94 µM. This contrasts sharply with the higher IC50 values observed in other models: 21.2 µM for Caco-2ATCC, 24.8 µM for Caco-2ECACC, and a substantially higher 137.8 µM for MDCK-BCRP cells. Interestingly, despite the Caco-2ECACC cells exhibiting a higher baseline efflux ratio without inhibitor compared to Caco-2ATCC cells, the IC50 values for verapamil in the two Caco-2 models remained quite similar. At the highest concentrations of verapamil tested, complete inhibition of efflux was observed across all models, demonstrating the full extent of P-gp blockade.

 

Regarding the ko143 inhibitor, complete inhibition of efflux was achieved at the highest concentrations in all models except for the Caco-2ECACC model (as shown in Fig. 9), which displayed incomplete inhibition, possibly due to higher endogenous BCRP expression or the presence of other efflux mechanisms. The lowest IC50 for ko143 was strikingly observed in the MDCK-BCRP cells, at a remarkably potent 0.026 µM. This is considerably lower than the IC50 values observed in Caco-2ATCC (0.52 µM), Caco-2ECACC (0.18 µM), and MDCK-MDR1 (7.44 µM) cells. Overall, the IC50 values for ko143 were substantially lower than those for verapamil, providing clear evidence for the more potent inhibitory effect of ko143 on both P-gp and BCRP transporters, especially BCRP. These functional data unequivocally demonstrate the distinct and quantifiable differences in transporter activity and sensitivity to inhibitors across the various cell models employed.

 

Discussion

The intricate involvement of ABC transporters, particularly P-glycoprotein (P-gp) and Breast Cancer Resistance Protein (BCRP), in the pharmacokinetics of rivaroxaban is a well-established scientific fact. These transporters significantly influence the disposition of rivaroxaban in the body. Consequently, the co-administration of rivaroxaban with other medications that modulate the activity of these transporters can substantially alter the systemic exposure of rivaroxaban. Specifically, inhibition of these transporters can lead to a decrease in the drug’s excretion, potentially resulting in its bioaccumulation within the body and, consequently, a heightened risk of adverse events, most notably bleeding complications.

 

To thoroughly investigate the role of P-gp and BCRP transporters in drug pharmacokinetics, various *in vitro* systems, including the widely utilized Caco-2 and MDCK cell models, have been extensively employed to assess the disposition of rivaroxaban. However, it is well-recognized that these cell models possess distinct characteristics, and as such, the interpretation of drug transport study results can vary considerably depending on the specific cell type chosen. Our present study deliberately focused on precisely quantifying the impact of ABC transporter expression levels on the assessment of *in vitro* rivaroxaban drug-drug interactions (DDIs) specifically mediated by P-gp and BCRP. We utilized the cell models recommended by the FDA for drug transport assays, including Caco-2ECACC, Caco-2ATCC, MDCK II, MDCK-MDR1, and MDCK-BCRP, to provide a comprehensive comparison. The pioneering aspect and originality of this work lie in its systematic integration of complementary analytical techniques. We combined advanced proteomic mass spectrometry for both absolute and relative quantification of transporters, alongside classical methods such as Western blotting, immunofluorescence staining, and functional bidirectional transport experiments. This multi-pronged approach aimed to unequivocally illustrate the critical value of accurately quantifying ABC transporters to comprehensively explain the observed variability in *in vitro* study results. While previous studies have indeed evaluated transporter quantities in different cell models using mass spectrometry, our research uniquely investigated the direct correlation between the variability in transporter expression levels and the corresponding variability in their functional activity across these diverse cell models, a crucial yet previously understudied relationship.

 

Given that rivaroxaban is known to be a substrate for both P-gp and BCRP, our study determined the half-maximal inhibitory concentrations (IC50s) of verapamil (a P-gp inhibitor) and ko143 (a BCRP inhibitor) on four commonly used cell models for bidirectional transport studies. The baseline efflux ratios for 10 µM rivaroxaban in MDCK-MDR1 and MDCK-BCRP cells were notably higher than those determined in Caco-2 cells (11.5 and 7.22 versus 3.13 and 5.75, respectively), clearly reflecting the engineered overexpression of the respective transporters in the MDCK transfected lines. When evaluating verapamil’s inhibitory potency, its IC50 was found to be approximately 20-fold lower in MDCK-MDR1 cells than in MDCK-BCRP cells (6.94 µM versus 137.8 µM, respectively). This marked difference strongly suggests that verapamil predominantly inhibits P-gp at concentrations below 100 µM, with a significantly weaker effect on BCRP. The IC50 values for verapamil in Caco-2 models were slightly higher than those determined in MDCK-MDR1 cells (21.2 µM for Caco-2ATCC and 24.8 µM for Caco-2ECACC). This slight difference can be partly attributed to the fact that Caco-2 cells endogenously express other ABC transporters, such as BCRP, that are also involved in rivaroxaban disposition. These other transporters can potentially compensate for P-gp inhibition, leading to a higher apparent IC50 for P-gp selective inhibitors in these more complex cellular systems.

 

Regarding the ko143 inhibitor, its IC50 was dramatically lower in MDCK-BCRP cells, being 286-fold more potent than in MDCK-MDR1 cells (0.026 µM versus 7.44 µM). For Caco-2 cells, the IC50s were respectively 7-fold (Caco-2ECACC) and 20-fold (Caco-2ATCC) higher than that in MDCK-BCRP cells. This exceptionally low IC50 value determined in MDCK-BCRP cells unequivocally confirms ko143′s status as a highly potent and selective BCRP inhibitor. Furthermore, its IC50 in MDCK-MDR1 cells was comparable to that of verapamil, indicating that while ko143 effectively inhibits BCRP at nanomolar concentrations, it also exerts an effect on P-gp at higher micromolar concentrations, a dual activity that has been reported in a few other studies. These findings highlight that the precise adjustment of inhibitor concentrations based on the specific cell model employed is an important parameter, particularly when attempting to determine whether a drug is a substrate of ABC transporters or when assessing the potency of an inhibitor.

 

To predict a potential clinically relevant DDI for orally administered drugs like rivaroxaban, the FDA recommends utilizing a validated predictor, specifically the [I]^2/IC50 ratio (where IC50 is derived from *in vitro* studies). Fenner et al. previously assessed the relevance of this ratio using the example of digoxin, another ABC transporter substrate. To minimize the bias introduced by interlaboratory variations, they evaluated 19 compounds known to modulate P-gp-mediated digoxin transport within the same laboratory using Caco-2 cells. Their findings suggested that an [I]^2/IC50 ratio less than 10 was predictive of negative clinical DDIs, implying no significant interaction *in vivo*. However, a considerable percentage of false-positives (approximately 40%) was observed, which significantly limited the establishment of a robust *in vitro-in vivo* correlation for predicting DDIs. This problem, where IC50 values derived from *in vitro* data exhibit significant inter- and intralaboratory variability, is a particular concern. This variability is partly attributable to differences in cell model characteristics, including variations in transporter expression levels.

 

Regarding rivaroxaban, our study revealed very substantial and clinically impactful differences in the IC50 values for specific inhibitors, depending directly on the cell model utilized. For verapamil, when the IC50 was determined in Caco-2 models, the [I]^2/IC50 ratio was less than 10 (8.7 in Caco-2ATCC and 7.4 in Caco-2ECACC cells), which, according to the FDA metric, would suggest no need for further *in vivo* clinical DDI studies. Conversely, when the IC50 was determined in MDCK-MDR1 cells, the [I]^2/IC50 ratio was greater than 10 (26.5), which would, according to the same metric, suggest that conducting clinical studies would be warranted. These contrasting results powerfully illustrate the inherent heterogeneity of the cell models commonly employed in drug transport studies. This heterogeneity is entirely expected, given that the cell lines originate from different tissues (intestinal for Caco-2 and renal for MDCK) and, consequently, display distinct physiological phenotypes. This is further corroborated by our relative quantification of phenotypic markers. For instance, the barrier properties of these models also differed significantly. The expression of the tight junction protein occludin and the adherens proteins cadherin-1 and cadherin-17 was notably lower in MDCK cells (approximately 10–100-fold lower for occludin and 5–10-fold lower for cadherins) compared to Caco-2 cells. This differential expression pattern of junctional proteins is perfectly consistent with the substantially lower TEER values observed for MDCK cells (approximately 80 Ω·cm^2 versus 600 Ω·cm^2 for Caco-2 cells), indicating a less restrictive barrier. In addition to their differing barrier properties, these cell models are also characterized by varying endogenous expression levels of ABC transporters, which can undoubtedly interfere with the accurate determination of IC50 values for specific inhibitors. Therefore, the ability to precisely and accurately quantify the expression levels of these transporters in each specific cell type is not merely beneficial but absolutely essential for reliable data interpretation.

 

To address this critical need, mass spectrometry emerges as a powerful reference technique. In recent years, a wide array of new mass spectrometry-based analytical platforms and experimental strategies have been developed, making this technology increasingly accessible for routine protein quantification in various research and industrial settings. Mass spectrometry offers the unique advantage of simultaneously and accurately quantifying multiple transporters of interest across diverse biological samples, including cultured cells, isolated membrane vesicles, and both animal and human tissues. Our study significantly contributed to this field by first developing and meticulously validating an LC-MS/MS method for the absolute quantification of P-gp and BCRP, adhering strictly to current FDA guidelines. As expected, and consistent with previous reports, P-gp and BCRP were found to be more highly expressed in their respective transfected-MDCK models than in the Caco-2 models. Specifically, P-gp expression was up to 7.5-fold higher in MDCK-MDR1 cells, and BCRP expression was approximately 2.3-fold higher in MDCK-BCRP cells. These higher expression levels directly correlated with the correspondingly higher efflux ratios observed for rivaroxaban in these engineered models, unequivocally demonstrating the functional impact of transporter abundance. Intriguingly, our study revealed a 1.7-fold higher expression of P-gp in Caco-2ECACC cells compared to Caco-2ATCC cells. This difference perfectly correlated with a 1.8-fold higher efflux ratio of rivaroxaban observed in Caco-2ECACC cells compared to Caco-2ATCC cells. These results definitively establish a direct link between the quantitative expression level of a transporter and the measured efflux ratio in the absence of inhibitors across the four cell lines. However, it is important to acknowledge that we did not fully determine the precise link between transporter expression and IC50 values in this study, which would necessitate additional studies with more extensive data to establish a more precise correlation.

 

Regarding quantitative comparisons with previous literature, the P-gp level determined in Caco-2ECACC cells in our study was found to be 10-fold lower than that reported by Uchida et al. This discrepancy may be related to our approach of using two peptides for protein quantification, whereas Uchida et al. (and most other studies) quantified only a single peptide. Our study specifically observed that the concentrations of different peptides derived from the same protein are not always equal. For instance, we detected a 1.8-fold higher level of the SSLLDVLAAR peptide compared to the VIQELGLDK peptide for BCRP. Conversely, we found a greater amount of P-gp than that reported by Zhang et al. in MDCK-MDR1 cells, even when quantifying the same peptide as used in the Uchida et al. study. These discrepancies suggest that the differing selection of quantification peptides alone is not sufficient to fully explain these variations in quantification. Other critical parameters, such as differences in cell culture conditions, variations in sample preparation protocols, and the specific mass spectrometry techniques employed, can significantly affect the accurate quantification of transporters in these diverse cell models. It is noteworthy that while the FDA provides guidance for acceptance criteria for the quantification of small molecules by LC-MS/MS or proteins by ligand-binding assays, no official guidance is currently available specifically for the quantification of proteins by LC-MS/MS. However, some valuable good practice recommendations, tailored to the type of analysis being performed, have been proposed by Carr et al. Nevertheless, these recommendations do not specify the minimum conditions required to obtain consistently reliable protein quantification.

 

To further validate our results obtained by absolute quantification, we also performed a relative quantification of various ABC transporters using mass spectrometry. The findings for P-gp and BCRP from this relative quantification were highly consistent with those obtained from absolute quantification, reinforcing the robustness of our measurements. This relative quantification, performed by data-dependent acquisition, also provided valuable additional information regarding the expression of other ABC transporters. For instance, the expression of MRP1 was found to be similar across all cell models, with the notable exception of MDCK II cells, which exhibited a 5–10-fold lower expression. Caco-2 cells consistently demonstrated a 10-fold higher expression of MRP2 compared to MDCK cells. Conversely, the expression of MRP3 and MRP4 was 10-fold lower in Caco-2 cells than in MDCK cells. Our results concerning the relative quantification of these ABC transporters are in excellent agreement with those reported by Maubon et al., who utilized mRNA quantification. They reported a 5-fold higher quantity of P-gp mRNA in Caco-2ECACC cells than in Caco-2ATCC cells, a 4.5-fold higher amount of MRP2 mRNA in Caco-2ECACC cells, and a similar expression of BCRP and MRP4 mRNA in both Caco-2 models, all consistent with our protein-level findings. As a final confirmation technique, Western blotting unequivocally highlighted the presence of P-gp and BCRP in the two Caco-2 models and specifically in MDCK-MDR1 and MDCK-BCRP cells, respectively. Moreover, immunofluorescence staining visually confirmed the expression and distinct localization of the tight junction protein ZO-1, as well as the two ABC transporters, P-gp and BCRP, within these cellular structures. Regarding ABC transporters, BCRP was consistently detected at the apicolateral side in all cell models, aligning with previous reports, and showed a more intense signal in the Caco-2 and MDCK-BCRP models, indicating higher expression. P-gp was found to be highly expressed in MDCK-MDR1 and Caco-2 cells, exhibiting a characteristic apical localization, as documented in earlier studies. Taken collectively, the convergence of all these diverse analytical results, from absolute and relative quantification to Western blotting and immunofluorescence, strongly and clearly demonstrated a significant and inherent variability in transporter expression levels among the different cell models commonly employed in drug transport research.

 

Conclusion

This original and comprehensive work has critically illuminated the direct relationship between the abundance of ATP-binding cassette (ABC) transporters within commonly employed *in vitro* cell models and the corresponding extent of rivaroxaban efflux transport observed in these systems. Our findings unequivocally demonstrated that the judicious choice of cell model for conducting drug transport studies exerts a significant influence on the determined half-maximal inhibitory concentration (IC50) values for well-known P-gp and BCRP inhibitors. Crucially, we have shown that this observed variability in IC50 values is, at least in part, directly attributable to the inherent differences in the expression patterns and absolute quantities of ABC transporters within these diverse models.

 

Our proteomic analysis, specifically employing mass spectrometry, provided highly accurate and precise quantification of P-gp and BCRP transporters. This quantification was performed in strict accordance with FDA guidelines, enhancing the reliability and comparability of our data. This approach revealed a compelling correlation: a higher level of ABC transporter expression in the cell models was directly associated with a correspondingly greater capacity for rivaroxaban efflux transport. This quantitative link suggests that the precise quantity of transporters, as determined by mass spectrometry, could serve as a valuable metric for standardizing the expression of IC50 values derived from cell models. Such standardization would significantly help to mitigate interlaboratory variations, thereby improving the consistency and reproducibility of *in vitro* transport data.

 

Ultimately, the precise quantification of transporters in *in vitro* cellular systems and, by extension, in *in vivo* tissues has profound implications for pharmaceutical development. This quantitative information can be strategically integrated into physiologically based pharmacokinetic (PBPK) models. By incorporating transporter abundance, these sophisticated models can be significantly refined, leading to a much improved interpretation of pharmacokinetic results and a more accurate extrapolation of drug behavior from preclinical *in vitro* and animal studies to clinical outcomes in humans. While this study has made significant strides, it is acknowledged that additional studies, accumulating a greater volume of data, will be necessary to establish a more precise and comprehensive correlation between the exact expression levels of specific transporters and their corresponding IC50 values across a broader range of substrates and inhibitors.

 

Acknowledgements

The authors express their sincere gratitude to Dr. Piet Borst from The Netherlands Cancer Institute for his generous gift of the MDCK II, MDCK-MDR1, and MDCK-BCRP cell lines, which were invaluable resources for this research.

 

Funding

This research was conducted without the receipt of any specific grant from funding agencies in the public, commercial, or not-for-profit sectors, demonstrating the intrinsic motivation and collaborative efforts of the research team.