FormalPara Key Points

P-glycoprotein (P-gp) is a drug transporting protein that physicians should be aware of when evaluating potential drug–drug interactions.

The evidence base for evaluating the role of P-gp in potential drug–drug interactions is varying and interpretation may be complex.

Potential drug–drug interactions should be identified and appropriate management strategies applied.

1 Introduction

Adverse drug events (ADEs) are a common cause of hospital admission [1, 2] and display a potential negative impact on patient morbidity and mortality, treatment costs and duration of hospitalization [3, 4]. In this context, interactions between concomitantly used drugs may be a cause of serious ADEs [5, 6]. Formally, a drug interaction is defined as an alteration in a clinically meaningful way of the effect of a drug (object or victim drug) as a result of co-administration of another drug (precipitant or perpetrator drug) [7]. While not all potential drug–drug interactions (DDIs) are clinically significant [810], the outcome (therapeutic failure or toxicity) when actual DDIs occur may be severe and sometimes fatal [9]. The clinical challenge lies in sorting out the unimportant interactions and identifying the DDIs that are most likely to be clinically manifest. The evidence base for particular DDIs may be of varying quality and whether or not a particular DDI is clinically meaningful may vary depending on a number of individual patient characteristics such as age, organ function, genetic makeup, comorbidity as well as concurrent medication.

The mechanisms responsible for DDIs may be pharmacodynamic or pharmacokinetic. An example of a pharmacodynamic interaction is excessive bleeding with concurrent use of warfarin, which is a vitamin K antagonist, and low-dose aspirin, where warfarin affects bleeding through decreased production of coagulation factors and aspirin through inhibition of thrombocyte aggregation [7, 11, 12]. Pharmacokinetic interactions can occur due to changes in the absorption, distribution, metabolism and elimination of a drug and may result in higher or lower drug levels leading to potential toxicities or loss of efficacy. These processes may be altered by drugs interfering with drug metabolizing enzymes such as those belonging to the cytochrome P450 system (CYP450), involved in phase I reactions, or uridine diphosphate glucuronosyltransferases (UGTs), involved in phase II reactions. Drugs may also affect the pharmacokinetics of other drugs by interfering with drug transporters such as the p-glycoprotein efflux transporter (P-gp), breast cancer resistance protein (BCRP) or the organic anion transporting polypeptides (OATPs) [13]. Individual drugs may thus be substrates, inhibitors or inducers of specific enzymes or transporters and concurrent administration of, for example, an inhibitor of a transporter or an enzyme may imply an increased concentration of a drug that is a substrate of the same transporter or enzyme and subsequently an increased risk of ADEs. Conversely, administration of an inducer may decrease the concentration of a substrate of a particular enzyme or transporter, and lead to decreased effect of the drug.

The focus of the present review is the P-gp drug transporter in relation to DDIs. Particularly, we aim to synthesize the evidence for classifying various drugs as substrates, inhibitors and inducers of P-gp according to in vitro and in vivo evidence. We searched the literature and drug labels/summaries of product characteristics (SmPCs) to identify human in vitro and in vivo evidence to support the particular P-gp modulatory status (substrate, inhibitor, inducer) of the various drugs. The review consists of the following main parts:

Section 2 is a short introduction to P-gp (gene, protein and transport molecule).

Section 3 includes methodological considerations for compilation of the list of drugs to be evaluated for P-gp modulatory properties and evaluation criteria. The compiled list of drugs was evaluated by means of a two-step procedure evaluating (1) in vitro evidence (human P-gp) and (2) if available, human in vivo studies.

Section 4 consists of five tables of drugs categorized according to therapeutic groups as P-gp modulators (substrates, inhibitors or inducers) for which we found compelling evidence of potential clinical relevance. The remaining drugs, for which we did not find compelling evidence of clinical relevance, are presented in five similar tables in the Electronic Supplementary Material (ESM, Supplementary Tables 1–5). All tables in the main paper and in the ESM provide the specific references forming the basis of the classification of the individual drugs.

Section 5 consists of general considerations regarding clinical management of potential DDIs including elaborations on drugs from the tables in Sect. 4 as well as a discussion of limitations of the suggested classification of drugs as P-gp modulators.

Section 6 consists of conclusions and perspectives.

2 Background

2.1 P-Glycoprotein: Gene, Protein and Transport Molecule

P-gp is a protein encoded for by the ABCB1 (MDR1) gene, which belongs to the group of adenosine triphosphate (ATP)-binding cassette (ABC) genes [14] encoding the widespread ATP-binding cassette transport proteins. At present, 49 members of the ABC superfamily have been identified and grouped into seven subfamilies: ABCA to ABCG [15]. In eukaryotes, all ABC proteins are efflux pumps and play a role in protecting the organism from noxious substances [16].

ABCB1 is located on chromosome 7q21.2 and encodes P-gp which consists of 1276–1280 amino acids and has a molecular mass of around 170 kDa. The topological structure of P-gp consists of two homologous halves, each with six highly hydrophobic transmembrane domains and a nucleotide-binding domain. ATP must bind to both nucleotide-binding domains for activity of P-gp to occur. Moreover, it contains multiple drug-binding sites and is able to simultaneously bind multiple substrates at overlapping binding sites [17].

P-gp exhibits polarized expression and is found in the plasma membranes of cells in barrier and elimination organs. The transporter is expressed on the luminal-facing epithelia of the gut and the bile-facing canaliculi of the liver, where it plays a role in first-pass drug metabolism. It is also found on the luminal membrane in the proximal tubules of the kidney, eliminating substances from the systemic circulation. Further, it is present in the blood-brain barrier and limits the permeability of many drugs into the brain as well as into other specific organs such as the placenta and testis [16].

P-gp substrates display a large diversity in structure including both small molecules (e.g. organic cations, carbohydrates and amino acids) and macromolecules (polysaccharides and proteins) [1719]. Drugs that are substrates of P-gp may or may not also be inhibitors or inducers of P-gp, and vice versa [17, 20]. A number of drugs that inhibit P-gp, however, have some chemical properties in common, such as aromatic ring structures, a tertiary or secondary amino group and high lipophilicity [17]. Regarding induction of P-gp, this does not occur by the drug binding directly to P-gp but instead by regulation at the transcriptional level by nuclear factors [21]. For instance, the xenobiotic nuclear receptor (PXR) regulates ABCB1 expression and may be activated by, for example, rifampicin [17]; in this context, animal studies demonstrating presence of PXR at the level of the blood–brain barrier might contribute to explaining the decreased efficacy of CNS-acting drugs that are also P-gp substrates [22].

Genetic polymorphisms of clinical importance exist for some of the drug metabolizing enzymes (CYP2C9, CYP2C19 and CYP2D6). These account for a substantial portion of interpersonal variability in drug metabolism [13]. For MDR1, a vast number of commonly occurring single nucleotide polymorphisms have been identified and characterized. However, here results have been inconsistent and no clear association with clinical risk profile has been established. Thus, present evidence does not yield support for recommendations of adjustment in drug dose according to specific ABCB1 sequence variants [16, 23, 24].

2.2 Overlap in Drugs That Interfere with Both P-Glycoprotein (P-gp) and CYP3A4

There is a substantial overlap in drugs that interact with CYP3A4 and P-gp [25]. However, the overlap is by no means complete, and no clear rules seem to exist for determining whether or not an overlap exists [26]. P-gp and CYP3A4 are found in many of the same organs and tissues [25, 27, 28] and seem to function in a complementary fashion to reduce systemic drug exposure: as a drug traverses down the intestinal tract, repeated cycles of P-gp extrusion followed by passive reabsorption facilitates repeated cycles of exposure of the drug to CYP3A4 present in the gut wall and thus to increased metabolism and reduced bioavailability [29]. The structure–activity overlap between CYP3A4 and P-gp may thus complicate a clear determination of the relative contribution of P-gp versus CYP3A4 to a particular DDI. However, various approaches may increase the likelihood of isolating an effect attributable to P-gp. These include use of an appropriate and validated cellular system for in vitro studies, appropriate choice of substrate and inhibitors, and performing in vivo studies when indicated by the results of in vitro studies [30]. Examples of probe drugs are midazolam (CYP3A4 substrate for which P-gp is not a barrier for absorption) and digoxin, dabigatran etexilate or fexofenadine (P-gp substrates that are not metabolized by CYP3A4) [31, 32]. With respect to inhibitors, it is important to note that dual inhibitors of CYP3A4 and P-gp do not per se have the same inhibitory potency towards P-gp and CYP3A4. For example, quinidine and amiodarone are potent P-gp inhibitors (as defined by a >1.5-fold change in fexofenadine or digoxin area under the concentration time curve [AUC]) but weak CYP3A4 inhibitors, whereas itraconazole is a potent inhibitor of both CYP3A4 and P-gp [32].

3 Methodological Considerations

3.1 Compilation of Drugs for Evaluation and Literature Search

First, we compiled a list of drugs to be evaluated for P-gp modulation. The list consisted of (1) drugs mentioned in previous reviews of P-gp: substrates [16, 3336], inhibitors [16, 33, 36, 37] and inducers [16, 33, 36]; (2) drugs in the list of newly marketed drugs within the last 6 months approved by either the Danish Medicines Agency as of January 2015 [38] and/or the European Medicines Agency (EMA) as of October 2015 [39] and where P-gp modulatory properties were mentioned in the SmPC; and (3) drugs mentioned with P-gp modulatory properties in papers identified in step 1 or references therein. Second, we sought to identify background evidence regarding the respective P-gp modulatory property for each drug in the compiled list. We did this by a reference search of the before-mentioned reviews and/or searched for information in the SmPC from either the Danish Medicines Agency [40], EMA [39] or in the drug label from the United States Food and Drug Administration (FDA) [41]. If approved by the EMA or FDA, we also searched supporting documents available online (e.g. from the EMA, the Scientific Discussion or the European Public Assessment Report (EPAR) and from the FDA, the Clinical Pharmacology and Biopharmaceutics Review(s) [39, 41]. Finally, if necessary we searched PubMed for additional evidence including by use of the following search strings for substrates, inhibitors and inducers, respectively. Substrates and inhibitors: ((“P-Glycoprotein/agonists” [Mesh] OR “P-Glycoprotein/antagonists and inhibitors” [Mesh] OR “P-Glycoprotein/metabolism” [Mesh] OR “P-Glycoprotein/pharmacokinetics” [Mesh] OR “P-Glycoprotein/pharmacology” [Mesh]) AND “P-Glycoprotein” [Mesh]) AND “Suspected Drug” [Mesh] AND “humans” [MeSH terms]. For inducers we searched on combinations of the suspected drug as a Mesh-term combined with “Digoxin” [Mesh], “talinolol” [Supplementary Concept] or “fexofenadine” [Supplementary Concept], respectively.

The evidence base for evaluating the potential for DDIs varies substantially for newer versus older drugs likely due to the increased regulatory focus on discovering potential for DDIs during recent years [42]. Previous reviews and database sources regarding DDIs involving P-gp vary quite substantially both with respect to which drugs are mentioned as substrates, inhibitors or inducers and to what constitutes potential clinical relevance. Explanations for this are likely related to differential target audiences of the data sources (e.g. a referential overview aiming to include all possible evidence as opposed to tools for clinical decision making, which may be more restrictive) as well as different approaches to evaluation and categorization of the available evidence. The scope of the present review is clinical, that is, we aim to provide the clinician with a compilation of the evidence base regarding the P-gp modulatory status for various drugs as well as a classification regarding clinical relevance. This is meant to be used as a supporting document when, for example, performing medication reviews, deciding on a new drug treatment or retrospectively when trying to elucidate a potential mechanistic basis for an observed DDI; that is, it should not be seen as a list of clear-cut contraindications. With this aim in mind, we chose a restrictive approach focusing on maximizing the likelihood of clinical relevance. In terms of classification, this implied disregarding evidence from studies using non-human P-gp and focusing on evidence from in vitro studies of human P-gp and corresponding in vivo studies. The final classification (defined in Sects. 3.2 and 3.3) contained one level for inducers (solely based on in vivo evidence) and two levels for substrates and inhibitors (one based on in vitro evidence and one on in vivo evidence). An initial evidence search and grading of the evidence for all drugs in the compiled list formed the basis for a discussion of the classification of each drug among co-authors and the final classification for each drug was based on a consensus agreement.

3.2 Substrates and Inhibitors

In the 2012 FDA draft guidance on investigation of potential DDIs, it is advised that all newly marketed drugs should be evaluated in vitro to determine whether they are a P-gp substrate and, depending on the therapeutic use (e.g. probability of being prescribed with digoxin, a P-gp substrate with a narrow therapeutic index), evaluation of P-gp inhibitory properties should be considered [32]. Flowcharts for assessment of both P-gp substrate and inhibitor status have previously been published [30] and incorporated by the FDA in their guidance [32]. A similar guidance is provided by EMA [31]. Aiming to construct a classification that would acknowledge and accommodate the differential levels of evidence available for newer versus older drugs, we used the FDA classifications (see Figs. 1, 2) to guide our criteria for evidence needed to obtain the different classifications described below. Overall, the classification was based on two steps for substrates and inhibitors: the first step was based on in vitro evidence and the second step on in vivo evidence.

Fig. 1
figure 1

Decision tree for determining status as P-glycoprotein substrate used for regulatory guidance (modified from [30]). aPreclinical and clinical information should be assessed to determine whether a new molecular entity is a P-gp substrate. Particularly, the relative contribution of the transporter-mediated pathway to the overall clearance of the drug is the primary determinant of whether an inhibitor will have a major effect on the disposition of the new molecular entity; i.e. an in vivo interaction study may not be needed for a drug that has high solubility, high permeability and/or is highly metabolized since it is less likely to be affected by a P-gp inhibitor [30]. DDI drug–drug interaction, P-gp P-glycoprotein

Fig. 2
figure 2

Decision tree for determining status as P-glycoprotein inhibitor used for regulatory guidance (modified from [30]). a[I]1 represents the mean steady-state total (free and unbound) maximum serum concentration (C max) following administration of the highest proposed clinical dose. [I]2 represents concentration of a therapeutic dose of the inhibitor dissolved in 250 mL. IC50 is defined as the concentration of the inhibitor needed to inhibit half the transport rate of the substrate. P-gp P-glycoprotein

In the first step, based on in vitro studies of human P-gp, we categorized evidence of clinical relevance for substrates and inhibitors of P-gp as ‘yes’, ‘no’ or ‘uncertain’. To qualify for categorization ‘yes’ as a substrate, we required evidence for polarized transport from a study using cells overexpressing P-gp in a bidirectional assay with a net efflux above 2. If, with the addition of a known P-gp inhibitor, the positive net efflux was not inhibited by 50% or more or the efflux ratio was not attenuated towards 1, we took this as evidence that other drug transporters were likely to be responsible for the observed polarized net efflux [43]. Acceptable cells for categorization ‘yes’ were (for both substrates and inhibitors) Caco-2 cells (human colon carcinoma cell monolayers) or either of the recombinant epithelial cell lines MDR1-MDCK (Madine-Darby canine kidney cells transfected with the human MDR1 gene) or L-MDR1 cells (porcine kidney epithelial cells, LLC-PK1, transfected with the human MDR1 gene). To qualify for categorization ‘yes’ as an inhibitor, we required evidence for likely clinically relevant inhibition based on calculation of [I]1/IC50 or [I]2/IC50 (as defined in Fig. 2), and if this was not found we classified as ‘no’. If evidence was in various ways insufficient or incomplete (e.g. by being based on other cell lines or by lacking a positive control but otherwise indicating that the drug was a P-gp substrate), we would classify as ‘uncertain’. We also used categorization as ‘uncertain’ if evidence from established cell lines was conflicting (for both substrates and inhibitors). If the drug was marketed within recent years and the label or SmPC explicitly mentioned that the drug had been shown to be an in vitro P-gp substrate or P-gp inhibitor but did not mention specifically which cells had been used, we classified as ‘yes’ assuming that this was likely to be one of the established cell lines (except if mentioned that this was not thought to be clinically relevant in which case we categorized as ‘no’).

In the second step, for drugs classified as ‘yes’ or ‘uncertain’ (as substrates or inhibitors) in the first step, we then classified evidence of clinical relevance as ‘yes’, ‘no’ or ‘uncertain’ based on in vivo evidence. For substrates, we classified evidence of clinical relevance based on changes in the AUC and/or adverse effects taking into account the specific pharmacokinetics of the particular object drug in question such as the therapeutic window (i.e. no global AUC cut-off for substrates). For a drug to qualify for categorization ‘yes’ as an inhibitor, we used an increase in AUC for a model P-gp substrate (digoxin, talinolol, fexofenadine or dabigatran etexilate) of at least 25% as cut-off for assessing clinical relevance [32]. For increases in AUC < 25% we classified as ‘no’, and if results were conflicting, based on a non-model substrate, or if no relevant in vivo evidence was identifiable or otherwise unclear, we categorized as ‘uncertain’.

3.3 Inducers

Since in vitro methods for studying P-gp induction do not currently provide clear evidence, we based classification of inducers solely on in vivo evidence of clinical relevance as ‘yes’, ‘no’ or ‘uncertain’—in the latter case, such as if a drug has been observed to be an inducer of enzymes via nuclear receptors such as PXR, but where no direct in vivo evidence was found [31, 43]. To qualify for classification as ‘yes’, we required a >20% reduction in AUC with digoxin, talinolol, fexofenadine or dabigatran etexilate as substrates [32]. If less was observed, we classified as ‘no’, and if evidence was conflicting or otherwise unclear, or other substrates had been used, we classified as ‘uncertain’. A few drugs were both inhibitors and inducers of P-gp with inhibition being dominant in the acute phase and induction in the chronic phase (rifampicin, ritonavir and tipranavir).

3.4 Notes on Interpretation of Tables

In the main tables (Tables 1, 2, 3, 4, 5), we highlight the drugs for which the evidence regarding the potential for clinically manifest P-gp-mediated DDIs was most compelling; that is, substrates and inhibitors classified as either (in vitro evidence–in vivo evidence): ‘yes’–’yes’, ‘uncertain’–’yes’, or ‘yes’–’uncertain’, and inducers classified as ‘yes’. Drugs with the remaining combinations of in vitro and in vivo classifications are presented in five similar tables in the ESM (Supplementary Tables 1–5). For drugs classified as ‘no’ in the first step based on in vitro evidence (substrates or inhibitors), no evaluation of in vivo evidence was undertaken (assuming that transporters other than P-gp would be responsible for any in vivo evidence).

Before using the results from the present review to guide daily clinical practice, a number of challenges regarding classification and interpretation merit discussion. For instance, an effect attributed to P-gp may in fact be driven by other (known or unknown) drug transporters or drug-metabolizing enzymes and isolating an effect attributable to P-gp may not always be possible. For this reason, if there was evidence qualifying the drug as a P-gp substrate in vitro (classification ‘yes’) and in vivo evidence of a clinically relevant change in AUC, which may be driven by P-gp but also by other enzymes or transporters, we classified as ‘yes’ in the second step with respect to clinical relevance. To acknowledge shared contribution from P-gp and CYP3A4 in vivo, a column denoting whether a drug is a known substrate, inhibitor or inducer of CYP3A4 is included in the tables presented in the main paper. One example of this is the protein kinase inhibitor, dabrafenib, which is an in vitro substrate of human P-gp. Dabrafenib has a high oral bioavailability of 94.5% and a high metabolic clearance. When dabrafenib is co-administered with ketoconazole, which is a potent inhibitor of both CYP3A4 and P-gp, the AUC of dabrafenib increased by 57% [44]. However, given the high oral bioavailability and high metabolic clearance of dabrafenib, this increase in AUC is likely driven by the inhibitory effects on CYP3A4 decreasing the clearance of dabrafenib rather than by the inhibitory effects of P-gp on extent of absorption (cf. Supplementary Table 1, ESM). On the other hand, aliskiren (an antihypertensive drug, cf. Table 2) is an example of a P-gp substrate that is eliminated mainly as an unchanged compound in the faeces and where only approximately 1.4% of the total oral dose is metabolized [45]. Thus, in this case the 4- to 5-fold increase in aliskiren AUC with co-administration of ciclosporin A is likely driven by P-gp as opposed to by CYP3A4.

Cobicistat, a CYP3A4 inhibitor used for pharmacokinetic boosting of HIV-protease inhibitors, is an example of a newly marketed drug where the information available on results of in vitro testing for P-gp and CYP3A4 modulatory properties is more extensive as compared with older drugs. In addition, cobicistat is an example of a drug where the interpretation of P-gp modulatory status is especially complex. In vitro cobicistat has been shown to be both a substrate and an inhibitor of P-gp and CYP3A4 and according to our classification qualifies for classification ‘yes’ in the first step based on in vitro data [46]. With respect to concurrent in vivo administration with digoxin, AUC of digoxin is, however, only increased by 8% with co-administration of cobicistat. Thus, in the second step based on in vivo data, cobicistat classifies as ‘no’ as an inhibitor and is consequently mentioned in Supplementary Table 3 (ESM) [46]. Of note, the manufacturer still recommends monitoring digoxin levels (and also levels of dabigatran etexilate) on co-administration [47]. While cobicistat is a (high-permeability) P-gp substrate in vitro [46], in vivo evidence is sparse. According to the SmPC, concurrent administration with inhibitors or inducers of CYP3A4 (i.e. potentially also P-gp) is warned against. These interactions have not always been studied in vivo (e.g. azole antibiotics) but are based on theoretical extrapolations and the disentangling of the interaction potential may further be complicated by the fact that cobicistat is administered for clinical purposes as a pharmacokinetic enhancer together with the HIV-protease inhibitors darunavir or azatanavir or with the HIV-integrase inhibitor elvitegravir [47, 48]. Thus, the individual effect of P-gp on cobicistat may be hard to isolate, and this is why, based on the current evidence, we have classified evidence of clinical relevance for its status as substrate as ‘uncertain’ (cf. Table 3).

Ketoconazole is an antifungal agent and a dual inhibitor of CYP3A4 and P-gp. However, the clinical relevance of this for potential in vivo DDIs appears to be dose dependant. For instance, when ketoconazole is administered with digoxin (200 mg per day for 4 days), a 9% increase in AUC has been observed [49]. Further, for administration of fexofenadine, an 8% decrease in fexofenadine AUC has been observed with pretreatment with ketoconazole 200 mg daily for 5 days [50], but a 164% increase with coadministration of ketoconazole 400 mg once daily [51] (cf. Table 3). Likewise, for coadministration of ketoconazole 400 mg with dabigatran etexilate, the AUC of the latter increases by approximately 2.5-fold and the combination is contraindicated according to the manufacturer [52].

4 Main Tables of Drugs with P-Glycoprotein-Modulatory Properties Classified According to In Vitro and In Vivo Evidence

See Tables 1, 2, 3, 4 and 5.

Table 1 Anticancer and immunomodulatory drugs with p-glycoprotein modulatory properties classified according to in vitro and in vivo evidence
Table 2 Cardiovascular drugs and anticoagulants with p-glycoprotein modulatory properties classified according to in vitro and in vivo evidence
Table 3 Drugs for infectious diseases with p-glycoprotein modulatory properties classified according to in vitro and in vivo evidence
Table 4 CNS drugs, drugs for gastrointestinal or endocrine disorders with p-glycoprotein modulatory properties classified according to in vitro and in vivo evidence
Table 5 Miscellaneous drugs with p-glycoprotein modulatory properties classified according to in vitro and in vivo evidence

5 Clinical Implications

Once a potential DDI has been identified, the next step is to predict the likelihood that this may result in a clinically manifest DDI and decide on the appropriate measures to be taken. These could span from no additional measures (i.e. if the potential DDI is rendered harmless), to additional monitoring (e.g. treatment with closer than usual follow-up with appropriate laboratory tests, physical exams or therapeutic drug monitoring), dose adjustment, staggered administration or in some cases frank contraindication coupled with a search for alternative treatments. Appropriate measures to be taken depend on a number of factors relating to characteristics of both the precipitant and object drugs as well as patient characteristics.

5.1 Considerations Relating to Individual Drug Characteristics

5.1.1 Dose

For P-gp, dose-dependent DDIs may occur. In particular, loperamide, an anti-diarrhoeal agent that reduces motility of the gut by interfering with the opioid receptor in the gut wall [320], is an example of a P-gp substrate for which the clinical relevance of a potential DDI is dose dependent. Sadeque et al. [276] conducted a study in eight healthy volunteers who received a single dose of loperamide 16 mg in the presence of either quinidine 600 mg (P-gp inhibitor) or placebo. The study showed a significant difference in the occurrence of respiratory depression in individuals who were administered quinidine as opposed to placebo. A subsequent study investigated the effects of quinidine and HM30181 (third-generation P-gp inhibitor with a very low bioavailability of 0.3%) on the pharmacokinetics and pharmacodynamics of a single oral dose of loperamide 16 mg [274]. Both inhibitors resulted in a clinically relevant increase in loperamide AUC as compared with loperamide on its own (HM30181 48% increase and quinidine 120% increase), and for concomitant administration of quinidine a marked reduction in pupil size was also seen. Concurrent administration of loperamide 16 mg as a single dose with ketoconazole resulted in a 5-fold increase in the plasma concentration of loperamide but no pharmacodynamic effect as assessed by pupillometry [320]. This may be explained by a differential level of P-gp inhibition in the gut and the CNS, due to differences in the inhibitor concentration at the two sites, implying that a clinically relevant increase in plasma AUC due to administration of an oral P-gp inhibitor may not transfer directly to a clinically manifest CNS affection. Thus, while the transport of loperamide across the blood–brain barrier and into the CNS is increased by P-gp inhibition [274, 321], the clinical relevance of this has been disputed [275]. Evidence to support the presence of a clinically relevant interaction between loperamide and P-gp inhibitors is insufficient when loperamide is taken at the recommended dose, which according to the SmPC is 2 mg at a time for multiple doses [320]. However, the half-life of loperamide is 11 h, which following multiple 2-mg doses may result in drug accumulation, and thus toxicity cannot be excluded in case of repeated administration or abuse [322]. We have therefore chosen to classify loperamide as ‘yes’ with respect to both in vitro and in vivo evidence (substrate, cf. Table 3).

Finally, for classification of some of the CNS drugs as P-gp inhibitors, based on in vitro studies, values of I2/IC50 may be above or below the cut-off of 10 (cf. Fig. 2) depending on the doses recommended for the various therapeutic indications. Thus, for higher doses, in vitro evidence may indicate potential for a clinically relevant DDI whereas this may not be the case for lower doses. In these instances, we have classified these drugs as ‘yes’ with respect to inhibitory status (in vitro evidence) but have mentioned in the table that this may indeed not be clinically relevant for indications with doses in the lower range (cf. Tables 2 and 4).

5.1.2 Therapeutic Index of Object Drug

Potential DDIs are more likely to become clinically manifest in the setting of a drug with a narrow therapeutic index, that is, a short span from occurrence of therapeutic effect to side effects. Examples of P-gp substrates with a narrow therapeutic index are digoxin and dabigatran etexilate. On interpretation of results presented in the tables of this paper, it is therefore important to take into account that while concurrent administration of a given P-gp inhibitor with digoxin may produce AUC changes determined to be clinically relevant for digoxin, this may not be the case for other P-gp substrates with a wider therapeutic index. Thus, results based on narrow therapeutic index substrates such as digoxin should not be over-extrapolated to other P-gp substrates [30, 323].

5.1.3 Potency of the Precipitant Drug in Inhibiting P-gp

Potency of the P-gp inhibitor relative to its unbound plasma levels or estimated intestinal levels after therapeutic doses is another factor influencing recommendations for prescription with P-gp substrates. For example, for edoxaban, a direct oral anticoagulant, dose reduction is recommended with potent P-gp inhibitors such as ciclosporin A and ketoconazole but not with the less potent inhibitor verapamil [162]. Also, for another direct oral anticoagulant, dabigatran etexilate, the impact on the AUC differed quite substantially according to the studied inhibitor in vivo. For clarithromycin, the AUC increased by 19% and C max (peak drug concentration) by 15%, substantially less than with ketoconazole, for example, for which the AUC increased by 138% [150].

5.1.4 Contributions of Other Enzymes or Transporters to the Pharmacokinetics of the Substrate Drug

For many drugs, more than one enzyme or transporter influences the pharmacokinetics of the drug and to a varying extent. In this context, the proportion of a P-gp substrate depending on P-gp transport and metabolization or transport by other enzymes or transporters, respectively, needs to be evaluated together with the modulatory properties of the precipitant drug on the same enzymes and transporters. Examples of this relating to P-gp and CYP3A4 are mentioned in Sect. 3.

5.1.5 Dual Effects Over Time

Some drugs may be both inhibitors and inducers of P-gp depending on the timing. Rifampicin is an example of such a drug. If, for example, rifampicin and digoxin are administered in temporal proximity, the inhibitory effect of rifampicin is clinically manifest whereas the induction effect takes over with time [240]. Likewise, for both fexofenadine and digoxin, the inhibitory properties of ritonavir are predominant in the acute phase and lessen with time as the induction takes over [242, 324].

5.1.6 Route of Administration

Route of administration of the drug might also play a role. If the site of potential interaction is predominantly at the intestinal level affecting absorption, this would not be expected to affect the pharmacokinetics of a drug administered parenterally.

5.1.7 Highly Permeable Drugs

By being an efflux pump present at the apical level of the intestinal membrane, P-gp plays a role in limiting drug absorption. However, the extent to which P-gp limits absorption of a given drug depends also on how much of the drug is absorbed by passive permeability (as well as transport by other transporters). Generally, a P-gp substrate that is highly soluble, highly permeable and/or highly metabolized is less likely to be affected to a clinically relevant extent by a P-gp inhibiting drug [30]. For example, midazolam, which is highly lipophilic, has generally not been thought of as a P-gp substrate, perhaps due to its high intestinal permeability. But there is some evidence that midazolam actually exhibits characteristics of a highly permeable P-gp substrate [325]. Still, in relation to the clinical setting, this would not be of importance with co-administration of a P-gp inducer or P-gp inhibitor since midazolam passes the intestinal membrane anyway due to the high permeability. Midazolam is, however, a CYP3A4 substrate, and in that context interactions would need consideration.

5.1.8 Examples of Drug–Drug Interactions with Complex Interpretation

The new oral hepatitis C drugs are examples where interpretation of an interaction potential may be particularly complex due to multiple drugs administered at the same time. For example, ledipasvir and sofosbuvir are antivirals administered together for the treatment of chronic hepatitis C [220]. Both drugs are in vitro substrates of P-gp, and ledipasvir is an in vitro inhibitor of P-gp. While the inhibitory properties of ledipasvir have not been studied with model substrates such as digoxin or dabigatran etexilate in vivo, it has been shown to cause a more than two-fold increase in sofosbuvir AUC, which is likely driven by the combined effects of P-gp and BCRP modulation [219]. Accordingly, we have classified the inhibitory status of ledipasvir as ‘yes’ with respect to both in vitro and in vivo evidence (cf. Table 3).

Another complex example is the interpretation of the potential for DDIs with ombitasvir, paritaprevir and ritonavir given together (with or without dasabuvir) as part of treatment for chronic hepatitis C. Paritaprevir, ritonavir and dasabuvir all inhibit P-gp in vitro. In vivo, only ritonavir has been administered with digoxin alone, which resulted in an 86% increase in digoxin AUC [243]. When the combination of paritaprevir, ritonavir, ombitasvir (not a P-gp inhibitor in vitro) and dasabuvir are administered together with digoxin in vivo, digoxin AUC increases by 16% compared with digoxin alone [326]. However, when paritaprevir, ritonavir and ombitasvir are administered with digoxin (but without dasabuvir), digoxin AUC increases by 36% compared with digoxin alone [237]. Since the digoxin AUC decreases compared with ritonavir administered alone, a clinically relevant inhibitory potential of dasabuvir and paritaprevir seems unlikely, which is why we have chosen to classify the P-gp inhibitory status of both as ‘yes’ with respect to in vitro evidence and ‘no’ with respect to in vivo evidence (cf. Table 3 and Supplementary Table 3, ESM).

5.2 Clinical Considerations Relating to Patient Characteristics

5.2.1 Organ Impairment and Special Populations

Renal and hepatic impairment are important considerations when evaluating potential DDIs. That is, in patients with impaired hepatic or renal function, the threshold for a DDI becoming clinically manifest is lower if the drug depends highly on that particular organ for elimination. Mirabegron, indicated for treatment of patients with overactive bladder, is an example of a P-gp substrate where the manufacturer recommends caution only in patients with impaired renal or hepatic function. In patients with normal renal or hepatic function, mirabegron can be taken with inhibitors of CYP3A4 and/or P-gp without dose adjustment [298]. Also, for the direct oral anticoagulants (dabigatran etexilate, rivaroxaban, apixaban and edoxaban), dose reduction is recommended in patients with impaired renal function due to a decreased elimination of the drug and thus increased risk of toxicity [138, 150, 162, 177].

Particular consideration should also be taken among special populations such as the elderly and paediatric patients, where individual pharmacokinetics are different compared with healthy adults [32]. When new drugs are approved they will usually not have been tested in an elderly or paediatric population, meaning that knowledge regarding safety and efficacy in these populations depends on post-marketing evidence. The elderly are at increased risk of manifest DDIs for a number of reasons such as age-related pharmacokinetic and pharmacodynamic changes but also because polypharmacy, comorbidities and malnutrition are more prevalent with age [327]. For apixaban, the SmPC states that the AUC was increased by 32% in individuals >65 years old compared with younger individuals (no difference in C max). If the patient also receives acetylsalicylic acid, this additionally increases the risk of bleeding [138]. For a more extensive review of managing DDIs among the elderly we refer to other reviews such as the one by Mallet et al. [327].

For the drug-metabolizing enzyme CYP3A4 there is emerging evidence of a distinct ontogeny of the enzyme, with low activity at birth reaching adult levels during the first years of life [328331]. For P-gp, evidence is sparser. Nevertheless, there is some human evidence of a pattern with high placental expression levels of P-gp in early pregnancy that decrease with increasing gestational age [332, 333]. Interestingly, paralleling this developmental decrease in placental expression, the P-gp expression of the human blood–brain barrier seems to increase with gestational age [334, 335]. Together this may serve to protect the foetal brain from noxious substances. However, further research is warranted to fully understand whether any difference in developmental P-gp expression in humans plays a role in drug disposition in the foetus and in neonates [336].

5.3 Ways to Handle Drug–Drug Interactions in Clinical Practice

When a DDI with potential clinical relevance has been identified, there are various ways to handle this. At the extremes, there is either no need for additional measures (the potential lack of treatment effect or increased drug effect is rendered harmless) or the medications should not be used concomitantly and the clinician should search for alternative treatments. However, in between those poles various other options for appropriate management exist. These are reviewed and exemplified in the following subsections.

5.3.1 Treatment with Additional Monitoring

Additional monitoring of either the main treatment effect or side effects is a possibility when a potential DDI has been identified. This can be by means of monitoring relevant paraclinical measures or performing additional physical examinations. For a few drugs (e.g. digoxin), an option may be therapeutic drug monitoring where the concentration of the drug is measured in plasma.

Carvedilol is an example of a drug where the evidence regarding the magnitude of the effect on digoxin AUC is heterogeneous. The changes in digoxin AUC with co-administration of carvedilol have been reported to range between a 19% and a 56% increase in various studies [146, 148]. According to the SmPC, the concentration of digoxin is increased by approximately 15% following co-administration with carvedilol, and increased monitoring of digoxin levels at onset, dose adjustment or cessation of carvedilol is recommended [144]. Thus, this is an example of a drug where the interpretation of the potential for a clinical manifest DDI is not straightforward and where increased monitoring may be a solution with respect to clinical management.

5.3.2 Dose Adjustment

Another option is dose adjustment. If an inhibitor or inducer of P-gp is administered with a substrate, then the dose of the substrate may be decreased or increased, respectively. Unfortunately, there are currently no expert dosing guidelines that provide clinicians with information on the dose to use in patients receiving P-gp substrates and concomitant P-gp inhibitors or P-gp inducers.

5.3.3 Staggered Administration

Staggered administration of the object and precipitant drug is another option. As an example, ibrutinib (irreversible Bruton’s tyrosine kinase inhibitor) is an in vitro inhibitor of P-gp with an IC50 of 2.15 µg/mL. At the proposed oral dose of 560 mg and assuming that the drug is taken with 250 mL of water, the concentration in the gut would be 560,000 µg divided by 250 mL and [I2]/IC50 would be 1042 (i.e. much greater than 10). Thus, the potential for a P-gp-mediated DDI cannot be excluded (cf. Fig. 2). However, ibrutinib is predicted to be quickly absorbed (in <2.5 h). Thus, the potential for a clinically relevant P-gp-mediated DDI would be minimized by staggering the dose of ibrutinib and a P-gp substrate by at least 2.5 h [96].

Co-administration of dabigatran etexilate, a direct oral anticoagulant, with the calcium antagonist verapamil is another example of a potential DDI involving P-gp that may be avoided with staggered administration. Härtter et al. have conducted a study to examine the potential for pharmacokinetic and pharmacodynamic interactions between dabigatran etexilate and verapamil [149]. The authors reported a 2.5-fold increase in dabigatran etexilate exposure (as determined by the AUC) when administered 1 h after a single dose of verapamil immediate-release formulation. This was reduced to an approximate 1.8-fold increase in AUC with the extended-release formulation of verapamil. However, administration of dabigatran etexilate 2 h before verapamil did not significantly increase exposure to dabigatran etexilate (<20% increase in AUC). This is explained by absorption of dabigatran etexilate being completed within 2 h [150].

5.3.4 Paused Treatment

Paused treatment may be a solution in those instances where lack of treatment with one of the drugs contributing to the DDI is rendered less harmful. An example of this could be the case of primary prevention with atorvastatin in a patient where a shorter course of a given drug with P-gp inhibitory properties is needed. In this case, paused treatment with atorvastatin during treatment with the perpetrator drug could be a solution. Anticancer drug interactions represent another case in which paused treatment may work. That is, during cancer treatment, perpetrator drugs can be stopped and then resumed following cessation of treatment.

5.4 Discussion of Limitations

We aimed to evaluate, as much as possible, a complete list of all drugs with P-gp modulatory properties. Nevertheless, with the approach described in the Sect. 3, there will inevitably be drugs that are P-gp modulators that have not been included for evaluation as part of this review. Many of these ‘false negatives’ probably represent newer drugs (except for very newly marketed drugs, as mentioned in Subsect. 3.1).

Also, the evidence base is of varying methodological quality; in general, the P-gp modulatory status of many of the drugs may not always have been investigated under ‘model conditions’. Regarding Caco2 cells, two aspects merit further discussion. First, variation in transporter expression and function of the cells may be seen depending on the exact laboratory conditions that the cells have been maintained under [337]. Second, Caco2 cells exhibit endogenous activity of transporters other than P-gp, for example multidrug resistance-associated proteins (MRPs) or BCRP that may thus have confounded the observed drug transport [338]. To address this issue, Caco2 cells transfected with short interfering RNA (siRNA) to knock down the contribution of other specific transporters may be used [337]. Of note, in the present review we did not make a distinction between transfected versus untransfected Caco2 cells for studies forming the basis of the classification of the individual drugs. However, at least for the underlying references from the scientific literature, all studies that included Caco2 cells were based on untransfected cells.

For the newer drugs, it may also be that specific combinations of drugs have been investigated together because that reflects the intended clinical use (e.g. see the new oral hepatitis C drugs as discussed in Sect. 5). Overall, our chosen classification strategy implies that classifications may change with the emergence of new evidence (e.g. from ‘uncertain’ to ‘yes’ with respect to both in vitro and in vivo evidence). Our chosen classification criteria were decided on with the aim of maximizing objectivity. However, since a consensus agreement was also included in the classification and since model evidence was not always found, this implies a certain degree of subjectivity which should be kept in mind on interpretation of the tables. We therefore provide the references that the classification is based upon for each drug along with the classification (see main tables and supplementary tables) and discuss examples of some of the interpretations of classifications in Sect. 3 (methods) and in Sect. 5 (clinical implications) of the paper.

6 Conclusions and Perspectives

Drug transporters such as P-gp have been increasingly recognized as active players in DDIs. Consequently, the potential for P-gp-mediated DDIs (together with the potential for interactions with other transporters or drug-metabolizing enzymes) is now more systematically investigated as part of the approval of new drugs. This has resulted in an increased amount of information on DDI potential available in the drug labels/SmPCs, including an increase in the complexity of the information. Prescription recommendations for individual drugs on the approved product label may not always reflect data from well characterized DDI studies; for example, a contraindication may be included in the label for various medico-legal reasons rather than due to clear-cut evidence of harm. The real challenge for the prescriber is therefore to navigate the various systems providing electronic decision support (causing potential alert fatigue) and strike the happy mean between an overcautious prescription practice and a practice where DDIs are not considered at all. Many interacting drugs can be safely taken together when appropriate measures are taken. The challenge still lies in narrowing the focus on those DDIs that are likely to be clinically manifest. In this context, more research in those drugs that interfere with more than one transporter or enzyme is needed to aid clinical decision making in this area. Further expert guidelines for dosing adjustments in the setting of concurrent administration of P-gp substrates and inhibitors and inducers, respectively, are needed.