Do differences in cell lines and methods used for calculation of IC50 values influence categorisation of drugs as P-glycoprotein substrates and inhibitors?

Abstract In vitro bidirectional assays are employed to determine whether a drug is a substrate and/or inhibitor of P-glycoprotein (P-gp) transport. Differences between cell lines and calculation methods can lead to variations in the determination of efflux ratios (ER) and IC50 values used to classify a drug as a P-gp substrate and inhibitor, respectively. Information was collected from the literature on ER and IC50 values with digoxin as the probe substrate using different cell lines and inhibition calculation methods. Predictive performance was evaluated by comparing [Igut]/IC50 ratios versus reported in vivo results. For known P-gp substrates, 50% of the drugs had their highest ER value in MDCK-MDR1 cells while 81% had their lowest ER value in Caco-2 cells. For 30 drugs with inhibition data, lower mean IC50 values were often observed with the Caco-2 cells and calculations based on ER. Based on the cut-off criteria of [Igut]/IC50 ≥ 10, there were no significant differences in positive or negative predictive values based on either cell line or calculation method for the drugs. Within this limited dataset, differences between cell lines or IC50 calculation methods do not seem to impact the prediction of in vivo P-gp inhibitor classification.


Introduction
Drug-drug interactions (DDI) occur when one drug ('perpetrator' or 'precipitant') affects the pharmacokinetics of another drug ('victim' or 'object'). The importance of comprehensively assessing metabolism-based DDIs is to ensure the safe and effective use of drugs has been extensively documented (Obach et al. 2006;Galetin et al. 2008). More recently, the critical role played by drug transporters in affecting the absorption and disposition of drugs and endogenous substances, and toxicity has been clearly recognised (Estudante et al. 2013;Arya and Kiser 2016;Cheng et al. 2016;Lee et al. 2017). The International Transporter Consortium (ITC) has proposed a clear framework regarding the prospective evaluation of transporters in drug development (Giacomini et al. 2010;Brouwer et al. 2013). In addition, regulatory guidances underscore the importance of assessing transporter-mediated DDIs during drug development (European Medicinal Agency 2012; Pharmaceuticals and Medical Devices Agency 2018; Food and Drug Administration 2020; ICH Harmonised Guideline 2022).
One of the clinically relevant transporters is P-glycoprotein (P-gp, ABCB1), or MDR1, a member of the ATP-binding cassette (ABC) family of efflux transporters. This transporter is found on the apical surfaces of enterocytes, hepatocytes, proximal renal tubular cells, and brain endothelial cells (Giacomini et al. 2010). The P-gp transporter plays an important role in (i) the efflux of drugs from the intestinal lumen (Estudante et al. 2013); (ii) the elimination of endogenous compounds/metabolites, drugs and metabolites into the bile (K€ ock and Brouwer 2012); (iii) the clearance of compounds into the urine (Morrissey et al. 2013); and/or (iv) the access of toxins and drugs into the central nervous system (Mahringer and Fricker 2016).
To assess whether a drug has the potential to inhibit P-gp transporters in the intestinal lumen, regulatory guidances recommend determining whether [I gut ]/IC 50 ratio is greater than or equal to 10 (European Medicinal Agency 2012; Pharmaceuticals and Medical Devices Agency 2018; Food and Drug Administration 2020). [I gut ] refers to the theoretical maximum gastrointestinal drug concentration calculated at the highest dose per administration (HDPA) divided by a volume of 250 mL and IC 50 refers to the concentration causing half-maximal inhibition of P-gp transporters in vitro. Of note, although there have been several efforts to assess the performance of other in vitro prediction methods, based on the available data, [I gut ]/IC 50 appears to reasonably predict the potential for orally administered drugs as P-gp inhibitors (Agarwal et al. 2013;Zhou et al. 2019). Digoxin is commonly used as a probe in vitro and in vivo substrate to evaluate whether a drug is a P-gp inhibitor (Taub et al. 2005;Rautio et al. 2006;Ma et al. 2010;Nader and Foster 2014).
In vitro transporter assays are used during drug development to assess whether a drug is a substrate and/or inhibitor of P-gp transporters. These include cell-based bidirectional assays that are the most direct and functionally relevant models available to evaluate whether a new drug is a substrate or inhibitor of efflux transporters (Brouwer et al. 2013;Volpe 2016). The basic setup of a cell-based assay employs a dual-chamber apparatus with a cell monolayer grown on a semi-porous membrane separating the apical (AP) and basolateral (BL) chambers. Based upon the apparent permeability (P app ) in the secretive (P app,BL-AP ) and absorptive (P app,AP-BL ) directions, an efflux ratio (ER) is calculated. A drug is generally considered to be a substrate of an efflux transporter if its ER is greater than or equal to 2 (Giacomini et al. 2010;Food and Drug Administration 2020). IC 50 values for the drugs are calculated from the inhibition of digoxin efflux in the citations based on the change of ER, net secretory flux (NSF), or P app,BL-AP over a range of drug concentrations (Balimane et al. 2008).
It is well known that the results of in vitro cell-based transporter assays are influenced by multiple factors such as physicochemical characteristics of test compounds (e.g. solubility, stability), test systems (e.g. cells), experimental design (e.g. drug concentration, buffer pH) (Bhoopathy et al. 2014;Volpe 2016), and calculation methods (ER, P app,BL-AP or NSF). Differences in these factors can lead to variability in the determination of IC 50 values (Balimane et al. 2008;Sugimoto et al. 2011;Bentz et al. 2013;Kishimoto et al. 2014;Volpe et al. 2014). The research presented here focuses on two of these factors that may influence the results of in vitro cellbased transporter assays and by extension, the reliability of predicting the in vivo potential of a drug to inhibit P-gp: cell lines and calculation methods.
The cell line factor was selected as the expression of efflux transporters can vary between cells (Shirasaka et al. 2008;Bentz et al. 2013) and this can lead to different efflux ratios of a substrate drug (Tang et al. 2002;Siissalo et al. 2007;Patil et al. 2011) and IC 50 values of an inhibitor drug (Tang et al. 2002). The human epithelial colorectal adenocarcinoma (Caco-2) and Madin-Darby canine kidney (MDCK) cells are two cell lines that are most commonly used to assess whether a new drug is a substrate and/or inhibitor of the P-gp efflux transporter in bidirectional cell assays (Volpe 2011(Volpe , 2016. Given the canine origin of MDCK cells and low expression of transporter proteins, they are often transfected with human transporter proteins such as P-gp (MDR1-MDCK cells) for use in transport assays (Tang et al. 2002;Volpe 2011). Additionally, Lilly Laboratories cell-porcine kidney 1 (LLC-PK1) cells are transfected with MDR1 for use in P-gp assays (Sugimoto et al. 2011).
The calculation method factor was selected as there are multiple methods reported in the literature for calculating IC 50 values for a drug in the bidirectional assay (Balimane et al. 2008). However, there is limited information in the literature on whether the type of method selected has an influence on the IC 50 value to the extent that it can differ in classifying a drug as a P-gp inhibitor or a non-inhibitor.
Three methods selected for this research were the ER, P app,BL-AP , or NSF methods (Balimane et al. 2008).
Overall, the purpose of our work was to assess whether differences between cell lines and calculation methods can lead to variations in the computation of ER and IC 50 values used to classify a drug as a P-gp substrate and inhibitor, respectively. Further, our goal was to assess whether a specific cell line and calculation method are more accurate in predicting if a drug is a substrate and/or inhibitor of P-gp in vivo.

Materials and methods
We conducted a literature survey to collect in vitro information regarding the use of Caco-2, LLC-PK1-MDR1, and MDCK-MDR1 cells to predict whether a new drug is a P-gp substrate or inhibitor and in vivo information from DDI trials conducted to evaluate the effect of drugs on digoxin pharmacokinetics. Sources of the in vitro and in vivo interaction studies included the University of Washington Drug Interaction Solutions (https://www.druginteractionsolutions. org/), PubMed (https://www.ncbi.nlm.nih.gov), EMBASE (https://www.embase.com), Google Scholar (https://scholar. google.com/), Web of Science (https://apps.webofknowledge. com), and Drugs@FDA (https://www.accessdata.fda.gov/ scripts/cder/daf/).
For in vitro studies, information on the cell line used, culture and transport conditions, and resultant ER or IC 50 values were compiled and tabulated. ER and NSF were defined for the in vitro inhibitors according to the equations in Table 1. IC 50 values were calculated in the citations for a drug in the bidirectional assay based on the ER, P app,BL-AP , or net secretory flux (NSF) methods (Balimane et al. 2008;Bentz et al. 2013).
For DDI trials, information included the dosing regimen of digoxin and the inhibitor drug as well as digoxin AUC (area under the concentration-time curve) values in the absence and presence of the inhibitor drug. The DDI studies included were those using the inhibitor drug at its HDPA and recommended dosing schedule. While compiling the dataset for this research, studies that evaluated the multiple doses of inhibitor drugs were selected over single-dose interaction studies as the inhibitor drug concentrations as the former are expected to be relatively better aligned with the drug's therapeutic use than the latter. Each drug's intestinal concentration ([I gut ]) was determined based on the respective drug's HDPA in 250 mL volume and these I gut values were compared with the pre-defined cut-off criteria. Specifically, a drug was classified as in vitro inhibitor if its [I gut ]/IC 50 ratio was greater than or equal to 10 (European Medicinal Agency 2012; Pharmaceuticals and Medical Devices Agency 2018; Food and Drug Administration 2020; ICH Harmonised Guideline 2022). A drug was included in the dataset if (1) clinical drug interaction data with digoxin AUC values were available and (2) in vitro IC 50 values in at least two of the cell lines and at least two calculation methods were available. A drug was classified to be an in vivo inhibitor if the ratio of digoxin AUC (AUCR) in the presence and absence of a drug was greater than or equal to 1.25-fold and an in vivo non-inhibitor when the AUCR was less than 1.25-fold (Agarwal et al. 2013;Zhou et al. 2019).
Predictive performance was evaluated by comparing in vitro and in vivo classifications. Specifically, to assess performance, true-positive (TP, in vitro and in vivo inhibitor), true-negative (TN, in vitro and in vivo non-inhibitor), falsepositive (FP, in vitro inhibitor and in vivo non-inhibitor), and false-negative (FN, in vitro non-inhibitor and in vivo inhibitor) results were calculated. From these values, accuracy, specificity, sensitivity, negative predictive value (NPV), and positive predictive value (PPV) were calculated (Trevethan 2017

Results
Thirty-two substrates with ER values in at least two cell lines were collected for a total of 1097 data points. The majority of the data originated from Caco-2 cells (64%), followed by MDCK-MDR1 (25%) and LLC-PK1-MDR1 (11%) cells. No one cell line produced the highest or lowest ER values across all the drugs. Over half of the substrates had their maximum ER value in MDCK-MDR1 cells while 38% had their highest ER value in Caco-2 cells. All the drugs had their minimum individual ER value in Caco-2 cells except colchicine, doxorubicin, erythromycin, etoposide, ritonavir, and vincristine. Where there were multiple ER values available for a given cell line, geometric mean ERs ranged from 45.14 for vinblastine in MDCK-MDR1 cells to 1.72 for doxorubicin in MDCK-MDR1 cells. Table 2 summarises the geometric mean ER values showing the variability in the efflux ratios based on the cell lines for the 32 substrates. Additional descriptive statistics for the ER values of these substrates are found in Supplemental file 1.
Data was collected for 30 drugs having a known positive or negative clinical interaction with digoxin, generating 691 in vitro IC 50 values with at least two cell lines and two calculation methods. The number of data points for a drug ranged from 79 (for verapamil) to 3 (for atorvastatin, linagliptin, and pantoprazole). Twelve of the drugs were inhibitory in vivo with 331 IC 50 data points (48%) and the remaining 18 drugs were not inhibitory in vivo with 360 data points (52%). Caco-2 was the most common cell line with 411 (59%) of the IC 50 values followed by LLC-PK1-MDR1 and MDCK-MDR1 with 148 (21%) and 132 (19%), respectively. The use of different calculation methods was more evenly distributed among NSF (274, 40%), P app,BL-AP (224, 32%), and ER (193, 28%). Supplemental file 2 summarises the in vitro data collected for the inhibitory drugs. There was variability in the overall geometric mean IC 50 values for the 30 drugs which ranged from 76.02 mM for pantoprazole to 0.18 mM for valspodar. Eighteen drugs had geometric mean IC 50 values less than 20 mM. There was no pattern in rank order for the geometric mean of IC 50 values for the drugs based on a cell line or calculation method. Based on the calculation method, the lowest geometric mean IC 50 values were from ER in 23 of the 30 drugs (77%). The lowest geometric mean IC 50 values were in Caco-2 cells in 22 of the 30 drugs (73%). Valspodar was the most potent inhibitor in Caco-2 and LLC-PK1-MDR1 cells (no data available in MDCK cells), while itraconazole was the most potent in MDCK-MDR1 cells (Table 3). Valspodar had the lowest IC 50 values across all three calculation methods (Table 3). There was no pattern in the rank order of IC 50 values for the drugs based on either the cell line or calculation method (rank order data not shown). Over two-thirds of the drugs had their lowest geometric mean IC 50 values in Caco-2 cells and with the ER calculation method for 80% of all the drugs.
With respect to the in vivo classification data for the 30 drugs, 12 (40%) drugs were inhibitory in vivo (i.e. AUCR !1.25) and while the other 18 (60%) drugs did not affect digoxin exposure (i.e. AUCR < 1.25). Of note, several of these drugs are also in vitro P-gp substrates including atorvastatin, lapatinib, linagliptin, omeprazole, pantoprazole, quinidine, quinine, ranolazine, ritonavir, and verapamil. Supplemental file 3 summarises the clinical drug-drug interaction data with digoxin for these drugs.
In general, no prominent differences were noted when the [I gut ]/IC 50 ! 10 cut-off criterion was compared based on a cell line or calculation method. The three cell lines had similar/comparable (difference 15%) accuracy and sensitivity while the specificity was lower in the Caco-2 cells (Table  4). This may be due to the differences in P-gp expression between the cell lines. MDCK-MDR1 cells had slightly higher NPV and PPV results. MDCK-MDR1 cells had numerically higher and comparable NPV and PPV results to results from Caco-2 cells, respectively. However, MDCK-MDR1 cells had comparable NPV and PPV results with LLC-PK1-MDR1 results. Looking at the IC 50 calculation methods, P app,BL-AP had a higher NPV than the ER and NSF (Table 5). There was a 16% difference in the three methods for PPV, accuracy specificity, or sensitivity.

Discussion
It is well known that variability in experimental conditions and methodology can influence outcomes in efflux and inhibition experiments for the evaluation of a drug as a substrate or inhibitor of an efflux transporter (Volpe 2011(Volpe , 2016. These include cell lines (Tang et al. 2002;Taub et al. 2005), P-gp expression levels (Hayeshi et al. 2008;Shirasaka et al. 2008    2007; Elsby et al. 2008;Kamiyama et al. 2009;Senarathna and Crowe 2015), buffer pH (Neuhoff et al. 2003;Crowe and Wong 2004;Korjamo et al. 2005;Crowe and Wright 2012), and data analysis methodology (Cook et al. 2010;Sugimoto et al. 2011;Volpe et al. 2014). For assays determining whether a drug is a P-gp substrate, investigators have found that efflux ratios can vary based on cell passage number, initial cell seeding, plate format, monolayer age, substrate concentration, and transport buffer pH (Neuhoff et al. 2003;Balimane et al. 2004;Crowe and Wong 2004;Korjamo et al. 2005;Siissalo et al. 2007;Elsby et al. 2008;Shirasaka et al. 2008;Kamiyama et al. 2009;Miliotis et al. 2011;Patil et al. 2011;Crowe and Wright 2012).
For the 30 P-gp substrates evaluated, geometric mean ERs ranged from 45.14 (vinblastine) to 1.72 (doxorubicin). Overall, there was no single cell line that yielded the highest or lowest ER values. An examination was also conducted on a small subset of these substrate drugs for those with low ($2-3) and higher (>10) geometric mean efflux ratios. There were no significant differences in logP values between the groups or physiological charge (data not shown). Similarly, there were no significant trends between the groups based on their permeability or solubility class according to the biopharmaceutics classification system (BCS; data not shown).
For assays determining whether a drug is a P-gp inhibitor, IC 50 values can vary across cell lines, substrates, and calculation methods (Balimane et al. 2008;Cook et al. 2010;Perloff et al. 2011;Sugimoto et al. 2011;Bentz et al. 2013;Kishimoto et al. 2014;Poirier et al. 2014;Volpe et al. 2014). However, a common observation was that IC 50 values based on ER were lower than those calculated from NSF or P app,BL-AP (Balimane et al. 2008;Cook et al. 2010;Perloff et al. 2011;Sugimoto et al. 2011;Bentz et al. 2013;Kishimoto et al. 2014;Poirier et al. 2014;Volpe et al. 2014). A study with 23 labs, using four systems (Caco-2, LLC-PK1-MDR1, MDCK-MDR1, vesicles) with 15 test drugs, found substantial variability in IC 50 values for the inhibition of digoxin efflux (Bentz et al. 2013). In the study, the lowest variability was seen for sertraline (20-fold) and isradipine (24-fold) with telmisartan (407-fold), and verapamil (796-fold) having the highest variability. The study concluded that a large amount of the variability was due to lab-to-lab procedural differences (e.g. cell source, passage number, culture/transport conditions) rather P-gp expression in the cell systems (Bentz et al. 2013). Variability within a laboratory may be limited by uniform practices in terms of these multiple factors including cell source and calculation methods when conducting in vitro experiments to determine whether a new drug is a substrate or inhibitor of a transporter (Brouwer et al. 2013;Volpe 2016).
Given the aforementioned reports of various factors shown to have an impact on ER and IC 50 values, we evaluated whether differences between cell lines and calculation methods can impact a drug's classification as a P-gp substrate or inhibitor. Our research shows that there were no differences in predictive performances in this small dataset based on a cell line or IC 50 calculation method used. It is also noteworthy that a number of drugs routinely had false or true predictions based on the [I gut ]/IC 50 ! 10 cut-off criterion, that is, bepridil, conivaptan, dipyridamole, ketoconazole, and troglitazone nearly always had FPs or FNs. Additionally, mibefradil, nicardipine, propafenone, sertraline, spironolactone, and telmisartan mostly had FPs or FNs with this criterion. It is also notable that bepridil, propafenone, and telmisartan had in vivo AUCRs that were close to the 1.25 cut-off. Conversely, amiodarone, atorvastatin, carvedilol, clarithromycin, diltiazem, isradipine, itraconazole, lapatinib, linagliptin, pantoprazole, quinidine, ranolazine, ritonavir, valspodar, and verapamil nearly always had TP or TN results.
Overall, the NPV (86%) was greater than PPV (48%) with the 30 drugs in the dataset. IC 50 values from the Caco-2 cells tended to result in an NPV and PPV numerically lower than the LLC-PK1-MDR1 and MDCK-MDR1 cells. Both the LLC-PK1-MDR1 and MDCK-MDR1 cells had the same NPV (100%) and sensitivity (100%) results. Using P app,BL-AP for calculations of IC 50 values resulted in somewhat better NPV and PPV than the ER or NSF methods. Even though there were differences in predictive performances in this small dataset, there were no significant differences in the PPV and NPV values regardless of the use of a particular cell line or IC 50 calculation method.
We do acknowledge several important limitations with our research presented here. First, this study was limited by the low number of evaluated drugs (30) of which 60% were in vivo non-inhibitors of P-gp in digoxin DDI studies. However, it should be noted that for the in vitro IC 50 values there was a similar number of data points for the in vivo inhibitors (48%) and non-inhibitors (52%). Secondly, the dataset collated was from various laboratories which could potentially introduce variability from the experimental procedures such as incubation time, inhibitor concentration, and digoxin concentration. However, this closely mimics the situation frequently encountered during drug development where in vitro characterisation of drugs is conducted in different laboratories under different experimental conditions. A third limitation is that the findings failed to identify a combination of cell line and calculation method as superior in reducing the FN rate. This could be due to the fact that the number of FN and TN results was relatively low compared to FP and TP results. The last limitation is that although digoxin is widely used as a probe substrate for P-gp clinical interaction studies and the majority of available in vitro and in vivo data is based on using digoxin as a substrate (Taub et al. 2005;Rautio et al. 2006), it has a few limitations. This includes high interlaboratory results with digoxin and uptake mechanisms of digoxin in the liver and kidney (Bentz et al. 2013;Lee et al. 2014;Nader and Foster 2014). Hence, if a different P-gp substrate was to be used in in vitro and in vivo studies, whether differences in cell lines or calculation methods will influence the categorisation of drugs as P-gp substrates or inhibitors, needs to be investigated.
Overall, we conducted an analysis of literature data of Pgp substrates and inhibitors to examine differences in ER or IC 50 values due to cell lines and calculation methods and to compare the resultant [I gut ]/IC 50 ratios to predict the potential for a drug to inhibit P-gp transporters. For 30 drugs, lower mean IC 50 values were often observed with the Caco-2 cells and calculations based on efflux ratios. Both positive and negative predictive values were comparable regardless of the use of a particular cell line or calculation method for a drug's IC 50 value. This has implications in drug development and regulatory review in that different cell lines or IC 50 calculation methods may be used in predicting P-gp-mediated drug-drug interactions.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
The author(s) reported there is no funding associated with the work featured in this article. Data availability statement