Comparison of In Vitro Assays in Selecting Radiotracers for In Vivo P-Glycoprotein PET Imaging

Positron emission tomography (PET) imaging of P-glycoprotein (P-gp) in the blood-brain barrier can be important in neurological diseases where P-gp is affected, such as Alzheimer´s disease. Radiotracers used in the imaging studies are present at very small, nanomolar, concentration, whereas in vitro assays where these tracers are characterized, are usually performed at micromolar concentration, causing often discrepant in vivo and in vitro data. We had in vivo rodent PET data of [11C]verapamil, (R)-N-[18F]fluoroethylverapamil, (R)-O-[18F]fluoroethyl-norverapamil, [18F]MC225 and [18F]MC224 and we included also two new molecules [18F]MC198 and [18F]KE64 in this study. To improve the predictive value of in vitro assays, we labeled all the tracers with tritium and performed bidirectional substrate transport assay in MDCKII-MDR1 cells at three different concentrations (0.01, 1 and 50 µM) and also inhibition assay with P-gp inhibitors. As a comparison, we used non-radioactive molecules in transport assay in Caco-2 cells at a concentration of 10 µM and in calcein-AM inhibition assay in MDCKII-MDR1 cells. All the P-gp substrates were transported dose-dependently. At the highest concentration (50 µM), P-gp was saturated in a similar way as after treatment with P-gp inhibitors. Best in vivo correlation was obtained with the bidirectional transport assay at a concentration of 0.01 µM. One micromolar concentration in a transport assay or calcein-AM assay alone is not sufficient for correct in vivo prediction of substrate P-gp PET ligands.


Introduction
P-glycoprotein (P-gp, ABCB1, MDR1) is an efflux transporter expressed with varying abundance in different tissues of the body, which affects the absorption, distribution, metabolism and elimination (ADME) of many drugs. P-gp is involved in liver canalicular biliary excretion, kidney apical renal secretion, intestinal luminal secretion and blood-brain barrier (BBB) efflux [1]. At the BBB, P-gp is of regional P-gp expression (as opposed to P-gp function), a PET tracer is required which binds tightly to P-gp, instead of being transported. Several attempts were made to radiolabel compounds that showed such inhibitory binding in vitro, but the PET tracers based on these compounds, such as [ 11 C]tariquidar and [ 11 C]elacridar, still showed substrate behavior in vivo [16]. In vitro assays usually are performed in the micromolar concentration range, whereas in vivo PET studies are carried out at nanomolar concentrations. A possible explanation is that the mechanism of interaction of the tested compounds with the P-gp protein is concentration dependent.
The purpose of the present study was to investigate the usefulness of in vitro assays for predicting the in vivo behavior of PET tracers, with a focus on the concentration dependency of ligand behavior. Previously, several fluorine-18 labeled PET ligands for in vivo measurements of P-gp function at the BBB have been developed in our labs. Fluorine-18 tracers have the advantage over carbon-11 due to a longer half-life (110 min vs. 20 min) that they can be transported to imaging facilities without an on-site cyclotron. We had existing in vivo data in mice and/or in rats for the standard substrate (R)-[ 11 C]verapamil [17,18] and its derivatives (R)-N-[ 18 [19][20][21][22]. In this study, we included also two new fluorine-18 labeled isoquinoline derivatives, [ 18   Originally, MC224 and MC225 were selected from large libraries, based on bi-directional transport data in Caco-2 cells and calcein-AM inhibition data in MDCKII-MDR1 cells, which had been acquired with non-radioactive molecules. In this study, by labeling all the molecules with tritium ( 3 H, half-life 12.3 years), we could have a measurable signal even in a very low test compound concentration. Bi-directional transport assays were performed with tritiated molecules in MDCKII-MDR1 cells at three different concentrations (0.01, 1 and 50 µM). The lowest concentration (0.01 µM) best reflects the conditions seen during in vivo PET studies, whereas the higher concentrations are generally used in in vitro experiments. This set-up made it possible to evaluate whether measured Originally, MC224 and MC225 were selected from large libraries, based on bi-directional transport data in Caco-2 cells and calcein-AM inhibition data in MDCKII-MDR1 cells, which had been acquired with non-radioactive molecules. In this study, by labeling all the molecules with tritium ( 3 H, half-life 12.3 years), we could have a measurable signal even in a very low test compound concentration. Bi-directional transport assays were performed with tritiated molecules in MDCKII-MDR1 cells at three different concentrations (0.01, 1 and 50 µM). The lowest concentration (0.01 µM) best reflects the conditions seen during in vivo PET studies, whereas the higher concentrations are generally used in in vitro experiments. This set-up made it possible to evaluate whether measured ER values decreased with increasing concentration of the test compound. If so, this would indicate dose-dependent substrate transport and saturation of P-gp. In parallel experiments, cells were treated with the P-gp inhibitors ketoconazole and tariquidar (TQD) for further study of the substrate behavior of the test compounds [23,24]. The standard substrate [ 3 H] digoxin was added as a control in this experiment [25]. In addition, MC224 was used as a negative control in all the transport experiments, as it did not show substrate behavior in vivo. Finally, the acquired in vitro data was compared with the in vivo data.

Organic Synthesis
As depicted in Scheme 1, compound 2 was prepared by reductive amination using 4-(2-aminoethyl)-2-methoxyphenol and aldehyde 1, which was synthesized in a stereoselective manner as described before [19]. ER values decreased with increasing concentration of the test compound. If so, this would indicate dose-dependent substrate transport and saturation of P-gp. In parallel experiments, cells were treated with the P-gp inhibitors ketoconazole and tariquidar (TQD) for further study of the substrate behavior of the test compounds [23,24]. The standard substrate [ 3 H] digoxin was added as a control in this experiment [25]. In addition, MC224 was used as a negative control in all the transport experiments, as it did not show substrate behavior in vivo. Finally, the acquired in vitro data was compared with the in vivo data.

Organic Synthesis
As depicted in Scheme 1, compound 2 was prepared by reductive amination using 4-(2-aminoethyl)-2-methoxyphenol and aldehyde 1, which was synthesized in a stereoselective manner as described before [19]. Compound 7a and MC224 were synthesized as reported in Scheme 2. Carboxylic acid 3 was transformed into the corresponding acyl chloride 4 that was reacted with hydroxyisoquinoline to obtain amide 5a. Final amine 7a was obtained by reduction of amide 5a, followed by fluoroethylation of 6a. Synthesis of compounds 5b-6b and MC224 has been reported previously and spectral data of these compounds were identical to those described previously [20]. Compounds 13a, MC225, 14a and MC198 were prepared as reported in Scheme 3. Synthesis of compounds 9, 10, 11b, 12b and MC225 has been reported previously and spectral data of these compounds were identical to those reported [20,26]. Compounds 11a,b were obtained by alkylation of chloride derivative 10 with isoquinoline. Compound 13a was obtained by hydrogenation of 11a and consequent fluoroethylation of compound 12a. Compounds 14a and MC198 were obtained by direct fluoroethylation of compounds 11a,b. Scheme 1. Synthesis of 2. Reagents: (i) NaBH(OAc) 3 , Na 2 SO 4 , MeOH.
Compound 7a and MC224 were synthesized as reported in Scheme 2. Carboxylic acid 3 was transformed into the corresponding acyl chloride 4 that was reacted with hydroxyisoquinoline to obtain amide 5a. Final amine 7a was obtained by reduction of amide 5a, followed by fluoroethylation of 6a. Synthesis of compounds 5b-6b and MC224 has been reported previously and spectral data of these compounds were identical to those described previously [20].  [23,24]. The standard substrate [ 3 H] digoxin was added as a control in this experiment [25]. In addition, MC224 was used as a negative control in all the transport experiments, as it did not show substrate behavior in vivo. Finally, the acquired in vitro data was compared with the in vivo data.
Compound 7a and MC224 were synthesized as reported in Scheme 2. Carboxylic acid 3 was transformed into the corresponding acyl chloride 4 that was reacted with hydroxyisoquinoline to obtain amide 5a. Final amine 7a was obtained by reduction of amide 5a, followed by fluoroethylation of 6a. Synthesis of compounds 5b-6b and MC224 has been reported previously and spectral data of these compounds were identical to those described previously [20]. Synthesis of compounds 16, 21a and KE64 is depicted in Scheme 4. Dihydroxybiphenyl was reacted with methyl-4-chlorobutanoate to obtain ester 17. Compound 18 was prepared by fluoroethylation of ester 17 and the ester function was then reducted into alcohol yielding compound 19. Compound 19 was mesylated (compound 20) and then condensed with 7-methoxy-1,2,3,4tetrahydroisoquinolin-6-ol or 6,7-dimethoxy-1,2,3,4-tetrahydroisoquinoline to obtain final compounds 21a and KE64, respectively. Synthesis of compounds 15 and 16 has been previously reported and spectral data of these compounds were identical to those reported previously [27].  Synthesis of compounds 16, 21a and KE64 is depicted in Scheme 4. Dihydroxybiphenyl was reacted with methyl-4-chlorobutanoate to obtain ester 17. Compound 18 was prepared by fluoroethylation of ester 17 and the ester function was then reducted into alcohol yielding compound 19.
Compound 19 was mesylated (compound 20) and then condensed with 7-methoxy-1,2,3,4-tetrahydroisoquinolin-6-ol or 6,7-dimethoxy-1,2,3,4-tetrahydroisoquinoline to obtain final compounds 21a and KE64, respectively. Synthesis of compounds 15 and 16 has been previously reported and spectral data of these compounds were identical to those reported previously [27]. Synthesis of compounds 16, 21a and KE64 is depicted in Scheme 4. Dihydroxybiphenyl was reacted with methyl-4-chlorobutanoate to obtain ester 17. Compound 18 was prepared by fluoroethylation of ester 17 and the ester function was then reducted into alcohol yielding compound 19. Compound 19 was mesylated (compound 20) and then condensed with 7-methoxy-1,2,3,4tetrahydroisoquinolin-6-ol or 6,7-dimethoxy-1,2,3,4-tetrahydroisoquinoline to obtain final compounds 21a and KE64, respectively. Synthesis of compounds 15 and 16 has been previously reported and spectral data of these compounds were identical to those reported previously [27].   , because, under basic reaction conditions, which are required for the labeling, the double bond of precursor 11b can isomerize from the alkyl chain into the tetrahydronaphthalene ring. When the amount of NaH was increased, the amount of isomerization also increased. On the other hand, when the amount of NaH was reduced, the radiochemical yield remained low. An amount of 2 mg of NaH for 2 mg of precursor 11b was found to yield a sufficient amount of [ 18 F]MC198 (2.5% radiochemical yield, calculated from end of bombardment of 18 F − ) to conduct the in vivo studies. Use of tetrabutylammonium hydroxide (TBAOH) solution (40% in water, 1 µL TBAOH and 0.5 mg of precursor) instead of NaH was also attempted, but no product was formed at all. Radiosynthesis was also attempted by starting with ethylene di(p-toluenesulfonate), which was first reacted with dried fluoride complex. However, the formed [ 18 F]fluoroethyl tosylate was difficult to purify and using it in unpurified form could lead to side products together with the precursor and unreacted ethylene ditosylate. A Sep-Pak Silica Plus cartridge, a Sep-Pak C-18 cartridge and HPLC were used for purification, but all methods were time consuming and HPLC and silica Sep-Pak produced the intermediate in large volume. In addition, the separation in the Sep-Paks was poor. Radiosynthesis progressed more easily and faster, and produced higher yields when the distillation employing bromoethyl tosylate as precursor was used. Using 0.1 M NaOAc/MeCN 6:4 (v/v) as an HPLC eluent, the product and the isomer could be partially separated, but baseline separation was , because, under basic reaction conditions, which are required for the labeling, the double bond of precursor 11b can isomerize from the alkyl chain into the tetrahydronaphthalene ring. When the amount of NaH was increased, the amount of isomerization also increased. On the other hand, when the amount of NaH was reduced, the radiochemical yield remained low. An amount of 2 mg of NaH for 2 mg of precursor 11b was found to yield a sufficient amount of [ 18 F]MC198 (2.5% radiochemical yield, calculated from end of bombardment of 18 F − ) to conduct the in vivo studies. , because, under basic reaction conditions, which are required for the labeling, the double bond of precursor 11b can isomerize from the alkyl chain into the tetrahydronaphthalene ring. When the amount of NaH was increased, the amount of isomerization also increased. On the other hand, when the amount of NaH was reduced, the radiochemical yield remained low. An amount of 2 mg of NaH for 2 mg of precursor 11b was found to yield a sufficient amount of [ 18 F]MC198 (2.5% radiochemical yield, calculated from end of bombardment of 18 F − ) to conduct the in vivo studies. Use of tetrabutylammonium hydroxide (TBAOH) solution (40% in water, 1 µL TBAOH and 0.5 mg of precursor) instead of NaH was also attempted, but no product was formed at all. Radiosynthesis was also attempted by starting with ethylene di(p-toluenesulfonate), which was first reacted with dried fluoride complex. However, the formed [ 18 F]fluoroethyl tosylate was difficult to purify and using it in unpurified form could lead to side products together with the precursor and unreacted ethylene ditosylate. A Sep-Pak Silica Plus cartridge, a Sep-Pak C-18 cartridge and HPLC were used for purification, but all methods were time consuming and HPLC and silica Sep-Pak produced the intermediate in large volume. In addition, the separation in the Sep-Paks was poor. Radiosynthesis progressed more easily and faster, and produced higher yields when the distillation employing bromoethyl tosylate as precursor was used. Using 0.1 M NaOAc/MeCN 6:4 (v/v) as an HPLC eluent, the product and the isomer could be partially separated, but baseline separation was Use of tetrabutylammonium hydroxide (TBAOH) solution (40% in water, 1 µL TBAOH and 0.5 mg of precursor) instead of NaH was also attempted, but no product was formed at all. Radiosynthesis was also attempted by starting with ethylene di(p-toluenesulfonate), which was first reacted with dried fluoride complex. However, the formed [ 18 F]fluoroethyl tosylate was difficult to purify and using it in unpurified form could lead to side products together with the precursor and unreacted ethylene ditosylate. A Sep-Pak Silica Plus cartridge, a Sep-Pak C-18 cartridge and HPLC were used for purification, but all methods were time consuming and HPLC and silica Sep-Pak produced the intermediate in large volume. In addition, the separation in the Sep-Paks was poor. Radiosynthesis progressed more easily and faster, and produced higher yields when the distillation employing bromoethyl tosylate as precursor was used. Using 0.1 M NaOAc/MeCN 6:4 (v/v) as an HPLC eluent, the product and the isomer could be partially separated, but baseline separation was not reached with this or with other eluents tested. Identification was based on the retention time of non-radioactive MC198 in HPLC. [ 18 F]MC198 was produced in 83 min with >95% radiochemical purity and >100 GBq/µmol molar radioactivity. Log D 7.4 was measured as 2.3 according to a method reported earlier [20]. A distillation method was used without any difficulties also for [ 18 F]KE64. It was generated with 6.7 ± 0.5% radiochemical yield, >97% radiochemical purity and 32 ± 17 GBq/µmol molar radioactivity in 72 min. Log D 7.4 was determined to be 3.2.
In addition, the bidirectional transport of the non-radioactive molecules was evaluated in Caco-2 cells (Figure 4a). All of the compounds showed very similar results, with ER values ≥2, indicating affinity for P-gp. Remarkably, MC224 showed lower A → B and B → A values than other molecules, but still the highest ER, indicating the strongest P-gp substrate potential, although this was not observed in the in vivo data or in vitro with the tritiated molecule. Bi-directional transport of the tested compounds across monolayers of Caco-2 and MDCKII-MDR1 cells was very different. ER values obtained in Caco-2 cells did not correlate with ER of MDCKII cells at test compound cells (mean ± SEM) using baseline concentration of 0.01 µM (n = 6) with co-incubation of the P-gp inhibitor ketoconazole at 25 µM (0.01 µM + keto, n = 3) and tariquidar at 10 µM (0.01 µM + TQD, n = 3). * Symbol represents a significant difference with 0.01 µM (* p < 0.05, ** p < 0.01, *** p < 0.001) and # symbol a significant difference with 0.01 µM + keto (# p < 0.05, ## p < 0.01, ### p < 0.001).
A similar behavior as with [ 3 H]verapamil was observed also for [ 3 H]N-FeVer. The only difference was that a significant change in B → A transport compared to 0.01 µM was observed only at 1 µM.
[ 3 H]O-FeVer also behaved like a P-gp substrate, with low A → B P app values (<2 × 10 −6 cm/s) at 0.01 µM and 1 µM, but this increased significantly at 50 µM and in the presence of inhibitors. The big difference in A → B and B → A resulted in a very high ER (>30) at 0.01 µM and 1 µM, and was still higher than 2 at 50 µM or after inhibition with ketoconazole.
In addition, the bidirectional transport of the non-radioactive molecules was evaluated in Caco-2 cells (Figure 4a). All of the compounds showed very similar results, with ER values ≥2, indicating affinity for P-gp. Remarkably, MC224 showed lower A → B and B → A values than other molecules, but still the highest ER, indicating the strongest P-gp substrate potential, although this was not observed in the in vivo data or in vitro with the tritiated molecule. Bi-directional transport of the tested compounds across monolayers of Caco-2 and MDCKII-MDR1 cells was very different. ER values obtained in Caco-2 cells did not correlate with ER of MDCKII cells at test compound concentrations of 0.01 and 1 µM. A poor correlation (r 2 = 0.35) was observed between data acquired at 50 µM in MDCKII cells and at 10 µM in Caco-2 cells (Figure 4b).  A calcein-AM inhibition experiment was performed in MDCKII-MDR1 cells with nonradioactive test compounds. The lowest EC50 value (0.014 µM) was obtained with KE64 and the highest with O-FeVer (2.4 µM, Table 1), indicating that KE64 has the greatest P-gp inhibition potential.  A calcein-AM inhibition experiment was performed in MDCKII-MDR1 cells with non-radioactive test compounds. The lowest EC 50 value (0.014 µM) was obtained with KE64 and the highest with O-FeVer (2.4 µM, Table 1), indicating that KE64 has the greatest P-gp inhibition potential. Definition as substrate (+) in vitro was ER > 2 and in vivo brain-to-plasma or AUC ratio > 2 in at least one strain.

In Vivo Evaluation of [ 18 F]MC198 and [ 18 F]KE64
Radiotracer uptake in the brain was evaluated in wild-type and Mdr1a/b (−/−) Bcrp1 (−/−) knockout mice with microPET imaging. Both [ 18 F]MC198 (control n = 5, knockout n = 4) and [ 18 F]KE64 (control n = 5, knockout n = 6) had higher brain uptake in the knockout mice than in wild-type, indicating that they are substrates for P-gp and/or Bcrp in vivo (Figure 5a,b). The difference in whole brain time-activity curves between strains, expressed in standardized uptake value (SUV), was 1.7-fold (P < 0.001, comparing areas under the curve (AUC) from 0 to 30 min) for [ 18 F]MC198 and 1.6-fold (P = 0.072) for [ 18 F]KE64. The biodistribution profile of both compounds (at 45 min post-injection (p.i.)) was similar, with the highest uptake in liver, kidney, pancreas and small intestine (Figure 5c,d). Radioactive metabolites in plasma and brain at 45 min p.i. were analyzed using radio-TLC. Both compounds showed significant and rather rapid metabolism, the fraction of intact parent compound in the plasma at 45 min was only 34% and 26% for [ 18 F]MC198 and [ 18 F]KE64, respectively. In the brain, more metabolites were detected for [ 18 F]MC198 than for [ 18 F]KE64, i.e., 11 versus 5% metabolites of total radioactivity, respectively.

In Vitro-In Vivo Correlation
In vivo evaluation was performed in control and P-gp knockout mice and for some compounds also in rats which were treated with P-gp inhibitors (Table 1) [19][20][21]. Experiments were performed in different institutions where different animal models were available and some compounds were involved in more experiments than others. The in vivo P-gp substrate potential was defined by difference in brain uptake between knockout and control animals. For radiolabeled compounds, the brain uptake can be determined either by PET imaging (time-activity curves) or by biodistribution studies (brain-to-plasma ratios). Brain-to-plasma ratios (Table 1) in knockout mice divided by brain-to-plasma ratios in control mice correlated nicely with in vitro ER values in MDCKII-MDR1 cells at both test compound concentrations of 0.01 µM (r 2 = 0.89, Figure 6a) and 1 µM (r 2 = 0.83, data not shown,). AUC ratios of whole brain SUV time-activity curves also correlated with ER values at these same concentrations. Correlation was slightly better at a concentration of 0.01 µM (r 2 = 0.79, Figure 6b) than at 1 µM (r 2 = 0.72, data not shown).

Discussion
This study investigated how accurately standard in vitro assays, i.e., bidirectional monolayer transport and calcein-AM assays, can predict in vivo behavior of potential P-gp substrates. For this purpose, seven different P-gp PET radiotracers were used: (R)-[ 11 KE64. In vivo evaluation was performed in transporter knockout and control mice and/or in rats treated with the P-gp inhibitor tariquidar. Initial in vitro characterization was performed using non-radioactive compounds in combination with both the calcein-AM assay in MDCKII-MDR1 cells and the bidirectional transport assay in Caco-2 cells. The main finding was that in vivo behavior was not the same for all compounds as that observed in vitro for their non-radioactive analogs. Therefore, an attempt was made to improve the predictive value of the in vitro data by identifying more appropriate assays, focusing on the use of a

Discussion
This study investigated how accurately standard in vitro assays, i.e., bidirectional monolayer transport and calcein-AM assays, can predict in vivo behavior of potential P-gp substrates. For this purpose, seven different P-gp PET radiotracers were used: (R)-[ 11 KE64. In vivo evaluation was performed in transporter knockout and control mice and/or in rats treated with the P-gp inhibitor tariquidar. Initial in vitro characterization was performed using non-radioactive compounds in combination with both the calcein-AM assay in MDCKII-MDR1 cells and the bidirectional transport assay in Caco-2 cells. The main finding was that in vivo behavior was not the same for all compounds as that observed in vitro for their non-radioactive analogs. Therefore, an attempt was made to improve the predictive value of the in vitro data by identifying more appropriate assays, focusing on the use of a bidirectional transport assay with tritium labeled compounds in MDCKII-MDR1 cells.
Due to differences between the Caco-2 and MDCK-MDR1 cell lines, the P-gp mediated transport of compounds may differ, as was observed for example with higher efflux of [ 3 H]digoxin in MDCK-MDR1 cells than in Caco-2 cells [6]. The transport assay in Caco-2 cells, performed only at a relatively high test compound concentration (10 µM), did not sufficiently assess the differences between test compounds and incorrectly classified MC224 as a P-gp substrate (Figure 4).

Verapamil, N-FeVer and O-FeVer Behavior
Verapamil and its derivatives N-FeVer and O-FeVer showed comparable results in the transport experiments. Passive diffusion (A → B transport) increased with increasing concentrations of the compound and after treatment with inhibitors, which is an indirect consequence of the inhibition of P-gp. More compound is able to pass through the cell, without interference of P-gp, since it is not directly transported back to the apical side, resulting in a higher A → B transport. The increased A → B transport caused a decrease in the ER (Figures 2 and 3). In the concentration assay, O-FeVer transport was affected substantially only at a concentration of 50 µM and in the Calcein-AM measurement it had the highest EC 50 value (2.4 µM. Table 1). P-gp efflux, the B → A transport, and ER values were the highest of all the tested compounds at concentrations of 0.01 and 1 µM. This means that O-Fever was the strongest substrate and weakest inhibitor of P-gp in vitro. Only at 50 µM a decrease in ER was found, which implies that, for a full assessment of the concentration dependent behavior, more concentrations between 1 and 50 µM would be needed for this compound. Compared with N-FeVer and verapamil, low A → B transport (<2 × 10 −6 cm/s) of O-FeVer at 0.01 and 1 µM indicates low passive diffusion and/or high P-gp affinity, which could be explained by the low log D 7.4 of 1.6. This effect is also observed in vivo with slowly increasing brain uptake and absence of an initial peak of activity [19]. However, verapamil was still the strongest substrate in vivo as it had the highest brain-to-plasma and AUC ratios, both in mice and rats (Table 1). N-FeVer actually showed a rather low AUC ratio in mice (1.2), but this was higher in rats (3.7), indicating species differences. With O-FeVer, it was the other way around. Both N-FeVer and O-FeVer showed rapid in vivo metabolism. After 5 min, only 46 and 20% of the total activity in plasma represented parent Nand O-FeVer, respectively, whilst for verapamil it was 88%. However, the formed metabolites are not expected to interact with P-gp, since they are not fluorine-18 analogs of the known [ 11 C]verapamil metabolite and P-gp substrate [ 11 C]D617 [19,29].

MC224, MC225, MC198 and KE64 Behavior
Compounds MC224, MC225, MC198 and KE64 are all structural analogs. The difference between MC225 and MC198 is only a double bond in MC198 in the middle of the molecule versus a single bond in MC225 (Figure 1). The transport assay in MDCKII-MDR1 cells yielded very similar ER values for both compounds, although A → B and B → A values were a bit higher for MC225 (Figures 2e,f and 3e,f). The calcein-AM assay identified MC198 as a slightly more potent P-gp inhibitor (Table 1). However, MC225 had higher knockout/control brain-to-blood and AUC ratios in mouse brain in vivo than MC198, it was metabolically more stable and showed fewer radioactive metabolites in the brain [20]. MC198 was also difficult to label due to isomerization of the double bond. Thus, MC225 had better characteristics for in vivo PET imaging than MC198, although the in vitro assays predicted them to be fairly equal.
MC224 has a shorter alkyl chain, with only one carbon atom separating the isoquinoline and biphenyl structures, than KE64 which has a 4-membered alkyl chain connected with an ether bond (Figure 1). In case of P-gp substrates, the in vitro ER values are expected to decrease after inhibition of P-gp. This was not observed for either MC224 or KE64. There was also no clear concentration dependent transport. The calcein-AM assay yielded EC 50 values for MC224, which were in the same range as for MC225 and MC198 (Table 1). KE64, on the other hand, had the lowest EC 50 value of all tested compounds (0.014 µM), suggesting the best inhibitory properties. MC224 was the only compound showing negative results both in vitro and in vivo. KE64, on the other hand, was the only compound out of the seven tested molecules that had conflicting in vitro and in vivo results. It was not a clear substrate in vitro, but at tracer concentrations, higher brain uptake (1.6-fold) was observed in Mdr1a/b (−/−) Bcrp1 (−/−) compared with wild-type mice, indicating that KE64 is a moderate P-gp substrate in vivo. However, the in vitro transport results of KE64 may not be very reliable due to low P app values obtained and extensive binding to cellular components and consequently low recovery values.

Calcein-AM Assay
A calcein-AM experiment alone would not have been sufficient to classify the molecules as P-gp substrates or inhibitors, as it can falsely indicate affinity for P-gp, especially in case of compounds with low P app values, such as for MC224 [11]. However, if the calcein-AM assay is performed in combination with a transport experiment at different concentrations of the test compound, the calcein-AM assay can give valuable additional information. Solely based on the calcein-AM experiment, KE64 would have been classified as a P-gp inhibitor. As no in vivo inhibition experiments with increasing concentrations of KE64 or any other compounds were performed, it is difficult to predict whether they could be used as inhibitors at high dose. In vitro almost all molecules showed a decrease in ER at high concentrations, indicating saturation of transport. When the concentration is high enough, P-gp will be saturated and efflux is decreased. At least 3 different concentrations should be used to see a trend in the ER values. An advantage of the calcein-AM assay is that it can be automated, as it produces a readout (fluorescence) that is suitable for high throughput screening. The transport assay is more laborious. However, tritiated molecules are very easy to use in this assay, as their concentration can be determined by liquid scintillation counting, which increases throughput. Due to the high sensitivity of this technique, it is possible to use very low concentrations of the tritiated molecules in the assays. In addition, calculation of the recovery of radioactivity after the experiment provides an indication of the extent of nonspecific binding. On the other hand, the synthesis of precursor molecules for tritium labeling and the development of a labeling method resulting in sufficient radiochemical purity and yield of the labeled product may be time consuming. Concentrations of the non-radioactive compounds usually are measured with liquid chromatography (LC) combined with mass spectrometry (MS), which requires the development of an analytical method and measurement of a standard curve. Of course, both transport and calcein-AM assays predict only interaction with P-gp, but they do not take into account other factors important for PET tracers such as BBB penetration and metabolism.

In Vitro and In Vivo Correlation
The in vitro transport assay in MDCKII-MDRI cells, in combination with an ER cutoff value of 2 at a test drug concentration of 0.01 µM, predicted substrate behavior in vivo. In vitro and in vivo ER valueswere nicely correlated (Figure 6), even though the cell line is transfected with a human P-gp gene and the in vivo data are obtained in rodents. Similar results have also been reported by Adachi et al. in Caco-2 and LLC-MDR1 cells [30]. Feng et al. reported a good correlation of ER values determined in human and mouse mdr-transfected MDCK cell lines [10], but Yamazaki et al. found that results obtained in mice were better correlated with apparent permeability determined in a mouse than in a human cell line [31]. It is unclear whether these results reflect true species differences or solely the absence of mdr1b in the cells, since the mouse P-gp isoforms mdr1a and mdr1b are known to possess distinct functional characteristics [32].
Deciding on a clear cutoff value for P-gp substrates was more difficult in vivo than in vitro, as only a small number of compounds was included in the present study and differences in brain-to-plasma and AUC ratios were observed not only between different compounds, but also between animal strains and species. All compounds had higher brain-to-plasma than AUC ratios. Brain-to-plasma values were calculated only at one time point, whereas AUC represents the whole time frame of the PET scan. In this study, a cutoff value of 2 was used for both brain-to-plasma and AUC ratios to identify P-gp substrates. N-FeVer and MC225 had AUC ratios lower than this in Mdr1a/b (−/−) mice, but since the brain-to-plasma ratios and AUC ratios in rats and in Mdr1a/b (−/−) Bcrp1 (−/−) mice were higher than 2, they can still be considered as substrates. MC198 and KE64 can be considered as weak substrates.

Chemicals
All reagents and solvents were obtained from commercial suppliers:

General Methods
Column chromatography was performed with 1:30 (mg crude product: g silica gel) silica gel 60 Å (63-200 µm, Merck, Darmstadt, Germany) as the stationary phase. Melting points were determined in an open capillary on a Gallenkamp electrothermal apparatus (Loughborough, UK). 1 H-NMR spectra were recorded in CDCl 3 at 300 MHz with a Mercury-VX (Varian, Palo Alto, CA, USA) or at 250 MHz on an Avance (Bruker Billerica, MA, USA) spectrometer. All chemical shift values are reported in ppm (δ) relative to the solvent peak (7.27 for CHCl 3 ). Recording of mass spectra was carried out on an HP 6890-5973MSD gas chromatograph/mass spectrometer (Agilent, Santa Clara, CA, USA); only significant m/z peaks, with their percentage of relative intensity in parentheses, are reported. Electrospray ionization mass spectrometry (ESI-MS) analyses were performed on an Agilent 1100LC/MSDtrap systemVL. Electrospray ionisation-high resolution mass spectrometry (ESI-HRMS) was carried out using a Bruker microTOF-Q instrument in positive ion mode (capillary potential of 4500 V). Elemental analyses (C, H, N) were performed on an Euro EA 3000 analyzer (Eurovector, Milan, Italy); the analytical results were within ±0.4% of the theoretical molecular formula values.

Cell Culture
Madin-Darby canine kidney (MDCK) II cells expressing cDNAs encoding human MDR1 (MDCKII-MDR1 cells) were obtained by TNO from the Netherlands Cancer Institute. Transfected MDCKII cells were seeded on semi-permeable (pore size 0.4 µm) filter inserts (12-well Transwell plates 3401, Costar Corp, Cambridge, MA, USA) at approximately 4 × 10 5 cells per filter insert (growth area of 1.13 cm 2 ) and cultured according to the standard procedure for this cell line [33]. Cells were counted using an automatic cell counter (Millipore, Bedford, MA, USA). Cells on the inserts were cultured for three days in a total volume of 2.3 mL culture medium per well (0.5 mL apical and 1.8 mL basolateral) at approximately 37 • C in a humidified incubator containing approximately 95% air/5% CO 2 . The medium was refreshed after 48 h. It has been shown, that under these conditions the human transporter protein MDR1 localizes on the apical plasma membrane of the cells [34,35].

Assessment of Monolayer Integrity
After three days of cell culture on Transwell, the MDCKII-MDR1 monolayers are expected to have developed a transepithelial electrical resistance (TEER) of approximately 70 Ωcm 2 higher than background (empty filter without cells) [36]. At the start of each study, TEER was measured to assess the integrity of the monolayers of MDCKII-MDR1 cells in culture medium using the Millicell-ERS epithelial voltohmmeter (Millipore). Monolayers with a TEER value less than 55 Ωcm 2 above background (empty filter) were excluded from the transport assay. TEER values ranged from 205 to 294 Ωcm 2 , with a mean of 253 Ωcm 2 , including the background of 125 Ωcm 2 . Cell monolayers were also inspected with a microscope before starting the experiments.

Bidirectional Transport of Tritiated Compounds in MDCKII-MDR1 Cells
Transcellular transport studies were performed as described previously with minor modifications [37]. In brief, a stock solution of digoxin (0.1 mM) was prepared in DMSO. [ 3 H]digoxin was mixed with a non-radiolabeled compound to achieve a final concentration of 0.05 µM digoxin in HBSS/HEPES in the transwells, and a radioactivity concentration of 10 kBq/mL. Stock solutions of test compounds verapamil, N-FeVer, O-FeVer, MC224, MC225, MC198 and KE64 (0.01, 1 and 100 mM) were prepared in DMSO, mixed with the tritium labeled analogs and diluted with HBSS/HEPES to final concentrations of 0.01, 1 and 50 µM, with a radioactive tracer concentration of 10 kBq/mL in the transwells.
After the integrity assessment, culture medium was removed from the filter inserts (apical compartment (A)). The inserts were washed once with warm (37 • C) HBSS, transferred to new 12-well plates which were preincubated with HBSS/HEPES for 18 h at 37 • C, to reduce non-specific binding to the plastic, and washed with warm HBSS. The MDR1-mediated apical to basolateral (A → B) transport in MDCKII-MDR1 cell monolayers was determined by addition of 0.65 mL of a tritiated compound in concentrations of 0.01, 1 and 50 µM to the apical compartment in triplicate. A volume of 1.8 mL of HBSS/HEPES was added to the basolateral compartment. To determine basolateral to apical (B → A) transport, 1.95 mL of a tritiated compound at concentrations of 0.01, 1 and 50 µM was added to the basolateral compartment and 0.5 ml of HBSS/HEPES to the apical compartment. A 150 µL aliquot of the apical (A → B) or basolateral (B → A) compartment was taken at the start of the experiment to measure the initial concentration and to calculate the recovery of radioactivity. Incubations were performed on a rocker platform (rotation approximately 60 rpm) at 37 • C for 2 h in a humidified incubator containing approximately 95% air/5% CO 2 . Aliquots of the apical and basolateral samples (150 and 1600 µL for A → B and 400 and 150 µL for B → A, respectively) and samples of the initial concentration were mixed with 10 mL of liquid scintillant Ultima GoldTM (PerkinElmer Inc.). Radioactivity was determined by LSC on a Tri Carb 3100TR liquid scintillation counter using QuantaSmart TM software (PerkinElmer Inc.) in which all counts were converted to disintegrations per minute (DPM) using tSIE/AEC (transformed spectral index of external standards coupled to automatic efficiency correction). Background values were measured for each sample sequence using liquid scintillant without test samples. Recovery of radioactivity was calculated as percentage of measured total radioactivity after the experiment versus initial radioactivity.
Subsequently, the effects of MDR1-inhibitors ketoconazole and tariquidar on the bidirectional transport of the tested tritiated compounds were examined. Stock solutions of the inhibitors ketoconazole (10 mM) and tariquidar (10 mM) were prepared in DMSO. On the day of the experiment, inhibitors were added to HBSS/HEPES solutions of the tritiated test compounds in a concentration of 0.01 µM or [ 3 H]digoxin in a concentration of 0.05 µM to reach a final concentration of 25 µM ketoconazole or 10 µM tariquidar. These concentrations were based on IC 50 values that are known to block P-gp completely [16,25]. The assay was performed in an identical manner as described above, except that all apical (in A → B experiment) and basolateral (in B → A experiment) solutions contained 25 µM ketoconazole or 10 µM tariquidar from the start of the experiment, keeping the concentrations of test compounds (0.01 or [ 3 H]digoxin 0.05 µM, 10 kBq/mL) and organic solvent concentration (≤1%) constant for all samples. where dQ/dt indicates the appearance rate of a test compound at the receiver side calculated from total DPM at t = 120 min, corrected for background (DPM/s). A is the surface area of the filter insert (cm 2 ) and C 0 the measured initial concentration, calculated from total DPM on the donor side at t = 0 min (DPM/L).
To determine the ER for the test compounds, the following equation was used:

In Vitro Experiments with Non-Radioactive Molecules
The transport assay with non-radioactive compounds in Caco-2 cells was performed in 96-well plates as described previously [38]. The initial concentration of the test compounds in the acceptor well was 10 µM and the concentration in the receiver well was measured after 2 h using UV spectroscopy. Samples from one compound were pooled in order to have a sufficient volume for the UV detection.
The calcein-AM experiment was performed in MDCKII-MDR1 cells according to methods described previously [39]. Calcein cell accumulation in the absence and in the presence of test compounds (0,1, 1, 10, 30, 50 and 100 µM) was evaluated and the basal level of fluorescence was measured in untreated cells. EC 50 values were determined by fitting the fluorescence increase (in %) versus log[dose].

Animal Experiments with [ 18 F]MC198 and [ 18 F]KE64
All animal studies were in compliance with the local ethical guidelines. Protocols (DEC 6456C) were approved by the Institutional Animal Care and Use Committee of the University of Groningen. Male FVB wild type mice (33 ± 2 g, 11-12 weeks) and Mdr1a/b (−/−) Bcrp1 (−/−) constitutive knockout mice (31 ± 3 g, 11-12 weeks) developed from the FVB line were purchased from Taconic (Hudson, NY, USA). After arrival, animals were housed individually and acclimatized for at least 7 days in the Central Animal Facility of the University Medical Center Groningen before experiments. Mice had access to food and water ad libitum and were kept under a 12 h light-dark cycle. During experiments, mice were anesthetized with 2% isoflurane in medical air and kept at a constant temperature using electronically controlled heating pads.

PET Procedure and Data Analysis
Mice were injected with radiotracers (3.5 ± 1.6 MBq, 0.1 mL, [ 18 F]MC198 or [ 18 F]KE64) in the penile vein under isoflurane anesthesia. Injections were performed with the mice in the microPET camera (microPET Focus 220, Siemens Medical Solutions, Malvern, PA, USA), simultaneously starting a 30 min dynamic emission scan. After each emission scan, a transmission scan of 515 s with a 57 Co point source was performed to correct the emission data for attenuation and scatter. Mice were terminated by cervical dislocation. Several organs and tissues were excised, weighed and radioactivity was measured in a γ-counter (LKB Wallac, Turku, Finland). Radioactivity accumulation in the organs was expressed as SUV, using the following formula: [tissue activity concentration (MBq/g)]/[injected dose (MBq)/body weight (g)]. Metabolites in the terminal plasma and brain samples were determined using a radio-TLC method described elsewhere [20]. TLC-plates were eluted with ethylacetate/methanol 9:1 v/v (R f [ 18 F]MC198 = 0.45, R f [ 18 F]KE64 = 0.62). PET data were reconstructed as described previously [20]. Inveon Research Workplace software version 4.0 (Siemens, Erlangen, Germany) was used for data analysis. All frames were summed, and PET images were co-registered with an MRI template [40]. A whole brain volume of interest (VOI) based on the MRI template was generated. Radioactivity concentrations were converted to SUV values and plotted as time-activity curves (TAC).

Statistical Analysis
The statistical significance of differences between two groups was calculated by two-sided unpaired Student's t test, using IBM SPSS Statistics version 22 (Armonk, NY, USA). One-way analysis of variance (ANOVA) with Bonferroni correction was used to assess differences between three or more groups. A p value of less than 0.05 was considered statistically significant.

Conclusions
In vitro prediction of in vivo results is always challenging, but this is especially the case for P-gp ligands, as their behavior appears to be concentration dependent. Neither the bidirectional transport assay in Caco-2 cells with the non-radioactive compounds verapamil, N-FeVer, O-FeVer, MC224, MC225, MC198 and KE64 in a single concentration nor the calcein-AM inhibition assay in MDCKII-MDR1 cells was sufficiently correlated with in vivo results in rodents by PET imaging. The predictive value of the in vitro data improved when the bidirectional transport assay in MDCKII-MDR1 cells was performed with tritium-labeled compounds at three different concentrations and in combination with P-gp inhibitors. Based on the data of all experiments, the order of in vitro substrate strength was: O-FeVer > verapamil > N-FeVer > MC225~MC198 > MC224~KE64 and in vivo: verapamil > O-FeVer > MC225 > N-FeVer > MC198 > KE64 > MC224.