Liquid Chromatography-Tandem Mass Spectrometry.

Total drug concentrations in plasma, blood, and tissues can be measured in various ways. Because of selectivity, sensitivity, and ease of use, liquid chromatography-tandem mass spectrometry (LC-MS/MS) is the most common analytical choice. For LC-MS/MS, analytes are extracted from plasma/blood and tissue homogenates by organic solvents or other sample preparation techniques (e.g., solid-phase extraction) before analysis.

Autoradiography.

Examining tissue-to-plasma distribution of total drug (K_p) can be done by quantitative whole-body autoradiography (Stumpf, 2005; Solon et al., 2010) and microquantitative autoradiography (Stumpf, 2005; Solon et al., 2010; Drexler et al., 2011). These methods allow determination of local radioactivity in specific regions. Three limitations of the technique include not being able to distinguish parent drug from metabolites, providing only the total concentration of drug-related materials and the need for radiolabeled material, which is usually not available in discovery.

Positron Emission Tomography and Other Imaging Techniques.

Techniques such as PET and magnetic resonance imaging have been used to estimate drug concentrations in tissues (Sasongko et al., 2005; Eyal et al., 2009; He et al., 2014; Sundelin et al., 2017; Kaneko et al., 2018). Like the limitations of autoradiography, they cannot differentiate parent from metabolites and require a radiolabel (¹¹C, ¹⁸F, etc.). Furthermore, for highly plasma protein drugs, corrections of tissue images for drug in the blood contained within the tissue should be performed to avoid overestimating total tissue drug concentration (Hsiao and Unadkat, 2014).

Matrix-Assisted Laser Desorption Ionization Mass Spectrometry Imaging.

More recently, matrix-assisted laser desorption/ionization mass spectrometry imaging (MALDI-MSI) has become an emerging cutting-edge tool for determining the spatial localization of molecules of interest within tissues (Schwamborn and Caprioli, 2010; Caprioli, 2016). Unlike other imaging techniques, such as autoradiography, MALDI-MSI does not require molecular tags or labels. Further, this technique can generate distribution profiles (ion intensity maps) of many hundreds of compounds, including endogenous molecules that have different mass/charge (m/z) ratios simultaneously (Angel and Caprioli, 2013). The application of MALDI-MSI to visualize the spatial distribution of the antiretroviral drug tenofovir (TFV) and its diphosphorylated metabolite, TFV-diphosphate (the pharmacologically active form), in human colorectal tissue after TFV administration (as a rectal enema) demonstrated the heterogeneity in the distributions of these analytes (Seneviratne et al., 2018) (Supplemental Fig. S2). In this study, the potential ion suppression effects were determined by calculating tissue extinction coefficients for both TFV and TFV-DP.

MALDI-MSI technology can be further extended for the quantitation of drugs and their metabolites as well. Although this quantification can be achieved by incorporating an internal standard and creating a calibration curve, performing absolute quantitation has some challenges. For example, variation of signal (pixel-to-pixel), ion-suppression effects, and interference from matrix-related ions (or ion adducts) are confounding factors that must be addressed by applying appropriate normalization strategies and quantitative software (Källback et al., 2012; Schulz et al., 2019).

Modeling Approaches to Estimate Drug Concentrations.

Modeling approaches can be used to estimate drug concentrations in plasma and tissues (Watanabe et al., 2009; Tsamandouras et al., 2015; Li et al., 2018a,b); however, this is still an evolving area, and further validation is needed. One example of this is a proteomics-informed bottom-up approach to predict total drug concentrations in tissues where there is active transport and/or metabolism (Prasad and Unadkat, 2014). Specifically, metabolism- and transport-mediated clearance of a drug can be measured in vitro (in human liver microsomes, cytosol, S9 fraction, transporter expressing cell lines or primary cells), and protein expression-based scaling factors can then be used to extrapolate in vitro clearance to in vivo (IVIVE). Using quantitative proteomics, transporter tissue abundance has been established in the liver, kidney, and intestine (Drozdzik et al., 2014; Wang et al., 2015; Prasad et al., 2016). As a proof-of-concept, rosuvastatin hepatobiliary clearance in rats has been reasonably predicted using both rat organic-anion-transporting polypeptide (OATP)-expressing cell lines and sandwich-cultured rat hepatocytes (Ishida et al., 2018). Whereas IVIVE using transfected cell lines and quantitative proteomics is a promising technique for predicting intracellular drug concentrations, further validation of this approach is needed. In another example, global scaling factors were developed for physiology-based pharmacokinetic (PBPK) modeling involving transporter-mediated processes (Li et al., 2014). The modeling approach is intended for extrapolation or prediction of K_p or K_p,uu for drug candidates that have not been administered to humans.

Measurement of f_u and Total Drug Concentration.

Since direct measurement of free drug concentrations is challenging, it is typically obtained indirectly by measuring f_u and total drug concentration. Measurement of total drug concentration was discussed earlier herein, and a number of methods are available to determine f_u (Di et al., 2011, 2012); equilibrium dialysis is a commonly used method (Di et al., 2017). An extensive discussion of K_p,uu and experimental methods for measuring unbound drug concentrations in tissue can be found in the International Transporters Consortium white paper (Guo et al., 2018). Although it is reasonable to assume that the ex vivo binding is representative of the in vivo situation in liquid matrices (plasma), the same equivalence may not be as generalizable for solid matrices (tissues) where homogenization is required before separation of free and total drug. In addition, some of the measurements can be influenced by the presence of contaminants, such as plasticizers. Various in silico models have been developed to predict plasma protein binding based on molecular properties (Sun et al., 2018).

Prediction of In Vivo K_p,uu from In Vitro K_p,uu.

A new method has recently been developed to estimate in vivo liver-to-plasma K_p,uu from suspension hepatocytes in 4% bovine serum albumin (which is similar to albumin context in plasma) for OATP substrates (Supplemental Table S2). Good IVIVE has been demonstrated in both rats and humans. This approach has the potential to estimate in vivo free drug concentration in the liver from free plasma concentration (Riccardi et al., 2016, 2017). The method has been shown to be applicable to a number of solute carrier family, such as OATPs, OAT2, sodium-taurocholate cotransporting polypeptide, and SLC13A (Riccardi et al., 2016, 2017, 2019). Prospective predictions of K_p,uu have been successful based on human PD data or based on clearance data using the K_p,uu method (Riccardi et al., 2016, 2017, 2019). Several hypotheses about the mechanisms of the effect of albumin have been developed, and increasing numbers of examples of more accurate predictions in the presence of albumin have been reported (Bowman and Benet, 2018; Mao et al., 2018; Miyauchi et al., 2018; Bowman et al., 2019). The theoretical bases for a direct IVIV translation of K_p,uu have been discussed; in most cases, the differences between in vitro and in vivo K_p,uu values are minimal when the metabolism- or transporter-mediated clearance is not extensive (e.g., extraction ratios are not too high) (Li et al., 2019).

Estimation Based on PD Endpoints.

Assuming in vitro potency is the same as in vivo potency, the activity shift between in vivo and in vitro can be used to estimate K_p,uu. An example is illustrated using rosuvastatin and pravastatin (Riccardi et al., 2017). By comparing the in vivo and in vitro IC₅₀ values of their inhibition potency for 3-hydroxyl-3-methylglutaryl coenzyme A reductase, liver-to-plasma K_p,uu values were derived. The key limitation for K_p,uu estimation is the underlying assumption that there is no potency difference between the in vitro and in vivo systems.

Microdialysis to Measure In Vivo Free Drug Concentration.

In vivo microdialysis can be used to measure free drug concentration. For example, cerebral microdialysis has been used to measure free drug concentration in brain (Tisdall and Smith, 2006). The limitations are that this method is technically challenging, it is low-throughput, and it suffers from high nonspecific binding to dialysis devices for lipophilic molecules.

Limitations of Current Approaches to Free Drug Estimation.

Most of the current methods to estimate free drug concentration use tissue homogenates to determine f_u and then a total drug concentration measurement, which destroys all subcellular structures. The free drug concentration determined using this approach thus represents the average free drug concentration from all the subcellular compartments. As efficacy or toxicity depends on the drug’s interaction with specific targets within the subcellular compartment, it is important to understand whether the average free drug concentration represents the subcellular compartment of interest. For acidic compounds, the average free drug concentration is mostly the cytosolic concentration, whereas for basic compounds, it might be mostly lysosomal or mitochondrial and also depends on the cell type since different cell types can have different volumes of subcellular compartment. An example of heterogeneous subcellular drug distribution is metformin. Whereas the average K_p,uu of metformin from all the subcellular compartments in human embryonic kidney HEK293-OCT1 cells is 53.6, the model prediction of mitochondrial and endoplasmic reticulum K_p,uu is two to three times greater than the cytoplasm, lysosome, and nucleus (Chien et al., 2016). It is therefore important to convert f_u from tissue homogenates to f_u,cell of the subcellular compartment of interest to obtain the free drug concentration at the site of action. As K_p,uu is defined as a steady-state parameter, it is critical to ensure that both in vitro and in vivo measurements are made under steady-state conditions (e.g., by using intravenous infusion or intravenous area under the concentration curve (AUC) and using multiple time points to make sure steady state has been achieved). In some cases, contamination of blood or bile will need to be considered to estimate liver exposure. Typically, liver blood volume is set as the lower limit of liver K_p. For compounds with high biliary clearance, correction may need to be applied to obtain liver concentration.

Substrates of Uptake and Efflux Transporters: Liver and Brain Examples.

Active transporters can create asymmetric drug distribution between tissues and plasma where K_p,uu 1. Different tissues can have different K_p,uu values for a particular compound. For example, an acidic compound can have a liver-to-plasma K_p,uu 1, whereas a brain-to-plasma K_p,uu 1. Tissue-to-plasma K_p,uu > 1 or <1 depends on a number of factors, including transporter expression difference. In a noneliminating tissue (e.g., brain), when active uptake is greater than efflux and other elimination mechanisms in the tissue (e.g., bulk flow), K_p,uu will be >1. In contrast, if active efflux from the tissue is predominant relative to passive diffusion, K_p,uu will be <1. For example, for drugs that are efflux transporter [polyglycoprotein (P-gp), breast cancer resistance protein (BCRP)] substrates, the brain-to-plasma K_p,uu is <1 (Feng et al., 2018). For an elimination organ (e.g., liver), K_p,uu is determined by multiple processes, including passive diffusion, active uptake and efflux, and metabolism based on the extended clearance concept (Yamazaki et al., 1996; Shitara et al., 2006; Watanabe et al., 2010; Patilea-Vrana and Unadkat, 2016). For instance, in vivo rat liver-to-plasma K_p,uu values of pravastatin, cerivastatin, fluvastatin, and rosuvastatin were 2.2, 29, 44, and 57, respectively (Riccardi et al., 2017). Of note, the rat liver samples collected in Riccardi et al. (2017) had all blood vessels removed to correct for blood but not bile contamination, and thus the K_p,uu values are likely reflective of true partitioning coefficients.

For compounds that have asymmetric tissue distribution owing to transporter involvement, free drug concentration in the liver should be used rather than plasma free concentration to understand drug-drug interaction (DDI) and the impact of genetic polymorphism owing to inhibition or induction of drug-metabolizing enzymes. The extended clearance model has been developed to understand the enzyme-transporter interplay on clearance, DDI, and genetic polymorphism (Yamazaki et al., 1996; Shitara et al., 2006; Watanabe et al., 2010; Patilea-Vrana and Unadkat, 2016). In the liver, when the metabolic plus biliary efflux clearances are much greater than sinusoidal efflux clearance, sinusoidal uptake clearance becomes the rate-determining step in the hepatic clearance of the drug. In other words, when the drug is eliminated out of the liver before it can return to plasma, the hepatic clearance is driven mainly by uptake clearance, even if there is significant metabolism and/or biliary efflux. In this case, inhibition of uptake transporters will increase plasma drug concentrations and exposure (AUC) but will not significantly impact the liver drug AUC, provided the drug is eliminated mainly by the liver. On the other hand, inhibition of metabolic enzyme or biliary transporters will increase the liver drug AUC while minimally impacting the plasma drug AUC (Watanabe et al., 2010). In both examples highlighted, the changes in the drug’s plasma concentrations will not reflect changes in the hepatic drug concentrations. This asymmetry may lead to misinterpretations of the impact of a DDI or genetic polymorphism on the drug’s PD effects when the target site of the drug is in the liver.

Atorvastatin, a permeable drug (logD_7.4 = 1.53), is a substrate of OATP transporters and is eliminated from the body primarily by hepatic CYP3A metabolism. In a DDI study with atorvastatin, coadministration with rifampin, but not itraconazole, OATP, and CYP3A inhibitors, respectively, led to a significant increase in the plasma AUC of atorvastatin (Maeda et al., 2011). Based on these data, hepatic uptake clearance of atorvastatin via OATPs is the rate-determining step in the clearance of the drug. Compared with wild-type allele, patients with OATP1B1 polymorphism c.521T > C (conferring reduced function) have increased plasma atorvastatin AUC; however, no significant change in the PD response (i.e., lowering of low-density lipoprotein cholesterol) of atorvastatin is observed in patients with reduced OATP1B1 function (Shitara et al., 2013). In contrast, patients with homozygous CYP3A5*3 allele (conferring reduced function) have demonstrated an increased PD response to atorvastatin (Kivistö et al., 2004). PET imaging of rosuvastatin in the rat showed that inhibition of OATP uptake by rifampin led to a 2.3-fold increase in blood AUC, but it showed no significant change to the hepatic AUC (He et al., 2014). PBPK analysis of rosuvastatin in humans further illustrates that whereas OATP1B1 polymorphism impacts rosuvastatin plasma concentrations, the predicted rosuvastatin liver concentrations are not significantly impacted, and therefore neither is its PD response (Rose et al., 2014). These highlighted results demonstrate how the changes to a drug’s plasma exposure do not reflect changes to liver exposure and, therefore, PD. This is because the rate-determining step in the hepatic clearance of atorvastatin and rosuvastatin is uptake clearance via OATPs, even though, for the two drugs, there is metabolism via CYP3A enzymes and biliary efflux via MRP2/BCRP, respectively.

Metformin, an antidiabetic agent, is an interesting example. The pharmacologic site of action of metformin is in the liver, but it is eliminated primarily via the kidney as unchanged drug. In the liver, metformin is transported into the hepatocytes by OCT1 and perhaps is excreted in the bile by multidrug and toxin extrusion 1 (MATE1) as a minor route (Liang and Giacomini, 2017). Patients with OCT1 and MATE1 genetic polymorphism have an decreased and increased PD effect of metformin, respectively, but with no significant change in metformin plasma AUC (Shu et al., 2007; Stocker et al., 2013). PET imaging of [¹¹C]metformin distribution into the liver demonstrated that patients with OCT1-reduced-function polymorphism had ∼30% lower metformin hepatic distribution versus those patients with wild-type OCT1 (Sundelin et al., 2017) without knowing whether this result is pharmacologically relevant. Likewise, in mice, coadministration with pyrimethamine (MATE inhibitor) results in a ∼2.5-fold increase in hepatic metformin AUC compared with controls (Jensen et al., 2016). Because most of metformin’s elimination is renal, hepatic transporters will impact its hepatic drug concentrations and therefore PD effect without impacting its plasma drug concentration.

The impact of efflux transporters on tissue-to-plasma K_p,uu is best demonstrated by interactions at the blood-brain barrier (BBB) (Supplemental Fig. S3). A critical functional component of the BBB that limits drug delivery is active efflux transport, mainly by P-gp and Bcrp, and brain drug delivery can be severely limited by this mechanism (Table 1) (Schinkel et al., 1996; Agarwal et al., 2011b; Gampa et al., 2017). For example, the brain distribution of [¹¹C]verapamil increased 88% when coadministered with cyclosporine (i.e., P-gp and weak CYP3A inhibitor) without significantly affecting the systemic [¹¹C]verapamil concentrations (Fig. 1) (Sasongko et al., 2005). In addition, reports indicate expression of efflux proteins in tumor cells, forming a secondary barrier to drug delivery to tumors in the brain (Demeule et al., 2001; Dean et al., 2005; Szakacs et al., 2006; Fattori et al., 2007; Luo et al., 2012). Even though the BBB/brain-tumor barrier may be relatively compromised in some regions of tumor, regions of brain that contain infiltrative tumor cells can have an intact and functional BBB. The development of systemic antitumor therapies (e.g., kinase inhibitors) that can adequately permeate an intact BBB and reach the target tumor cells has proven challenging (Table 1). On the other hand, restricted brain penetration owing to transporter-mediated drug efflux may be advantageous in some cases by reducing central nervous system toxicity.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Impact of efflux transporter inhibition on tissue drug concentrations. Increased [¹¹C]verapamil distribution in the human brain (increased red color) in the presence of P-gp inhibition by cyclosporine. Representative brain magnetic resonance (left panel) and PET (middle and right pane) images before (middle panel) and during (right panel) cyclosporine infusion (2.5 mg/kg per hour) are shown (Sasangko et al., 2005). Whereas the [¹¹C]verapamil concentrations in the brain increased (∼2-fold increase in brain AUC) when cyclosporine was coadministered, the systemic radioactive content of verapamil changed minimally (Sasangko et al., 2005). Figure reprinted with permission from Sasangko et al. (2005).

Many examples in the literature support the utility of K_p (when binding factors cancelled out) and K_p,uu in characterizing the brain distribution (Table 1). The following practical example also demonstrates the utility of K_p and K_p,uu in characterizing the brain distribution of a novel MEK inhibitor E6201 (Gampa et al., 2018). E6201 demonstrates low nanomolar potency in several BRAF mutant melanoma cell lines (Byron et al., 2012; Narita et al., 2014; Babiker et al., 2018). In vitro accumulation experiments in Madin-Darby canine kidney cell II cells show that E6201 is not a substrate of P-gp or Bcrp. After a 40-mg/kg single intravenous bolus dose of E6201, the total concentrations were greater in brain compared with plasma in four genotypes of mice (wild-type or efflux transporter–deficient); the systemic drug concentrations were similar across the four genotypes. As a consequence, the brain partitioning represented by the brain-to-plasma AUC ratios (K_p) were >1: 2.66, 4.37, 3.72, and 5.40 in wild-type, P-gp-deficient (Mdr1a/b^−/−), Bcrp-deficient (Bcrp1^−/−), P-gp- and Bcrp-deficient (Mdr1a/b^−/− Bcrp1^−/−) mice, respectively. The brain K_p after single oral and intraperitoneal doses of 40 mg/kg E6201 were comparable to the K_p observed in intravenous dosing studies. The free fractions (f_u) of E6201 were determined using in vitro rapid equilibrium dialysis experiments and were 0.14% and 3.4% in brain and plasma, respectively. The K_p,uu values after a single 40-mg/kg intravenous dose of E6201 in wild-type, Mdr1a/b^−/−, Bcrp1^−/−, and Mdr1a/b^−/− Bcrp1^−/− mice were 0.14, 0.24, 0.2, and 0.29, respectively. These findings demonstrate that the delivery of E6201 to the brain is minimally impacted by P-gp or Bcrp-mediated active efflux at the BBB. For E6201, the brain K_p in wild-type mice is considerably higher than that of trametinib and cobimetinib, and the K_p,uu is higher than that of cobimetinib, although similar to that of trametinib (Table 2).

The following example demonstrates that the efflux transporters could affect drug distribution in tissues differently. Apixaban is a potent and highly selective Factor Xa inhibitor, which is widely used for treatment of arterial and venous thrombotic diseases (Wong et al., 2011). Apixaban is a substrate for P-gp and BCRP (Zhang et al., 2013b). In addition, BCRP appears to play a more important role for absorption and intestinal and renal elimination of apixaban than P-gp in transporter-KO rats after oral and intravenous dosing (Zhang et al., 2013a). The distribution of [¹⁴C]apixaban was investigated in rats after single oral doses in which tissue distribution of radioactivity was measured using quantitative whole-body autoradiography (Wang et al., 2011). After oral administration, radioactivity was distributed quickly in rats with C_max at 1 hour for most tissues in lactating rats. The radioactivity concentration was lowest in brains, which was detectable only within the first 1 hour after dose (K_p < 0.1 by concentration, K_p < < 0.1 by AUC). In pregnant rats, low levels of radioactivity were present in fetal blood, liver, and kidney, which were much lower (up to 20-fold) than that in the respective maternal organs (K_p < 0.1). In lactating rats, apixaban showed extensive lacteal excretion; values were 15-fold higher for C_max and 55-fold greater for AUC than those in blood (K_p > 15 by C_max, K_p > 50 by AUC) (Wang et al., 2011). Apixaban showed K_p,uu < 0.1 in brain, K_p,uu > 50 in milk, K_p,uu ∼1 in blood in rats, assuming similar protein binding in plasma and brain or milk. In this case, there was a direct correlation between blood concentrations and apixaban anticoagulant activity (K_p = 1 at the target tissue) since Factor Xa is mostly present in blood, the target for pharmacologic activity of apixaban (He et al., 2011; Wong et al., 2011). This provides a good example for a small-molecule drug whose tissue-to-plasma ratios could be greater, equal to, or less than that in blood (K_p < 1 to >1), depending on the specific tissue considered.

Lysosomal Trapping and Albumin-Assisted Uptake Leading to Higher Free Drug Concentrations in Tissue.

The free drug hypothesis can be complicated by lysosomal trapping and protein-assisted drug uptake. These two scenarios can lead to higher tissue free drug concentrations than plasma and are described here.

In the first scenario, there can be pH differences between plasma, tissue, and cellular organelles (Berezhkovskiy, 2011). The pH in lysosomes is 4.5–5 and, therefore, much lower than plasma, having a significant impact on the cellular levels of basic compounds. Even though there is still equilibrium of the neutral form of a molecule, basic compounds (pKa > 5) will be ionized in the lysosomes, and their charged nature will greatly reduce diffusion out of the lysosome. This phenomenon, called lysosomal trapping, results in K_p,uu > 1 (Ufuk et al., 2017). Generally, the impact on efficacy is small if the molecular target is in the cytosol, endoplasmic reticulum, or nucleus because the pH of those organelles is similar to the pH of plasma; however, if the target is in the lysosome itself or on the luminal side of the lysosomal membrane, it can greatly enhance efficacy, and it may appear that efficacy occurs at lower unbound plasma or tissue concentrations than anticipated.

In the second scenario, compounds, usually acids, that are highly bound to albumin can be prone to “protein-facilitated uptake” and, in particular, “albumin-facilitated uptake” (Fig. 2). This phenomenon was observed as early 1966 (Baker and Bradley, 1966), and an elegant study described it in more detail in 1988 (Tsao et al., 1988). Subsequently, the phenomenon was forgotten while academia and the industry focused on drug-metabolizing enzymes and, thereafter, transporters. More recently, the hypothesis was applied (Poulin et al., 2012, 2016) to improve in vivo-in vitro correlation for highly plasma protein-bound compounds. Using traditional models, highly bound compounds tend to underestimate in vivo clearance based on in vitro microsomal or hepatocyte data. Other examples have also been presented and reviewed related to albumin-facilitated uptake (Bowman and Benet, 2018). For example, hepatic uptake experiments were performed with hepatocytes in the presence and absence of plasma or albumin (Kim et al., 2019b). Considering the free drug hypothesis, uptake should be identical in the presence and absence of plasma or albumin after correcting for plasma protein binding; however, the uptake was substantially greater in the presence of plasma after correcting for plasma protein binding, and this phenomenon was generally more pronounced for highly bound compounds (f_u < 0.05, especially f_u < 0.01); this results in K_p,uu > 1 in the presence of plasma or albumin. In another study, it was shown that increased hepatic uptake clearance (V_max/unbound K_m) was observed with increasing albumin concentrations, and this phenomenon was explained by albumin-facilitated uptake (Fukuchi et al., 2017). Although a substantial amount of data have been presented that seems to support this albumin-facilitated uptake hypothesis (Bowman and Benet, 2018), there is still a lot of speculation about the exact mechanism; several hypotheses have been proposed (Poulin et al., 2016; Bowman and Benet, 2018), including the following:

1. The presence of an albumin receptor (e.g., FcRn) on hepatocytes and enhanced drug cellular uptake from the receptor-albumin-drug complex (Christensen and Birn 2013)

2. Interaction between albumin and the hepatocyte cell surface accompanied by a conformational change or driven by ionic interaction between the albumin-drug complex and the negatively charged groups on the hepatocyte cell membrane surface

3. Rate-limiting dissociation of drug from the albumin-drug complex whereby the bound (instead of free) drug determines the overall rate of hepatic uptake

4. Rate-limiting diffusion of unbound drug through the unstirred water layer (i.e., the albumin-drug complex enhances diffusion through the unstirred water layer and increases uptake for highly lipophilic compounds with lower diffusional flux)

5. Involvement of a high-affinity uptake transporters that strip drug from the albumin-drug complex; this may explain why most drugs for which this effect has been invoked are both highly bound to albumin and uptake transporter substrates

In addition to the preceding hypotheses, allosterism is a possible mechanism as the presence of an endogenous compound in plasma can change the transport clearance of a drug, and many transporters have been shown to be allosteric, such as OATPs and MRP2 (Gerk et al., 2004; Kindla et al., 2011). Although albumin-facilitated uptake has been mainly invoked to explain higher than anticipated uptake of drug into the liver, it could also occur in other organs, especially if the phenomenon is general and does not involve an albumin receptor or transporters (e.g., ionic interaction between the albumin-drug complex and the negatively charged groups on any cell surface). Assuming the molecular target of the drug is not in the liver, what would be the net effect of albumin-facilitated uptake? If this is a general phenomenon happening across tissues, the net effect on the efficacious dose may well be quite small because 1) increased uptake at the site of action leads to enhanced local unbound concentrations and presumably efficacy (K_p,uu > 1), but 2) enhanced uptake in the liver leads to higher drug clearance (for low ER drugs) and a lower unbound plasma concentration. Thus, the difference between the efficacious dose for compounds with and without albumin-facilitated uptake may well be quite small.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Protein-facilitated drug uptake mechanism.

Targeted Covalent Inhibitors.

The free drug hypothesis is more difficult to apply to targeted covalent inhibitor drugs because of PK and PD disconnect common with this mechanism of action. Since reactive moieties covalently bind to the target, target interaction is different from drugs that bind reversibly (Baillie, 2016). The following equation (eq. 1) describes the mechanism of how a covalent inhibitor (labeled C) modulates target enzyme or protein (labeled R):(1)The binding affinity of the covalent inhibitor to target proteins comes from both noncovalent (K_I) and subsequent covalent (k_inact) interactions. Permeation of free circulating covalent inhibitor through the cellular membrane and the reaction of free covalent inhibitors with either on-target or off-target enzymes are two different processes that are driven by concentration gradient (permeation) and reaction energy (covalent bonding). The differences in the reactive nucleophiles across the cellular membrane can alter free drug concentrations as a result of covalent bonding, assuming permeation is dominated by passive permeability. Depending on the molar ratio of intracellular free drug versus the target protein, the covalent interaction between inhibitor and target enzyme could play a role in the in vivo disposition of covalent inhibitors. With adequate inhibitor concentrations exceeding the target covalent binding threshold, an excess amount of free drug can re-establish a different free concentration gradient across cellular membranes. Several covalent inhibitors have demonstrated such capacity-limited nonlinear pharmacokinetic profile (Landersdorfer et al., 2012; Li et al., 2012).

Besides covalent interaction with targeted enzymes, it is common for covalent inhibitors to react covalently with off-target enzymes both in circulating blood and intracellular tissues, which also contributes to free drug disposition at steady state. The potential to form such covalent interaction is highly associated with the reactivity of different warhead moieties and their corresponding promiscuity toward other nucleophiles. Aspirin was shown to acetylate hemoglobin and serum albumin decades ago (Bridges et al., 1975; Burch and Blazer-Yost, 1981; Liyasova et al., 2010). Acrylamide-containing covalent inhibitors are intrinsically reactive to intracellular glutathione and free thiol groups in cysteine residues of proteins and enzymes. Numerous covalent inhibitors (e.g., neratinib, osimertinib, ibrutinib) show degrees of endogenous proteinaceous adduct formation. Neratinib formed covalent adducts on Lys-190 of human serum albumin (Wang et al., 2010). Similarly, [¹⁴C]osimertinib revealed covalent binding in incubations with hepatocyte, plasma, and serum albumin. In the case of ibrutinib, in vivo, more than 50% of total plasma radioactivity was attributed to covalently bound materials from 8 hours onward; as a result, covalent binding accounted for 38% and 51% of total radioactivity AUC_{0–24 h} and AUC_{0–72 h}, respectively (Scheers et al., 2015). The elimination of covalently bound radioactivity in human plasma was slower than the elimination of total radioactivity, indicating that ibrutinib likely has different reactivity toward protein targets in different tissues. Alternatively, the turnover rate of the proteins may differ.

In the free drug hypothesis, the interaction between drug and receptor is reversible. Within the time scale of receptor turnover half-life, covalent inhibitors can bind to receptors reversibly or irreversibly. The outcome of the covalent inhibition results in removal of receptor targets from the active naïve target pool. Thus, the pharmacologic activity from covalent inhibition is determined by both free drug concentration and the turnover of target protein or enzyme. If the time to resynthesize target protein is longer than covalent inhibitor in vivo half-life, a sustained pharmacology effect can occur even after the covalent inhibitor is eliminated from systemic circulation (Fig. 3A). After disappearance of drug, recovery to baseline level is dependent on resynthesis of the proteins, irrespective of drug concentration. In Fig. 3B, the animal model PKPD results for PRN1008, a reversible covalent inhibitor, showed a clear disconnect between PK exposure profile and BTK target occupancy. Clinically, covalent drug pharmacologic effect can last multiple days or even weeks after drug treatment discontinuation. Finasteride, a 5α-reductase inhibitor, prevents conversion of testosterone to dihydrotestosterone. Finasteride has a terminal pharmacokinetic half-life of 6–8 hours. After discontinuing daily finasteride administration, dihydrotestosterone concentrations do not return to pretreatment levels until approximately 2 weeks after withdrawal of therapy (https://www.accessdata.fda.gov/drugsatfda_docs/label/2014/020180s044lbl.pdf).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

PKPD profiles of a covalent drug. (A) Modeled PKPD plot of a hypothetical covalent inhibitor alkylating a target with turnover half-life of 24 hours. (B) Exposure profile disconnect from BTK target occupancy for covalent BTK inhibitor PRN1008 in a rat collagen-induced arthritis model.

Small-Molecule Prodrugs.

For prodrugs that form the active drug in tissue, it is difficult to describe their pharmacologic activity using plasma free drug concentration. A prodrug is an administered molecular entity that is converted to the active form in the body (Testa, 2009). It is usually easier to measure the drug in plasma when the active drug is quickly formed from the prodrug by carboxylesterases in plasma, as is the case for drugs such as clofibrate and oseltamivir; however, for prodrugs like sofosbuvir, the active form is not permeable and is formed through biosynthesis or biotransformation intracellularly and may not result in a meaningful level in plasma (K_p,uu > > 1). The rationale for design of prodrugs has been extensively and well established in the literature; these concepts are beyond the scope of this review (Stella, 2010). The compound administered is chemically distinct from the compound driving the pharmacologic response for a prodrug. Consequently, both the mechanism of release and the physiologic distribution and expression of activating enzyme(s) are of great importance in the selection and development of prodrugs (Stella, 2010). This tissue drug concentration asymmetry is anticipated even under PK steady state.

Prodrugs represent approximately 12% of all new drugs approved in last decade (Rautio et al., 2018). Clofibrate, oseltamivir, tenofovir, and sofosbuvir are used in the treatment of dyslipidemia, influenza, HIV, and hepatitis C, respectively (Oberg, 2006; Menendez-Arias et al., 2014; Rivero-Juarez et al., 2018). Beyond the obvious commonality that each of these drugs is administered orally, there are significant differences, as illustrated in Fig. 4, in the activation and hydrolysis mechanism, enzyme kinetics of activation/hydrolysis, target tissue, and pharmacologic mechanism of target inhibition. Whereas the relatively simple ester hydrolysis of clofibrate and oseltamivir might suggest that an obvious relationship exists between the plasma exposure of the liberated acid and pharmacologic response, this would ignore the confounding effects of the site of hydrolysis and the endpoints used to assess pharmacologic activity (Chasseaud et al., 1974; Miller and Spence, 1998; Shi et al., 2006; Wattanagoon et al., 2009). These relationships between target loading, hydrolysis, and pharmacologic mechanism become even more complicated considering tenofovir and sofosbuvir (Oberg, 2006; Ray et al., 2016; Rivero-Juarez et al., 2018). Both these compounds undergo multistep activation to their ultimate pharmacologically active species (Fig. 4). Although it is likely that the prodrug is the chemical species entering the target cells, there are significant differences in the distribution of the target cells, lymphocytes (tenofovir), and hepatocytes (sofosbuvir), that would further complicate deconvolution of the relationship between systemic concentrations and pharmacologic effect (Oberg, 2006; Ray et al., 2016; Rivero-Juarez et al., 2018). Furthermore, it is conceivable that both compounds could load their target cells locally, in the gastrointestinal-associated lymph tissue in the case of tenofovir and the liver for sofosbuvir such that portal venous concentrations would be related more to cellular uptake than to systemic concentrations. Additionally, the pharmacologic mechanism of these compounds is DNA (tenofovir) or RNA (sofosbuvir) chain termination by the triphosphorylated metabolite of each compound that is impermeable (Oberg 2006). This process is irreversible and further obscures any relationship with the plasma concentrations of the administered compound. For prodrugs that form the active drug in tissue, dynamic modeling will need to be used to develop PK/PD relationships.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Biotransformation of prodrugs. A-E are examples.

Antibody-Drug Conjugates.

An antibody-drug conjugate (ADC), in general, is itself inactive and could be considered a prodrug that requires a transformation to expose or release the payload (active drug). ADCs consist of an antibody and a payload with a chemical linker and have proven an effective modality of selectively delivering a small-molecule payload to targeted cells. The ADC marketing approval of ADCETRIS, KADCYLA, and BESPONSA has drawn wide interest in expanding the applications of these powerful agents (Doronina et al., 2003; LoRusso et al., 2011; Ricart, 2011; Sliwkowski and Mellman, 2013; Chari et al., 2014; Shor et al., 2015; Beck et al., 2017). The payload exerts its biologic activities after ADC is internalized and the payload is released into cells at the site of action (e.g., in tumor cells). For example, different efficacy profiles were obtained for two ADCs (CD22 conjugates of cyclopropyl-disulfide-PBD and cyclobutyl-disulfide-PBD) after administration in xenograft mice (Ma et al., 2016; Zhang et al., 2016). The cyclobutyl-disulfide-PBD released the active PBD payload, but the former ADC released an inactive payload metabolite. For these two ADCs, the tumor and plasma exposures of total antibody (i.e., monoclonal antibodies) were similar in plasma and tumors; the drug-to-antibody ratios of the two conjugates were also similar for 7 days after dosing. The cyclobutyl-disulfide-PBD ADC showed K_p > 150 in tumors for its payload 96 hours after dose. Figure 5 illustrates hypothetical ADC payload concentration-time profiles in tumors, liver, and blood, in which tumor payload concentration is much greater than that in livers with very low concentration in plasma. The tumor growth inhibition correlated with intratumor PBD exposures, but not with systemic exposures of ADCs or the payload after administration of HER2-disulfide-PBD conjugates in xenograft mice (Zhang et al., 2018). These results indicate that analysis of ADC species in circulation is insufficient to explain or predict ADC efficacy and suggested that payload/catabolite identification and quantitation in the tumors of xenograft mice are important to understand the ADC efficacy. As a consequence of the cellular uptake, biotransformation, and pharmacologic mechanisms of ADCs, the concentration of the active molecule (payload) in tumors could be much higher than that in plasma.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Hypothetical ADC payload concentration-time profiles in tumor, liver, and plasma. The figure was depicted partially based on recently published data in xenograft mice after intravenous administration of the CD22-conjugates of cyclobutyl-disulfide-PBD (Ma et al., 2016; Zhang et al., 2016).

Local Drug Delivery (Inhalation).

Local delivery is a direct way to increase the free drug concentration at a given tissue relative to plasma (K_p,uu > 1). There are many modes of local drug delivery, including dermal delivery to the skin, oral ingestion of gastrointestinal-restricted agents, intranasal delivery to the nasal cavity, and oral inhalation to the lungs. With multiple blockbuster therapies having benefited millions of patients (Stein and Thiel, 2017), inhalation to the lungs will be used to highlight the principles of local drug delivery for this review.

In preclinical species, inhalation doses are typically administered by intranasal delivery, intratracheal instillation, or tower dosing of dry powders or nebulized solutions and suspensions (Turner et al., 2011; Price et al., 2019). Lung tissue can be routinely harvested from preclinical species, allowing total drug levels to be measured directly and relationships between lung and plasma concentrations to be established. In humans, inhaled drugs are commonly delivered as dry powders from breath-actuated devices or as solutions or suspensions administered via pressurized metered devices or nebulizers (Stein and Thiel, 2017; Kaplan and Price, 2018). It is typically not feasible to measure drug concentrations directly in human lung tissue because of the invasive nature of lung biopsy (Esmailpour et al., 1997). Instead, it is common to measure drug concentrations in plasma and then infer lung concentrations based on lung-to-plasma correlations established in preclinical species.

A key advantage of inhaled and lung-targeted drugs is that they are typically administered in lower doses than are oral systemic drugs. This is the case for two reasons. First, an inhaled drug is not subject to first-pass metabolism by the liver or gut wall before reaching the lung as would be the case with an oral drug. Second, an inhaled drug is immediately concentrated at the site of action (the lung), as opposed to an oral drug that reaches the lung only after significant dilution in blood and other tissues.

In addition to minimizing the total dose, inhalation can lead to a lung-biased drug distribution. Therapeutic index may thereby be increased by maximizing pharmacology in the lung while minimizing its exposure elsewhere in the body; however, many drugs, particularly those that are highly soluble or highly intrinsically permeable, pass rapidly through the lung after inhalation and enter the systemic circulation via the pulmonary vasculature. Inhaled and lung-targeted drugs must therefore be designed with specific properties to slow permeation and prolong retention in the lung. These properties generally include one or more of the following: 1) low solubility/slow dissolution rate, 2) high tissue affinity imparted by strongly basic center(s) or high lipophilicity, 3) low membrane permeability, or 4) slow off-rate from the target of interest. Specific examples of inhaled drugs using these lung-retention strategies are illustrated in Supplemental Fig. S4 and Table 3. High systemic clearance is typically another desirable feature for lung-targeted drugs since, even with inhalation dosing, a compound may be absorbed systemically (Millan et al., 2011; Jones et al., 2017). As discussed already herein, the lung-deposited portion of the dose is subject to pulmonary absorption into systemic circulation. Additionally, a large fraction of a human inhaled dose is typically swallowed, which may allow it to be absorbed from the gastrointestinal tract (Taburet and Schmit, 1994; Hochhaus et al., 2015). High systemic clearance will rapidly eliminate any systemically absorbed compound and will maximize the ratio of drug in the lung relative to the blood and peripheral organs.

Figure 6A shows an example of a compound with high systemic clearance that is also poorly retained in the lungs after inhalation dosing. The compound diffuses quickly through the lung into systemic circulation, and there is negligible separation between lung and plasma concentrations. Lack of lung retention and high systemic clearance result in a short half-life in the lungs, and this profile is suboptimal for a lung-targeted drug. In contrast, Fig. 6B shows the profile of a compound with high systemic clearance that is well retained after inhalation dosing. The compound diffuses only slowly across the lung and, once eventually absorbed into blood, is metabolized rapidly. The result is a desirable profile for a lung-targeted inhaled therapy characterized by a long half-life in the lung and a marked separation between lung concentrations relative to plasma (K_p > > 1).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

Typical concentration-time profiles after inhalation of (A) a poorly lung-retained drug with high clearance in plasma; (B) a significantly lung-retained drug with high clearance in plasma.

Although determining K_p between lung and plasma in preclinical species is straightforward, measuring K_p,uu is more challenging because of the uncertainty in quantifying “free” drug in the lung. Indeed, concentrations measured in the lung represent an aggregate of all states of the drug, and it is not typically possible to differentiate drug that is unbound and soluble from drug that is bound to lung tissue or insoluble. For that reason, a K_p,uu > 1 is typically inferred only indirectly from preclinical in vivo PKPD models. Typically, this is accomplished either by 1) demonstrating a potent PD response in the lung but a diminished response in other tissues with inhaled dosing (Hemmerling et al., 2017) or 2) demonstrating equivalent lung PD via both intravenous and inhaled dosing but at lower plasma concentrations with inhaled dosing (Shaw et al., 2016). Finally, in the clinical setting, a K_p,uu > 1 is inferred by observing the desired PD effect or disease improvement in the lung, with a corresponding lack of on-target pharmacology in the periphery.

Nanoparticles.

Nanoparticles (NPs) have applications in targeted drug delivery that can create asymmetric drug distribution between tissues and plasma. Several categories of NP have been developed and are depicted as cartoons in Supplemental Fig. S5. NPs have unique disposition properties. Most engineered NPs are in the size range from 30 to 400 nm and as such do not undergo renal clearance or suffer from extensive metabolism by liver enzymes that require 5.5 nm for these clearance mechanisms (Choi et al., 2007); however, when NPs are intravenously administered, serum proteins known as opsonins adhere to the surface of the NPs, thereby making them amenable to uptake from mononuclear phagocyte cells that are a part of the reticuloendothelial system. This results in selective uptake of NPs in the liver and spleen.

Tumor and other inflamed tissues have a leaky vasculature through which larger-sized molecules can either enter or leave. This principle is used for the extravasation of NPs from the systemic circulation to tumor tissues. The effect has been labeled as enhanced permeability and retention (EPR) and has been the bedrock for passive targeting using NPs. The EPR effect is accumulative and is therefore significantly enhanced by increasing systemic circulation of the NPs. From a drug-distribution standpoint, the drug concentration in the tumor clearly will differ from other normal tissues and plasma owing to the EPR effect.

For therapeutic benefit, the NP cargo (drug) should be readily released at the site of action (i.e., the tumor cell). For active targeting, the NP is decorated with a ligand that will bind with a receptor expressed on the cell surface of the target cell. Once the ligand binds to the receptor, it facilitates receptor-mediated endocytosis transporting the drug into the cell. For example, Lippard and coworkers showed that a lethal dose of cisplatin could be safely delivered to prostate cancer cells by using an aptamer-decorated poly(lactic-co-glycolic acid-b-PEG NP that encapsulates cisplatin (Dhar et al., 2008). Park et al. (2002) synthesized long-circulating immunoliposomes of doxorubicin that were decorated with PEG chains and anti-HER2 monoclonal antibody fragments. These immunoliposomes were administered intravenously for targeting tumor cells. Given the mechanism of action, when using NPs either for passive or targeted delivery, the tissue concentrations can be substantially different from the plasma concentrations.