Results and Discussion

As noted in the Introduction, TKIs are typically lipophilic organic bases that are neutral or partially ionized at pH 7.4. The aqueous solubility of TKIs is poor, although, at least at relatively low concentrations (<100 µM), solubilization is achieved in in vitro incubations containing HLM and DMSO (1% w/v). Initial studies of the NSB of lapatinib, pazopanib, regorafenib, and sorafenib to HLM, using either equilibrium dialysis cells or a commercial RED device, demonstrated that equilibrium was established in buffer-buffer and HLM-HLM (0.25 mg/ml) controls only for pazopanib. Increasing the concentration of DMSO present in the dialysis medium to 2% or using other organic solvents (acetonitrile, methanol, and 0.95% acetonitrile plus 0.05% DMSO, as reported by Kenny et al. (2012)] did not result in the attainment of equilibrium in control experiments with lapatinib, regorafinib, and sorafenib. The mean (± S.D.) fu_mic of pazopanib obtained by conventional equilibrium dialysis (in the presence of 1% DMSO) was 0.75 ± 0.03 for five concentrations in the range of 0.1–2 µM (Table 1). Similarly, the use of ultracentrifugation with lapatinib, regorafinib, and sorafenib resulted in almost immeasurable concentrations in the reservoir compartment and low recoveries (<50%).

Thus, we explored the use of the detergents Brij-58, CHAPS, and Triton X-100 to aid the solubilization of TKIs in NSB experiments. Each of the detergents used differ in terms of physicochemical properties and critical micelle concentration (CMC). Since detergents may disrupt the microsomal membrane and hence potentially NSB, equilibrium dialysis experiments were performed with eight validation set basic drugs, for which attainment of equilibrium could be demonstrated in buffer-buffer and HLM-HLM (0.25 mg/ml) controls in the absence of detergent. Pazopanib (see above), felodipine, isradipine, loratidine, midazolam, and nifedipine are weak organic bases that are nonionized in phosphate buffer at pH 7.4, whereas amitriptyline and chlorpromazine are lipophilic bases that are essentially completely ionized at pH 7.4. Zidovudine, a zwitterionic compound that is extensively ionized and known not to bind nonspecifically to HLM (Uchaipichat et al., 2006), served as a “negative” control. Table 1 lists the percentage of ionization and the calculated log P for each of the control drugs and TKIs investigated here, whereas Supplemental Fig. 1 shows the structures of the validation set drugs and TKIs. It is noteworthy that calculated log P varied by as much as a factor of 2 (i.e., 2 log orders) using the four algorithms employed here (Table 1). Thus, Table 1 gives the range (lowest to highest) of the calculated log P values.

NSB of the validation set drugs amitriptyline, chlorpromazine, felodipine, isradipine, loratidine, midazolam, nifedipine, and zidovudine was measured in duplicate at five concentrations in the range of 1–100 µM, whereas the binding of all TKIs (including pazopanib) was measured at five concentrations in the range of 0.1–2 µM, which we have found to be relevant for inhibition studies (unpublished data). Binding of each of the validation set drugs was independent of concentration in the absence of detergent. Brij-58 (0.007 mM; CMC, 0.007–0.07 mM) disrupted the binding of all eight basic drugs; fu_mic values were higher with added detergent compared with the absence of detergent (data not shown). Despite the attainment of equilibrium in buffer-buffer and HLM-HLM controls in NSB experiments conducted in the presence of Triton X-100 (CMC, 0.2–0.9 mM), the recoveries of each compound were low (<50%). By contrast, the presence of CHAPS (6 mM; CMC, 6–10 mM) had no effect on the binding of pazopanib, felodipine, isradipine, loratidine, midazolam, and nifedipine to HLM, although binding of the fully charged bases amitriptyline and chlorpromazine to HLM was disrupted (Table 1). No binding of zidovudine was observed in the absence or presence of CHAPS. Importantly, recoveries of all nine drugs from the equilibrium dialysis cells were acceptable (>70%) and attainment of equilibrium (fu_mic values of 0.85–1.05) was demonstrated for buffer-buffer and HLM-HLM control experiments.

Based on these data, the microsomal binding of eight TKIs (axitinib, erlotinib, gefitinib, ibrutinib, nilotinib, regorafenib, sorafenib, and trametinib) that are completely or largely unionized at pH 7.4 and three TKIs that are partially ionized (39–64%) at pH 7.4 (dabrafenib, lapatinib, and nintedanib) was characterized in the presence of CHAPS (6 mM) (Table 1). NSB of the TKIs was concentration independent over the range of 0.1–2 µM. Measured fu_mic values ranged from 0.14 (regorafenib and sorafenib) to 0.93 (nintedanib). Recoveries of each of the TKIs from equilibrium dialysis cells were >70%, and attainment of equilibrium was demonstrated in buffer-buffer and HLM-HLM control experiments (fu_mic values of 0.85–1.05). Subsequent experiments demonstrated that equilibrium was also attained in dialysis experiments with the neutral TKI erlotinib and the partially charged dabrafenib in the absence of CHAPS; fu_mic values generated in the presence and absence of detergent were essentially identical (Table 1).

There was no clear relationship between the extent of ionization at pH 7.4 and fu_mic for the drugs investigated here. While the two most extensively charged compounds (dabrafenib and nintedanib) exhibited low NSB, the moderately ionized (38%) lapatinib bound extensively to HLM. Although the three highest binding TKIs (lapatinib, regorafenib, and sorafenib) have high log P values (4.2–5.2), the binding of axitinib (log P of 4.2) was comparatively low. Thus, log P is clearly not the sole determinant of extensive NSB of TKIs to HLM. We and others have shown previously that NSB is generally highest for lipophilic organic bases with molecular masses >200 Da (Austin et al., 2002; Hallifax and Houston, 2006; Sykes et al., 2006; McLure et al., 2011). However, other physicochemical properties also influence the degree of NSB. These include the presence of halogen atoms (especially the trifluoromethyl group), polar surface area, charge distribution, and number of hydrogen bond acceptors/donors [summarized by McLure et al. (2011)]. Not only does the addition of fluorine or the trifluoromethyl groups to an aryl ring or adjacent to a heteroatom-containing functional group increase lipophilicity (Hagmann, 2008), but halogenation of drugs is more generally known to enhance drug membrane binding and permeation. The importance of the trifluoromethyl group in particular has been highlighted in this regard (Gerebtzoff et al., 2004). With respect to the TKIs investigated here, nilotinib, regorafenib, and sorafenib all possess a trifluoromethyl group (Supplemental Fig. 1). However, the two TKIs that bind most extensively to HLM, sorafenib and regorafenib, additionally contain one and two additional halogen atoms, respectively.

Kenny et al. (2012) reported the NSB of a number of TKIs to HLM (0.6 mg/ml) using a commercial RED device. The TKIs investigated included several of those studied here: lapatinib, nilotinib, pazopanib, and sorafenib. As noted previously, in our hands, only pazopanib attained equilibrium in HLM-HLM and buffer-buffer controls using a RED device and conventional equilibrium dialysis. The previous study did not report data for HLM-HLM and buffer-buffer controls. Since different HLM concentrations were employed in the two studies, the results are not directly comparable. However, the rank order of binding reported by Kenny et al. (lapatinib > nilotinib > pazopanib > sorafenib) differs markedly from that found here.

The SimCYP ADME calculator and equations published by Austin et al. (2002) and Hallifax and Houston (2006) that were used here to predict fu_mic values take into account only log P and charge state/ionization. The predicted fu_mic values are shown in Table 1 as a range since the calculated log P values differed among the algorithms. Indeed, as highlighted previously, the calculated log P values surprisingly differed by as much as a factor of 2. Despite using a range of log P values for the calculation of fu_mic, this parameter was generally underpredicted for the most highly bound compounds using the SimCYP calculator and the equation published by Hallifax and Houston. Although the fu_mic values predicted for the highly bound compounds regrofenib and sorafenib using the equation of Austin et al. spanned the experimentally determined fu_mic, this equation tended to overpredict the binding of other TKIs. No single algorithm for calculated log P performed consistently well for predicting fu_mic (data not shown). More generally, the choice of log P algorithm used for the prediction of NSB and other parameters warrants further investigation.

Available evidence demonstrates that NSB, particularly of lipophilic basic drugs, arises largely from the partitioning of drugs into phospholipid membranes (Margolis and Obach, 2003; Nussio et al., 2007; McLure et al., 2011; Nagar and Korzekwa, 2012). Neutron diffraction studies with the charged lipophilic base propranolol showed that that the lipophilic naphthalene moiety partitions into the lipid bilayer, whereas the charged amine group apparently associates with the phospholipid head group (Herbette et al., 1983). CHAPS was selected as the detergent for the solubilization of TKIs here because it retained all compounds in an aqueous solution and, since the concentration employed in this work (6 mM) is at the lower end of the CMC range, it was expected not to affect dialysis. This was confirmed by the attainment of equilibrium in control experiments. Moreover, CHAPS (6 mM) was shown to have no effect on the NSB of the six neutral basic validation set drugs: felodipine, isradipine, loratidine, midazolam, nifedipine, and pazopanib. Later experiments further demonstrated that CHAPS (6 mM) is similarly without effect on the NSB of erlotinib and the partially charged dabrafenib (Table 1). CHAPS concentrations below 6 mM (viz. 2 and 4 mM) appeared to give incomplete solubilization of the more lipophilic TKIs, whereas fu_mic values of representative TKIs and “control” drugs (felodipine, lapatinib, pazopanib, regorafenib, and sorafenib) were shown to increase as the CHAPS concentration was increased to 10 mM (data not shown). The latter observation appears to be consistent with the known effects of CHAPS on membrane structure. At low concentrations, detergent molecules tend only bind to the surface of the membrane, but as the CHAPS concentration increases, the membrane bilayer becomes disrupted (Kalipatnapu and Chattopadhyay, 2005). Such a model further explains why CHAPS disrupts the binding of the fully charged amitriptyline and chlorpromazine to HLM.

In summary, conventional equilibrium dialysis performed in the presence of CHAPS permits measurement of the NSB of neutral TKIs to HLM. NSB data were also generated for partially charged TKIs, which appeared consistent with the physicochemical properties of these compounds (and confirmed in the case of dabrafenib). It is emphasized that the method has been developed specifically for uncharged and weakly charged lipophilic organic bases, and it is unlikely to be applicable to other chemical classes. Moreover, experiments were performed at a single HLM concentration (0.25 mg/ml) and further method optimization will be required for other HLM concentrations. Since detergents are known to affect cytochrome P450 and UDP-glucuronosyltransferase enzyme activities in a concentration-dependent and variable manner, the addition of CHAPS to microsomal incubations is generally not recommended. However, as noted earlier, the solubility of relatively low concentrations of TKIs in incubations of HLM containing DMSO appears to be acceptable. NSB has generally been ignored in previous in vitro kinetic and inhibition studies of TKIs. The substantial microsomal binding of many of the TKIs investigated here indicates that the K_m and K_i values will be significantly overestimated if NSB is not accounted for, which in turn impacts the accuracy of in vivo intrinsic clearance values and the drug-drug interaction potential predicted using in vitro–in vivo extrapolation approaches.