Abstract
Evaluating biliary excretion, a major elimination pathway for many compounds, is important in drug discovery. The bile duct-cannulated (BDC) rat model is commonly used to determine the percentage of dose excreted as intact parent into bile. However, a study using BDC rats is time-consuming and cost-ineffective. The present report describes a computational model that has been established to predict biliary excretion of intact parent in rats as a percentage of dose. The model was based on biliary excretion data of 50 Bristol-Myers Squibb Co. compounds with diverse chemical structures. The compounds were given intravenously at <10 mg/kg to BDC rats, and bile was collected for at least 8 h after dosing. Recoveries of intact parents in bile were determined by liquid chromatography with tandem mass spectrometry. Biliary excretion was found to have a fairly good correlation with polar surface area (r = 0.76) and with free energy of aqueous solvation (ΔGsolv aq) (r = −0.67). In addition, biliary excretion was also highly corrected with the presence of a carboxylic acid moiety in the test compounds (r = 0.87). An equation to calculate biliary excretion in rats was then established based on physiochemical properties via a multiple linear regression. This model successfully predicted rat biliary excretion for 50 BMS compounds (r = 0.94) and for 25 previously reported compounds (r = 0.86) whose structures are markedly different from those of the 50 BMS compounds. Additional calculations were conducted to verify the reliability of this computation model.
Biliary excretion is a major elimination pathway for many drugs and discovery compounds both in humans and in preclinical animals. For example, pravastatin and losoxantrone were found to be mainly eliminated as intact parent through biliary excretion in humans (Hatanaka 2000; Joshi et al., 2001). In rats, pravastatin and methotrexate were minimally metabolized and were primarily excreted intact into bile (Masuda et al., 1997; Kurihara et al., 2005). Extensive biliary excretion can be linked to a high clearance (Arimori et al., 2003), enterohepatic recirculation (Caldwell and Cline 1976; Rollins and Klaassen 1979), toxic gastrointestinal side effects (Kato et al., 2002), and potential drug-drug interactions (Luo et al., 2007). As a result, most lead discovery compounds are assessed for biliary excretion in selected preclinical animals early in the drug discovery and development process.
Among the preclinical animal models, rats are the most commonly used model species for pharmacology, pharmacokinetics, and toxicology. The existing experimental models for determining rat biliary excretion include bile duct-cannulated rats and isolated perfused rat liver. However, these models are very time-consuming and cost-ineffective because of complicated preparation of test models and difficulty in bile sample analyses.
Undoubtedly, a computational model for predicting rat biliary excretion could significantly reduce laboratory efforts and, consequently, cost. Furthermore, a computational model could enable scientists to determine the potential for biliary excretion of virtual compounds, thereby helping to prioritize synthetic efforts in drug discovery programs. However, such a model has not yet been reported, despite efforts to identify factors that critically influence rat biliary excretion. In previous work, molecular weight was commonly identified as a dominant factor influencing biliary excretion (Millburn et al., 1967; Abou-El-Makarem et al., 1967a,b; Hirom et al., 1972a,b; Hughes et al., 1973a,b; Wright and Line 1980; Proost et al., 1997; Han et al., 2001), and a hypothesis of “molecular weight threshold” was proposed. For example, Wright and Line (1980) demonstrated in their study with 18 cephalosporin derivatives that a molecular weight of 450 was the threshold for rat biliary excretion; above that molecular weight threshold, biliary excretion increased in a generally progressive way and became the principal route of excretion of the higher-molecular-weight derivatives. Nevertheless, molecular weight alone cannot predict rat biliary excretion although it may indicate a trend toward increased biliary excretion. For example, the carboxylate and lactone forms of irinotecan have little difference in molecular weights, but the carboxylate exhibits much more biliary excretion than does the lactone (Arimori et al., 2003; Itoh et al., 2004).
In the present study 50 discovery compounds from Bristol-Myers Squibb Co. (BMS) with diverse chemical structures were evaluated in bile duct-cannulated (BDC) rats after intravenous administration. Predictions of rat biliary excretion were then made based on polar surface area (PSA), free energy of aqueous solvation (ΔGsolv aq), and the presence or absence of carboxylic acid moieties in these molecules. The prediction was highly correlated with the observed in vivo biliary excretion. When the same prediction method was applied to 25 compounds whose biliary excretion was published previously (Hirom et al., 1972b; Russell and Klaassen 1973; Fahrig et al., 1989; Monsarrat et al., 1990; Masuda et al., 1997; Hinchman et al., 1998; Payan et al., 1999; Song et al., 1999; Arimori et al., 2003; Chong et al., 2003; Funakoshi et al., 2003; Moriwaki et al., 2003; Kamath et al., 2005a,b, 2008; Kurihara et al., 2005; Takayanagi et al., 2005; Akashi et al., 2006; Beconi et al., 2007), a high correlation was also seen between the predicted and observed values. After the validation work, the two sets of compounds were combined into a single combined data set that was then used to generate a final model equation to predict biliary elimination.
Materials and Methods
Chemicals.
Fifty discovery compounds were prepared by Bristol-Myers Squibb Co. (Pennington and Lawrenceville, NJ, and Wallingford, CT).
Preparation of Bile Duct-Cannulated Rats.
Male Harlan Sprague-Dawley rats (250–300 g) were purchase from Harlan (Indianapolis, IN). Rats were anesthetized by inhalation of isoflurane (administered at 2–3% using oxygen as the carrier) before surgery. Although the rats were under isoflurane anesthesia, the abdominal and dorsal neck areas were shaved, rinsed with isopropanol, and then scrubbed with Nolvasan scrub. The sites were then swabbed with Betadine. The rats were covered with sterile drapes and sterile instruments were used. The abdominal cavity was opened, and a small polyethylene catheter was inserted into the common bile duct. Another catheter was inserted into the duodenum at the level of the common bile duct for the recirculation and infusion of bile. Both cannulae were passed through the abdominal musculature and then routed subcutaneously to the dorsal neck area and exteriorized through a small incision. The abdominal cavity was sutured shut, and the skin was closed with wound clips. All rats received 25 ml of warm sterile saline during surgery (applied directly into the abdominal cavity) and another 10 ml of saline or 5% dextrose subcutaneously postoperatively. Postsurgical analgesia (0.1 mg/kg buprenorphine s.c.) was administered. The rats were then allowed to recover. This surgery was done 2 days before the study. Control bile was collected the day after surgery and used the following day during the course of the experiment to infuse as a replacement for the bile collected after dosing. A jugular vein and femoral vein were also cannulated to allow intravenous administration of dose solution and blood sample collection. Upon completion of the study the rats were euthanized by CO2 inhalation overdose.
Studies in Bile Duct-Cannulated Rats.
Test compounds were dosed intravenously to two or three rats, either in a cassette dosing format (maximum of four compounds per dose group with each compound dosed at less than 2 mg/kg to minimize potential drug-drug interactions) or in a discrete dosing format (dosed at less than 10 mg/kg to avoid saturation of transporter(s) involved in biliary excretion). In general, bile samples were collected at 0 to 1, 1 to 2, 2 to 4, and 4 to 8 h intervals after dose administration.
Determination of Parent Compounds in Bile Samples.
Bile samples were diluted with plasma (10–50-fold) and analyzed against a plasma standard curve. A Packard MultiPROBE liquid handler (PerkinElmer Life and Analytical Sciences, Waltham, MA) was used to transfer 25 to 50 μl of each standard, quality control, or bile sample diluted in plasma to a 96-well plate for protein precipitation extraction. After the addition of 120 μl of acetonitrile containing the internal standard, the samples were vortex mixed and the resulting supernatant was separated from the precipitated proteins by centrifugation for 5 min. An aliquot of the supernatant was transferred using a Tomtec automated liquid handler to a second clean 96-well plate. The mixture was diluted five times with 0.1% formic acid in water, and the plate was capped and vortex mixed. An aliquot of 5 to 10 μl of supernatant was injected into liquid chromatography with tandem mass spectrometric detection (LC-MS/MS) for analysis. The high-pressure liquid chromatography system consisted of LC10ADvp pumps (Shimadzu, Columbia, MD) and an HTC PAL autosampler (Leap Technologies, Cary, NC) linked to a Synergi Hydro-RP analytical column (2.0 × 50 mm, 4 μm; Phenomenex, Torrance, CA). Mobile phase A consisted of 0.1% formic acid in water. Mobile phase B was 100% acetonitrile. The LC flow rate was 0.3 ml/min. The initial mobile phase composition was 20% B for 1 min, followed by a linear gradient to 80% B over 1 min. The mobile phase composition was held at 80% B for 1 min, and then returned to the initial conditions over the next 0.1 min, and re-equilibrated for 0.5 min. Total analysis time for most compounds was 3.5 min. The high-pressure liquid chromatography system was interfaced to an API3000 or API4000 Q trap spectrometer (MDS Sciex, Toronto, ON, Canada) equipped with a TurboIonSpray source. The source temperature was set at 450°C, and the ion spray voltage was set to 4.8 kV. UHP nitrogen was used as nebulizer and auxiliary gas.
Cumulative biliary excretion of parent compound was calculated from the bile volume and corresponding concentrations at each time interval. The biliary excretion was expressed as percentage of dose excreted into bile as parent.
Calculation of Physicochemical Properties.
Various physicochemical properties, including molecular weight, ACD logD at pH 6.5, ACD logD at pH 7.4, ACD logP, clogP, free energy of aqueous solvation (ΔGsolv aq), free energy of solvation in dimethyl sulfoxide (ΔGsolv DMSO), and number of rotational bonds, were generated with Abacus, an in-house application containing various property calculators from a number of vendors. PSA was computed using the TPSA algorithm in SciTegic Pipeline Pilot (version 51; Accelrys, Inc., San Diego, CA). For ΔGsolv aq, a conformational search was performed with MacroModel (Schrodinger, Inc., New York, NY) using the OPLS2005 force field with implicit solvation. The 10 lowest energy conformers were each submitted to Jaguar (Schrodinger, Inc.) for single point energy calculations with B3LYP/6-311+G**/water. The calculated ΔGsolv aq of the lowest energy conformation from this self-consistent reaction field calculation was the value used in regression studies.
The structural similarity among compounds was analyzed using the AtomPair fingerprints with the Tanimoto similarity coefficient. Pairs of compounds with a Tanimoto similarity greater than 0.7 were considered to be very similar in structure.
Prediction of Biliary Excretion.
Correlations were calculated using a simple linear least-squares regression between in vivo biliary excretion and individual physicochemical properties. The individual physicochemical properties included ACD logD at pH 6.5, ACD logD at pH 7.4, ACD logP, clogP, molecular weight, number of rotational bonds, ΔGsolv aq, ΔGsolv DMSO, PSA, and presence/absence of carboxylic acid moieties. Based on these correlations, a multiple linear regression model was generated for the prediction of biliary excretion using three physicochemical properties (PSA, ΔGsolv aq, and presence/absence of carboxylic acid moieties). Each of these parameters was highly correlated with the observed extent of rat biliary excretion.
Results
Rat Biliary Excretion of 50 BMS Compounds.
Biliary excretion, expressed as a percentage of the intravenous dose, of 50 BMS compounds was obtained experimentally in BDC rats (Table 1). The biliary excretion of this set of compounds ranged from 0.1 to 100% of the intravenous dose. It is noteworthy that biliary excretion of the test compounds in BDC rats was observed to occur mostly during the first 2 h and was almost complete within the first 4 h after drug administration regardless of total cumulative biliary excretion (Fig. 1). This observation is consistent with findings from other laboratories (Itoh et al., 2004) and suggests that biliary excretion of small molecules in rats is an efficient process, and high biliary excretion may result in high clearance, at least for small molecules such as those tested in the present study.
Biliary excretion in BDC rats and physicochemical properties of 50 BMS compounds
Typical examples of biliary excretion time courses in BDC rats. Test compounds were given intravenously to two BDC rats in cassette dosing format at a dose of approximately 0.9 mg/kg for each compound. The bile samples were continually collected up to 9 h after dosing, and the concentrations of intact parent in bile samples were determined using LC-MS/MS.
Correlation between Rat Biliary Excretion and Individual Physicochemical Properties.
Various physicochemical properties, including ACD logD at pH 6.5 and ACD logD at pH 7.4, ACD logP, clogP, molecular weight, number of rotational bonds, ΔGsolv aq, ΔGsolv DMSO, and PSA were generated for the 50 compounds investigated (Table 1). Simple linear regression analyses were performed to evaluate the relationships between the observed rat biliary excretion (expressed as a percentage of dose) and the individual physicochemical properties to identify those physicochemical properties that significantly influenced rat biliary excretion of the test compounds. As shown in Table 1 and Fig. 2a, molecular weight correlated with rat biliary excretion (correlation coefficient r = 0.60). This moderate correlation agrees with previous findings from other laboratories (Millburn et al., 1967; Abou-El-Makarem et al., 1967a; Hirom et al., 1972a,b Hughes et al., 1973a,b; Wright and Line, 1980; Proost et al., 1997; Han et al., 2001). However, PSA and ΔGsolv aq exhibited even greater degrees of correlation (r = 0.76 and r = −0.67, respectively), as shown in Table1, and Fig. 2, b and c. ΔGsolv DMSO also correlated with biliary excretion but to a slightly lower degree than ΔGsolv aq (data not shown). In addition, the number of rotational bonds also exhibited a moderate correlation (r = 0.42). The presence of a carboxylic acid moiety in the test compounds seems to play a marked role in rat biliary excretion (r = 0.87), as shown in Table 1. None of the other physicochemical properties examined exhibited significant correlations.
Correlation between observed percentage of dose excreted in rat bile and individual physicochemical properties of test compounds. The correlation between rat biliary excretion expressed as a percentage of the intravenous dose and molecular weight (a), PSA (b), and ΔGsolv aq (c) of 50 BMS compounds was calculated using simple linear regression of least squares. Lines indicate equations of best fit with corresponding N, r, and r2 values.
Prediction of Rat Biliary Excretion for 50 BMS Compounds.
A multiple linear regression analysis was performed between rat biliary excretion and the three physicochemical properties (PSA, ΔGsolv aq, and presence/absence of carboxylic acid moieties) that seemed to influence rat biliary excretion significantly. Equation 1 was then established for predicting rat biliary excretion (percentage of dose) from the regression as follows:
where acid = 1 indicates the presence of an acid moiety and acid = 0 indicates the absence of acidic functionality.
Based on eq. 1, predicted rat biliary excretion was calculated for each of the test compounds (Table 2). Comparisons of observed and calculated values indicate that this computational model fits the training data well (Table 2; Fig. 3). The correlation was 0.94, whereas the mean absolute difference between the predicted and observed values was only 7.8% of dose. There were no obvious outliers in this data set.
Observed and predicted biliary excretion of 50 BMS compounds
r = 0.94 between observation and prediction. The mean of the differences is 0, whereas the mean of the absolute differences is 7.8% of the dose.
Correlation between predicted and observed rat biliary excretion of 50 BMS compounds using eq. 1. Rat biliary excretion of 50 BMS compounds was predicted (Table 2) using the equation biliary excretion = 0.245 × PSA + 50.289 × acid (1 or 0) − 0.616 × ΔGsolv aq − 29.395 and the corresponding physicochemical properties of BMS compounds (Table 1). A correlation between the predicted and observed rat biliary excretion was calculated using a simple linear regression of least squares.
Prediction of Rat Biliary Excretion for Compounds in Literature.
The computational model was also applied to compounds whose rat biliary excretion was published previously. The criteria for inclusion of these data were mol. wt. <1000 and relatively low intravenous bolus dose (<50 mg/kg). Table 3 summarizes results for 25 compounds described in literature reports (Hirom et al., 1972b; Russell and Klaassen 1973; Fahrig et al., 1989; Monsarrat et al., 1990; Masuda et al., 1997; Hinchman et al., 1998; Payan et al., 1999; Song et al., 1999; Arimori et al., 2003; Chong et al., 2003; Funakoshi et al., 2003; Moriwaki et al., 2003; Kamath et al., 2005a,b, 2008; Kurihara et al., 2005; Takayanagi et al., 2005; Akashi et al., 2006; Beconi et al., 2007). The reported rat biliary excretion values ranged from 0.9 to 90% of the intravenous dose. PSA and ΔGsolv aq were calculated as described above. Rat biliary excretion was predicted from eq. 1, and the predicted values are listed in Table 4. As shown in Table 4 and Fig. 4, the biliary excretion values predicted from eq. 1 agreed well with the reported in vivo results. The only obvious deviations were cephradine and paclitaxel (Taxol) for which the predictions (66 and 44%, respectively) were much higher than observed values (27 and 12%, respectively). Overall, there was a very good correlation between the predicted and observed values (r = 0.86, r2 = 0.73), with a mean absolute difference between the predicted and observed values of only 13% of dose.
Dose, duration of bile collection, and observed biliary excretion of 25 literature compounds in BDC rats
The mean of the differences is 3.9%, and the mean of the absolute differences is 13% of dose.
Physiochemical properties and observed and predicted biliary excretion of 25 literature compounds in BDC rats
Correlation between predicted and reported rat biliary excretion of 25 literature compounds using eq. 1. Rat biliary excretion of 25 compounds was predicted using the equation biliary excretion = 0.245 × PSA + 50.289 × acid (1 or 0) − 0.616 × ΔGsolv aq − 29.395 and the corresponding physicochemical properties (Table 4). A correlation between the predicted and observed rat biliary excretion was calculated using a simple linear regression of least squares.
Analysis of Structural Diversity for Compounds.
The 50 BMS compounds used in training mode were from 14 different discovery programs (Table 1). The physiochemical properties shown in Table 1 also indicate significant structure diversity. These compounds differ in molecular weight (278–738), number of rotational bonds (1–16), PSA (46.9–183.4 Å2), and ΔGsolv aq (−53.0 to −15.7 kcal/mol) as well as in ACD logP, clogP, and ACD logD at pH 6.5. Furthermore, structural similarity of the compounds against each other among 50 BMS compounds was analyzed using AtomPair fingerprints with the Tanimoto similarity coefficient. With use of a threshold of Tanimoto similarity of 0.70, 28 similarity pairs (including 4 single similarity pairs from 8 compounds and 5 similarity clusters from 20 compounds) were identified out of 1225 total pairs (Table 5). Structure similarity was only observed between or among compounds from same discovery program. The remaining 22 compounds were found not to have significant similarity with any other compounds. This analysis indicated that the 50 BMS compounds used in the study are diverse in chemical structure, representing 31 different structural clusters as described by AtomPair fingerprints.
Structural similarity among 50 BMS compounds
The threshold of Tanimoto similarity coefficient for structure similarity used in the present study is 0.70.
Structure similarity among the 25 compounds from literature reports was also examined as described above. Seven similarity pairs (including 4 single similarity pairs from 8 compounds and a similarity cluster from 3 compounds) were identified of 300 total pairs (Table 6). The remaining 14 compounds were found not to have significant similarity with any other compounds. Therefore, the 25 compounds from literature reports should represent 19 diverse chemotypes. Furthermore, structure similarity was examined comparing the 25 literature compounds against the 50 BMS compounds. The highest Tanimoto coefficient observed was only 0.36, indicating very low structural similarity between the BMS compounds and the literature compounds.
Structure similarity among 25 compounds from literature reports
The threshold of Tanimoto similarity coefficient used in the present study is 0.70.
Establishment of Similar Prediction Equations.
A second prediction equation (eq. 2) was generated using multiple linear regression as described above, but applied to the data (observed biliary excretion, PSA, ΔGsolv aq, and presence/absence of a carboxylic acid moiety) obtained from the 25 literature compounds:
where acid is defined in the same way as for eq. 1. The predicted biliary excretion of 25 literature compounds based on eq. 2 showed a reasonably good correlation with the reported in vivo biliary excretion (Fig. 5; r = 0.86 and r2 = 0.74). The predicted biliary excretion values obtained by applying eq. 2 to the 50 BMS compounds also agreed well with the observed biliary excretion values (Fig. 6; r = 0.94 and r2 = 0.89), similar to the agreement observed with predictions from eq. 1.
Correlation between predicted and reported rat biliary excretion of 25 literature compounds using eq. 2. Rat biliary excretion of 25 compounds was predicted using the equation biliary excretion = 0.148 × PSA + 41.572 × acid (1 or 0) − 0.529 × ΔGsolv aq − 12.674 and the corresponding physicochemical properties (Table 4). A correlation between the predicted and observed rat biliary excretion was calculated using a simple linear regression of least squares.
Correlation between predicted and observed rat biliary excretion of 50 BMS compounds using eq. 2. Rat biliary excretion of 50 BMS compounds was predicted using the equation biliary excretion = 0.148 × PSA + 41.572 × acid (1 or 0) − 0.529 × ΔGsolv aq − 12.674 and the corresponding physicochemical properties (Table 1). A correlation between the predicted and observed rat biliary excretion was calculated using a simple linear regression of least squares.
A third equation (eq. 3) was generated based on data from the combined data set of the 50 BMS compounds and the 25 literature compounds:
Predictions based on the model developed using the combined 75-compound data set exhibited a high correlation with the observed biliary excretion values (Fig. 7; r = 0.92 and r2 = 0.84).
Correlation between predicted and observed rat biliary excretion of 75 compounds using eq. 3. Rat biliary excretion of 75 compounds was predicted using the equation biliary excretion = 0.169 × PSA + 47.382 × acid (1 or 0) − 0.701 × ΔGsolv aq − 23.527) and the corresponding physicochemical properties (Tables 1 and 4). A correlation between the predicted and observed rat biliary excretion was calculated using a simple linear regression of least squares.
Impact of a Different PSA Calculation on Prediction Model.
PSA appears to be the critical physiochemical property in the current prediction models as demonstrated above. To explore whether PSA data generated from a different calculation method have a significant impact on prediction of biliary excretion, a second set of PSA data was generated for all compounds using the TPSA algorithm in the OEChem Toolkit (OpenEye Scientific Software, Inc., Santa Fe, NM). The two sets of PSA data were highly correlated (r = 0.97 and r2 = 0.94), indicating that the prediction model should not be limited by the software used to calculate the physicochemical properties.
Discussion
In drug discovery, a computational model for predicting rat biliary excretion based on physicochemical properties would allow the estimation of biliary excretion for both existing and virtual compounds. A reliable model could significantly reduce the number of high-cost BDC rat studies that are currently needed. Previous efforts to explore the factors influencing rat biliary excretion have met with limited success (Millburn et al., 1967; Abou-El-Makarem et al., 1967a; Hirom et al., 1972a,b; Hughes et al., 1973a,b; Wright and Line 1980; Proost et al., 1997; Han et al., 2001). Previous efforts probably failed to establish a highly predictive computational model because of the small sets of test compounds and lack of diverse chemical structures in each of those studies. In addition, those efforts tended to focus on single physicochemical properties (for example, molecular weight) instead of a combination of multiple physicochemical factors (Millburn et al., 1967; Abou-El-Makarem et al., 1967a; Hirom et al., 1972a,b; Hughes et al., 1973b; Wright and Line 1980).
In the present study, 50 BMS compounds were evaluated in BDC rats. Observed biliary excretion, expressed as percentage of the intravenous dose, ranged from 0.1 to 100%. These compounds were assessed under the same experimental conditions, including identical strain and gender as well as comparable body weights, a standard surgical procedure for preparation of BDC rats, an adequate duration (at least 8 h) of bile collection, relatively low intravenous doses of test compounds (to avoid saturation of biliary transport), and similar LC-MS/MS methods for quantitation of intact parent in bile samples. In addition, this study included a relatively large set of test compounds covering a wide diversity of chemical structures. The BMS compounds were from 14 different discovery programs. The compounds covered a range of physicochemical properties including molecular weights (278–739), numbers of rotational bonds (1–16), PSA (46.9 to 183.4 Å), and ΔGsolv aq. (−53.0 to −15.7 kcal/mol) as well as differences in ACD logP, clogP, and logD (pH 6.5). Analysis of structural similarity confirmed diversity of structure and indicated that the compounds represent 31 different chemotypes. As a result, the test compounds and the results of biliary excretion generated in the present study should be adequately qualified for establishing a computational model.
Using a multiple linear regression method to correlate rat biliary excretion and three physicochemical properties (PSA, ΔGsolv aq, and the presence/absence of carboxylic acid moieties), an equation for predicting rat biliary excretion was established. The observed and calculated rat biliary excretion of 50 internal BMS training compounds showed good correlation (r = 0.94, r2 = 0.89). The mean of the absolute differences between the observed and the predicted biliary excretion values was only 7.8% with no obvious outliers.
When the same computational model was applied to 25 structurally diverse compounds whose rat biliary excretion data were published by different laboratories (Hirom et al., 1972b; Russell and Klaassen 1973; Fahrig et al., 1989; Monsarrat et al., 1990; Masuda et al., 1997; Hinchman et al., 1998; Payan et al., 1999; Song et al., 1999; Arimori et al., 2003; Chong et al., 2003; Funakoshi et al., 2003; Moriwaki et al., 2003; Kamath et al., 2005a,b, 2008; Kurihara et al., 2005; Takayanagi et al., 2005; Akashi et al., 2006; Beconi et al., 2007), the predicted and observed biliary excretion values were again very close for most compounds. The clear exceptions were cephradine and paclitaxel (Monsarrat et al., 1990; Moriwaki et al., 2003). The correlation coefficient was 0.88, and the mean of the absolute differences between the observed and predicted values was only 13%. Such a successful prediction is significant because the 25 literature compounds are structurally distinct from the 50 BMS compounds, and the rat biliary excretion data were generated in approximately 20 different laboratories using various experimental conditions.
A similar prediction model was established using PSA, ΔGsolv aq, presence/absence of carboxylic acid, and the reported rat biliary excretion (percentage of intravenous bolus dose) from 25 literature compounds alone or from the combined 75-compound data set that includes both the literature and in-house data sets. Those models successfully predicted the in vivo rat biliary excretion, verifying the reliability of the prediction model reported in the present study. In addition, variations in PSA values calculated by a different method should not affect the biliary elimination prediction model.
In the present study, the top three critical physicochemical properties influencing biliary excretion in rats were found to be PSA, ΔGsolv aq, and presence/absence of carboxylic acid moieties. Note that molecular weight was found to play an important but less critical role in biliary excretion. The correlation coefficient between molecular weight and biliary excretion was 0.60, less than that between PSA (or ΔGsolv aq) and biliary excretion. For example, irinotecan exists stably in lactone and carboxylate forms. Although the molecular weights of the two forms differ by only 16 atomic mass units, the percentage of dose excreted in bile was quite different (7 and 60%, respectively) (Arimori et al., 2003). Based on eq. 1 described above, biliary elimination is predicted to be 15 and 79% of dose, respectively, close to the observed values.
In the present study, the presence or absence of a carboxylic acid was noted as a key factor for biliary excretion. This finding is in agreement with many well known examples of carboxylate-containing compounds that exhibit high levels of biliary elimination such as irinotecan (carboxylate form), methotrexate, bile acids (such as taurocholic acid), and glucuronide, glutathione, and sulfation conjugates derived from the parent compounds.
Active transport by specific canalicular membrane transporters such as P-glycoprotein (ABCB2), multidrug resistance-associated protein 2, breast cancer resistance protein (ABCG2), and bile salt export pump (Chandra and Brouwer 2004) is generally believed to contribute significantly to biliary elimination. Although the predictive model for biliary excretion generated in the present study cannot distinguish contributions from specific transporters, it may reflect a generalized physicochemical space encompassing common structural characteristics recognized by the canalicular membrane transporters that are involved in biliary elimination.
Biliary excretion of a compound may also be related to its hepatic metabolism. The relationship between biliary excretion and metabolism seems to be inversely correlated. Compounds with high biliary excretion are often poorly metabolized [such as digoxin (Caldwell and Cline 1976; Funakoshi et al., 2003), methotrexate (Henderson et al., 1965a; Fahrig et al., 1989), and pravastatin (Hatanaka 2000)], whereas those that are highly metabolized generally show poor biliary excretion [such as dasatinib and BMS-182874) (Chong et al., 2003; Kamath et al., 2008)]. Furthermore, many carboxylic acid-containing compounds in the present study were poorly metabolized by rat liver microsomes (data not shown) and showed a high biliary excretion. Although the relationship between biliary excretion and hepatic metabolism is beyond the scope of the present study, the predictive model of rat biliary excretion described here may be of use in efforts to estimate the relative contributions of rat biliary excretion and hepatic metabolism to the overall elimination of test compounds in rats.
The predictive model generated from the present study probably applies only to rat biliary excretion because significant species differences in biliary excretion have been observed (Henderson et al., 1965a,b; Abou-El-Makarem et al., 1967b; Bertagni et al., 1972; Gregson et al., 1972; Hirom et al., 1972b; Hughes et al., 1973a). For instance, biliary excretion of methotrexate was 57 to 72% in rats, 7% in dogs, 16% in monkeys, and 10% or less in humans (Henderson et al., 1965a,b; Masuda et al., 1997; Luo et al., 2007). In general, biliary excretion often plays a more significant role in the elimination of xenobiotics in rats than in other species, possibly because of the high expression of transporters on the canalicular membrane of rats coupled with high bile flow relative to body weight. The relationship of biliary elimination trends across species remains to be investigated adequately.
This predictive model estimates the percentage of intravenous dose excreted into bile but not the biliary clearance as in units of milliliters per minute per kilogram. To calculate biliary clearance, corresponding pharmacokinetic data are needed. For most compounds used in the present study, the pharmacokinetic data from rats were not available.
Acknowledgments.
We thank the following colleagues at Bristol-Myers Squibb Co. for providing rat biliary excretion data for the current study: Laishun Chen, Aberra Fura, Christine Huang, Hongjian Zhang, Ming Zheng, and Yang Zheng. We also thank colleagues in the Technical Support Unit at Bristol-Myers Squibb Co. for preparing and dosing BDC rats and collecting bile and urine samples. We also thank Dr. Wen Chyi Shyu for her critical review and comments on the manuscript. Finally, we thank the anonymous reviewers who made helpful suggestions to improve this manuscript.
Footnotes
Article, publication date, and citation information can be found at http://dmd.aspetjournals.org.
doi:10.1124/dmd.108.026260.
-
- BMS
- Bristol-Myers Squibb Co.
- BDC
- bile duct-cannulated
- PSA
- polar surface area
- ΔGsolv aq
- free energy of aqueous solvation
- LC
- liquid chromatography
- MS/MS
- tandem mass spectrometry
- ΔGsolv DMSO
- free energy of solvation in dimethyl sulfoxide
- BMS-182874
- 5-(dimethylamino)-N-(3,4-dimethyl-5-isoxazolyl)-1-naphthalene sulfonamide.
- Received December 18, 2008.
- Accepted December 3, 2009.
- Copyright © 2010 by The American Society for Pharmacology and Experimental Therapeutics
