Abstract
The fraction of unbound drug (fuinc) in in vitro intrinsic clearance (CLint) incubation is an important parameter in the pursuit of accurate clearance predictions and is often predicted using algorithms based on drug lipophilicity measures. However, analysis of an AstraZeneca database suggests that simple lipophilicity alone is a relatively poor predictor of fuinc measured using equilibrium dialysis. He fuinc value can also be measured directly in CLint assays using multiple concentrations of hepatocytes or microsomal protein. Since this approach informs of the unbound drug concentration in the assay used to predict in vivo clearance, it should be considered the gold standard method. As a starting point for building better predictive algorithms we aimed to determine if equilibrium dialysis really is an appropriate assay for assessing fuinc. Employing a large number of compounds with a wide range of lipophilicities, experiments were performed to measure fuinc using rat hepatocytes (RH) and human liver microsomes (HLM) in both assay formats. A high percentage (94% and 93% for HLM and RH, respectively) of the fuinc values were within 2-fold when the compound distribution coefficient describing the ratio of compound concentration in octanol and pH 7.4 buffer when the test system is at equilibrium (lipophilicity measure) (logD7.4) values were less than 3.5. However, with logD7.4 values greater than these, the agreement was considerably worse. Additional experimental data generated indicated that this discrepancy was likely due to failings in the direct method when drug binding is high. Thus, we conclude that unbound CLint can be indeed calculated indirectly by incorporating equilibrium dialysis data with measured CLint but that simple lipophilicity descriptors alone may be inadequate for predicting fuinc.
Introduction
Prediction of in vivo hepatic metabolic clearance from in vitro data uses measured hepatocyte or microsomal metabolic intrinsic clearance (CLint); fraction of unbound drug in incubation (fuinc), which is assumed to be at 1 mg/ml microsomal protein or 1 million hepatocytes/ml unless otherwise stated; and fraction of drug unbound in blood (Obach, 1996; Ito and Houston, 2005). To add confidence to human clearance predictions for any given candidate drug, observed unbound in vivo CLint should first be predicted to within 2-fold in two preclinical species (Grime et al., 2013). To meet such stringent criteria, attention to fine detail on experimental conditions and physiologic relevance of the in vitro measurements is paramount. In a drug discovery setting, fuinc is typically measured using equilibrium dialysis in which partitioning of drug between two chambers either side of the dialysis membrane (one holding inactivated microsomes or hepatocytes and one holding buffer) is determined. This assay allows high throughput of compounds but does not necessarily represent the conditions experienced by a drug at the site of drug metabolism in a CLint assay, and thus assessment of unbound CLint directly in the metabolic stability assay may be viewed as the ideal. With this in mind, an approach for determining fuinc using metabolic stability data generated at multiple concentrations of hepatocytes or microsomal protein can be used, since the rate at which a compound is metabolized is proportional to the unbound concentration available to enzymes (Giuliano et al., 2005; Grime and Riley, 2006; Nordell et al., 2013). The pharmaceutical industry is under great pressure to reduce costs and as such it is an appropriate aim to minimize wet screening where possible. As such, fuinc in silico prediction algorithms based on ion class and lipophilicity [the partition coefficient or distribution coefficient at pH 7.4 (logD7.4, i.e., the distribution coefficient describing the ratio of compound concentration in octanol and pH 7.4 buffer when the test system is at equilibrium—the lipophilicity measure) octanol/water partition coefficients] (Austin et al., 2002, 2005; Kilford et al., 2008) as well as other molecular descriptors (Gao et al., 2008, 2010; Abraham and Austin, 2012; Nair et al., 2016) have become widely used. On review of an in-house database, we noted that simple lipophilicity based predictions were within 2-fold of equilibrium dialysis measured values for only 64% human liver microsomes (HLM) and 62% rat hepatocytes (RH) of compounds (Fig. 1). As a starting point for building better fuinc predictive algorithms we considered the validity of equilibrium dialysis measures and compared HLM and RH fuinc values generated by this approach and directly from the CLint assays. The data sets cover 58 and 74 AstraZeneca proprietary compounds for HLM and RH, respectively, spanning over five orders of magnitude in measured lipophilicity.
Material and Methods
Materials.
UltraPool HLM (mixed gender, lot number 38289) was purchased from BD Gentest (San Jose, CA) and cryopreserved Han Wistar RH (mixed gender, lots RJQ and RAL) was purchased from Bioreclamation IVT (Brussels, Belgium). L-15 Leibovitz medium and phosphate buffers were obtained from Life Technologies (Grand Island, NY). NADPH was obtained from Sigma (St. Louis, MO). Compound stock solutions of dimethylsulfoxide (DMSO) (10 mM) were obtained from AstraZeneca Compound Management.
Measurement of logD7.4 and Determination of fuinc by Equilibrium Dialysis.
The logD7.4 value was measured as previously described (Austin et al., 2002). Equilibrium dialysis was performed to determine extent of compound binding to HLM (1 mg/ml in 0.1 M, pH 7.4 phosphate buffer) and freshly isolated RH (1 million cells/ml in Leibovitz media). Compound stocks (100 μM in DMSO) were added at a ratio of 10 μl/ml HLM or RH suspensions. Inactivation of drug metabolizing enzymes was performed in accordance with previous studies (Austin et al., 2005) using 1-aminobenzotriazole (400 mM in DMSO) and salicylamide (300 mM in DMSO), added to the preparations at a ratio of 2.5 and 5 μl/ml, respectively, followed by 1 hour of preincubation at 37°C. Samples were checked for drug metabolism by analyzing parent drug pre- and postdialysis experiments. Experiments were performed in triplicate. Dialysis membranes were soaked in ultrapure water for 60 minutes, in 20% ethanol for 20 minutes, and finally in phosphate buffer (0.1 M, pH 7.4) for 20 minutes prior to loading into 96-well dialysis devices (HTDialysis LLC, Gales Ferry, CT). An aliquot (150 μl) of HLM spiked with test compound (1 μM) and drug metabolism inhibitors was added to donor wells of the dialysis device with phosphate buffer (150 μl, 0.1 M, pH 7.4) added to acceptor wells. The dialysis device was loaded onto a rotor housed within an air bath. After 4 hours of incubation (37°C, 300 rpm), aliquots (50 μl) from each donor and acceptor well were transferred to a 96-deep well plate. Samples were quenched with acetoniltrile (100 µl), diluted, analyzed by liquid chromatography–tandem mass spectrometry, and quantified from a standard curve. Free fraction of drug was determined as previously described (Austin et al., 2002). RH was isolated from adult male Sprague-Dawley rats (250–300 g) as described previously (Kenny and Grime, 2006) and resuspended in Leibovitz media at 1 million cells/ml. Leibovitz medium (150 μl) was added to the acceptor chamber and hepatocyte suspension (150 μl) containing compound (1 μM) and drug metabolizing enzyme inhibitors was added to the donor chamber, following the dialysis procedure as previously described. All liquid chromatography–tandem mass spectrometry analyses were performed on a Waters triple quadrupole mass spectrometer (Waters, Milford, MA), coupled to a liquid chromatography sample and solvent manager system, as described previously (Nordell et al., 2013). To investigate if nonlinear drug binding, with respect to concentration of enzymes, was likely to influence the estimation of fuinc, equilibrium dialysis experiments were performed as described previously, at HLM concentrations of 0.125, 0.5, 1, and 4 mg/ml for 25 compounds.
Metabolic Stability at Multiple Microsomal Protein or Hepatocyte Concentrations.
The CLint value was determined in duplicate for each of the HLM or RH concentrations used. HLM stock solutions were thawed and diluted in phosphate buffer (0.1 M, pH 7.4) containing NADPH (1 mM) to the following protein concentrations: 0.14, 0.21, 0.28, 0.41, 0.56, 0.83, 1.11, 2.22, and 4.44 mg/protein/ml. Suspensions were mixed with compound stock solutions (50 μM in 50% acetonitrile) to a final concentration of 0.5 μM and incubated in a 96-well plate at 37°C with plate shaking. Reactions were terminated at 0.5, 5, 10, 15, 20, and 30 minutes by quenching with cold acetronitrile containing 0.1% formic acid, and subsequently analyzed by liquid chromatography–tandem mass spectrometry. RH was suspended in Leibovitz media to the following cell concentrations: 0.13, 0.19, 0.25, 0.38, 0.50, 0.75, 1.0, 2.0, and 4.0 million cells/ml. Cell suspensions were mixed with compound solutions (50 μM in 50% acetonitrile) to a final concentration of 0.5 μM and incubated in a 96-well plate at 37°C with plate shaking. Reactions were terminated (0.5, 5, 15, 20, 30, 45, 60, 80, 100, and 120 minutes) and analyzed by as described previously.
Model of Incubational Binding.
Incubational binding, treated as an equilibrium between drug in aqueous and the membrane phases, is described by a partitioning constant Kp, which is related to the fuinc: Kp = drug bound to the membrane phase/unbound in the aqueous phase = (1 − fuinc/fuinc) (Austin et al., 2002). Since only unbound drug is available to metabolizing enzymes, the measured CLint becomes a function of the microsomal or hepatocyte concentration, related to the unbound metabolic intrinsic clearance (CLint,u): CLint = fuinc × CLint,u. The fraction unbound at standard conditions (microsomal protein or hepatocyte concentration of 1 mg/ml or 1 million cells/ml, respectively), fuinc,0, is related to the fraction unbound and CLint at a second enzyme concentration, allowing definition of the true unbound intrinsic clearance, CLint,u (where CLint represents the observed intrinsic clearance at any enzyme concentration C).
(1)Direct Determination of fuinc from In Vitro Depletion Curves.
Analytical peak areas of samples, withdrawn from drug incubations at the various HLM or RH concentrations, were loge transformed (as exemplified in Fig. 2A) and elimination rate constants k (min−1) were derived from the slopes of the loge[substrate]-time plots. Since CLint,= k × V, where V represents the incubation volume (ml/mg protein or ml/million cells), the entire data set from multiple microsomal protein or hepatocyte concentrations was used to define fuinc and CLint,u (from eq. 1) as follows: using the nonlinear least squares solver lsqnonlin in MATLAB 2014b (MathWorks, Inc., Natick, MA), the sum of squares of the vertical distances between the loge-transformed percentage substrate remaining and the linear best fit at all HLM or RH concentrations was minimized and the intercept (from the family of CLint values analyzed with their associated microsomal protein or hepatocyte concentrations, Fig. 2B) was allowed to vary freely for optimal fit. Prior to summation, squared residuals were weighted to account for the number of data points included at each enzyme concentration.
Predicted In Vivo Rat Clearance.
Clearance predictions were made using a regression line approach, whereby an existing in vitro-in vivo unbound CLint data set (for which in vivo CLint values represent metabolic clearance only) was used as a framework for predicting the in vivo clearance for novel compounds (Grime and Riley, 2006; Sohlenius-Sternbeck et al., 2012).
Results and Discussion
With ionic class and lipophilicity being identified as the main drivers of incubational binding (Austin et al., 2002, 2005; Kilford et al., 2008) a set of compounds was selected from the AstraZeneca compound library comprising acids (A), bases (B), and neutrals (N) that each covered a broad range of logD7.4 values. The fraction unbound was determined in triplicate by equilibrium dialysis. For HLM, 58 (A: 14; B: 20; N: 22) compounds spanning logD7.4 values of 0.5–5.3 (A: 1.9–4.2; B: 0.5–4.8; N: 0.2–5.3) had measured fuinc values (at 1 mg/ml HLM protein) of 0.001–0.95 (A: 0.02–0.75; B: 0.001–0.87; N: 0.001–0.95) defined. For RH, 74 (A: 21; B: 20; N: 20) compounds spanning logD7.4 values of −0.3 to 5.3 (A: −0.4 to 4.5; B: −0.3 to 4.1; N: 0.9–5.3) had fuinc values (at 1 million cells/ml) of 0.004–0.996 (A: 0.009–0.71; B: 0.03–0.91; N: 0.004–0.995) defined. The fuinc value was also determined directly from the CLint assays as described previously. In the RH equilibrium dialysis experiments, 1-aminobenzotriazole was used as a cytochrome P450 (P450) inhibitor. Although a very effective CYP3A4 inhibitor, other enzymes may be less effectively inhibited including CYP2C9, for which over 60% of activity may remain using 1 mM and 30-minute preincubation (Linder et al., 2009). This was a compromise to avoid the use of several individual P450 inhibitors designed to remove all P450 activity, but at a cost of greater organic solvent addition. The preincubation was 60 minutes and since the HLM equilibrium dialysis incubations contained no NADPH and the results were in line with the RH experimental data, it is apparent that the use of 1-aminobenzotriazole as a P450 inhibitor did not compromise the results. In confirmation of this, biotransformation assessment of all equilibrium dialysis incubations revealed no drug metabolism in the RH incubates. The DMSO concentration of 1% v/v used in all metabolic stability incubations may give rise to a suboptimal rate of metabolism for some drug metabolizing enzymes. However, since the relative contribution of individual P450 to the overall CLint is not relevant in determining fuinc by eq. 1, this compromise was made to ensure the solubility of all compounds since many were relatively lipophilic.
Figure 3 shows a log-scale representation of the agreement between incubational binding estimates given by the two assays for acids, bases, and neutrals in RH and HLM incubations. Only 65% (HLM) and 77% (RH) of the fuinc values from the direct assay fell within 2-fold of the values provided by the equilibrium dialysis assay. However, 94% and 93% (HLM and RH, respectively) of the fuinc values were within 2-fold when compound logD7.4 values were less than 3.5. Although this is a somewhat arbitrary cutoff, it allows some quantification of the lipophilicity effect. When considering compounds with logD7.4 > 3.5 only 25% of HLM and 27% of RH fuinc values were within 2-fold. For these compounds (logD7.4 > 3.5), there was significant bias (P = 1 and 7 × 10−4 for HLM and RH, respectively), with fuinc values determined by equilibrium dialysis consistently being lower than estimates from direct assessment using substrate depletion data. Compounds with logD7.4 < 3.5 showed no indication of bias. Using HLM concentrations of 0.125, 0.5, 1, and 4 mg/protein/ml in equilibrium dialysis experiments, we confirmed that nonlinear drug binding, with respect to HLM concentration, was not an influencing factor in the difference in fuinc values from the two assay formats; for 25 compounds, equilibrium dialysis incubational binding increased proportionally with increasing HLM concentration (data not shown). Overall the analysis indicates that more lipophilic compounds may have a strong bias toward equilibrium dialysis yielding lower fuinc estimates than the direct approach, independent of whether whole cells or microsomes are studied. In line with this, Giuliano et al. (2005) identified two outliers, both extensively bound, deviating from the otherwise good correlation between equilibrium dialysis and metabolic stability based fuinc when investigating HLM-drug binding behavior. A possible explanation for these observations is that when incubational binding is high, the rapidly changing CLint-enzyme concentration curve (Fig. 2B) is not sufficiently well-described by the input data at low hepatocyte or microsomal protein concentrations to allow appropriate back-extrapolation to the true CLint,u value (Y-axis intercept), making the fuinc direct from CLint approach less reliable for more lipophilic compounds.
To substantiate the idea that for molecules with higher hepatocyte binding equilibrium dialysis gives a more accurate description of binding, we investigated the in vivo rat clearance of two of the compounds with a big discrepancy in fuinc from the two assays. Using the previously described method for predicting clearance at AstraZeneca with input rat hepatocyte CLint, fuinc, and plasma protein binding data (Sohlenius-Sternbeck et al., 2012), the clearance predictions were 37 and 1 ml/min/kg when using fuinc determined from the equilibrium dialysis method and the direct CLint approach, respectively, for the first compound. The observed rat plasma clearance was 65 ml/min/kg. Similarly, for the second compound the predicted clearance values were 48 and 10 ml/min/kg using equilibrium dialysis and the direct method fuinc, respectively, while the observed clearance was 51 ml/min/kg. Although such an assessment is not completely robust since there can be other explanations for poor clearance prediction, this evidence does point to the greater general utility of equilibrium binding data for making accurate clearance predictions. This study supports the long-held assumption that unbound CLint, required for predicting in vivo hepatic metabolic clearance, can indeed be calculated indirectly using equilibrium dialysis data and CLint taken under standard conditions. Additionally, our analysis of an in-house database indicates that prediction of incubational binding requires more sophisticated prediction approaches than simple lipophilicty algorithms, as discussed elsewhere (Gao et al., 2010; Nair et al., 2016).
Authorship Contributions
Participated in research design: Grime, Nordell, Bergström, Chen, Prieto Garcia.
Conducted experiments: Chen, Prieto Garcia.
Performed data analysis: Chen, Prieto Garcia, Bergström, Nordell.
Wrote or contributed to the writing of the manuscript: Grime, Nordell.
Footnotes
- Received October 26, 2016.
- Accepted January 18, 2017.
Abbreviations
- Clint
- metabolic intrinsic clearance
- CLint,u
- unbound metabolic intrinsic clearance
- DMSO
- dimethylsulfoxide
- fuinc
- fraction of unbound drug in incubation
- HLM
- human liver microsomes
- logD7.4
- distribution coefficient at pH 7.4
- P450
- cytochrome P450
- RH
- rat hepatocytes
- Copyright © 2017 by The American Society for Pharmacology and Experimental Therapeutics