SEARCH

SEARCH BY CITATION

Keywords:

  • hepatocytes;
  • intrinsic clearance;
  • unbound fraction;
  • computational ADME;
  • in vitro–in vivo extrapolation;
  • in vitro–in vivo correlation;
  • IVIVE;
  • pharmacokinetics;
  • physiologically based pharmacokinetics;
  • PBPK modeling

Abstract

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHOD
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

The objective of this study was to follow up a previous study on a comparative analysis of diverse in vitroin vivo extrapolation (IVIVE) methods used for predicting hepatic metabolic clearance (CL) of drugs from intrinsic clearance (CLint) data determined in microsomal incubations, but using hepatocyte data instead. Six IVIVE methods were compared: the “conventional and conventional bias-corrected methods,” the “regression equation method,” the “direct scaling method,” the “Berezhkovskiy's method,” and the “novel IVIVE method of Poulin et al.” offering a new paradigm. A large and diverse dataset of 49 drugs were collected from the literature for hepatocyte data in human. Based on all statistical parameters, this study confirms that the novel IVIVE method of Poulin et al. shows the greatest prediction performance among the IVIVE methods tested by using hepatocyte data. The superior prediction performance of this novel IVIVE method is again most pronounced for (a) drugs highly bound in blood, (b) drugs bound to albumin, and (c) low CL drugs. Because the novel IVIVE method has been developed particularly to improve the prediction accuracy for drugs with such properties, this study confirms its utility. Furthermore, the results of the current comparative analysis performed using hepatocyte data confirm the findings of a previous analysis made with microsomal data. Overall, the proposed novel IVIVE method offers a new paradigm for the prediction of hepatic metabolic CL particularly for drugs, which have the aforementioned properties, and, hence, this would contribute to a more accurate CL prediction for small molecules in drug discovery and development, interspecies scaling, and can potentially be used for the optimization of driving factors of CL in an attempt to facilitate the simulation of drug disposition by using the physiologically based pharmacokinetics (PBPK) model. © 2013 Wiley Periodicals, Inc. and the American Pharmacists Association J Pharm Sci 102:3239–3251, 2013

INTRODUCTION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHOD
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

Hepatic metabolic clearance (CL) is one of the more essential pharmacokinetics (PK) parameters to estimate in drug discovery and development. It is common practice to use in vitroin vivo extrapolation (IVIVE) methods to scale up the intrinsic clearance (CLint) determined in vitro in the human liver preparations for predicting CL in vivo of drugs that are mainly eliminated by metabolism.1,2 Recently, a comparative analysis using a dataset of 139 drugs obtained in preclinical species and human was performed for five promising IVIVE methods2; (i) the “conventional bias-corrected method” correcting for the average-fold error of deviation (AFE) observed with the conventional method, (ii) the “conventional method” using the measured unbound fraction in blood (fub) as the sole correction factor, (iii) the “Berezhkovskiy's method” using an apparent fub value, which is fub adjusted for drug ionization on either side of the plasma membrane based on pH differences, and the unbound fraction in the incubation medium (fuinc) as the correction factors, (iv) the “Poulin et al. method” offering a new paradigm by using fuinc and the apparent unbound fraction in the liver (fuliver), which accounts for both the pH differences and the protein-facilitated uptake due to potential ionic interactions between the protein-bound drug complex and cell surface of the hepatocytes, and finally (iv) the “direct scaling method” that does not consider any binding corrections. This comparative analysis confirmed the superior accuracy of the novel method developed by Poulin et al.1,2 for calculation of hepatic metabolic CL particularly for the following drug properties; (a) drugs highly bound in blood, (b) drugs bound to albumin (AL), and (c) low CL drugs, which corroborates a previous sensitivity analysis illustrating that the novel IVIVE method differed to other IVIVE methods particularly for such drug properties.2 Recently, the novel IVIVE method was also compared with the empirical conventional bias-corrected method proposed by Halifax and Houston3 using another dataset. This second comparison exercise demonstrated, once more, that the novel mechanistic IVIVE method of Poulin et al.1,2 had superior precision and lower bias in the majority of cases.3

So far, the prediction performance of these IVIVE prediction methods has been assessed by using microsomal data only.1,2 Because in vitro hepatocyte data are also important data generated in drug discovery for CL prediction, an extension of the recently published comparative analysis of IVIVE methods from microsomal data to hepatocyte data was necessary. Very recently, Sohlenius-Sternbeck et al.4 presented an additional IVIVE method for removing the systematic bias through application of empirical correction factors derived from regression analyses between the in vitro hepatocyte data and in vivo CL data for a defined set of reference compounds. Indeed, this method has never been compared with other IVIVE methods. A weakness of the regression equation of Sohlenius-Sternbeck et al.4 and the conventional bias-corrected method of Halifax and Houston3 are that their development required previous optimization from in vivo CL data of several datasets, which depends purely on the composition of the datasets (i.e., number and disposition of compounds). The objective of this study was therefore to follow up a previous study on a comparative analysis of diverse IVIVE methods used for predicting CL of drugs from microsomal data, but using hepatocyte data instead. Hence, the performance of the Poulin et al., method1,2 will be challenged to five other IVIVE methods for the same datasets and the most accurate IVIVE method will be identified for hepatocyte data.

METHOD

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHOD
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

The methodology used to attain our goal consisted of comparing predicted CL derived from in vitro hepatocyte data and in vivo CL values observed in humans for several drugs. Five IVIVE methods that have undergone previous comparative assessments1–3 were the focus of further evaluation in this study. In addition, the recently published “regression method” of Sohlenius-Sternbeck et al.4 was also included in the current comparative analysis. Therefore, a total of six IVIVE methods were challenged in this study. Overall, this study consists of an independent validation exercise of several IVIVE methods with a series of drugs covering a broad range of physicochemical and plasma protein-binding properties.

Datasets and Assumptions

The dataset consist of drugs for which hepatocyte data in humans were available in the literature.4–9 The drug dataset and the corresponding experimentally determined input parameters are presented in Table 1. Data on CLint and fuinc were obtained experimentally by using blood free viable suspended hepatocyte incubations under in vitro conditions. The total number of drugs studied here is 49 (14, acids; 21 bases; 14 neutrals) covering a large range of drug properties. The drugs used in this study are thought to be eliminated by hepatic oxidative metabolic CL under in vivo conditions in humans. It was assumed that distribution of the current drugs into the hepatocytes was not impeded by limited diffusion and/or transport processes as reported in the literature.10 However, four drugs, namely, cyclosporine, montelukast, prazocin, and ritonavir, were not considered for the purpose of this study because their CLint in the hepatocytes could be determined by permeability and/or transport processes rather than by the metabolism, at least under in vitro conditions.11 Furthermore, two other drugs (irbesartan and troglitazone) were also not considered in this study. First, irbesartan is metabolized via glucuronide conjugation, and oxidation by the cytochrome,12 and, hence, it was not considered in the current analysis because the importance on in vivo blood CL of the glucuronidation in the liver versus renal and intestinal glucuronidation is unknown. For example, the glucuronidation of drugs in kidneys might be significant.13 Second, troglitazone undergoes extensive hepatic metabolism; sulfation, glucuronidation, and oxidation,14 and, consequently, it was also not used for the analysis for the same reason. The additional modeling assumptions are the same as previously published by Poulin et al.1,2

Table 1. Test Set of Drugs Used in this Study1,2,4–9,15
  Physicochemical DataHuman Data
DrugsMain Binding ProteinClasspKa1,acidpKa2,acidpKa1,basepKa2,basefubloodfuincRange of CLint obtained from the literature1,4–9 [µl/(min 106 cells)]Range of observed blood CL obtained from the literature1,4–9 [ml/(min kg)]
  1. aEstimated from the unbound CLint corrected for fub and fuinc presented in Riley et al.6

  2. A, acid; B, base; N, neutral; DA, diprotic acid; DB, diprotic base; AL, albumin; AAG, alpha1-acid glycoprotein.

Acids
DiclofenacALA4   0.00580.841507.33
         32.76.4
DiflunisalALA3.3   0.00530.9240.18
         2.10.2
EtodolacALA4.7   0.01690.91291.31
       0.008 7.91.2
FenoprofenALA4.5   0.00850.26211.69
         7.21.7
FurosemideALA3.8   0.02450.9220.59
GemfibrozilALA4.7   0.00470.671183.1
         30.7 
GlipizideALA5.9   0.01670.3330.96
         0.901
IbuprofenALA4.4   0.01820.150.591.4
       0.01 3.7 
IndomethacinALA4.5   0.01730.95102.24
KeteprofenALA4.45   0.0240.6582.22
         2.31.7
OxaprozinALA4.3   0.00110.2980.10
         1.60.071
TenoxicamALDA1.15  0.01640.7830.07
         0.7 
TolbutamideALA5.27   0.0310.8130.3
       0.021 0.70.4
WarfarinALA5   0.0180.810.08
         0.2 
Bases
BuspironeAAGB  7.32 0.050.6115715.9
CarvedilolALB  8.1 0.02950.68428.7
       0.009 4411
ChlorpromazineAAGB  9.7 0.050.8818.18.6
       0.030.33a1311
CimetidineAAGB  7.1 0.90.8213.2
DesipramineAAGB  10.3 0.170.291310.3
       0.16 3.911
DiltiazemAAGB  7.7 0.410.42612.8
       0.25 7.413
GranisetronAAGB  9.4 0.700.95911
       0.72 2.59.1
ImipramineAAGB  9.5 0.1220.9559.5
       0.125 7.38.8
LidocaineAAGB  8 0.30.96714.9
MetroprololAAGB  9.7 0.750.69212.2
       0.88 6.212
NaloxoneAAGB  7.9 0.560.634419.3
         37.916.1
PindololAAGB  8.8 0.90.7314.2
       0.84 2.16.1
PropranololAAGB  9.5 0.1190.81916.4
       0.205 19.914
QuinidineAAGB  105.40.150.5435.33
       0.18 3.24.5
RanitidineAAGB  8.2 0.77160.5912.9
         1.43.2
SildenafilAAGB  7.6 0.040.2766
         13.410
TimololAAGDB  9.88.20.50.7129.2
       0.77 2.89.3
TiprolidineAAGB  9.33.60.10.7548
VerapamilAAGB  8.5 0.1150.421814.8
       0.1290.9223.115.3
PhRMA #63AAGB  6.5 0.040.1010.37.1
PhRMA #64AAGB  9.9 0.27517.026.2
Neutrals
AcetaminophenALN    0.790.9914
       0.89 1.3 
AntipyrineALN    0.940.850.30.6
CaffeineALN    0.6850.9611.4
       0.742 1.21.3
DiazepamALN    0.01870.9710.43
       0.013 1.20.49
LorazepamALN    0.0940.960.31.1
MethylprednisoloneALN    0.230.84117.5
       0.26 1.9 
MidazolamALN    0.04680.92136.6
NifedipineALN    0.0620.91137.8
OndansetronALN    0.67510.2615.9
OxazepamALN    0.030.9221.1
         3.31
PhenatecinALN    0.5940.672619.4
PrednisoloneALN    0.260.15118.7
         1.72.9
TheophyllineALN    0.40.6510.65
ZidovudineALN    0.80.8412.4

In compiling the dataset, more than one source of data (e.g., CLint and/or fub) was available for several drugs. In other words, for some drugs more than one value on the input parameters was presented in the cited sources. Because this interlaboratory variability cannot be totally explained,6 no set of data was disregarded, and, hence, the lowest and highest values were presented and used for predicting drug blood CL from each IVIVE method. When data came from different sources, the in vitro-derived prediction was always compared with the in vivo value coming from the same reference. However, prediction of blood CL by considering one dataset of input parameters compared with another was also investigated. Therefore, the prediction performance of the IVIVE methods for only one dataset in Table 1; the one that provided the more accurate predictions of blood CL for each method was also presented (i.e., only one data of CLint and/or fup was used if a drug have two data, and the one that gave the ratio between predicted and observed CL the closest to unity was considered). Exceptionally, in the case of caffeine, we had to disregard one in vitro data because the measured value of the unbound fraction in the incubation medium (fuinc) varied largely in the literature; Krime et al.5 reported a value of 0.18, whereas Riley et al.6 reported a value of 0.96, and the value of 0.96 was preferred in this study for each IVIVE method because more accurate predictions were obtained. To use the Poulin et al. method, the main binding protein [(AL or alpha1-acid glycoprotein (AAG)] was identified for literature drug on the basis of published binding studies.1,2,15 When this information was not available, binding to AAG was assumed to be preferred for relatively strong basic drugs, whereas binding to AL was assumed to be preferred for acidic and neutral drugs.1,2

Comparative Analysis of IVIVE Methods

The equations of the six IVIVE methods tested in this study were obtained from Poulin et al.2 and Sohlenius-Sternbeck et al.4. As mentioned these six IVIVE methods are the (1) “Poulin et al. method,” (2) “regression equation method,” (3) “conventional bias-corrected method,” (4) “conventional method,” (5) “Berezhkovskiy's method,” and (6) “direct scaling method.” The corresponding main equations are displayed in Table 2. For detailed informations on these equations and their assumptions, please refer to the original sources.1,2,4 Briefly, the IVIVE methods of Sohlenius-Sternbeck et al.4 is a regression equation previously established from the correlation of in vitro with in vivo CLint data. The conventional method assumes that only the free drug in plasma is available for drug metabolism in the liver, and hence, fub is used as the only correction factor. The conventional bias-corrected method involves multiplying the CL values predicted from the conventional method with the corresponding average bias of under-prediction to reduce the under-prediction. The average bias of under-prediction was obtained from the AFE value observed from each dataset (i.e., each prediction scenario) investigated in this study. Moreover, the conventional bias-corrected method required analysis of several datasets to first determinate the AFE correction factor, which depends on the composition of the datasets (i.e., number and disposition of compounds) as shown by the variability in the AFE values across the various datasets.2 The Berezhkovskiy's method uses the in vitro value of fuinc and an apparent fub value that is a correction of the true in vitro fub value for a difference of drug ionization between the plasma and intracellular water of liver. The Poulin et al. method assumes that the delivery of additional free drug to the hepatocytes may also occur via a protein-facilitated uptake, which is estimated by correcting the apparent fub values of each drug for the differences of AL concentration between the plasma and liver, and, hence, fuliver is used as the correction factor with fuinc determined in vitro. The utility of fuliver is applied only for a drug bound mainly to AL because for a drug principally bound to AAG, the apparent fub value was used instead.1,2 Finally, the direct scaling method does not consider any binding corrections.

Table 2. IVIVE Methods Tested in this Study for the Prediction of Blood CL as Reported from the Literature1–4
MethodsModel Equations
  1. aAFE obtained from the conventional method predictions for each dataset studied; therefore, a different AFE value was used for each dataset.

  2. bThe estimation of fub-app and fuliver is well detailed in two previous manuscripts,1,2 whereas fuinc was determined experimentally under in vitro conditions as mentioned in the method.

  3. CLint, intrinsic clearance; Q, liver blood flow rate; fub, unbound fraction in blood; fub-app, apparent unbound fraction considering the pH gradient; fuliver, unbound fraction in liver considering the protein-facilitated uptake and pH gradient; fuinc, unbound fraction in incubation medium (hepatocytes); SF, scaling factor (i.e., the physiological SF was (99 × 106 cells/g liver) × (1799 g liver/70 kg body weight).8

Intrinsic CLequation image
Conventionalequation image
Conventional bias correctedaequation image
Berezhkovskiybequation image
Poulin et al.bequation image
Direct scalingequation image
Regression equationequation image

The well-stirred liver model was used to calculate the blood CL for the purpose of this study. The strategy consists of evaluating the effect of various scenarios for which the prediction performance significantly differed across the IVIVE methods, as previously reported1,2: (a) drugs with low in vitro fub (≤0.05), (b) drugs bound to AL, and (c) drugs with low or very low CL in vivo (≤20% and ≤5% of the liver blood flow rate, respectively). We hypothesized that the Poulin et al. method may be advantageous for these scenarios because this method was efficient over other methods as seen in two previous comparative analyses with microsomal data.1,2 In other words, because the novel IVIVE method has been developed particularly to improve the prediction accuracy for drugs with such properties,1,2 it might also have the predictive advantage when using hepatocyte data.

Evaluation of Predictive Performance

The statistical analyses performed in the present study were the same as already described by Poulin et al.1,2. Briefly, the predicted CL was compared with the observed human CL to determine the predictability of each method using standard statistical techniques for accuracy, precision, and correlation. The AFE and root mean-squared error (RMSE) were calculated and are presented for each prediction method. Furthermore, the global concordance coefficient of correlation (CCC) is presented, which evaluates the global degree to which pairs of predicted and observed data fall on the line of unity passing through the origin. Specific fold errors of deviation between the predicted and observed values (percentage of fold error ≤2, ≤5, and ≤10) were also calculated. Finally, plots of predicted versus observed CL values are also presented for each IVIVE method.

RESULTS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHOD
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

Comparative Assessment for Various IVIVE Methods for Predicting CL

Six IVIVE calculation methods of CL were compared using the same drug dataset, and the comparative assessment was made on the basis of several statistical parameters. The overall statistical summary in terms of accuracy, precision, and correlation is listed in Tables 3 and 4 for the different scenarios of prediction. The plots of predicted versus observed blood CL values for each method are shown in Figures 1a–1e, whereas Figure 2 compares the precision bias across several IVIVE methods.

thumbnail image

Figure 1. Predicted CL versus observed CL for various IVIVE methods. (a) Poulin et al. method, (b) Regression equation method, (c) Berezhkovskiy's method, (d) Conventional method, and (e) Direct scaling method. The solid line indicates the best fit (unity). Short and long dashed lines on either side of unity represent twofold and threefold error, respectively. All of the datasets are included (n = 82). Blood CL is in mL/(min kg). The corresponding statistics are presented in Table 3.

Download figure to PowerPoint

thumbnail image

Figure 2. AFE values of various IVIVE methods obtained for different scenarios (datasets). Black squares, direct scaling method; red circles, Poulin et al. method; green triangles, Berezhkovskiy's method; and pink diamonds, regression equation method. The AFE values are presented in Table 3.

Download figure to PowerPoint

Table 3. Comparative Assessment of Six IVIVE Methods Used to Predict Blood CL in Humans from Hepatocyte Data for the Current Datasets
Scenarios% ≤ Twofold Error% ≤ Fivefold Error% ≤ 10-Fold ErrorAFERMSECCC
  1. aReferring to blood CL in vivo.

  2. AFE, average-fold error; RMSE, root-mean-squared error; CCC, concordance correlation coefficient global.

  3. LBF, liver blood flow rate in human [20.3 mL/(min kg].8

Low CL (≤20% LBF)a (n = 40)
Poulin et al.701001001.360.330.83
Regression equation5383982.220.470.63
Conventional bias-corrected4088981.010.470.74
Conventional3058780.250.760.52
Berezhkovskiy2348650.230.820.53
Direct scaling2848784.610.890.10
Very Low CL (≤5% LBF)a (n = 17)
Poulin et al.651001001.590.370.70
Regression equation3565943.140.610.31
Conventional bias-corrected53821001.00.460.56
Conventional2453770.220.800.30
Berezhkovskiy035530.150.930.31
Direct scaling0185312.61.200.02
Highly Bound (fub ≤ 0.05) (n = 39)
Poulin et al.741001001.230.300.90
Regression equation6488971.550.420.76
Conventional bias-corrected6492971.00.380.86
Conventional1344690.170.860.55
Berezhkovskiy1544640.170.860.62
Direct scaling2646775.510.900.32
Albumin Bound (n = 48)
Poulin et al.711001001.310.310.89
Regression equation6085981.740.430.74
Conventional bias-corrected4690980.990.440.82
Conventional2558770.240.760.61
Berezhkovskiy2359690.230.810.62
Direct scaling3356813.770.820.29
All Predictions (n = 82)
Poulin et al.731001001.00.280.90
Regression equation7292991.310.350.81
Conventional bias-corrected4894991.00.390.86
Conventional2766840.270.690.67
Berezhkovskiy4571820.360.640.75
Direct scaling5974892.100.640.40
Table 4. Prediction Performance of the IVIVE Methods for Only One Dataset in Table 1; the One that Provided the More Accurate Predictions of Blood CL for Each Method (i.e., Only One Data of CLint and/or fup Was Used If a Drug Have Two Data, and the One That Gave the Ratio Between Predicted and Observed CL the Closest to Unity Was Considered)
Scenarios% ≤ Twofold Error% ≤ Fivefold Error% ≤ 10-Fold ErrorAFERMSECCC
  1. AFE, average-fold error; RMSE, root mean-squared error; CCC, concordance correlation coefficient global.

  2. LBF, liver blood flow rate in human [20.3 mL/(min kg)].8

Low CL (≤ 20% LBF) (n = 24)
Poulin et al.881001001.020.240.91
Regression equation63881001.790.380.73
Conventional bias-corrected63921000.990.370.81
Conventional4279920.380.580.61
Berezhkovskiy2967880.330.630.62
Direct scaling3858883.140.610.23
Highly Bound (fub ≤ 0.05) (n = 22)
Poulin et al.1001001000.910.140.98
Regression equation82911001.220.270.90
Conventional bias-corrected6877961.060.390.85
Conventional2323960.280.640.67
Berezhkovskiy2341910.280.640.74
Direct scaling3246863.980.740.48
Albumin Bound (n = 29)
Poulin et al.901001001.040.200.96
Regression equation76901001.440.330.85
Conventional bias-corrected69931000.980.350.88
Conventional3883930.360.580.72
Berezhkovskiy3169900.340.610.73
Direct scaling4266902.790.660.46
All Predictions (n = 49)
Poulin et al.881001000.920.200.95
Regression equation8494981.240.310.86
Conventional bias-corrected74961000.920.300.90
Conventional3984960.370.540.74
Berezhkovskiy5382940.480.490.82
Direct scaling6580941.770.520.59

Predictivity of Human Dataset

On the basis of all statistical parameters, this study confirms that the novel IVIVE method of Poulin et al.1,2 shows the greatest prediction performance among the IVIVE methods and scenarios tested. The superior prediction performance of this novel IVIVE method is again most pronounced particularly for (a) drugs highly bound in blood, (b) drugs bound to AL, and (c) low CL drugs (Table 3). For these scenarios, the novel IVIVE method shows the greatest accuracy based on the fold errors of deviation between the predicted and observed values (% fold error ≤2, ≤5, and ≤10). Accordingly, the global CCC value is closest to unity for the novel IVIVE method of Poulin et al.1,2 and, hence, the RMSE value is the lowest. In other words, the Poulin et al. method provides AFE values close to unity, whereas the Berezhkovskiy and conventional methods provide lower AFE value, and the direct scaling method and regression equation method give higher AFE value. This is also corroborated graphically (Figs. 1a–1e and 2).

No relevant bias (no outliers) is observed with the novel IVIVE method of Poulin et al.1,2, because it is the only IVIVE method that did not provide CL predictions greater than fivefold error or more compared with the observed CL values in humans, whereas for the other IVIVE methods tested 6%–82% of the predicted CL values are greater than fivefold error compared with observed values (Table 3). Furthermore, the prediction performance of the novel IVIVE method is relatively constant over the scenarios (datasets) tested by contrast to the other IVIVE methods for which it significantly varies (e.g., decreases) from one scenario to another. Finally, it is not surprising to observe that the resulting AFE values of the conventional bias-corrected method and regression equation method are relatively close to unity because these two empirical methods correct for the under-prediction observed with the conventional method; however, these corrections do not appreciably improve the other statistical parameters comparatively to the novel IVIVE method of Poulin et al.1,2 (Tables 3 and 4). The conventional bias-corrected and regression equation methods were optimized by using almost the same drug dataset to that one of this study; therefore, it is not surprising that these two empirical methods sometimes performed better than other methods because the training and test sets are almost identical.

Predictivity According to Plasma Protein Binding

The novel IVIVE method of Poulin et al.1,2 shows superior prediction performance for the drugs bound mainly to AL and those drugs highly bound in blood (Table 3). For these two scenarios, the novel IVIVE method provides AFE values ranging from 1.23 to 1.31, whereas the conventional and Berezhkovskiy methods obtains much lower AFE values (0.17 to 0.24). The direct scaling method not performs better as AFE values ranged from 3.77 to 5.51. Finally, the regression equation method provides AFE values ranging from 1.55 to 1.74 for these two scenarios.

Predictivity According to the Magnitude of Drug CL

The most important difference in predictivity among the IVIVE methods is observed when the magnitude of CL in vivo of the drugs was considered in the prediction (Table 3). For drugs with low CL in vivo, the novel IVIVE method of Poulin et al.1,2 is by far the most accurate prediction method. The superiority of the novel IVIVE method is emphasized with the results obtained for the very low CL drugs. Again, the global CCC value is closest to unity for the novel IVIVE method and, hence, the RMSE value is the lowest. A systematic under-prediction of CL in vivo for the lowest CL compounds is observed for the Berezhkovskiy and conventional methods (AFE range from 0.15 to 0.25), whereas the direct scaling (AFE range from 4.61 to 12.6) and regression equation (AFE range from 2.22 to 3.14) methods over-predict the CL in this scenario compared with the Poulin et al. method (AFE range from 1.36 to 1.59). Furthermore, 65%–70% of CL predictions are within twofold error compared with observed values with the novel IVIVE method of Poulin et al.1,2, whereas the estimates range from 0% to 53% for the other IVIVE methods. Additionally, 100% of predictions fall within a fivefold error for the novel IVIVE method for the low and very low CL compounds, whereas for the other IVIVE methods this falls between 18% and 88%.

Outlier Predictions

For drugs for which more than one dataset was reported in the literature, in general, the blood CL predictions are more accurate by using one of the datasets. For example, for oxaproxin the Poulin et al. method calculated blood CL values of 0.081 mL/(min kg) [CLint = 1.6 µL/(min 106 cells)] and 0.4 mL/(min kg) [CLint = 8 µL/(min 106 cells)], respectively, whereas the observed value ranged from 0.071 to 0.1 mL/(min kg). This highlights the challenges in compiling datasets from several sources.6 Consequently, when only one dataset of input parameters in Table 1 was used in the calculations of CL (i.e., one data on CLint and/or fuinc was favored compared with the others; the one that gave the ratio between predicted and observed CL the closest to unity), the prediction accuracy increased compared with when all datasets were compiled for the predictions (Table 4 versus Table 3). Again, the novel IVIVE method demonstrates the greatest prediction performance among the IVIVE methods tested (Table 4).

DISCUSSION

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHOD
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

The main focus of this study was to provide a comparative analysis between recently published IVIVE methods by predicting blood CL in vivo in humans from hepatocyte data for several drugs. The findings of the current comparative analysis made from hepatocyte data are similar to three other assessments made previously from microsomal data1–3; the novel IVIVE method of Poulin et al.1,2 offers a significant improvement for the prediction of blood CL over other IVIVE methods. This is particularly confirmed when a drug had particular properties (i.e., low CL, low fub, and/or AL bound) (Tables 3 and 4 as well as Fig. 2). In other words, by using different datasets of drugs and various types of in vitro data (hepatocytes and microsomes), the conclusion is the same. Therefore, justification for the mechanistic modeling proposed by Poulin et al.1,2 is well supported. It is of interest to note that Halifax and Houston,3 Sohlenius-Sternbeck et al.4, and Krime et al.5 stated that their empirical methods have a superior predictivity to the novel mechanistic IVIVE methof of Poulin et al.1,2 without ever actually showing results that prove these claims over diverse scenarios and by using various statistical parameters. Our two previous comparative analyses1,2 and this present study do exactly that and demonstrate without a doubt that the consideration of additional mechanisms of CL yields greater success in CL predictions. Therefore, we have demonstrated that we can obtain comparable or superior degree of accuracy by using a mechanistic framework based solely on in vitro data compared with previous empirical regression analyses that need in vivo data first for their development; this mechanistic framework based on in vitro data is exactly the kind of predictive approach that is needed in drug discovery and development. And the novel IVIVE method uses strictly (a) physiological data that are readily available in the literature, (b) drug physicochemical properties, which are usually available or can be estimated in silico, and c) experimentally obtained in vitro data in microsomal or hepatocyte incubations.

The novel IVIVE method advocated by Poulin et al.1,2 can be considered as a mechanistic method compared with other IVIVE methods, particularly, the conventional bias corrected method and regression method. In this case, some authors suggested that prediction of CL from microsomes, and particularly from hepatocytes, might be improved beyond any of the methods assessed in this study through the use of an empirical correction factor to eliminate both the average bias and the CL dependency bias.3–5 As mentioned, the development of the conventional bias-corrected method of Halifax and Houston3 and the regression method of Sohlenius-Sternbeck et al.4 required previous optimization from in vivo CL data of several datasets,3–5 which depends purely on the composition of the datasets (i.e., number and disposition of compounds). Moreover, the drugs used to optimize such empirical IVIVE methods are almost the same drugs that those used in the present study. Therefore, it is not surprising to observe that the resulting AFE values of the conventional bias-corrected method and regression method are also relatively close to unity for the scenarios studied here; however, the values of the other statistical parameters obtained with these two empirical methods are definitively not as good as those by the novel IVIVE method of Poulin et al.1,2 (Tables 3 and 4). In contrast, the Poulin et al. method, which only requires compound-specific input and no prior analysis of a large dataset to provide CL predictions, represents a first step toward the development of an efficient IVIVE method based strictly on mechanisms of drug CL.

In this context, the novel IVIVE method accounts for additional mechanisms potentially occurring under in vivo condition, namely, the pH differences in extracellular and intracellular water of liver as well as the protein-facilitated uptake due to potential ionic interactions between the protein-AL bound drug complex and cell surface of the hepatocytes, which may explain why this new paradigm shows improved predictive performance for drugs with such properties. Accordingly, for the current dataset, low CL compounds are generally highly bound to AL, which might explain the greater accuracy of the novel IVIVE method of Poulin et al.1,2 for these drug properties because they are taken into account by this method. In other words, because the novel IVIVE method has been developed particularly to improve the prediction accuracy for drugs with such properties,1,2 this study confirms that this improvement provides an advantage over other IVIVE methods. This is also explained by previous sensitivity analyses, which demonstrated the importance of compound properties and binding parameters that are reflective of specific mechanistic determinants relevant to CL prediction using the IVIVE methods tested.2 Accordingly, the IVIVE methods differ particularly at low CLint and low fup values. Furthermore, the class of drug had an effect on the CL predictions. The difference between the IVIVE methods is more noticeable for an acidic drug highly bound to AL than for a basic drug highly bound to AAG.2 Thus, the importance of CLint and fup as input parameters is obvious, but it should be highlighted that it is experimentally hard to determine CLint and fup accurately for low CL and highly bound drugs, respectively. Furthermore, the novel IVIVE method of Poulin et al.1,2 is relatively sensitive to drug ionization (i.e., ionization constant; pKa) because it also considers a pH difference between the liver cells and extracellular water. The latter is also important because calculated pKa values are used in early drug discovery and they are not always accurate because significant differences between calculated and measured values are common.

In contrast, according to the conventional theory of cellular uptake, only unbound ligand participates in the uptake process, and, hence, the metabolism in liver. Therefore, it is commonly assumed that virtually instantaneous equilibration occurs between bound and unbound ligand, and, hence, the plasma unbound drug concentrations can be used to indicate the amount of drug metabolized in the liver as described by the conventional IVIVE method. However, a very recent paper had highlighted various examples from the literature where tissue unbound drug concentrations have demonstrated a superior correlation, at least with efficacy, compared with the plasma unbound drug concentrations.16 Therefore, it seems that referring to fup for predicting biological processes in tissues might not be always valid. In this case, kinetic analysis of drug transfer from serum to hepatocytes indicates that the delivery of additional free drug to the hepatocytes may also occur via a protein-facilitated uptake due to ionic interactions between the protein-bound drug complex and hepatocyte cell surface, and/or by the presence of AL binding sites on the hepatocytes,13,17–30 which is the main mechanisms described by the apparent fuliver in the novel IVIVE method of Poulin et al.1,2 In this context, several mechanistic studies published in the literature were discussed just below in support of this novel IVIVE method.

First, Burczynski et al.17,18 reported that palmitate CL was greatest when AL was used as the extracellular binding protein in the incubation medium and lowest in the presence of AAG. The main reason was the difference in the isoelectric point between AL and AAG because positive charge on the surface of a binding protein might lead to ionic interactions with the cell membrane due to the presence of negatively charged groups that decorate the lipid bilayer surface of most mammalian cells.1,17,18 Indeed, it has been observed that the interaction of AL with the cell surface may be instantaneous.19 Accordingly, the effects of charge on antibody tissue distribution and PK demonstrate that more the protein is charged positively more important its uptake by the liver will be.20,21 Second, Blanchard et al.22 observed that the blood CL calculated from CLint in vitro obtained in the presence of AL in the incubation medium increased up to 18-fold the CL values compared with plasma-free incubations. Third, Wattanachai et al.23 showed that supplementation of human liver incubations with bovine serum albumin resulted in a 3.6-fold increase in the CLint for paclitaxel 6α-hydroxylation, due mainly to a reduction in the Michaelis–Menten constant (Km). A lower Km value is in accordance with additional interactions between the protein–drug complex and liver cell surface compared with when no AL was added in the incubation medium. However, it has been demonstrated that the AL of rat mimics much better the human AL compared with bovine serum albumin, at least with regard to bilirubin binding,24 which could be of relevance in the in vitro metabolic studies and for accurate IVIVE's.

Fourth, Gill et al.13 demonstrated that the use of in vitro data obtained in the presence of bovine serum albumin and inclusion of renal CL improved the IVIVE of glucuronidation CL for several drugs. Fifth, Mitchell et al.,25,26 who studied the defenestration of the liver sinusoidal endothelium in the isolated perfused rat liver model for diazepam, a drug highly bound to AL in plasma, showed that the protein-bound drug complex is highly extracted by the liver. Sixth, the AL binding sites of hepatocytes might also explain the delivery of additional drugs (bound to AL) to the human hepatocytes.27,28 Other authors suggested that the drug delivery does not require a direct interaction between the protein-bound drug complex and native hepatocyte plasma membrane vesicles, at least under in vitro conditions.24 However, these authors used vesicles made of a neutral phospholipid only (i.e., phospatidylcholine) rather than of acidic phospholipids (e.g., phosphatidylserine),24 which are also constituents of the hepatocyte membranes in vivo,1 and, hence, this probably limited the importance of the ionic interactions between the protein-bound drug complex and the surface of the vesicles in their in vitro studies. Overall, the resulting effect would be that AL enhances the hepatocyte uptake whereas AAG limits that uptake particularly for the highly bound drugs,29,30 and these mechanisms are now well covered by the Poulin et al. method.1,2

Developing additional mechanistic metabolic in vitro assays to evaluate the importance of the protein-facilitated uptake on the metabolism of drugs in hepatocytes would be a valuable next step. Similarly, mechanistic plasma protein- binding assays are also required to identify the major binding protein to apply the novel IVIVE calculation method of Poulin et al.1,2 prospectively. In this context, we have proposed a new experimental setting to define whether AL or AGG is the main binding protein in plasma for 21 compounds (i.e., the binding ratio of human AAG to human AL was determined).2 Many highly binding drugs may bind to both proteins (AL or AAG). We assumed either one or the other protein, but in reality for compounds that bind equally to both proteins, we will explore that further in a subsequent analysis. Furthermore, any significant errors in experimental assessment of fub would confound the predictability of IVIVE, particularly for highly bound drugs.1,2 We also explored the use of this IVIVE method for prediction of drug CL in preclinical species; in general, the novel IVIVE method is comparable or superior to the other methods,2 which again demonstrates its usefulness. For prediction of CL values in any species, the required input parameters were presented in a previous manuscript.2 The current study also involved consideration of the interlaboratory variability in the input parameters for predicting blood CL clearance of several drugs (Table 3 vs. Table 4). Relevant differences in the prediction performance were observed among the datasets used; therefore, any variability needs to be covered for a more accurate prediction of CL in humans.6,31

Halifax et al.10 demonstrated that prediction accuracy of the IVIVE methods was not dependent on the relative permeability in hepatocytes indicating the absence of a general rate limitation by passive hepatocyte uptake on metabolic CL for almost the same drugs than those used in this present study. However, Huang et al.11,32 recently reported that the accuracy of CL prediction could be influenced by permeability and efflux status. Therefore, it has been demonstrated that permeability and transporters might contribute to the lack of correlation between in vitro CLint and in vivo CL in the rat. Blood CL in vivo could be predicted reasonably well from in vitro metabolic CLint for compounds that displayed high passive permeability (>5 × 10−6 cm/s) and that were not good substrates (efflux ratio <5) for efflux transporters such as P-glycoprotein and breast cancer resistance protein.32 For compounds with other permeability and transport properties in vitro metabolic CLint was unlikely to be predictive of in vivo blood CL.32 For example, the prediction of the in vivo blood CL was improved significantly when using uptake CLint for the compounds with low permeability. In this context, it has also been concluded by other authors that the potential interplay of transporters and enzymes must be considered in defining the CLint in the liver for particular drugs across species (e.g., bioclassification system class 2 drugs).2,32–34 In this case, Umehara and Camenisch35 quantified the impact of transport processes in the prediction of CL from hepatocyte data. However, coupled with the IVIVE method proposed by Huang et al.11,32 and Umehara and Camenisch,35 who quantified the impact of permeability and transport processes in the prediction of CL, the novel IVIVE method for metabolism proposed by Poulin et al.1,2 takes into consideration various mechanistic determinants of hepatic metabolic CL. Saturable uptake into the hepatocytes can also be taken into account in the prediction of in vivo CL.36 Finally, instead of the well-stirred model, a physiologically based liver model comprising seven compartments was also previously proposed by Haddad et al.,37 which enable the IVIVE of drug CL as well as drug–drug interactions. This liver model that can integrate data on permeability, transport, and nonlinear kinetics could also be adapted to consider the additional mechanisms of metabolic CL covered by the novel IVIVE method of Poulin et al.1,2

CONCLUSIONS

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHOD
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

This study demonstrated the prediction performance of six IVIVE methods for the prediction of blood CL in vivo of drugs in human from hepatocyte data. The results of the current comparative analysis performed from hepatocyte data confirm the findings of a previous analysis made from microsomal data; therefore, the novel IVIVE method proposed in two previous manuscripts1,2 was the most successful in various prediction scenarios over other methods either based on microsomal or hepatocyte data. The difference across these methods is most pronounced for (a) drugs highly bound in blood, (b) drugs bound to AL, and (c) low CL drugs. Therefore, a new paradigm is emerging in the prediction of metabolic CL particularly for the drugs that have the aforementioned properties, which would contribute to a more accurate CL prediction in drug discovery and development. Consideration of additional mechanistic studies should be encouraged to further support the conclusions of this study. However, despite all the caveats, the development of the novel IVIVE method of Poulin et al.1,2, which is adapted to additional problematic drug properties, can potentially be useful in the interpretation and prediction of CL and, hence, the potential improvement of high-throughput screening in drug discovery and development for small molecules metabolized essentially by the liver.

Acknowledgements

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHOD
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES

This work represents an initiative undertaken in collaboration as a part of the research program of Dr. Poulin, and of Dr. Haddad's program, which is supported by a Discovery Grant from the National Sciences and Engineering Research Council of Canada (NSERC). The authors wish to thanks Dr. Cornelis Hop and Dr. Jane Kenny, at Genentech Inc., as precursors of fruitful discussions that have strongly contributed to the conduct of this work.

REFERENCES

  1. Top of page
  2. Abstract
  3. INTRODUCTION
  4. METHOD
  5. RESULTS
  6. DISCUSSION
  7. CONCLUSIONS
  8. Acknowledgements
  9. REFERENCES
  • 1
    Poulin P, Kenny JR, Hop CE, Haddad S. 2012. In vitro–in vivo extrapolation of clearance: Modeling hepatic metabolic clearance of highly bound drugs and comparative assessment with existing methods. J Pharm Sci 101:838851.
  • 2
    Poulin P, Hop CE, Haddad S, Kenny JT. 2012. Comparative assessment of in vitroin vivo extrapolation methods used for predicting hepatic metabolic clearance of drugs. J Pharm Sci 101:43084326.
  • 3
    Halifax D, Houston BJ. 2012. Evaluation of hepatic clearance prediction using in vitro data: Emphasis of fraction unbound in plasma and drug ionization using a dataset of 107 drugs. J Pharm Sci 101:26452652.
  • 4
    Sohlenius-Sternbeck AK, Jones C, Ferguson D, Middleton BJ, Projean D, Floby E, Bylund J, Afzelius L. 2012. Practical use of the regression offset approach for the prediction of in vivo intrinsic clearance from hepatocytes. Xenobiotica 42:841953.
  • 5
    Krime K, McGinnity DF, Barton P. 2013. Application of in silico, in vitro and pre-clinical pharmacokinetic data for the effective and efficient prediction of human pharmacokinetics. Mol. Pharm. DOI: 10.1021/mp300476z.
  • 6
    Riley RJ, McGinnity DF, Austin RP. 2005. A unified model for predicting human hepatic, metabolic clearance from in vitro intrinsic clearance data in hepatocytes and microsomes. Drug Metab Dispos 33:13041311.
  • 7
    Ekins S, Obach RS. 2000. Three-dimensional quantitative structure activity relationship computational approaches for prediction of human in vitro intrinsic clearance. J Pharmacol Exp Ther 295:463473.
  • 8
    Ring B, Chien JY, Adkison KK, Jones HM, Rowland M, Jones RDO, Yates JWT, Ku MS, Gibson CR, He H, Vuppugalla R, Marathe P, Fischer V, Dutta S, Sinha VK, Björnsson T, Lavé T, Poulin P. 2011. PhRMA CPCDC initiative on predictive models of human pharmacokinetics. 3. Comparative assessment of prediction methods of human clearance. J Pharm Sci 100:40904110.
  • 9
    Naritomi Y, Terashita S, Kagayama A, Sugiyama Y. 2003. Utility of hepatocytes in predicting drug metabolism: Comparison of hepatic intrinsic clearance in rats and humans in vitro and in vitro. Drug Metb Disp 31:580588.
  • 10
    Halifax D, Turlizzi E, Zanelli U, Houston BJ. 2012. Clearance-dependent underprediction of in vivo intrinsic clearance from human hepatocytes: Comparison with permeabilities from artificial membrane (PAMPA) assay, in silico and caco-2 assay, for 65 drugs. Eur J Pharm Sci 45:570574.
  • 11
    Huang L, Berry L, Ganga S, Janosky B, Chen A, Roberts J, Colletti AE, Lin MHJ. 2010. Relationship between passive permeability, efflux, and predictability of clearance from in vitro metabolic intrinsic clearance. Drug Metab Disp 38:223231.
  • 12
    Product monograph of Irbesartan. http://www.bmscanada.ca/static/products/en/pm_pdf/Avapro_EN_PM.pdf. Assessed 18 of February 2013.
  • 13
    Gill KL, Houston BJ, Galetin A. 2012. Characterization of in vitro glucuronidation clearance of a range of drugs in human kidney microsomes: Comparison with liver and intestinal glucuronidation and impact of albumin. Drug Metab Dispos 40:825835.
  • 14
    Loi CM, Young M, Randinitis E, Vassos A, Koup JR. 1999. Clinical pharmacokinetics of troglitazone. Clin Pharmacokinet 37:91104.
  • 15
    Morgan T. 1994. Clinical pharmacokinetics and pharmacodynamics of carvedilol. Clin Pharmacokinet 26:335346.
  • 16
    Mariappan TT, Mandlekar S, Marathe P. 2013. Insight into tissue unbound concentration: Utility in drug discovery and development. Curr Drug Metab 14:324-340.
  • 17
    Burczynski FJ, Wang GQ, Elmhadoun B, She YM, Roberts MS, Standing KG. 2001. Hepatocyte palmitate uptake: Effect of albumin surface charge modification. Can J Physiol Pharmacol 79:868-875.
  • 18
    Burczynski FJ, Wang GQ, Hnatowich M. 1997. Effect of binding protein surface charge on palmitate uptake by hepatocyte suspensions. Br J Pharmacol 120:12151220.
  • 19
    Osterloh K, Ewert U, Pries AR. 2002. Interaction of albumin with the endothelial cell surface. Am J Physiol 283:H398H405.
  • 20
    Boswell AC, Tesar DB, Mukhyala K, Theil FP, Fielder PJ, Khawli LA. 2010. Effects of charge on antibody tissue distribution and pharmacokinetics. Bioconjug Chem 21:21532163.
  • 21
    Igawa T, Tsunoda H, Tachibana T, Maeda A, Mimoto F, Moriyama C, Nanami M, Sekimori Y, Nabuchi Y, Aso Y, Hatorri K. 2010. Reduced elimination of IgG antibodies by engineering the variable region. Protein Eng Des Sel 23:385392.
  • 22
    Blanchard N, Hewitt NJ, Silber P, Jones H, Coassolo P, Lavé T. 2006. Prediction of hepatic clearance using cryopreserved human hepatocytes: A comparison of serum and serum free incubations. J Pharm Pharmacol 58:633641.
  • 23
    Wattanachai N, Polasek TM, Heath TM, Uchaipichat V, Wongwiwat T, Tassaneeyakul W, Miners JO. 2011. In vitro–in vivo extrapolation of CYP2C8-catalyzed paclitaxel 6α-hydroxylation: Effects of albumin on in vitro kinetic parameters and assessment of interindividual variability in predicted clearance. Eur J Clin Pharmacol 67:815824.
  • 24
    Zucker SD, Goessling W, Gollan JL. 1995. Kinetics of bilirubin transfer between serum albumin and membrane vesicles. J Biol Chem 270:10741081.
  • 25
    Mitchell SJ, Huizer-Pajkos A, Cogger VC, McLachlan AJ, Le Couteur DG, Hilmer SN. 2011. Poloxamer 407 increases the recovery of paracetamol in the isolated perfused rat liver. J Pharm Sci 100:334340.
  • 26
    Mitchell SJ, Huizer-Pajkos A, Cogger VC, McLachlan AJ, Le Couteur DG, Jones D, de Cabo R, Hilmer SN. 2012. The influence of old age and poloxamer-407 on the hepatic disposition of diazepam in the isolated perfused rat liver. Pharmacology 90:233241.
  • 27
    Thung SN, Gerber MA. 1983. Albumin binding sites of human hepatocytes. Liver 3:290294.
  • 28
    Wright TL, Lysenko N, Ockner RK, Weisiger RA. 1987. Interaction of natural and synthetic albumin polymers with hepatocytes. Hepatology 7:294301.
  • 29
    Qin M, Nilsson M, Øie S. 1994. Decreased elimination of drug in the presence of alpha1-acid-glycoprotein is related to a reduced hepatocytes uptake. J Pharmacol Exp Ther 269:11761181.
  • 30
    Bilello JA, Biello PA, Stellrecht K, Leonard J, Norbeck DW, Kempf DJ, Robins T, Drusano GL. 1996. Human serum alpha acid glycoprotein reduces uptake, intracellular concentration, and antiviral activity of A-80987, an inhibitor of the human immunodeficiency virus type 1 protease. Ant Microb Agent Chemother 40:14911497.
  • 31
    Cubitt HE, Yeo KR, Howgate EM, Rostami-Hodjegan A, Barter ZE. 2011. Sources of interindividual variability in IVIVE of clearance: An investigation into the prediction of benzodiazepine clearance using a mechanistic population-based pharmacokinetic model. Xenobiotica 41:623638.
  • 32
    Huang L, Chen A, Roberts J, Janosky B, Be X, Berry L, Lin MHJ. 2012. Use of uptake intrinsic clearance from attached rat hepatocytes to predict hepatic clearance for poorly permeable compounds. Xenobiotica 42:830840.
  • 33
    Ménochet K, Kenworthy KE, Houston BJ, Galetin A. 2012. Use of mechanistic modeling to assess interindividual variability and interspecies differences in active uptake in human and rat hepatocytes. Drug Metab Disp 40:17441756.
  • 34
    Chu X, Bleasby K, Evers R. 2013. Species differences in drug transporters and implications for translating preclinical findings to humans. Expert Opin Drug Metab Toxicol 9:237-252.
  • 35
    Umehara K, Camenisch G. 2012. Novel in vitro-in vivo extrapolation (IVIVE) method to predict hepatic organ clearance in rat. Pharm Res 29:603617.
  • 36
    Halifax D, Houston BJ. 2007. Saturable uptake of lipophilic amine drugs into isolated hepatocytes: Mechanisms and consequences for quantitative clearance prediction. Drug Metab Dispo 35:1325137.
  • 37
    Haddad S, Poulin P, Funk C. 2010. Extrapolating in vitro metabolic interactions to isolated perfused liver: Prediction of metabolic interactions between bufarolol, bunitrolol and debrisoquine. J Pharm Sci 99:44064426.