Introduction

Top Abstract Introduction Results and Discussion Conclusions Appendix 1 Appendix 2 References

The modern pharmaceutical industry has adopted high-throughput screening for the identification of lead molecules (Fernandes, 1998) and subsequent optimization of chemical structure, leading to clinical candidates with desirable biopharmaceutical and pharmacokinetic properties (e.g., clearance, half-life, and oral bioavailability) (Tarbit and Berman, 1998). Orally active, once-a-day drugs are desirable because of their clinical and commercial advantages. In recent years, several pharmaceutical companies have published reports on the simultaneous administration of several compounds to a single animal (cassette dosing or "N-in-One" dosing) (Berman et al., 1997; Olah et al., 1997; Allen et al., 1998; Frick et al., 1998a; Shaffer et al., 1999; Wu et al., 2000) as a means to rapidly rank-order compounds on the basis of their pharmacokinetics. Cassette dosing is used to screen compounds in two general ways: for systemic clearance (i.v. dosing) and for oral plasma drug levels (p.o. dosing). Compared with conventional pharmacokinetic studies, this method has the advantage of speed, because the slow steps of animal dosing, blood collection, and sample analysis are minimized. Another advantage is that animal usage is greatly reduced, which is particularly important when dog or monkey is the test species. Cassette dosing also avoids the problem of in vitro-in vivo correlation, which is always present with in vitro methods of rapidly assessing pharmacokinetics. The enabling technology for cassette dosing is liquid chromatography coupled to tandem mass spectrometry which allows many compounds to be simultaneously assayed in a single sample (Berman et al., 1997; McLoughlin et al., 1997; Beaudry et al., 1998; Frick et al., 1998b). The degree of acceleration depends on the number of coadministered compounds (n), but there are practical limitations on n. Most applications of cassette dosing use n of 10 or less (Bayliss and Frick, 1999).

Cassette dosing has been controversial because of concerns over whether there are serious errors. Some reports claim that reliable pharmacokinetic data are obtained (Berman et al., 1997; McLoughlin et al., 1997; Olah et al., 1997; Allen et al., 1998; Frick et al., 1998a; Bayliss and Frick, 1999; Shaffer et al., 1999; Rano et al., 2000; Wu et al., 2000), but only a few have actually demonstrated reasonable correspondence of pharmacokinetic parameters obtained from cassette dosing and conventional single-compound dosing (Berman et al., 1997; Frick et al., 1998a; Shaffer et al., 1999). However, in some of the other reports, large errors are evident (Allen et al., 1998; Bayliss and Frick, 1999), and in several studies no attempt was made to actually assess the reliability of the results (McLoughlin et al., 1997; Rano et al., 2000; Tong et al., 1999; Wu et al., 2000). Although many investigators are aware of the potential for the occurrence of drug-drug interactions to compromise the results, there has been no published assessment of these interactions in terms of their nature, frequency, magnitude, and direction. A majority of the papers focus on the analytical challenges of simultaneously assaying many compounds in a single sample. In the absence of theoretical guidance, a set of intuitive assumptions has arisen regarding the nature of the errors and how to avoid them.

These assumptions are

1.	Drug-drug interactions only occur when one of the dosed compounds is a potent inhibitor of drug-metabolizing enzymes;
2.	One may guard against competitive inhibition of a shared metabolic enzyme by keeping doses small;
3.	The size of the cassette (n) is limited only by the sensitivity of the assay and the solubility of the compounds;
4.	Errors can be detected by including a benchmark compound with known pharmacokinetic characteristics;
5.	Drug-drug interactions can lead only to false positives, which will be discovered later; and
6.	Even if the absolute values are wrong, the correct rank order will be observed.

We applied pharmacokinetic principles to analyze and understand the kinetics of cassette dosing. We then evaluated the above assumptions, both from the viewpoint of theory and by reference to published experimental data. We found that none of these assumptions is always valid.

	Results and Discussion

Top Abstract Introduction Results and Discussion Conclusions Appendix 1 Appendix 2 References

Types of Drug-Drug Interactions. In the following discussion, we will examine the pharmacokinetics of drug-drug interactions following coadministration of many compounds. To begin, let us enumerate the drug-drug interactions that potentially have pharmacokinetic consequences, as follows:

1.	Competition for clearance pathways (mutual inhibition of metabolic enzymes and transporter proteins);
2.	Competition for net absorption (mutual inhibition of influx and efflux transporter proteins);
3.	Competition for plasma protein binding;
4.	Heteroactivation of clearance pathways (activation of a metabolic enzyme by a second drug acting through an allosteric mechanism);
5.	Pharmacological and toxicological effects on organ blood flows and clearances; and
6.	Enzyme induction phenomena.

Pharmacokinetic Theory. We can consider cassette dosing to be an extreme case of multiple drug therapy. Our approach is similar to the pharmacokinetic theory of drug-drug interactions in multiple drug therapy elaborated by Aarons (1981). A rigorous treatment of drug-drug interactions due to metabolic enzyme inhibition has also been given by Ito et al. (1998). Let us consider the case of oral bioavailability, which is often the effective endpoint of cassette dosing screening. Absolute oral bioavailability (F¹) is sensitive to effects 1 through 5, as shown in eq. 1:

(1)

where f_a is the fraction of dose absorbed from the intestinal lumen into the gut wall, E_G (gut wall extraction ratio) is the fraction eliminated by metabolism in the gut wall, and E_H (hepatic extraction ratio) is the fraction eliminated by the liver during the first pass. Passive absorption is not expected to be much affected by cassette dosing, but the active uptake and efflux components of f_a and the first pass extractions (i.e., E_G and E_H) can be substantially affected. We will limit this discussion to effects at the level of E_H. However, at the end of the discussion, it will be intuitively clear how to extend the analysis to f_a and E_G.

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

(11)

Evaluation of Common Assumptions. In the light of eq. 11, we can now examine the intuitive assumptions that were mentioned in the introduction as the supposed operational characteristics of cassette dosing.

Assumption 1. Drug-drug interactions only occur when one of the dosed compounds is a potent inhibitor of drug-metabolizing enzymes. Many variations of this very intuitive assumption have been presented in the literature (McLoughlin et al., 1997; Olah et al., 1997; Frick et al., 1998a; Tarbit and Berman, 1998; Rano et al., 2000). While a potent inhibitor can cause serious drug-drug interactions, we will show here that this is not the only source of these interactions. In fact, inhibition of a similar magnitude can be expected whenever a sufficient number of nonpotent inhibitors are present. This number will be seen to be in the range of the number of drugs in a typical cassette.

Assumption 2. One may guard against competitive inhibition of a shared metabolic enzyme by keeping doses small. This assumption is almost universally cited (Berman et al., 1997; McLoughlin et al., 1997; Olah et al., 1997; Adkison et al., 1998; Allen et al., 1998; Frick et al., 1998a; Tarbit and Berman, 1998; Bayliss and Frick, 1999). We will show that the use of low doses will tend to reduce drug-drug interactions, but meaningful curtailments of these interactions are probably not realized with the typical "low" doses (1 mg/kg) used in reported studies.

View larger version (13K): [in this window] [in a new window]   Fig. 1.   Dependence of fractional intrinsic clearance on concentration of the individual compounds in a cassette of 10 drugs. K values of all drugs were set to 1. CL* was calculated according to eq. 11. Drug concentration is expressed as multiples of the K value of Drug 1. Thus, a drug concentration of 0.1 means that each drug in the cassette was present at a concentration equal to 10% of K1. The horizontal line corresponds to 50% inhibition of intrinsic clearance. The intersection of this line with the inhibition curve (indicated by the arrow) shows that 50% inhibition occurs when concentrations of the 10 drugs are only 10% of K1.

Assumption 3. The size of the cassette (n) is limited only by the sensitivity of the assay and the solubility of the compounds. Restrictions on the size of the cassette are sometimes explicitly stated (Olah et al., 1997; Adkison et al., 1998; Bayliss and Frick, 1999) but are more often implicit. Some investigators have used large cassettes and seem to recognize no real restrictions (Frick et al., 1998a; Beaudry et al., 1998; Rano et al., 2000; Wu et al., 2000). In all cases, however, the aggregate inhibitory effect of n coadministered compounds, as predicted by eq. 11, was not appreciated.

View larger version (16K): [in this window] [in a new window]   Fig. 2.   Dependence of fractional intrinsic clearance on the total number of drugs that are coadministered. Three cases are shown in which the reference compound is the most potent inhibitor (Min; open circles), the weakest inhibitor (Max; closed triangles), or all compounds have equal inhibitory potency (Equal K's; closed circles). For these simulations, a set of 10 compounds was chosen with K values distributed randomly above and below K1 (K1 ... K10 = 1, 0.1, 0.2, 0.3, 0.5, 1, 2, 3, 5, 10) and concentrations equal to 10% of K1. For the specific n values, subsets of the full set were taken, as follows: (n = 2: 0.1 and 10); (n = 3: 0.1, 1, and 10); (n = 6: 0.1, 0.3, 1, 2, 3, and 10); (n = 10: 1, 0.1, 0.2, 0.3, 0.5, 1, 2, 3, 5, and 10); (n = 20, 30, 60, and 1000, multiples of the entire set).

Assumption 4. Errors can be detected by including a benchmark compound with known pharmacokinetic characteristics. Including a benchmark compound (also called a biological internal standard) is commonly used to safeguard the accuracy of the results (McLoughlin et al., 1997; Olah et al., 1997; Adkison et al., 1998; Allen et al., 1998; Frick et al., 1998a; Bayliss and Frick, 1999; Shaffer et al., 1999; Tong et al., 1999; Rano et al., 2000). If the benchmark compound gives comparable data in the cassette experiment versus individual dosing, then it is presumed that no drug-drug interactions occurred. However, this assumption is not always true. Referring again to Fig. 2, we can see that a benchmark compound may show little change in its own clearance if it happens to have one of the lower K values for the drug-metabolizing enzyme involved. With n = 3, the Min curve shows only about 5% error, even though other members of the same cassette could have experienced intrinsic clearance reductions as large as 52%, as shown by the Max curve.

Assumption 5. Drug-drug interactions can lead only to false positives, which will be discovered later. A false positive is a result in which a compound appears to have acceptable pharmacokinetic characteristics when dosed in a mixture but would be identified as unacceptable if dosed singly. Mutually competitive inhibitors of elimination pathways tend to decrease clearance and thereby increase plasma levels and AUC. This may be acceptable in a screening mode, since it will tend to produce false positives, which will be corrected in the later single-compound pharmacokinetics determination.

1.

For drugs that are restrictively cleared (i.e., only the nonprotein bound fraction is subject to clearance), clearance can be overestimated by displacement from protein binding leading to higher fu and higher effective clearance (see further discussion of this phenomenon below);

2.

If both i.v. and p.o. doses are used, F will be underestimated if AUCiv is increased proportionally more than AUCpo, implying that CLiv is decreased proportionally more than CLpo. Unfortunately, this is exactly what happens when one or more of the drugs has less than complete oral absorption because less total drug is delivered systemically after the oral dose. In that case, less inhibition of intrinsic clearance occurs after the oral dose because the p.o. inhibitor concentration term in eq. 11 is less than the i.v. concentration term, i.e., (12)

3.

Clearance can be higher in cassette dosing than in single-compound dosing if the compounds being screened are agents, such as vasodilators, that increase liver blood flow. As seen in the clearance equation below, for high-clearance drugs the actual clearance is determined by hepatic blood flow (QH) (Gibaldi and Perrier, 1982). Thus, any effect that increases blood flow will also increase clearance; (13)

4.

If a component of the cassette is an activator of a drug-metabolizing enzyme, the clearance of other components can be increased. For example, Tang et al. (1999) reported an increase in clearance of diclofenac when codosed with quinidine, which was attributed to an allosteric effect of quinidine on CYP 3A;

5.

The clearance of a component could appear to increase in a cassette if it exhibits nonlinear pharmacokinetics and if the cassette dose is much lower than a subsequent single-component dose. For example, a compound with a low KM for the drug-metabolizing enzyme might be in the linear range at the low cassette dose but in the saturated range when dosed singly at a higher, more pharmacologically relevant dose. This apparent clearance increase is not a direct consequence of cassette dosing, since it would occur even if the compound were dosed singly at the low dose. Nonetheless, because the cassette-dosing procedure encourages the use of very low doses, we increase the risk of being misled.

Assumption 6. Even if the absolute values are wrong, the correct rank order will be observed. In recognition of the unreliability of the parameter values, a frequently discussed tactic is to use cassette dosing merely to rank-order the compounds. Then, the ones with the best plasma levels are taken forward in the drug discovery program, with the assumption, implicit or explicit (Adkison et al., 1998; Allen et al., 1998), that the best compound in the cassette actually is the best compound. It should be clear from the discussion above that the drug-drug interactions have differential effects, so that each compound experiences a different reduction in clearance. Thus, eq. 11 predicts that the correct rank order of AUC values will not necessarily be maintained in cassette dosing. We can compare this prediction experimentally. Berman et al. (1997) reported a reasonable maintenance of the correct order of clearances comparing cassette with individual dosing. They found that the correct order (high to low) of 2, 4, 3, 1, 5 was modified to 3, 2, 4, 1, 5 in the cassette experiment. Allen et al. (1998) observed that the correct order (high to low) of rat plasma AUC values for a series of compounds (43, 15, 41, R1, 7, 2, 13, 44, 42) was distorted to 43, R1, 41, 42, 2, 44, 7, 15, 13. Shaffer et al. (1999) also reported some distortion of the order (high to low) of clearance values, from 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 to 1, 3, 4, 2, 7, 5, 8, 6, 10, 9. Finally, considerable discrepancy in order was seen in the work of Bayliss and Frick (1999); the compound with the highest plasma level in the cassette experiment was only the ninth highest according to individual dosing. These experiments verify the prediction of eq. 11 that the correct rank order will not necessarily be preserved in cassette dosing.

High- versus Low-Clearance Drugs. Are the drug-drug interactions predicted by eq. 11 more important for high-clearance or low-clearance drugs? Returning to our original endpoint of absolute bioavailability (eq. 4) and assuming for simplicity no plasma protein binding, we get

(14)

In the multidrug situation, we must correct CL_i by the factor CL*.

(15)

Now let us take a hypothetical drug that actually has a medium extraction ratio (i.e., with CL_i equal to Q_H). When the drug is dosed singly, the bioavailability can be calculated to be
When the drug is dosed in a mixture of 10 drugs, we can use the value of CL* calculated in Case 3 above (i.e., CL* = 0.33) to calculate the altered bioavailability
By repeating this calculation at several values of CL_i spanning a range from low to high clearance, we can assess where the error is most serious.

View larger version (23K): [in this window] [in a new window]   Fig. 3.   Error in estimates of bioavailability of a reference drug in the absence (F) or presence (F*) of nine other competing drugs as a function of intrinsic clearance of the drug. For this simulation, a set of 10 compounds was chosen with K values distributed randomly above and below K1 (K1 ... K10 = 1, 0.1, 0.2, 0.3, 0.5, 1, 2, 3, 5, 10) and concentrations equal to 10% of K1. Bioavailability for each value of CLi was calculated as shown in the text. The dashed horizontal line corresponds to an arbitrary screening cutoff criterion of 10% bioavailability. Compounds falling below this line would be eliminated by the screening procedure. Inset, relative error was calculated by (F*  F)/F × 100%.

Plasma Protein Binding. Cassette dosing can potentially cause an increase in clearance if one or more compounds in the cassette are restrictively cleared (i.e., only the unbound fraction in plasma is subject to clearance). This effect arises because competition for binding sites will occur with every protein with which a set of similar compounds interacts. Obviously, just as with the metabolic enzymes, compounds with higher affinity for plasma proteins will displace those with lower binding. However, in exact analogy to binding of compounds to the metabolic enzymes, competition among the cassette components for binding to plasma proteins such as serum albumin will also be governed by a C/K term. Thus, even compounds with lower affinity than Drug 1 will tend to displace when their aggregate concentration is high. Provided that the drug concentration is in the constant protein-binding range, the unbound fraction (f_u) is given by eq. 16 (see Appendix 2).

(16)

Figure 4 shows an example of the increase in unbound fraction as the number of competing drugs increases, in the special case in which all K_d values are equal. Obviously, if one or more members of the set are much more tightly bound to plasma protein, then other members of the set will be strongly displaced. The effect of the increase of unbound fraction on restrictively cleared drugs will be an increase in clearance, according to eq. 13.

View larger version (10K): [in this window] [in a new window]   Fig. 4.   Increase of unbound fraction (fu) of a reference drug due to competition by n coadministered drugs. The value of fu at each value of n was calculated according to eq. 16 by setting all concentrations and all Kd values at 10 µM and assuming 600 µM plasma protein. These values of C and Kd correspond to 98.4% protein binding in the absence of other competing drugs (i.e., n = 1).

View larger version (12K): [in this window] [in a new window]   Fig. 5.   Increase of CL of a restrictively cleared reference drug due to increase of unbound fraction caused by n coadministered drugs. Simulation parameters for protein binding were as in Fig. 4. Clearance was calculated according to eq. 13 assuming CLi = 2 QH. Dashed line, the asymptote of 0.67 QH discussed in text.

Reconciliation with Published Results. The preceding section shows the potential for large errors when applying cassette dosing. However, several accounts have been published in which good results have been claimed, as judged by comparison of pharmacokinetic parameters calculated from multiple- and single-compound dosing (Berman et al., 1997; Olah et al., 1997; Allen et al., 1998; Frick et al., 1998; Shaffer et al., 1999). The available data are compiled in Table 1. How can we reconcile the conclusions from eq. 11 and these published results? To begin, let us explicitly recognize that the derivation of eq. 11 assumed that clearance of all drugs in the set is due to metabolism. Therefore, if other clearance pathways are available to some drugs in the set, net clearance of those drugs may be relatively insensitive to competitive inhibition of only a single enzyme. In addition, eq. 11 refers only to intrinsic clearance, so drugs whose clearance is limited by blood flows may show little effect on net clearance even though their intrinsic clearances have been reduced. Also, many other physiological processes that contribute to the absorption, distribution, and elimination of drugs are potentially subject to the same drug-drug interactions that have been described here for metabolism (see Table 2). Each process that is mediated by a protein (i.e., each saturable process) may potentially be modulated by competition between different drugs, and the net result will be difficult to predict.

"Right Box" Analysis. Another measure of success of a screening method is the placement of the test compound into the correct category, something we can call the Right Box approach. A pharmacokinetic parameter that can be interpreted in an absolute sense, such as clearance, is divided into three categories: low, medium, and high. For instance, we can (somewhat arbitrarily) define low, medium, and high clearances as <10, 10 to 20, and >20 ml/min/kg, respectively, in dogs based on the hepatic extraction ratios falling into the ranges of 0 to 0.3, 0.3 to 0.7, and 0.7 to 1.0 according to the relationship E_H = CL/Q. The observed clearance values calculated from both individual and cassette-dosed experiments are then plotted as shown in Fig. 6. Boxes are defined by the areas enclosed by the boundaries for low, medium, and high clearances. Data points that fall within these areas (the right box) are scored as successful. Three of the published papers provided enough data for a Right Box analysis on clearances⁵. As seen in Table 1, Right Box success rates were 80, 86, and 100% in those studies. This suggests that if one is willing to accept gross categorization as an endpoint of the screening process, then cassette dosing may be an adequate procedure despite large absolute errors. Unfortunately, too little data have been published to allow us to judge whether cassette dosing is always successful in the Right Box approach. This is clearly an area where additional data sets would be a valuable contribution to the literature.

View larger version (15K): [in this window] [in a new window]   Fig. 6.   Right Box analysis. Success of cassette dosing in correct clearance classification of compounds. In this case, only 3 of 21 points did not fall into a box, corresponding to 86% success (data points taken from Frick et al., 1998a).

	Conclusions

Top Abstract Introduction Results and Discussion Conclusions Appendix 1 Appendix 2 References

Although cassette dosing has been reported to yield useful results when used as a screen, especially to rank-order drug candidates, we have shown both theoretically and experimentally the potential for large errors. Consequently, under no circumstances can the pharmacokinetic parameters derived from cassette dosing be accepted as accurate. Potentially affected parameters include F, CL, AUC, t_1/2, mean residence time, V_d. High-clearance compounds have the greatest potential for a serious screening error (i.e., false positive, false negative). To detect serious errors, we could use a second dosing episode, either in a different cassette or as a single-compound dose. Obviously, a second dosing defeats the productivity gain of cassette dosing. A better way to detect errors is to include a benchmark compound with known in vivo pharmacokinetics, but we have shown that this is no guarantee. To minimize the potential for errors, one should use the smallest doses detectable and keep the total number of coadministered compounds small.