Chemicals.

Diclofenac, gemfibrozil, MPA, naloxone, propofol, alamethicin (from Trichoderma viride), UDP-glucuronic acid, EDTA, BSA, and saccharic acid lactone were purchased from Sigma-Aldrich (Gillingham, Dorset, UK). Ezetimibe was purchased from Toronto Research Chemicals Inc. (North York, ON, Canada). Telmisartan was purchased from Sequoia Research Products (Pangbourne, UK). All other reagents were of the highest grade available.

Source of the Microsomes.

Pooled HLM were purchased from BD Gentest (Woburn, MA). Pooled HKM and HIM were purchased from XenoTech (Tebu-Bio Ltd., Peterborough, Cambridgeshire, UK). Microsomes were stored at −80°C. HLM were pooled from 150 white donors, with 50% female and a mean age of 53 years (range, 18–79 years). HKM were pooled from 8 donors; with 88% white, 50% female, and a mean age of 61 years (range, 48–69 years). HIM were pooled from 18 donors; with 72% white, 39% female, and a mean age of 43 years (range, 21–67 years). Activity for pooled HLM was reported as 0.93, 0.85, 16.0, 2.00, and 0.72 nmol · min⁻¹ · mg protein⁻¹ for UGT1A1-, UGT1A4-, UGT1A6-, UGT1A9-, and UGT2B7-specific substrates. For HKM and HIM pools, activity was only reported for 4-methylumbelliferone (a substrate of multiple UGTs); with values of 125 and 8.44 nmol · min⁻¹ · mg protein⁻¹, respectively.

Microsome Experimental Conditions.

Incubations were performed in duplicate using a Thermomixer (Eppendorf, Hamburg, Germany). The final substrate concentration used was 1 μM, with the exception of propofol for which 5 μM was used. Substrate concentrations were >5-fold below the reported K_m values, when these data were available (Kiang et al., 2005); therefore, the enzyme kinetics assessed are expected to be in the linear range. Microsomal protein concentrations ranged from 0.25 to 1.0 mg/ml; details for individual drugs and tissues are provided in Supplemental Table 1. Microsomes were activated by incubation with alamethicin (50 μg/mg protein) on ice for 15 min (Fisher et al., 2000; Cubitt et al., 2009; Kilford et al., 2009). Substrates were preincubated with activated microsomes and 0.1 M phosphate buffer, pH 7.1, containing 3.45 mM magnesium chloride, 1.15 mM EDTA, and 115 μM saccharic acid lactone, for 5 min at 37°C shaken at 900 rpm (Fisher et al., 2001; Cubitt et al., 2009; Kilford et al., 2009). The reaction was initiated by addition of glucuronic acid (5 mM in incubation) to give a final incubation volume of 1 ml (Cubitt et al., 2009; Kilford et al., 2009). Stability of acyl glucuronides at pH 7.1 was tested using gemfibrozil as a representative example. Gemfibrozil acyl glucuronide was stable at pH 7.1 over the incubation time, and, therefore, this pH was considered to be suitable for microsomal assays of drugs that form this type of glucuronide in the current study. Concentration of the organic solvent used (methanol) was <0.6% of the incubation media. Control incubations were performed for each drug with no cofactor present to account for any potential cofactor-independent loss of the drug over the incubation time. The total length of the incubations ranged from 30 to 60 min (Supplemental Table 1). Samples of 60 to 100 μl of the incubation were removed at each time point and added to an equal volume of ice-cold methanol containing the internal standard as specified in Table 1 to terminate the reaction. Samples were frozen at −20°C for at least 1 h and then thawed and centrifuged (MSE Mistral 3000i centrifuge; MSE, London, UK) at 4°C and 2500 rpm for 30 min. An aliquot of the supernatant (10 or 20 μl) was analyzed by LC-MS/MS for parent drug concentration. Experiments were repeated on three separate occasions with the exception of those with HIM because of limited availability of the microsomal batch investigated.

Incubation conditions for experiments including BSA were comparable to those for experiments without BSA. In preliminary experiments, minimal depletion was observed for the acidic drugs when 2% BSA was used, and, therefore, CL_{int, UGT} could not be determined. This is most likely because acids bind strongly to BSA (Rowland et al., 2009), and the remaining free drug for depletion is minimal. Nonspecific binding of gemfibrozil and diclofenac at BSA concentrations ranging from 0.5 to 2% were determined (data not shown); nonspecific binding was reduced at lower BSA concentrations, but the resultant fraction unbound from protein in the incubation (f_{u, inc}) values remained <0.1. Previous research showed a plateau in the increase of CL_{int, u, UGT} results at 1% BSA (Rowland et al., 2007, 2009). Therefore, to allow quantifiable depletion and adequate sequestration of free fatty acids, a concentration of 1% BSA was used for the acidic drugs (diclofenac, gemfibrozil, MPA, and telmisartan), and 2% BSA was found to be suitable for bases/neutral drugs (ezetimibe, naloxone, and propofol). BSA concentration, microsomal protein concentrations, and the total length of the incubations are provided in Supplemental Table 1. The reaction was terminated by adding samples of the incubation to a double volume of ice-cold acetonitrile containing the internal standard. Samples were refrigerated for at least 10 min and then centrifuged (Mini Spin; Eppendorf) at 13,400 rpm and room temperature for 5 min. An aliquot of the supernatant (10 or 20 μl) was analyzed by LC-MS/MS for parent drug concentration. Experiments were repeated 3 times with the exception of those for HIM for which limited availability prevented repeat analysis.

LC-MS/MS.

All compounds were analyzed on a Waters 2790 with a Micromass Quattro Ultima triple quadruple mass spectrometer in negative mode with the exception of naloxone and telmisartan, which were analyzed on a Waters 2790 with a Micromass Quattro Micro triple quadruple mass spectrometer in positive mode. Source temperature was 125°C, desolvation temperature was 350°C, desolvation gas rate was 600 l/h, and cone gas rate was 150 l/h, with the exception of MPA and ezetimibe for which cone gas was 50 l/h. The capillary voltage was 3.25 kV for the Micromass Quattro Ultima and 3.5 kV for the Micromass Quattro Micro. Propofol was analyzed using single ion recording owing to its limited fragmentation. Analytes were separated using a Luna C18 (3 μm, 50 × 4.6 mm) column (Phenomenex, Macclesfield, UK) with the exception of naloxone, for which a Luna Phenyl Hexyl (3 μm, 50 × 4.6 mm) column (Phenomenex) was used. Four mobile phases were used, with varying gradients for each drug: A, 90% water, 10% methanol, and 0.05% formic acid; B, 10% water, 90% methanol, and 0.05% formic acid; C, 90% water, 10% methanol, and 1 mM ammonium acetate; and D, 10% water, 90% methanol, and 1 mM ammonium acetate. The flow rate was 1 ml/min, splitting to 0.25 ml/min before entry into the mass spectrometer. Mass spectrometric conditions for each drug are detailed in Table 1.

Correction for Nonspecific Protein Binding.

The high throughput dialysis method (Gertz et al., 2008) was used to determine the f_{u, inc} values for all drugs at 0, 0.1, 0.25, 0.5, and 1.0 mg/ml HLM protein concentrations in the presence and absence of 1 or 2% BSA. Al-Jahdari et al. (2006) reported similar f_{u, inc} values for propofol in HLM and HKM; therefore, nonspecific binding was assumed to be comparable across systems at the same protein concentration. At the low substrate concentrations used herein, nonspecific binding is expected to be below the saturation limit and therefore independent of substrate concentration. This assumption is supported by previous data showing that nonspecific binding of propofol (in the presence and absence of BSA) and MPA (in the absence of BSA) is independent of substrate concentration when determined over a range of concentrations exceeding those used in the current study (Bowalgaha and Miners, 2001; Rowland et al., 2008, 2009). Dialysis membranes (12–14 kDa molecular mass cutoff) were purchased from HTDialysis, LLC (Gales Ferry, CT). Phosphate buffer (0.1 M, pH 7.1), containing 3.45 mM magnesium chloride, 1.15 mM EDTA, 115 μM saccharic acid lactone, and, where applicable, 1 or 2% BSA was added to the acceptor side of the membrane, and the relevant concentration of HLM in buffer was added to the donor side. Each HLM protein concentration was tested in triplicate. A 1 μM concentration of drug (5 μM for propofol) was spiked into each side of the membrane. The plate was left to equilibrate for 6 h on a plate shaker (450 rpm) at 37°C. Then samples from both the acceptor and donor sides of the membrane were transferred to Eppendorf tubes containing an equal volume of methanol with the relevant internal standard for non-BSA samples or a double volume of acetonitrile with internal standard for BSA samples. Sample preparation and LC-MS/MS methods were the same as those for microsomal incubations.

Recovery from equilibrium dialysis was greater than 70% for all drugs with the exception of propofol. Although recovery of propofol was low (<30%), f_{u, inc} values determined both in the presence and absence of BSA were comparable to values reported by Rowland et al. (2008), and so the experimental data have been used. The f_{u, inc} values for the other drugs investigated (both in the presence and absence of BSA) were comparable to values reported in the literature (where available) (Bowalgaha and Miners, 2001; Al-Jahdari et al., 2006; Gertz et al., 2008; Cubitt et al., 2009; Kilford et al., 2009). From preliminary experiments, the volume shift between the donor and acceptor side of the membrane was generally less than 5%, and therefore correction for volume shift had minimal impact on the experimental f_{u, inc}.

Nonspecific binding data were analyzed across the range of protein concentrations in GraFit 5 (Erithacus Software, Horley, UK) to determine the binding constant (K_a) for each drug. The K_a was then used to calculate the f_{u, inc} at the protein concentrations used for the glucuronidation depletion assays (eq. 1). The experimentally determined f_{u, inc} values in the presence of BSA were constant over the range of microsomal protein concentrations and, therefore, K_a could not be calculated. In these cases, the mean of the experimental f_{u, inc} values was used for all protein concentrations with BSA. The CL_{int, UGT} values from HLM, HKM, and HIM were corrected for f_{u, inc} (CL_{int, UGT}/f_{u, inc}) to generate the unbound intrinsic clearance (CL_{int, u, UGT}) (microliters per minute per milligram of protein). where C is the microsomal protein concentration.

Data Analysis.

The mean concentration of the duplicate samples at each time point was analyzed using GraFit 5 to determine the elimination rate constant (k). A nonlinear single exponential fit was used, with the exception of diclofenac in HLM without BSA for which a double exponential fit was used. The rate constant was used to calculate the CL_{int, UGT} (eq. 2) and corrected for nonspecific binding to give CL_{int, u, UGT} (microliters per minute per milligram of protein).

To compare CL_{int, u, UGT} values from different tissues, CL_{int, u, UGT} was expressed per gram of tissue by correcting the values for the microsomal protein yield, resulting in scaled CL_{int, u, UGT.} Microsomal protein yields of 40, 12.8, and 20.6 mg of protein/g tissue were used for hepatic, renal, and intestinal data, respectively (Al-Jahdari et al., 2006; Barter et al., 2007; Cubitt et al., 2009). In contrast with hepatic microsomal recovery, only one estimate of microsomal recovery for HKMs could be found, based on data from only five donors and an unspecified kidney region (Al-Jahdari et al., 2006). Data from rats support this report; relative microsomal recovery for kidney is 30% of that for liver. The mean scaled CL_{int, u, UGT} and S.D. have been presented for each drug. The mean scaled clearance values were compared using ratios.

Prediction of Intravenous Glucuronidation Clearance.

Ezetimibe cannot be dosed intravenously, and the lack of human fraction absorbed and fraction escaping gut metabolism data prevent the conversion of oral plasma clearance data to CL_{int, UGT}; therefore, ezetimibe was not included in IVIVE. In vivo pharmacokinetic parameters including intravenous plasma clearance (CL_i.v.), plasma binding (f_{u, p}), blood/plasma ratio (R_B), and renal clearance were collated from the literature for the remaining six drugs. In cases for which multiple clinical studies were available, the mean was calculated, weighted for the number of subjects in each study; references for the in vivo parameters and values from individual studies are shown in Supplemental Tables 2 to 6. Clinical data from Japanese, Korean, and Chinese populations, disease populations, subjects receiving concomitant medication, elderly or obese populations, extended release formulations, and cases in which the area under the curve was reported over an insufficient time period were excluded. No nonlinearity issues were observed across multiple dose levels, so all data were included. Only plasma clearance data were used, with the exception of propofol for which blood data were reported. CL_i.v. data for each drug were collated from up to 12 clinical studies (up to 121 subjects in total) (Supplemental Table 2). Data for renal clearance were more limited and for the majority of drugs were available from only one report in the literature. For many of the drugs R_B values were determined in vitro and intersubject variability in this parameter was not assessed. The f_{u, p} data were more readily available in the literature, and values for individual drugs were collated from up to four studies (<59 subjects in total per drug). Intersubject variability for CL_i.v. and f_{u, p} was generally <30% within individual studies; values varied to a similar extent between studies, with the exception of f_{u, p} for mycophenolic acid for which values from different studies varied by 48%. The ranges of values from individual studies collated for each parameter are shown elsewhere (Supplemental Tables 2–6).

Plasma CL_i.v. of all drugs was corrected for R_B and renal clearance to give blood metabolic clearance (CL_H) (Ito and Houston, 2005; Gertz et al., 2010). Clearance via metabolism was corrected for the fraction metabolized by glucuronidation (f_{m, UGT}) to give the apparent in vivo glucuronidation clearance (CL_UGT). Where available, data were corrected using in vitro f_{m, UGT} from HLM data in the presence or absence of BSA (from in-house data), which were generated in a pool of HLM with UGT and cytochrome P450 characterization similar to that used for the current work. In the case of MPA and telmisartan, no in vitro f_{m, UGT} data were available; however, for these two drugs in vivo data suggest almost complete glucuronidation (f_{m, UGT} = 1 for telmisartan or 0.95 for MPA), and this assumption was applied in IVIVE.

Currently, there are no kidney-specific models available for IVIVE; therefore, perfusion-limited kinetics and assumptions of the well stirred model were applied for the kidney, which may have biased the IVIVE results presented herein. In vitro HLM and HKM CL_{int, u, UGT} data (per gram of tissue) were scaled to give CL_{int, u, UGT} (per kilogram body weight) using average organ weights of 21.4 g of tissue/kg (Ito and Houston, 2005) and 4.5 g tissue/kg for liver and kidney, respectively. The average kidney weight (range from 4.1 to 5.0 g tissue/kg) was estimated from the collated literature reports (Supplemental Table 7). These data were then scaled up using the well stirred model to give apparent glucuronidation clearance for the respective organ (eq. 3) (Houston, 1994). The intestine was expected to have minimal impact on glucuronidation clearance after intravenous drug administration; therefore, HIM CL_{int, u, UGT} data were not included during IVIVE. where f_{u, B} represents fraction unbound in blood, Q_h is hepatic blood flow (Q_h = 20.7 ml · min⁻¹ · kg⁻¹), and CL_UGT represents apparent in vivo glucuronidation clearance. An analogous model was used for the kidney where Q_h was replaced with Q_r (average renal blood flow of 16.4 ml · min⁻¹ · kg⁻¹) and in vitro HKM data were used. The values reported for renal blood flow, gender differences, and methods used to estimate this parameters were collated from >50 studies reported in the literature (Supplemental Table 8).

For IVIVE using both the liver and kidney data, combined CL_UGT was calculated as the sum of scaled hepatic and renal CL_UGT. Predicted CL_UGT values from HLM data alone or in combination with HKM data were compared with the observed CL_UGT values from in vivo data. Both observed CL_UGT and predicted CL_UGT values were calculated using f_{m, UGT} and in vitro scaled CL_{int, u, UGT} data determined in the presence of BSA (where available). Similar comparisons were made using data in the absence of BSA. Prediction bias and precision were assessed using geometric fold error and root mean squared error, respectively, as described previously (Gertz et al., 2010).