When is protein binding important?


  • Dedicated to Leslie Z. Benet


The present paper is an ode to a classic citation by Benet and Hoener (2002. Clin Pharm Ther 71(3):115–121). The now classic paper had a huge impact on drug development and the way the issue of protein binding is perceived and interpreted. Although the authors very clearly pointed out the limitations and underlying assumptions for their delineations, these are too often overlooked and the classic paper's message is misinterpreted by broadening to cases that were not intended. Some members of the scientific community concluded from the paper that protein binding is not important. This was clearly not intended by the authors, as they finished their paper with a paragraph entitled: “When is protein binding important?” Misinterpretation of the underlying assumptions in the classic work can result in major pitfalls in drug development. Therefore, we revisit the topic of protein binding with the intention of clarifying when clinically relevant changes should be considered during drug development. © 2013 Wiley Periodicals, Inc. and the American Pharmacists Association J Pharm Sci 102:3458–3467, 2013

Abbreviations used:

AL, albumin; AAG, alpha-1-acid glycoprotein; CLH, hepatic clearance; CLint,in vitro, intrinsic clearance in vitro; CLint,in vivo, intrinsic clearance in vivo; DDI, drug–drug interaction; fub, unbound fraction in blood; fuinc, unbound fraction in incubation; fuliver, unbound fraction in liver; fup, unbound fraction in plasma; fup-app, unbound fraction in plasma apparent; IVIVE, in vitroin vivo extrapolation; Ki, inhibition constant; Ki,u, unbound inhibition constant; MIC, minimum inhibitory concentration; PLR, plasma-to-liver concentration ratio; Q, liver blood flow; RBP, blood-to-plasma ratio.


Over 10 years ago, Benet and Hoener1 published a citation classic paper entitled, “Changes in plasma protein binding have little clinical relevance”. In this paper, the authors elegantly presented the case that changes in plasma protein binding will usually not affect the exposure of free, pharmacologically active drug. They convincingly concluded that drug–drug interactions (DDIs) due to displacement from plasma protein binding sites may still occur but will not result in changes in unbound drug exposure and have consequently little clinical relevance. Before the Benet and Hoener publication, Clinical Pharmacology textbooks included a chapter on DDIs caused by protein binding displacement and presented the warfarin–phenylbutazone interaction as a typical example.2 In this example, a causal link between in vitro experiments and clinical bleeding events was established, arguing that coadministration of phenylbutazone leads to an increase in free, active warfarin concentrations due to protein binding displacement. However, Benet and Hoener provided in their paper a theoretical argument, which showed that this conclusion was wrong and that protein binding displacement of warfarin does not result in free, unbound drug concentrations and pharmacological activity. In fact, it turned out that there is a second DDI involving enzyme inhibition by phenylbutazone that was responsible for the resulting increase in unbound drug exposure.3–5

Benet and Hoener were also very detailed in pointing out the assumptions and limitations of their conclusions. Unfortunately, the title of the paper was—and still is—sometimes misinterpreted by those who overlook the detailed presentation of the underlying assumptions and limitations of the conclusions that are discussed in the seminal work.3,4 The misinterpretation is to erroneously think that protein binding in general is of little relevance. Benet and Hoener anticipated this wrong interpretation and stated: “This conclusion should not be extrapolated to suggest that measurements of protein binding are not important in drug development.” These limitations are discussed in the last paragraph of their paper entitled “When is protein binding important?”. It is the purpose of this paper to revisit these limitations and expand on some protein-binding related aspects that can be major pitfalls in drug development if ignored.


Benet and Hoener pointed out that there is one exception to the rule that unbound drug exposures are not significantly altered by changes in plasma protein binding, that is, high-extraction drugs following parenteral administration.1 For high extraction drugs with hepatic clearance, total drug clearance is approximately equal to liver blood flow and, hence, independent of plasma protein binding. Average total drug concentrations and total area under the curve (AUC) will not change if protein binding is altered. However, the scenario changes when considering unbound drug concentrations as they are proportional to the fraction unbound in plasma and, hence, sensitive to changes in plasma protein binding. One example of this concept is the inhaled corticosteroid, ciclesonide. The drug is an inactive prodrug that is administered by inhalation and quickly converted into its active metabolite desciclesonide.6–8 Ciclesonide's clearance is close to liver blood flow, in spite of its very high plasma protein binding of approximately 99%. Consequently, it is a clear example illustrating that high plasma protein binding does not automatically result in clearance restrictions. This example further shows that intrahepatic reequilibration between bound and free drug occurs so fast that the vast majority of drug can be metabolized as the blood crosses the liver. Although protein binding is not rate-limiting for hepatic clearance for this type of drug, it still controls the magnitude of the unbound plasma concentrations. Concentrations are much lower than those of other inhaled corticosteroids with comparable clearance,6–8 which leads to much lower systemic side effects as indicated by lower suppression of endogenous cortisol as well as lack of growth retardation in children.9


Although it is true that the total average steady-state concentration after multiple dosing as well as the total AUC will not change when clearance is unaltered, the shape of the plasma profile will be different if the volume of distribution changes due to protein binding changes. This change in volume of distribution will depend on the relationship of drug binding in plasma and in tissues. If the fraction bound in plasma increases more than that in tissues, the volume of distribution will go down. As a result, total peak plasma concentrations will increase, the half-life will be shorter and total trough concentration in plasma will be lower. The opposite will be true if the fraction bound in the tissues increases more than that in plasma: the volume of distribution will increase, total peak plasma concentrations will be lower, the half-life will be longer and total trough concentrations in plasma will be elevated. In all of these scenarios the total average steady-state concentration and the total AUC will remain unaltered if clearance does not change. A more detailed analysis of these scenarios can be found elsewhere.10 Depending on the relationship between pharmacokinetics and pharmacodynamics (PK/PD), these different plasma drug profiles may result in different clinical outcomes as it is not always the AUC that correlates best with clinical outcome. For example, it is well known that the antiinfective activity of beta-lactam antibiotics correlates best with the time that the unbound plasma concentrations remain over the respective minimum inhibitory concentration (MIC) of the microorganism that is to be eradicated rather than the AUC.11


Therapeutic drug monitoring (TDM) is another potential source for misinterpreting the impact of plasma protein binding as was pointed out already in great detail by Benet and Hoener.1 TDM is usually performed and therapeutic target ranges are defined on the basis of total (free + bound) plasma concentrations. Although it is now well established that only the unbound drug concentration is responsible for the PD activity.10–13 In most cases, reporting total drug levels does not pose a problem as total concentrations (e.g., 10–20 mg/L for total phenytoin) directly correspond to unbound target concentrations (e.g., 1–2 mg/L for unbound phenytoin based on a fraction bound of 0.9) if the fraction bound in plasma is constant over the therapeutic range. However, if protein binding changes due to disease or DDIs, this relationship is no longer valid. Although the unbound target range is still valid, the total target range will differ. The simple solution in this situation is to monitor unbound drug concentrations, which will give unambiguous values to directly compare with the desired unbound target range. However, not all clinical laboratories are prepared or willing to measure unbound concentrations. As a second-best solution for this situation, Winter proposed a work-around using the measured albumin concentration to estimate the degree of plasma protein binding and adjust the measured total phenytoin concentration in a patient with lowered albumin concentrations to the respective concentration in a patient with normal albumin concentrations so that the number then can be compared with the established target range for total phenytoin concentrations.14

However, this approach has many assumptions and is clearly inferior to the direct assessment of unbound drug levels. In any case, the recommended target ranges for drug level monitoring are no longer valid in cases when protein binding changes.


Probably the most problematic and most common pitfall from ignoring protein binding occurs when extrapolating PK/PD parameters from in vitro studies and preclinical studies to the clinical situation. This problem is also mentioned in the last part of the original paper by Benet and Hoener, but is frequently forgotten or not sufficiently appreciated.1 This issue has been most controversial in the arena of antiinfective drug development compared with other drug classes. The explanation is historical: The PD properties of new antibiotics are usually assessed by microbiologists and provided to the rest of the drug development team in the form of MICs. MICs are defined as the drug concentration that prevents visible growth of responding bacteria after overnight incubation. These MIC values are then used as target concentrations in plasma. However, it is crucial that in these extrapolations the target concentration is based on the unbound drug concentration, not on the total drug concentration.12,15 Clinical trials have failed in cases where protein binding was not considered or was considered too late in development.16 This is not new and has been reported for a long time.13,17 However, it is surprising that this relationship has appreciated very slowly. Indeed, relatively new antibiotics have been introduced with the claim that “protein binding does not matter.” The typical experiment to support this claim is an MIC or a microbiological kill curve for the drug of interest with and without the presence of protein (usually albumin), where no difference in the bactericidal behavior is observed. Unfortunately, there are some potential pitfalls that can lead to this conclusion. Figure 1 shows a set of kill curves that were obtained after incubation of Staphylococcus aureus with different concentrations of the beta-lactam antibiotic oxacillin.18 This drug has a plasma protein binding of 70%. Plotted are the numbers of bacteria (colony-forming units, cfu/mL) as a function of time. Three drug concentrations were investigated that correspond to 0.25×MIC, 1×MIC, and 4×MIC. The line without symbols shows the control growth of the bacteria in absence of the drug. The filled squares show that both 1×MIC and 4×MIC showed a bactericidal effect, which were of comparable magnitude. 0.25×MIC had a very minor effect. The outcome changed significantly when a physiological amount of albumin was added (40 g/L) as represented by the filled circles. Clearly, the effect of a concentration equal to 1×MIC was attenuated in the presence of albumin and absent in a concentration equal to 0.25×MIC. However, no difference was observed for the high concentration of 4×MIC when albumin was added. The explanation can be shown by performing an additional experiment. The open squares represent the kill curves that are obtained when drug is added to the bacteria in a concentration equivalent to the unbound concentration in the albumin experiment. Hence, these samples do not contain albumin. In all three scenarios, the resulting kill curves (open squares) are virtually identical to those observed with the higher total concentration in the presence of albumin (filled circles) supporting the hypothesis that it is the unbound drug concentration that drives bacterial kill. The data also show that in case of the highest concentration studied (4×MIC), the total and the unbound concentration result in the same outcome indicating that the maximum kill rate has been reached. It is not surprising that in this scenario, the addition of albumin will not have an impact because the effect is at a maximum anyway. Hence, the conclusion that the effect of the drug is not affected by protein binding is not correct, as can be seen by the data after incubation with a concentration of 1×MIC.

Figure 1.

Bacterial kill curves of S. aureus after incubation with different concentrations of oxacillin (left panel 0.25×MIC; middle panel 1×MIC, right panel 4×MIC; MIC, minimum inhibitory concentration).18 Shown are the number of baceteria (colony forming units, cfu/mL) over time. For each concentration, four curves are shown: control growth (no symbols); drug without albumin (filled squared symbols); drug with albumin (filled circles); drug without albumin in a concentration equivalent to the unbound concentration in the experiment of drug with albumin.

There have been many other studies showing that the driver of pharmacologic activity is the unbound drug concentration and not the total concentration. To show this unambiguously requires scenarios where total and unbound concentrations are not proportional, as only then the correlation between exposure and response will differ. One example is a study by Tawara, who investigated the therapeutic efficacy of a series of cephem antibiotics in mice with experimentally induced intraperitoneal infections with Klebsiella pneumoniae.19 Figure 2 shows the correlations that were obtained when comparing the clinical efficacy as expressed by 1/ED50 with the MIC-normalized PK AUC based on total drug concentrations (left plot, poor correlation) and unbound drug concentrations (right plot, excellent correlation). Similar results were obtained for Staphylococcus aureus infections. The results indicate that unbound drug exposure is superior in predicting therapeutic outcome to total drug exposure.

Figure 2.

Correlation of therapeutic potency (ED50) of a number of cephem antibiotics in an intraperitoneal Klebsiella pneumoniae infection model in mice with MIC-normalized total drug exposure (left) and unbound drug exposure (right).19

Figure 3 shows similar examples comparing two PK/PD indices to both total and unbound exposure of two classes of antibiotic.11,20 For quinolones, the most appropriate PK/PD index is the ratio of AUC/MIC. Figure 3a (top) shows the necessary AUC/MIC (both total and unbound exposure) for a static dose of different quinolones to treat an experimental Streptococcus pneumoniae infection. The data show that the higher plasma protein bound quinolones (gemifloxacin, garenoxacin) require higher total drug exposure than the other quinolones, whereas the unbound exposure that correlated with stasis was identical for all quinolones investigated. The magnitude of the unbound AUC24/MIC was approximately 30, which corresponds to an average unbound drug concentration slightly above the MIC. The AUC24 is the steady-state AUC over 24 h. A similar picture can be seen in Figure 3b for a different class of antibiotics (cephalosporins) and a different PK/PD index (Time>MIC in% of dosing interval) and different bacteria (Klebsiella pneumoniae). The data show that the higher plasma protein bound cephalosporins (ceftriaxone, cefonicid) require longer times above the MIC than the other cephalosporins, whereas the time above MIC for the unbound plasma concentration that corresponded to stasis was identical for all cephalosporins at approximately 35% of the dosing interval. Again, these data are consistent with the hypothesis that it is the unbound drug concentration that drives the pharmacological activity.

Figure 3.

PK/PD indices resulting in stasis for a number of quinolones (top) and cephalosporins (bottom).11 For quinolones, total and free AUC/MIC ratios for S. pneumoniae infections are displayed. For cephalosporins, the time above MIC within a dosing interval is displayed for total and free drug concentrations in plasma to treat Klebsiella pneumoniae infections.

Similar relationships have also been reported from other pharmacological classes. For example, Mandema compared the bound and unbound EC50 values of a number of benzodiazepines with their respective receptor affinity.21 The results are shown in Table 1 and indicate that a much better correlation is observed when unbound concentrations are used. For example, flunitrazepam has a fourfold lower EC50 than midazolam based on total concentration suggesting higher potency. However, receptor binding studies revealed that midazolam has a higher affinity than flunitrazepam. Furthermore, midazolam has a much higher plasma protein binding than flunitrazepam. When correcting for protein binding, this difference in receptor affinity is reflected in the lower unbound EC50 for midazolam in comparison to flunitrazepam. These data also make clear the difference of interpreting protein binding data when looking at the effects on enzymatic or transporter-mediated turnover versus a reversible receptor binding interaction. In the first case, the enzyme or the transporter causes a change in unbound drug concentration, which in turn will cause a rapid reequilibration with release from previously bound drug that now is available for metabolism or transport. This reequilibration can occur very rapidly as seen in the example mentioned above of ciclesonide as a drug with 99% plasma protein binding and yet a hepatic clearance close to liver blood flow. In this case, the rapid metabolism produces a sink that strips the drug off the protein. It is conceptually important to see the difference between this scenario and that of a reversible receptor binding event (on/off diffusion) that does not cause to change the drug concentration significantly and does not produce a sink unless there is an excess of high affinity binding sites. This same difference is relevant in in vitro systems where both scenarios may exist. A competitive receptor binding assay or a typical MIC determination occur at constant drug concentrations with very little reequilibration, whereas an in vitro metabolism experiment with a rapid turnover rate may produce sink conditions that change the degree of protein binding during the experiment.

Table 1. Pharmacodynamic (PD) Potencies Expressed as EC50 of Unbound and Total Drug for a Number of Benzodiazepines21
 EC50(total) (ng/mL)EC50(free) (ng/mL)Ki (ng/mL)
  1. EC50 were determined from using EEG as a biomarker for PD activity. Also shown, benzodiazepine receptor affinities (Ki) and correlation with total and unbound EC50.

Correlation with Ki0.9160.998 

Just as in the body, also in in vitro systems drug binding can occur to change the unbound drug concentration and, hence, the PK or PD read-out. It has long been known that addition of proteins to in vitro samples can produce a “protein-shift”. Figure 4 shows the effect of adding albumin on the determination of the MIC of a number of penicillins.22 For each compound, the bars on the left reflect the MIC in broth, that is, in absence of proteins, and the bars in the middle reflect the MIC in presence of albumin. As albumin is added, the MICs will go up proportionally to the degree of protein binding (fb, fraction bound). However, if the unbound concentrations are determined for each of the MICs in presence of albumin (bars on the right), they agree well with those on the left representing absence of protein. Interestingly, it could be shown that adding albumin or other proteins in vitro is a risky business as the quality and properties of commercial protein can vary significantly.23,24 Therefore, the only reliable way to assess the degree of in vitro binding is experimental measurement. In vitro microdialysis seems to be a flexible method that allows measurement of in vitro unbound concentrations.23

Figure 4.

Effect of adding albumin on the determination of the MIC of a number of penicillins.22 For each compound, the bars on the left reflect the MIC in broth, that is, in absence of proteins, and the bars in the middle reflect the MIC in presence of albumin. As albumin is added, the MICs will go up proportionally to the degree of protein binding (fb, fraction bound). However, if the unbound concentrations are determined for each of the MICs in presence of albumin (bars on the right) they agree well with those on the left representing absence of the protein.


A popular strategy depending on these in vitro systems adopted by the pharmaceutical industry is high-throughput screening of new chemical entities (NCE) to efficiently identify new safe, effective compounds. Several criteria need to be met for such an NCE to be considered as a lead compound that can go into further development, as this next stage of research and development (R&D) is time consuming and expensive; failures need to be identified in the earliest stage possible. One of the most important PK parameters that can be evaluated for such compounds is hepatic metabolic clearance, as this parameter is linked to exposure to the drug, dosing interval, possible DDIs, and therefore all kinds of different efficacy and safety considerations. To meet this need, the industry has developed methods to efficiently determine hepatic metabolic clearance and evaluate clinical DDI.25 These are mainly in vitro assays, which are then used for in vitroin vivo extrapolations (IVIVE). However, a fundamental problem is that there does not seem to be a consensus on how to scale up in vitro measurements to human in vivo clearance. One of the major inconsistencies is the use of parameters for unbound fraction of drug, fu. It has been widely accepted in pharmacology that it is only the unbound drug that can exert biological effects, so in these IVIVE studies too, unbound concentrations in plasma, tissue, and assay need to be used for any kind of scale-up.26,27 However, recent publications on IVIVE are still not consistent in their use of these parameters, especially regarding nonspecific binding in the assay. Moreover, the guidelines from the United States Food and Drug Administration (US FDA) and European Medicines Agency (EMA) on IVIVE for DDI are not clear regarding this issue, and therefore it is important to get a consensus on this.28,29


The use of in vitro assays for extrapolation to in vivo clearance was first suggested by Rane et al.,30 who expressed the activity of the drug-metabolizing enzyme measured in the assay as intrinsic clearance (CLint,in vitro), calculated by Eq. 1.

equation image(1)

where Vmax is the theoretical maximum velocity of the metabolism and KM the Michaelis–Menten constant, the concentration that corresponds to half-maximum velocity. This CLint,in vitro in μL/(min mg proteins) is then scaled to in vivo intrinsic clearance, CLint,in vivo in mL/(min kg body weight) by a physiologically based scaling factor for 45 mg protein/g liver and 20 g liver/kg body weight. To calculate in vivo hepatic plasma clearance (CLH) from this parameter the authors incorporate factors to correct this metabolizing capacity for rate of drug delivery to the liver by the blood resulting in the following Eq. 2:

equation image(2)

where Q is the hepatic blood flow, RBP the blood-to-plasma ratio and fub and fup the fraction unbound drug in blood and plasma, respectively.

Thus, the importance of taking into account the unbound fraction of drug in blood or plasma in CLH was already suggested several decades ago. However, considering the paradigm that only unbound drug is available for the biological system, Obach suggested some 20 years later that not only binding in blood in vivo, but also binding of the drug in the in vitro assays might have an important effect on the extrapolation.31,32 As in vitro assays are not pure systems and nonspecific binding seems likely, the assumption that not all substrate molecules will bind the enzyme in a way permitting catalysis is reasonable. Obach argues that the KM in Eq. 1 measured in these assays is that based on free drug concentration, as bound drug is unavailable for metabolism. Thus, KM should be corrected for this nonspecific binding of drug to the assay by a binding factor termed “unbound fraction of drug in the incubation” (fuinc), which­ can be measured using equilibrium dialysis, ultracentrifugation or microdialysis. If we implement this factor in Eqs. 1 and 2, the result is Eq. 3 for CLH:

equation image(3)

In these publications, the correction for nonspecific binding seems to improve predictions of human clearance over the prediction using only a correction for fup.31,32 Further investigation33 systematically compared three approaches of calculating in vivo clearance, either (1) disregarding all binding, (2) only correcting for fup, or (3) correcting for both fup and fuinc. By now, it had become clear that the structural type of the drug (basic, neutral, or acidic) had an impact on the human clearance, and the three methods were evaluated per type of drug. The conclusion of the authors is that for neutral and basic compounds method (1) was preferred slightly over (2), and for acidic compounds method (3) was preferred. However, overall predictions when correcting for both binding factors are still not optimal and show a trend of underprediction.

Other groups later confirmed the impact in vitro nonspecific binding can have on the determination of for example KM and CLint,in vitro.34,35 Similar findings were performed relating to DDI studies, where nonspecific binding in vitro was found to affect Ki and IC50 values, which will ultimately affect the estimation of inhibitory potency in vivo.36–38 To assist IVIVE in a fast and effective way, some groups then developed models to predict incubational binding.39–41 These models have a relatively high accuracy and show that highest nonspecific binding is found in lipophilic compounds, especially when cationic at physiological pH. However, robust experimental measurements of fuinc are preferred.

Despite this reasonable rationale and increasing knowledge about nonspecific binding and evidence for the importance of incorporating the term fuinc into IVIVE, recent papers still question whether corrections for nonspecific binding should be made.42,43 This is due in part to suboptimal prediction from IVIVE with all binding factors incorporated. Additionally, contradictory publications have perhaps clouded the understanding of the subject, as argued by Grime and Riley.44 This has led to a situation where IVIVE calculations are carried out with inconsistent use of correcting for unbound drug concentrations. IVIVE is an important tool to predict clearance and DDIs in early development of drugs. But to minimize false positives (and kill promising compounds) or false negatives (and waste time and effort), IVIVE has to be accurate and consistent. Therefore, we propose to systematically adopt an accurate method with factors for unbound fractions for the use of IVIVE, optimizing the screening of NCEs. A similar appeal has been made by Grime and Riley in 2006, however, this does not seem to have become standard, and as shown earlier the inconsistencies remain to date.44 Recently, progress has been made in the methods to estimate in vivo CLH from in vitro experiments by taking a more mechanistic approach. A promising method has been published recently showing highly accurate results for predictions of CLH.45 This method should aid in the implementation of consistent use of unbound fraction in in vitro calculations of parameters. The authors suggest correcting the calculation for CLH not only for fup and fuinc, but also for a difference in drug ionization and concentration of binding proteins in plasma and liver. They state that IVIVE calculations of CLH should follow unbound drug concentrations in liver rather than plasma. The equation to calculate the related factor, termed fuliver, is given by 4:

equation image(4)

where (fup-app = in vitro fup * F1) and F1 is an ionization factor described by Berezhkovskiy.46 PLR is the plasma to liver concentration ratio of the respective binding proteins. For albumin bound drugs this value is 13.3. The final Eq. 5 for AL-bound drugs as described by Poulin et al. then is denoted as:

equation image(5)

For alpha-1-acid glycoprotein (AAG) bound drugs, facilitated hepatocyte uptake is significantly lower than for albumin and therefore fuliver = fup-app. When compared with the three methods proposed by Obach,33 the performance of the Poulin et al. method is superior in prediction of CLH in a selection of 55 compounds, taken from different publications,45,47 as can be seen in Figure 5. It should be mentioned that there are additional factors that may need to be considered and will make these relationships more complex such as nonlinearity in binding, reequilibration times for transporters and variability of protein concentrations.

Figure 5.

Comparison of different approaches to extrapolate in vitro metabolic turnover rates to in vivo predicted clearance values. Data from Ref.45,47, human liver blood flow (Q) was set at 21 mL/(min kg). (a) No correction (Eq. 2 with fup = 1), (b) correction for fup (Eq. 2), (c) correction for fup and fuinc (Eq. 3), and (d) correction for fuliver and fuinc (Eq. 5).

These examples from drug development indicate that by no means can we assume that “protein binding does not matter”. Indeed, the data indicate that it is prudent to account for different binding conditions and binding proteins when extrapolating to in vivo predictions.


The consistent use of unbound factors should also be reflected in the guidelines by the US FDA and EMA on DDIs. However, the newest guidelines remain somewhat ambiguous about this issue.28,29

The EMA addresses the issue of nonspecific binding in the following way: “It is recommended to use the estimated or determined unbound drug concentration in the in vitro system. In situations, where it is important to have a precise value on fumic (unbound microsomal fraction), such as estimations of inhibition or induction potential not followed by an in vivo study, determining the fraction (experimentally) is recommended. This also applies if there are reasons to believe that the free inhibitor concentration is markedly lower than the total concentration in the incubation, that is, if the substance binds covalently to proteins or may adsorb to the walls of the test tube.” And: “The in vitro study needs to be carefully performed and factors affecting the results should be taken into account. See the scientific literature for relevant protocols. Please note the need to determine non-specific binding at the initial step due to the general use of high protein concentrations.” The guideline then continues by describing a basic model, indicating possible interaction if the following condition is fulfilled 6:

equation image(6)

where [I] is the unbound mean Cmax obtained during treatment with the highest recommended dose, Ki is the inhibition constant and x a certain cut-off value.

So in the EMA guideline, [I] is corrected for unbound concentration. However, although from the previous text the EMA recommends to account for nonspecific binding, from this equation it is not clear that unbound inhibition constant (Ki,u) and not total inhibition constant (Ki) should be used. This might result in researchers using Ki determined from total and not unbound concentration in vitro in these equations, giving rise to overestimation of Ki in compounds with (high) binding. This might lead to false negatives when it comes to evaluation of DDIs.

The US FDA has slightly different recommendations, but uses a similar Eq. 7 as a basic model, where interaction is indicated when R > 1.1:

equation image(7)

where Ki is the unbound inhibition constant determined in vitro and [I] can be estimated by the maximal total (free and bound) systemic inhibitor concentration in plasma.

So contradictory to the EMA recommendations, the US FDA advises to use the total and not the unbound maximal systemic concentration of the inhibitor in this equation. Additionally, it unambiguously states the unbound Ki (i.e., Ki,u) should be used. This discrepancy in use of bound or unbound parameters in recommendations for DDI evaluations from the regulatory agencies is highly unwanted.

Both US FDA and EMA also recommend an alternative model to evaluate DDI, proposed by Fahmi et al.48 This mechanistic static model, which indicates possible interaction if AUCR is outside 0.8–1.25, is more complex (Table 2) than the basic model. However, it contains similar terms such as [I]h, the maximal unbound concentration in the portal vein, and Ki. Again, it is not particularly clear in the two guidelines or the original paper by Fahmi et al.48 that the inhibition constant used should be Ki,u to correct for nonspecific binding. The same is true for other parameters such as EC50.

Table 2. Algorithms Used in the Current US FDA Guidance for industry
equation image
  1. Drug Interaction Studies—Study Design, Data Analysis, Implications for Dosing, and Labeling Recommendations to assess significance of in vitro drug interaction studies.28

  2. aWhere fm is the fraction of systemic clearance of the substrate mediated by the CYP enzyme that is subject to inhibition/induction; Fg is the fraction available after intestinal metabolism; subscripts “h” and “g” denote liver and gut, respectively; equation image; equation image. In these equations, fub is the unbound fraction in blood; [I]max,b is the maximal total (free and bound) inhibitor concentration in the blood at steady state; Fa is the fraction absorbed after oral administration; ka is the first order absorption rate constant in vivo; and Qen and Qh, are blood flow through enterocytes (e.g., 18 L/h (70 kg)] and hepatic blood flow [e.g., 97 L/h (70 kg)], respectively. EC50 is the concentration causing half maximal effect; Emax is the maximum induction effect; Kdeg is the apparent first order degradation rate constant of the affected enzyme; Ki is the unbound reversible inhibition constant determined in vitro; kinact and KI are maximal inactivation rate constant and apparent inactivation constant, respectively; d is a scaling factor of 0.3 determined with linear regression of the control data set.

Time-dependent inhibitionequation imageequation image
Inductionequation imageequation image
Reversible inhibitionequation imageequation image

The guidelines on DDIs provided by the US FDA and EMA are not unambiguous when it comes to using unbound drug parameters. Moreover, at least for the basic model the two guidelines contradict each other on using bound or unbound concentrations for [I]. This lack of clarity and inconsistency is confusing and problematic and should be corrected soon. Binding factors should always be taken into account, as it is the unbound fraction of the drug that is available for biological processes, and this practice should be portrayed in both literature and guidance documents.


More than a decade after Benet's and Hoener's landmark paper, there is still considerable confusion about the significance of protein binding. The issue is frequently ignored or belittled but can be a project-breaker when misinterpreted. Although the fog is slowly lifting, rational and objective decision making is sometimes limited from available experimental. This is also true for the measurement of in vivo unbound drug concentrations. Microdialysis has opened the door for the experimental measurement of extracellular drug concentrations, both in humans and in animals. Imaging techniques have become very sensitive and powerful, however they usually lack the ability to differentiate between free and bound drugs. Measurement of intracellular unbound drug concentrations is presently limited to indirect assessment using biomarkers or response read-outs. Hopefully, 10 years from now, new technologies will have emerged that allow direct measurements of unbound intracellular drug concentrations.


Jules Heuberger was supported by a fellowship of the Saal van Zwanenberg Stichting in the Netherlands.