Predicting efficacious concentrations in vivo in humans using potency measured in vitro (i.e., potency parameters such as IC₅₀, EC₅₀, etc.) and unbound drug concentrations has been used widely at the early discovery or preclinical stages of drug development. The free drug hypothesis, which states that only unbound (free) drug molecules exert effects by binding to targets, has been dogma in pharmacology. If the free drug hypothesis is valid and potency measurements in vitro are well correlated with the in vivo effects in humans, the steady-state unbound average concentrations (C_{u_ss.avg}) achieved in humans by taking the approved dosage regimens should be higher than or at least comparable to the potency parameter values. For example, Smith et al. [1] reported that all of the ratios (C_{u_ss.avg}/in vitro potency) were > 0.5 (0.5–10) in the exemplified 16 drugs of 10 classes that they cited. In this review, these ratios were further surveyed for major therapeutic classes of drugs using published pharmacokinetic (PK) parameters, dose information in labels, and in vitro potency parameters. The ratios in 54 drug entities (13 classes) analyzed were highly variable (0.002–240) compared with the ratios reported by Smith et al.[1] (0.5–10). Although our exploration was not exhaustive, our results appear sufficient to suggest that the in vitro potency and protein binding characteristics of drugs may not always be useful to predict their efficacious dose in vivo in humans.

The C_{u_ss.avg}/potency ratios were extremely variable (about 100- to 100,000-fold differences even between drugs in the same class) despite there being trends by classes. In contrast to the general expectation (that ratios would be ≥ 1), the ratios were < 1 for 38 (69%) and < 0.1 for 22 (34%) of the 54 drugs (Fig. 1A). Even if we used the average total concentration (C_{tot_ss.avg}), the ratios of 17 drugs (26%) were < 1 (data not shown).

In the case of proton-pump inhibitors (PPIs), the low ratios may be explained by their irreversible binding to the proton pump, and the low ratios of statins by the fact that their target is intracellular within hepatocytes. The exposure of the hepatocytes to statins may be sufficiently high because the portal venous blood carries high concentrations of statins, which are higher than the peripheral venous concentrations, to the liver immediately after passing the intestine (first-pass effect). Thus, at the effect site (inside the hepatocyte), statin concentrations may be much higher than those in the peripheral venous blood, which in the present analysis were found to be much lower than the potency parameters. This explanation may also be applicable to other drugs that target intracellular receptors of hepatocytes.

When the ratios were plotted against f_u (Fig. 2), the “ratio < 1” phenomenon was distinct for drugs with high protein binding (27 of 30 [90%] drugs with f_u 5% vs. 10 of 24 [42%] drugs with f_u > 5%). The average ratios (unbound) of statins and CCBs were < 0.1. The finding that a ratio < 1 was most common for drugs with f_u ≤ 5% (Fig. 2) implies that the accuracy of the protein-binding assay needs to be improved, especially at the upper extremes. Because 3 of the 5 statins and all of the 3 PPIs listed in this review had f_u ≤ 5% (Table 1) [6789101112131415161718], their low ratios mentioned above may be also related to the low f_u (Fig. 2).

Because the potency measured in vitro varies according to assay methods and laboratories, the ratio for each drug reported herein may not be dependable. However, the trend observed across the 13 classes indicates that the traditional, free drug hypothesis-based approaches may be misleading when the data are obtained from in vitro studies alone without corroborating data from in vivo studies in animals or humans. An example of a study that took a similar approach to ours is the report of Smith et al. [1] concerning the effect of plasma protein binding on drug efficacy in vivo in humans [1]. In that study, unlike in ours, the ratios (C_{u_ss.avg}/potency) of all 16 drugs analyzed were > 0.5. Because the 16 drugs used in that study, with the exception of nifedipine and amlodipine, do not overlap with the drugs in our list, the two reports cannot be compared directly. However, one possible explanation of the differences is that the measured potency values tend to be highly variable because of the different methods used. For example, the ratios of nifedipine and amlodipine cited by Smith et al. [1] were 1.5 and 0.5, respectively. However, as indicated in Fig. 1B in our report, they were 0.44 (nifedipine) and 0.001 (amlodipine) when the IC₅₀s for rat tail artery contraction were measured [6]. Indeed, in the same report, the ratios calculated using potency values obtained from other methods such as IC₅₀ for rat papillary muscle contraction or K_i for rat heart muscle binding (radioligand) [6] were highly variable: 0.07–0.45 for nifedipine and 0.00005–0.05 for amlodipine. Of the several methods for measuring potency described in that article, the method of tail artery contraction (leftmost column in Fig. 1A) was chosen because it was considered to emulate most closely the situation in vivo in humans. Even when the same type of method (ligand binding assay) was used, the resulting potency values (K_i) varied by 100-fold depending on the concentrations of potassium used in their experiment (5.8 mM vs. 50 mM) as illustrated in Fig. 1B (method 4 and method 5).

The causes of in vitro–in vivo discrepancies in humans can be discussed from a few viewpoints. There are many cascading steps between target occupation in vitro and measurable responses in vivo in humans. Thus, the signal initiated by the occupation of a target molecule may be amplified or reduced while passing through these cascade steps; e.g., if the signals are amplified, the effect in vivo in humans may be high despite low target occupancy (and thus low potency in vitro) unless the potency marker exists downstream of the cascade. When the target binding site is intracellular, the target occupancy will be more closely correlated with the intracellular unbound concentration. In theory, the unbound concentrations in intracellular space are influenced by the plasma unbound concentration, the proportion of the drug ionized (pK_a), lipid solubility, and the presence and activity of membrane transporters. However, unless there is evidence for huge discrepancies between the intracellular and extracellular unbound drug concentrations, the existence of an intracellular target (e.g., for statins, EGFR inhibitors, and PPIs on our list) alone is not sufficient to justify the in vitro–in vivo discrepancies in humans reported here.

In vitro–in vivo discrepancies in humans may also arise from the inherent limitations of all in vitro biological studies. In vitro studies may provide clues about efficacy, but their implications should not be overemphasized. When a test is performed in vitro using a test molecule and another known competitor drug that acts on the same target, the results may give some indication as to the relative potency of the test molecule that can be used for decision-making. However, if the test molecule acts on a novel target for which no known drug exists, in vitro results alone for the test molecule should not be used when making critical decisions about prediction of efficacious concentrations in humans.

Because, as shown in Fig. 1, discrepancies between the potency in vitro and effects in vivo in humans are so widespread, caution is needed in the interpretation of data obtained in vitro; all the in vitro methods currently used for assaying potency should be regarded as ancillary screening tools that cannot be used to infer the in vivo effects in humans dependably. Predicting efficacious unbound concentrations in vivo in humans using potency data obtained in vitro is not a reasonable approach, particularly for molecules with high protein binding (f_u < 5%). The fate of candidates should not be determined by the f_u and potency data from in vitro studies alone, before comparing their effects in animal models in vivo. These observed discrepancies may not be new to experienced researchers working in the areas of drug discovery or early preclinical development. However, to our knowledge, the existence of in vitro–in vivo discrepancies in humans or their implications have never been properly addressed. Therefore, this review is the first to raise the issue of this discrepancy and its relationship to plasma protein binding and the caveats associated with it.