Quantitative Analysis of the ABCG2 c.421C>A Polymorphism Effect on In Vivo Transport Activity of Breast Cancer Resistance Protein (BCRP) Using an Intestinal Absorption Model

INTRODUCTION

Breast cancer resistance protein (BCRP), encoded by the ABCG2 gene, belongs to the ATP-binding cassette family transporters, most of which actively pump out a variety of substrates from cells by utilizing energy from ATP hydrolysis. BCRP is so-called a half-transporter with six putative transmembrane domains and one ATP binding cassette, and functions as a homodimer.1, 2 BCRP transports structurally diverse compounds, including many clinically used drugs, such as anti-cancer drugs (imatinib and irinotecan), antivirals (lamivudine and zidovudine), and statins (pitavastatin and rosuvastatin).3 BCRP is expressed in various tissues, such as the gastrointestinal tract, liver, kidney, blood–brain barrier, and placenta, thereby limiting the intestinal absorption and tissue distribution of BCRP substrates.4 Several non-synonymous single nucleotide polymorphisms (SNPs) have been identified in the ABCG2 gene.2 The c.421C>A SNP located in the ATP binding domain is associated with an amino acid change (141Q>K), and has been the most extensively investigated because of its high allele frequency with a notable ethnic difference (Asians: 29%–36%, Europeans: 4.5%–12%, and Africans: 0%–2.3%).2

Several clinical studies have investigated the effect of c.421C>A in ABCG2 on the pharmacokinetics and subsequent pharmacological/toxicological effects of BCRP substrate drugs. For example, the plasma area under the concentration–time curve (AUC) values of sulfasalazine,5 simvastatin,6 and rosuvastatin7 were more than twofold higher in subjects with the 421AA compared with those with the 421CC genotype. This polymorphism was also significantly associated with the low-density lipoprotein-cholesterol lowering effect of rosuvastatin8 and the occurrence of diarrhea caused by gefitinib.9

Several studies have investigated the decrease in the apparent transport activity of mutated BCRP. Kondo et al.10 suggested that the intrinsic transport activity normalized by BCRP protein expression level was almost equal in membrane vesicles expressing wild-type and 141Q>K BCRP, which implies that reduced surface membrane BCRP expression might be a major cause of the apparent decrease in transport activity. This hypothesis is supported by reports demonstrating that placental BCRP expression in subjects with the 421AA genotype was approximately 50% of that in subjects with the 421CC genotype,11 and that liver samples obtained from subjects with at least one mutated (c.421C>A) allele showed significantly lower BCRP protein expression than those from subjects with the wild-type allele, as determined using liquid chromatography coupled with tandem mass spectrometry.12 Other researchers suggested that intrinsic transport activity was affected by the mutation as ATPase activity was lower in mutant 141Q>K BCRP as compared with the wild-type without decreasing BCRP protein expression.13, 14 Despite these efforts, the impact of ABCG2 c.421C>A on BCRP transport activity in vivo is still undetermined.

The aim of this study was to investigate the quantitative impact of this mutation on BCRP transport activity in vivo based on clinical pharmacokinetic data obtained by pharmacogenetic studies. We previously reported an intestinal model to evaluate the effects of gut metabolism and efflux transport on the overall absorption of a drug from the lumen to the portal blood.15 Assuming that the pharmacokinetic alteration caused by c.421C>A is accounted for by intestinal BCRP, this model was selected for this study. In this model, F_aF_g was described as the following equation:

(1)

where PS_inf, PS_eff, CL_ab, CL_m, and LF represent the influx clearance from the lumen to epithelial cells, apical efflux clearance from cells to lumen, basolateral efflux clearance from cells to blood, metabolic clearance inside the cells, and luminal flow rate, respectively. The ratio of F_aF_g in subjects with the 421AA genotype to that in 421CC subjects (F_aF_g ratio_421AA/CC) for each BCRP substrate was calculated assuming that the ratio of the transport activity of the mutated BCRP to that of the wild-type BCRP (R_act,421C>A) was constant irrespective of the drug. This assumption would be appropriate as previous in vitro studies suggested that the ATP-dependent transport activities of four BCRP substrates by mutated 141Q>K BCRP were almost equal to those by wild-type BCRP when normalized by protein expression level of BCRP10 and that cell surface expression of mutated BCRP was markedly lower than wild-type BCRP in in vitro expression system, although total expression amount of BCRP protein in biopsy samples of human small intestine was not affected by ABCG2 c.421C>A genotype.16 Next, R_act,421C>A was optimized by comparing F_aF_g ratio_421AA/CC with the ratio of clinically observed AUCs in subjects with the 421AA and 421CC genotypes. By this approach, we tried to estimate the decrease in the function of mutated BCRP in the intestine.

METHODS

Model for Intestinal Drug Absorption

The intestinal drug absorption model, incorporating the active and passive membrane transport processes from lumen to epithelial cells, from epithelial cells to lumen, from epithelial cells to portal blood, and metabolism inside epithelial cells was described as Eq. 1. PS_inf, PS_eff, CL_ab, and CL_m can be further described as follows:

(2)

(3)

(4)

(5)

where P_inf, P_dif, P_BCRP, P_others, and P_b,eff represent the permeability coefficients of influx transport, passive diffusion, BCRP-mediated active transport, active efflux transport mediated by transporters other than BCRP, and basolateral active efflux via transporters, respectively. S, f, A_R, and CL_met are the basal surface area, protein unbound fraction of drugs in cells, apical/basolateral area ratio (A_R = 20),17 and metabolic intrinsic clearance inside cells, respectively. Finally, using Eqs. 2–5, Eq. 1 can be converted to Eq. 6.

(6)

To simplify this equation, it was assumed that intestinal metabolism does not occur, and influx and efflux transporters other than BCRP are not involved in the overall intestinal absorption of tested drugs from the lumen to the portal vein (CL_met = 0, P_inf = 0, P_others = 0, and P_b,eff = 0). Under these assumptions, the F_aF_g value can be described as the following equation:

(7)

In Silico Prediction of P_dif

The permeability coefficient of passive diffusion (P_dif) was predicted in silico using Eq. 8 reported by Winiwarter et al.18

(8)

where PSA, log D_5.5, and HBD represent the polar surface area (Unit: Ų), octanol/water distribution coefficient at pH 5.5, and the number of hydrogen bond donor atoms, respectively. PSA and HBD were calculated with the Discovery Studio 3.1.1 software (Accelrys, San Diego, California) and log D_5.5 was calculated with the Pallas 3.0 software (CompuDrug, South San Francisco, California).

Determination of LF/S in Intestinal Drug Absorption Model

The F_aF_g values of 25 reference drugs were collected from Kadono et al.19 These drugs were thought to be mainly absorbed through passive diffusion across the gastrointestinal membrane with minimal involvement of intestinal metabolism and active uptake/efflux transport20 (CL_met ∼ 0, P_inf ∼ 0, P_others ∼ 0, P_b,eff ∼ 0, and P_BCRP ∼ 0). Then, for the reference drugs, Eq. 7 can be transformed as follows:

(9)

LF/S was optimized using Eq. 9 with the predicted P_dif and clinically observed F_aF_g values for 25 reference drugs with the non-linear least-squares method using the computer program “Napp” (version 2.31).21 The predicted P_dif and F_aF_g values of reference drugs are summarized in Supplementary Table S1.

Calculation of In Vivo F_aF_g,421CC Values

The CL_tot, renal clearance (CL_r), and blood-to-plasma concentration ratio (R_B) of 16 drugs were obtained from the previous studies. CL_r was calculated using the following equation:

(10)

where f_e is the fraction of unchanged drug excreted into urine after intravenous administration. CL_r was regarded as 0 when f_e was less than 0.05. Assuming that CL_tot was not affected by the BCRP mutation, F_aF_g,421CC was calculated using the following equations. Eq. 12 was used when CL_oral,421CC was not reported in the literature.

(11)

(12)

(13)

(14)

(15)

where CL_h and F_h are the hepatic clearance and hepatic availability, respectively. CL_oral,421CC and F_421CC represent the oral clearance and absolute bioavailability, respectively, in subjects with the 421CC genotype. Q_h is the hepatic blood flow rate (1.24 L*h⁻¹*kg⁻¹).23

Two pharmacogenetic studies of rosuvastatin have been performed in Caucasian7 and Chinese subjects.24 The Chinese study was excluded because of the absence of intravenous administration and ethnicity-related differences in the pharmacokinetics of rosuvastatin.25

The F_aF_g,421CC values of simvastatin, sunitinib, and telatinib could not be calculated because of the lack of pharmacokinetic information after intravenous administration. The R_B values of sulfasalazine and diflomotecan were not reported. The R_B of sulfasalazine was estimated using the human protein unbound fraction in plasma (f_p) using a method reported by Uchimura et al.,26 whereas the R_B of diflomotecan could not be estimated with the same method because of the lack of human f_p data. There was no information that telmisartan was recognized by BCRP as a substrate. Thus, 11 out of 16 drugs were further analyzed in this study. The calculated P_dif, clinically observed F_aF_g,421CC for each BCRP substrate drug, and information about substrate recognition by BCRP are summarized in Table 1.

Table 1. Physicochemical Parameters and P_BCRP,421CC of BCRP Substrates

BCRP substrates	Log D5.5a	PSA (Å2)b	HBDb	Pdifc (10−4 cm/s)	PBCRP,421CCd (10−4 cm/s)	Information about Substrate Recognition by BCRP
Atorvastatin	4.8	112	3	1.6	99	27
Fluvastatin	3.5	83	2	3.1	0	27, 28
Pitavastatin	1.9	91	2	1.2	0	28
Pravastatin	1.3	124	3	0.25	0.28, 0.51	28, 29
Rosuvastatin	1.4	149	2	0.26	0.93	27, 28
Erlotinib	3.5	75	1	6.3	0	30
Gefitinib	1.9	86	2	1.4	6.6	30
Imatinib	1.4	69	1	2.9	0	31
Lamivudine	−0.57	113	2	0.25	0.14	32
Nitrofurantoin	−0.23	121	1	0.42	0	33
Sulfasalazine	2.6	150	2	0.44	14, 15	16

• HBD, number of hydrogen bond donor atoms; Log D_5.5, distribution coefficient in octanol/water at pH 5.5; P_dif, passive diffusion coefficient; PSA, polar surface area.
• ^a Calculated by Pallas 3.0 software (CompuDrug).
• ^b Calculated by Discovery Studio 3.1.1 software (Accelrys).
• ^c Calculated by the equation reported by Winiwarter et al.34: log P_dif = −2.883–0.010 PSA + 0.192 log D_5.5–0.239 HBD.
• ^d Calculated by Eq. 7.

Optimization of the Ratio of BCRP Transport Activity in Homozygous Mutation Carriers and Wild-Type Subjects (R_act,421C>A)

From 11 BCRP substrates analyzed in this study, erlotinib, gefitinib, and imatinib were excluded from the optimization of R_act,421C>A because of the lack of data from homozygous mutation carriers. Atorvastatin was also excluded because an increase of plasma AUC caused by co-administration with grapefruit juice, which is a known selective inhibitor of intestinal CYP3A4, has been reported in several clinical studies,34-36 suggesting its intestinal metabolism by CYP3A4 had to be considered in the estimation of its F_aF_g value. Consequently, 7 BCRP substrates (fluvastatin, pitavastatin, pravastatin, rosuvastatin, lamivudine, nitrofurantoin, and sulfasalazine) were analyzed for the determination of R_act,421C>A. P_BCRP for each BCRP substrate in subjects with the 421CC genotype (P_BCRP,421CC) was calculated based on Eq. 7 using the optimized LF/S value. When the calculated P_BCRP,421CC was less than 0, P_BCRP,421CC was treated as 0 for the subsequent analysis. Assuming that R_act,421C>A is constant irrespective of the BCRP substrate, P_BCRP in homozygous allele carriers (P_BCRP,421AA) can be described as the following equation:

(16)

By incorporating Eq. 16 into Eq. 7, F_aF_g,421AA can be described as Eq. 17. Finally, the F_aF_g ratio_421AA/CC can be described as Eq. 18.

(17)

(18)

Assuming that the increase of plasma AUC in subjects with the 421AA genotype is solely accounted for by the increase of F_aF_g, R_act,421C>A was optimized with the observed AUC ratios_421AA/CC and calculated F_aF_g ratios_421AA/CC of 7 BCRP substrates with a non-linear least-squares method using the computer program “Napp” (version 2.31).21

Calculation of F_aF_g Ratio_421CA/CC

The F_aF_g ratio_421CA/CC was calculated for 11 BCRP substrates. Assuming that P_BCRP in heterozygotes is the average of that in homozygotes with the wild-type and mutant alleles, P_BCRP in heterozygous carriers (P_BCRP,421CA) can be described as the following equation:

(19)

By incorporating Eq. 19 into Eq. 7, F_aF_g,421CA can be described as Eq. 20. Finally, the F_aF_g ratio_421CA/CC can be described as Eq. 21.

(20)

(21)

RESULTS

Determination of a Parameter Related to Luminal Flow Rate (LF/S) for the Intestinal Model

To quantitatively investigate the impact of c.421C>A on in vivo BCRP activity, an intestinal model was used as described in Methods section. The active transport by BCRP was included in this model, but those by other transporters and intestinal metabolism were neglected. The permeability coefficient of passive diffusion (P_dif) was calculated in silico from physicochemical properties. A parameter related to luminal flow rate (LF/S) was set by comparing F_aF_g with P_dif of 25 reference drugs predominantly absorbed by passive diffusion. The P_dif of the reference drugs was calculated from physicochemical parameters by using Eq. 8 (Supplementary Table S1 online). Then, LF/S was optimized by using Eq. 9 with 25 paired F_aF_g and P_dif values. The best-fitted curve was obtained when LF/S was set to 0.17 × 10⁻⁴ cm/s (Fig. 1).

Quantitative Impact of ABCG2 c.421C>A on In Vivo Transport Activity of Intestinal BCRP

The impact of ABCG2 c.421C>A on in vivo transport activity of intestinal BCRP was estimated using the clinical pharmacokinetic data of 7 BCRP substrates (fluvastatin, pitavastatin, pravastatin, rosuvastatin, lamivudine, nitrofurantoin, and sulfasalazine; see Methods section). Assuming that the increase in AUC is solely caused by the increase of F_aF_g owing to decreased function of mutated (c.421C>A) BCRP, the AUC ratio_421AA/CC should be equivalent to the F_aF_g ratio_421AA/CC. R_act,421C>A was optimized by using 7 sets of F_aF_g ratios_421AA/CC calculated with Eq. 18 and AUC ratios_421AA/CC from clinical data. The best correlation was obtained when R_act,421C>A was 0.23 (Fig. 2).

Calculation of F_aF_g Ratios in Heterozygotes Using Optimized R_act,421C>A

The F_aF_g of heterozygous (421CA) subjects relative to that of wild-type (421CC) subjects (F_aF_g ratio_421CA/CC) was calculated with Eq. 21 using R_act,421C>A optimized with data from homozygous subjects. Most calculated F_aF_g ratios correlated well with clinically observed AUC ratios even when data from heterozygous subjects were analyzed (Fig. 3).

DISCUSSION

The ABCG2 c.421C>A mutation has been shown to alter the pharmacokinetics of BCRP substrates in several clinical reports. However, the extent to which ABCG2 c.421C>A affects in vivo BCRP efflux transport activity remains unknown. In this study, we developed a method to quantitatively estimate the decrease in mutated (c.421C>A) BCRP function in the intestine. The optimized R_act,421C>A was determined to be 0.23 by comparing the calculated F_aF_g ratios_421AA/CC with clinically observed AUC ratios_421AA/CC, suggesting that in vivo intestinal BCRP transport activity in 421AA subjects is approximately 23% of that in 421CC subjects. This result was consistent with the previous in vitro and ex vivo studies, which demonstrated that the mutated BCRP protein has lower apparent transport activity than the wild-type protein,10, 13, 31, 57 and that subjects with this polymorphism have a lower BCRP protein expression level.11, 12

The intestinal model selected in this study has been used only for theoretical consideration and its application for the prediction of F_aF_g of clinically used drugs has not been reported. Thus, a parameter related to luminal flow rate (LF/S) had to be set with appropriate reference drugs before the practical use of this model. An optimized LF/S value (0.17 × 10⁻⁴ cm/s) obtained from reference drugs was used in the model. To validate the optimized LF/S, physiological LF/S was calculated from physiological parameters with Eq. 22, where r is the radius of small intestine (0.82–1.75 cm),58, 59 SAF is the surface area factor attributed to plicae circulares and villi (30),17 and T is the small intestine transit time (3.2–3.3 h).59, 60

(22)

The physiological LF/S was calculated to be 0.012–0.025 × 10⁻⁴ cm/s, which is 6–15-fold lower than the optimized LF/S. However, it made a marginal difference in the estimation of R_act,421C>A because the estimated R_act,421C>A was 0.27 even when the lowest physiological LF/S (0.012 × 10⁻⁴ cm/s) was used instead of the optimized value (0.17 × 10⁻⁴ cm/s).

Two assumptions were used to estimate the effect of ABCG2 c.421C>A on in vivo intestinal BCRP transport activity. The first assumption was that the AUC increase in subjects with the mutated BCRP is solely accounted for by the F_aF_g increase. The relationship between AUC ratio_421AA/CC and F_aF_g,421CC for each drug is shown in Figure 4a. Drugs with a high AUC ratio_421AA/CC had low F_aF_g,421CC values. The clinically observed AUC ratios_421AA/CC were lower than the 1/F_aF_g curve. The curve shows the AUC ratio_421AA/CC with the assumption that F_aF_g is equal to 1 in homozygous mutation carriers and that systemic and renal clearance is not affected by the mutation. Figure 4b shows a similar tendency for heterozygous mutation carriers, where the 1/F_aF_g curve was replaced with a (1 + 1/F_aF_g)/2 curve, assuming the decrease of BCRP activity in heterozygotes was the average of those in homozygotes for the wild-type and mutant alleles. These results should be provided if the first assumption is appropriate. Moreover, the half-life (T_1/2) values of all drugs, except simvastatin, were not significantly affected by the BCRP mutation (Table 2), suggesting that the change of intestinal absorption is responsible for the AUC increase as the T_1/2 values should be higher if hepatic or renal BCRP is responsible for the decrease in the systemic clearance and thereby the increase in the plasma AUC. Therefore, it is likely that the increased AUC in subjects with the 421AA genotype is mainly because of the increase in F_aF_g.

The second assumption was that intestinal metabolism and the contribution of influx and efflux transporters other than BCRP are negligible in the overall intestinal absorption process from the lumen to the portal vein. The impact of intestinal metabolism, apical influx transport, apical efflux transport not mediated by BCRP, and basolateral efflux transport on the optimization of R_act,421C>A were investigated by sensitivity analysis (see Supplementary Data 1 online). The results suggested that apical influx transport and basolateral efflux transport made a small difference in the F_aF_g ratios_421AA/CC calculation, as long as P_dif was higher than 0.24 × 10⁻⁴ cm/s (Supplementary Fig. S1 online). Actually, the P_dif of each drug investigated in this study was higher than 0.24 × 10⁻⁴ cm/s. On the other hand, the involvement of intestinal metabolism and apical efflux transport not mediated by BCRP could over-estimate the functional change caused by the mutation. However, BCRP should be mainly involved in the apical efflux of the drugs used in this study, and an involvement of intestinal metabolism in intestinal absorption has not been reported for any of the drugs, except for atorvastatin. A further analysis was conducted after the intestinal model was modified to address the intestinal metabolism of atorvastatin (see Supplementary Data 2 online). The F_aF_g ratio calculated with the modified model was similar to the clinically observed AUC ratio (Supplementary Fig. S2 online).

Based on R_act,421C>A estimated in this study, the maximum extent of AUC increase caused by ABCG2 c.421C>A can be predicted if the F_aF_g value of a BCRP substrate is available. The AUC increase for erlotinib, gefitinib, and imatinib, with F_aF_g,421CC values of 1, 0.75, and 1, respectively, was only 1-, 1.3-, and 1-fold at a maximum, respectively. A pharmacogenetic study showed that the c.421C>A polymorphism was not associated with gefitinib skin toxicity, but it was associated with diarrhea.9 Skin toxicity and diarrhea would be associated with the plasma concentration and the intestinal concentration of gefitinib, respectively. The lacking association between skin toxicity and the mutation is consistent with the small increase in the AUC predicted by our method. In contrast, the drug concentration in intestinal epithelial cells can be much higher in homozygous mutation carriers than in wild-type subjects because decreased BCRP function can affect the concentration in intestinal epithelial cells even when F_aF_g is not affected. However, it should be noted that some clinical studies showed opposite results. Akasaka et al. and Tamura et al. reported that the c.421C>A polymorphism was not associated with diarrhea.^{61, 62}

CONCLUSIONS

This study showed that in vivo intestinal BCRP transport activity of the homozygous subjects (421AA) was approximately 23% of that in wild-type (421CC) subjects. To our knowledge this is the first report, which tries to estimate the impact of the ABCG2 c.421C>A polymorphism on intestinal BCRP transport activity in vivo.