Expression, Up-Regulation, and Transport Activity of the Multidrug-Resistance Protein Abcg2 at the Mouse Blood-Brain Barrier

Much of the specificity of the endothelium forming the blood-brain barrier (BBB) is due to the tight junctions between adjacent brain endothelial cells. These restrict the entry of compounds by paracellular diffusion from the blood to the brain and vice versa. The transcellular diffusion of compounds through the membrane bilayer of the brain endothelial cells may also be limited by transmembrane efflux transporters that belong to the ABC superfamily, such as P-glycoprotein (P-gp). P-gp was first described as a component of the acquired multidrug resistance (MDR) mechanism of tumor cells, but it is now known to be present physiologically at the luminal membrane of the brain capillary endothelial cells (1). P-gp/mdr1a actively expels a wide spectrum of amphiphilic drugs from the brain into the blood, leading to a decrease in their uptake by the brain (1). Another member of the ABC superfamily, which was first called BCRP or MXR and is now designated ABCG2 in the HUGO¹ nomenclature, confers resistance to many drugs, including mitoxantrone (MX), prazosin (PRA), the anthracyclines, and some camptothecin derivatives, and its range may overlap with the substrates of other P-gp and/or MDR-associated proteins (MRPs; Refs. 2, 3, 4, 5, 6). ABCG2 was recently identified in human and porcine brain capillary endothelial cells (7, 8, 9, 10) and found to be responsible for the in vitro export of daunorubicin from porcine brain capillary endothelial cells (9). In vivo studies on Abcg2-deficient knockout mice have shown that Abcg2 is involved in the handling of and protection from porphyrins, such as chlorophyll and heme. Its ability to limit the oral availability of topotecan also suggests that it plays a role in intestinal absorption (11).

Whereas the action of ABCG2 in making cells drug resistant has been explored to some extent, knowledge of how ABCG2 mediates drug transport at the BBB in vivo could provide important information on the overall mechanisms underlying drug transfer, drug delivery and CNS toxicity. Recent in vitro studies showing that ABCG2 is present in brain endothelial cells raise the question of how it influences the efflux of drugs at the BBB in vivo (7, 8, 9, 10).

The in situ brain perfusion method developed in the rat (12) and more recently in mice (13, 14, 15) is a very sensitive method for assessing drug transport at the luminal membrane of the endothelial cells of the BBB in vivo. It can be used to measure both the rate and mechanisms of transport. This technique respects the physiological properties of transport by maintaining the physiological properties of the BBB (13, 16). One of its great advantages is that there is no systemic exposure before its transport through the BBB, thus allowing the drug concentration in the perfusion fluid to be controlled and avoiding confounding factor of systemic disposition such as the effect of metabolism. We used the in situ brain perfusion technique in wild-type (wt) and P-gp-deficient mutant mdr1a(−/−) mice to check and exclude P-gp mechanisms from the interpretation of the data. The presence of Abcg2-mediated transport at the mouse BBB was assessed in vivo by measuring the brain transport of MX and PRA, which interact with P-gp and Abcg2 in vitro. Unlike PSC833, which inhibits P-gp transport, GF120918 inhibits both P-gp- and Abcg2-mediated transport (17, 18). The transport of MX, PRA, and vinblastine (VBL), which is not transported by ABCG2 (19), were then measured in wt and mdr1a(−/−) mice with and without GF120918 or PSC833. These multiple experimental strategies enabled us to show that the uptake of MX and PRA by the brain is not significantly affected by P-gp, unlike that of VBL. Real-time quantitative reverse transcription-PCR (RT-PCR) of brain parietal cortex homogenates and a capillary-enriched fraction isolated by the capillary depletion method confirmed that there is more abcg2 mRNA in brain capillaries than in the cortex. We also show that the concentration of abcg2 mRNA in the brain capillaries of mice lacking mdr1a is above normal. This increased abcg2 mRNA in mdr1a(−/−) mutant mice was correlated with a greater in vivo brain efflux transport of PRA and MX in P-gp-deficient mice than in wt mice. These results therefore indicate that Abcg2 is normally present and functional at the luminal membrane of the brain endothelial cells of both wt and mutant mdr1a(−/−) mice, where it limits the uptake of its substrates by the brain.

Adult male CF-1 mice [wt and mutant mdr1a(−/−); 30–40 g, 7–9 weeks of age] were bred in house from progenitors, initially obtained from Charles River Laboratories (Wilmington, MA), genotyped for mdr1a P-gp. The mice were housed in a controlled environment (22 ± 3°C with 55 ± 10% relative humidity) and maintained under a 12-h dark:light cycle (light from 6:00 a.m. to 6 p.m.). They had access to food and tap water ad libitum. All experimental procedures complied with the ethics rules of the French Ministry of Agriculture for experimentation with laboratory animals (law no. 87-848).

The transport of [³H]MX (Moravek, Brea, CA), [³H]PRA and [³H]VBL (Amersham Pharmacia Biotech, Orsay, France) into the brain was measured by in situ brain perfusion (13). Briefly, mice were anesthetized by i.p. injection of ketamine-xylazine (140 and 8 mg/kg, respectively). The right external carotid was ligated rostral to the occipital artery at the level of the bifurcation of the common carotid. The caudal side of the right common carotid was ligated, and the cranial side was catheterized (polyethylene tubing; 0.30 mm inner diameter × 0.70 mm outer diameter; Biotrol Diagnostic, Chenevières-les-Louvres, France). The syringe containing the perfusion liquid was placed in an infusion pump (Harvard PHD 2000; Harvard Apparatus, Holliston, MA) and connected to the catheter. The thorax was opened, and the heart was cut before starting the perfusion. Perfusion was at a flow rate of 2.5 ml/min. The perfusion fluid consisted of bicarbonate-buffered physiological saline containing 128 mm NaCl, 24 mm NaHCO₃, 4.2 mm KCl, 2.4 mm NaH₂PO₄, 1.5 mm CaCl₂, 0.9 mm MgCl₂, and 9 mm d-glucose. The solution was gassed with 95% O₂–5% CO₂ for pH control (7.4) and warmed to 37°C in a water bath. Each mouse was perfused with tritiated tracer and [¹⁴C]sucrose (0.2 μCi/ml; Perkin-Elmer, Paris, France) added to the perfusion fluid. Brain perfusion was terminated by decapitating the mouse at a selected time (30, 60, 90, or 120 s). The brain was removed from the skull and dissected out on ice. The right cerebral hemisphere and aliquots of the perfusion fluid were placed in tared vials and weighed. Samples were digested in 2 ml of Solvable (Packard, Rungis, France) at 50°C and mixed with 9 ml of Ultima Gold XR (Packard). Dual label counting was performed in a Packard Tri-Carb 1900TR.

All of the calculations have been described previously (13, 14, 15, 16). The brain vascular volume of each animal was assessed with the vascular marker [¹⁴C]sucrose. This, like inulin, does not measurably cross the BBB during short perfusions (12). In the perfusion experiments, [¹⁴C]sucrose was perfused together with the tritium-labeled substrates to estimate the brain vascular volume (V_vasc; μl/g) as:

where X* (dpm/g) is the amount of sucrose in the right brain hemisphere and C*_perf (dpm/μl) is the concentration of [¹⁴C]sucrose in the perfusion fluid.

The apparent brain distribution volume was calculated from the amount of radioactivity in the right brain hemisphere by the formula:

where X_brain (dpm/g) is the calculated amount of tritium-labeled compound in the right cerebral parenchyma, and C_perf (dpm/μl) is the concentration of tritium-labeled tracer in the perfusion fluid. Brain tissue total radioactivity was corrected for vascular contamination with the following formula:

where X_tot (dpm/g) represents the total quantity of tritium-labeled tracer trapped within the brain parenchyma and cerebrovasculature capillary bed. Subtraction of the intravascularly located tritium-labeled compound from the total amount of tritium-labeled compound yielded the amount in the brain parenchyma.

Initial brain transport was expressed as a K_in (μl/s/g) and was calculated from the formula:

where T is the perfusion time (s).

A time-course study of the apparent distributional brain volume of the tritiated drugs (0.3–0.4 μCi/ml) was done in wt mice to select an appropriate brain perfusion time. The initial uptake of tritiated drugs into the mouse brain after 120 s of perfusion was measured in wt and mdr1a(−/−) mice with and without inhibitors. Inhibitors were added to the perfusion fluid to control exposure at the BBB and to avoid systemic metabolic processes. The inhibitors were PSC833 (3 μm; kindly provided by Novartis, Basel, Switzerland) and GF120918 (2 μm; kindly provided by Glaxo SmithKline, Les Ulis, France). The uptake of labeled MX, PRA, and VBL was also measured in the presence of unlabeled MX (100 μm; Sigma, St Quentin, France), PRA (30 μm; Sigma), or VBL (150 μm; Lilly, Saint Cloud, France), respectively. The stock solutions of PRA, GF120918, and PSC833 were prepared in DMSO. MX and VBL were dissolved in 0.9% NaCl. All solutions were prepared fresh on the day of the experiment and diluted with the bicarbonate-buffered saline used for perfusion. The final DMSO concentration in the perfusion fluid was always 0.5% (v/v); this concentration did not alter the integrity of the BBB, as measured by the permeability of [¹⁴C]sucrose.

The blood was removed from the brains of the mice before microvessels were extracted to minimize contamination with blood cells and hemoglobin in the quantitative RT-PCR study. The mice were perfused with saline by in situ brain perfusion for 30 s. They were then decapitated, and the brain was removed and cleaned of meninges on ice. The capillary depletion method of Triguero et al. (20) was used with some modifications (21) to obtain capillary-enriched fractions of wt and mdr1a(−/−) mouse brains. The blood-free brains of six wt or mdr1a(−/−) mice were pooled and homogenized with a glass homogenizer (15 pestle strokes) in 3.5 ml of buffer [10 mm HEPES, 141 mm NaCl, 4 mm KCl, 1 mm NaH₂PO₄, 2.8 mm CaCl₂, 1 mm MgSO₄ and 10 mm d-glucose (pH 7.4)] at 4°C. Chilled 37% dextran solution was then added to obtain a final dextran concentration of 18.5%. All homogenizations were performed rapidly at 4°C. The homogenates were centrifuged at 5400 × g for 15 min at 4°C in a swinging-bucket rotor, and the supernatants were carefully removed. Brain capillaries are separated from the parenchymal elements by a well-developed collagenous basal membrane as in the vasculature of peripheral organs, but the neural and glial elements of the brain lack significant amounts of nonvascular collagenous matrix, whereas most peripheral organs have abundant mesenchymal collagen networks. Thus, physical fractionation of this soft tissue efficiently separates central nervous system microvessels from the brain parenchyma (22, 23). The homogenates were centrifuged through a dextran gradient, and the pellet (composed of blood-free intact microvessels segments and brain nuclei) was collected (20). Mechanical homogenization of brain tissue rapidly provided brain capillaries, minimizing mRNA denaturation due to RNase activity and changes in cell phenotype, unlike time-consuming enzymatic methods (23). Microvessel purity was checked by measuring γ-glutamyl transpeptidase activity because this enzyme is abundant in the endothelial cells of the BBB (22). The enzyme activities in aliquots of the supernatant and pellet confirmed that there was ∼30 times more activity in the pellet than in the supernatant. The pellet was removed and placed in RNAlater (Qiagen, Courtaboeuf, France) at −20°C until RNA extraction.

Total RNA was extracted from isolated mouse brain capillaries and parietal cortex by use of the RNeasy Mini (Qiagen) extraction kit according to the manufacturer’s instructions. The RNA concentration and integrity of each sample were evaluated with the RNA 6000 LabChip kit (Agilent Biotechnologies, Massy, France) with the Agilent 2100 Bioanalyzer. The RNA samples (5 μg) were made up to 31 μl with PCR water, heated at 65°C for 10 min, and cooled on ice (4 min) before reverse transcription. We then added 2 μl of Oligo(dT) (Roche Diagnostic, Meylan, France) to each tube to obtain a final volume of 33 μl [with 10 mm Oligo(dT)], and this volume was transferred to a tube of First-Strand Reaction Mix Beads (Amersham Pharmacia Biotech). The tubes were placed on ice, homogenized by gentle vortexing for 10 min, and then incubated at 37°C for 60 min.

The sequences of the primers for the target gene abcg2 (accession no. AF140218) were obtained with Primer Express software (Applied Biosystems, Courtaboeuf, France) to minimize the formation of internal structures. The primers themselves were synthesized by Genset (Paris, France).

The forward primer was 5′-AGCTCCGATGGATTGCCAG-3′ (nucleotides 1146–1164), and the reverse primer was 5′-GAGGGTTCCCGAGCAAGT-TT-3′ (nucleotides 1196–1177). The annealing temperature was 81°C, and the amplicon size, evaluated with the DNA 500 Bio Sizing kit (Agilent Biotechnologies) and Agilent 2100 Bioanalyzer, was 51 bp. TaqMan Rodent gapdh Control Reagents (Applied Biosystems) were used as normalization controls. The products of RT-PCR were diluted to have the same cDNA concentration (10 ng/μl) in all samples. One μl of each RT-PCR dilution, 1 μl of water (no template control), or 1 μl of 10 ng/μl RNA pool (no amplification control) was added to 49 μl of mixture containing 25 μl of SYBR Green PCR Master Mix (Applied Biosystems), 300 nm sense and antisense primer, and PCR water for abcg2 and 25 μl of TaqMan PCR Master Mix (Applied Biosystems), 0.5 μl of primer 1, 0.5 μl of primer 2, 0.5 μl of probe, and PCR water. Samples were placed in a thermal cycler and cycled 40 times (denaturation at 95°C for 15 s and 60°C for 60 s, and polymerization at 81°C for 60 s).

Increasing amounts of standard cDNA were used with gene-specific primers and fluorescent probes to generate a standard curve displaying a linear relationship between the threshold cycle (C_T; number of cycles at which the fluorescence signal reaches a defined threshold) and the logarithm of the initial template concentration. The equation describing the standard curve for abcg2 was Y = −2.05X + 30.4 and that for gapdh was Y = −2.2X + 28.8. The efficiencies for abcg2 and gapdh were approximately equal. The absolute value of the slope of log input versus ΔC_T (C_T abcg2 − C_T gapdh) was 0.023 (<0.1 theoretical).

Dissociation curve analysis was performed after amplification to confirm that the expected product was generated and to distinguish specific from nonspecific products and primer-dimers. The thermal protocol for the dissociation was 15 s at 95°C, followed by 20 s at 60°C and a slow ramp (1.75°C/min) over 20 min from 60°C to 95°C. The data for the dissociation curve were captured during the slow ramp. We also analyzed the PCR product by agarose gel electrophoresis with the DNA Bio Sizing kit to determine the quality of the PCR.

The K_in values are the means ± SD of four to six mice unless specified otherwise. Student’s unpaired t test was used to identify significant differences between groups when appropriate. The tests were all two-tailed, and statistical significance was set at P < 0.05. Three experiments were performed for RT-PCR, and the data are means of the abcg2 mRNA concentration divided by the gapdh mRNA concentration ± SD. Data were analyzed by two-way ANOVA followed by Dunnett’s test, and statistical significance was set at P < 0.01 or P < 0.001.

The mouse brain vascular volume (16 ± 2 μl/g) agreed with normal distribution volume for sucrose (12, 13, 14); in all brain transport experiments. A brain vascular volume that exceeded 20 μl/g was considered to reflect a loss of BBB physical integrity; in those cases brain transport parameters were not calculated (13, 16). Brain perfusions performed with MX concentrations >100 μm were associated with a loss of BBB integrity, preventing measurement of the [³H]MX brain transport at the highest MX concentrations.

The uptake of MX, PRA, and VBL by the brains of wt mice, expressed as apparent brain distributional volume (V_brain), was measured at different brain perfusion times to select an appropriate perfusion time (Fig. 1). The perfusion time selected for single-time brain uptake studies ensured that at least 40% of the total radioactivity of the tritiated drug in the tissue resided outside the vascular space (12). Given a vascular volume of 14–18 μl/g, a drug brain volume V_brain at 9–12 μl/g is equivalent to 40% of the radioactivity in the extravascular space. This requirement was strictly met at 30 s for MX and PRA and at 60 s for VBL (Fig. 1). Therefore, a perfusion time of 120 s was used for all of the single-time uptake studies because it provided optimum sensitivity for transport measurements in various experimental situations, whereas uptake remained linear (16).

Because many of the drugs transported by Abcg2 are, like MX and PRA, also carried by other transporters, such as P-gp, the effect of P-gp on the transport of the three selected drugs was also measured in the two mouse strains. Animals lacking specific transporters are ideal and efficient tools for evaluating the contribution of transporters such as P-gp to the transport of drugs in the brain without the need for inhibitors (14). We measured the brain transport of the three selected substrates in wt and P-gp-deficient mice to assess the influence of P-gp on their brain uptake. The uptake of VBL was ∼2-fold greater in mice lacking P-gp than in wt mice, confirming that P-gp limits initial VBL uptake at the mouse BBB (Fig. 2,A). The transport of PRA was not affected by P-gp (Fig. 2,B), as shown by the data for mdr1a(−/−) (1.14 ± 0.06 μl/s/g) and wt (0.96 ± 0.09 μl/s/g) mice. The brains of mdr1a(−/−) mice took up significantly lesser MX (1.55 ± 0.05 μl/s/g; P < 0.01) than did those of wt mice (1.78 ± 0.06 μl/s/g; Fig. 2 C). This result gave no clear information about the influence of P-gp on the brain transport of MX. It probably indicates that a non-P-gp efflux, perhaps Abcg2 transport (see below), is enhanced in the mdr1a(−/−) mice.

Chemical modulation could be a way to determine the existence of a carrier-mediated transport, despite the lack of specificity. PSC833, a cyclosporin analog, inhibits P-gp efflux but not Abcg2 transport in vitro (2). We therefore could use PSC833 to measure the P-gp transport component in wt mice and help to confirm our results obtained with mdr1a(−/−) mice. PSC833 increased the brain uptake of VBL in wt mice to be significantly greater than in wt mice not given PSC833 (Fig. 2,A), but the brain uptake of VBL in wt mice given PSC833 was not different from that in P-gp-deficient mice (Fig. 2,A). The uptake of PRA and MX by the brains of wt mice was not significantly different with or without PSC833 (Fig. 2, B and C). Thus, P-gp does not significantly limit the brain uptake of PRA or MX at the mouse BBB.

GF120918, an effective inhibitor of P-gp and Abcg2 transport, was perfused with MX, PRA, and VBL in wt and mdr1a(−/−) mice to determine the influence of Abcg2 on their brain transport. Because of the lack of P-gp transport of MX and PRA at the mouse BBB, GF120918 inhibited only Abcg2-mediated transport of these two substrates in wt mice. In contrast to MX and PRA, VBL was significantly transported by P-gp at the mouse BBB. The uptake of VBL by the brains of wt mice was significantly increased by GF120918 over that of wt mice not receiving GF120918 (Fig. 2,A). Moreover, the uptake of VBL by the brains of mdr1a(−/−) mice was not affected to any great extent by GF120918. The uptake of VBL at the mouse BBB is thus restricted by P-gp and not by Abcg2. The brain transport of PRA in wt mice was increased 1.5-fold by perfusion together with GF120918, and that of MX was increased 2-fold, compared with control wt mice not receiving inhibitor (Fig. 2, B and C). The brain transport of PRA was increased 2.1-fold and that of MX 3.0-fold in mdr1a(−/−) mice when the drugs were perfused together with GF120918 compared with P-gp-deficient mice not receiving perfusions containing this inhibitor. GF120918 had a statistically greater effect on the brain transport of PRA (0.67-fold) and MX (0.71-fold) in mdr1a(−/−) mice than on their transport in wt mice (Fig. 2, B and C).

The brains of P-gp-deficient and/or wt mice were perfused with labeled PRA, MX, or VBL plus high concentrations of the unlabeled drug to determine whether efflux could be saturated at the mouse BBB. PRA and VBL efficiently inhibited their own efflux at the BBB, but unlabeled VBL increased the uptake of [³H]VBL (Fig. 2,A) by completely saturating P-gp transport. PRA also inhibited its own efflux in P-gp-deficient mice with the same efficiency as GF120918 in P-gp-deficient mice (Fig. 2 B). MX (100 μm) did not inhibit its own brain transport, probably because of a lower affinity and despite the fact that its transport at the mouse BBB is carrier-mediated.

The abcg2 mRNA values were normalized with reference to the gapdh mRNA in each sample. The dissociation curves show the products from a one-step RT-PCR for a abcg2 target as well as for a no-template control (water), indicating the specificity of the PCR. The amplicon from the main product was detected with a melting temperature of 77°C. We also analyzed a PCR product of the capillary-enriched wt brain fraction by agarose gel electrophoresis to determine the quality of the PCR (Fig. 3,A). Only two PCR products were observed in the lane, and their molecular sizes were 51 bp for abcg2 and 177 bp for gapdh, as expected (Fig. 3,A). The brain capillaries of wt mice had significantly more (∼700 times) abcg2 mRNA (675 ± 50; n = 3) than the brain cortex (1 ± 0.2; n = 3; P < 0.001). There was 3 times more (P < 0.01) abcg2 mRNA in the brain capillaries of mdr1a(−/−) mice (2254 ± 181; n = 3) than in wt capillaries (Fig. 3 B).

In this study, we found that the transport of VBL in the brain is mediated by P-gp, whereas the transport of MX and PRA across the BBB depends mainly on the presence of Abcg2 at the luminal membrane of the mouse brain microvessels. PRA, like progesterone, interacts with P-gp via the P-gp regulatory/allosteric site, which does not allow any apparent transport of drugs (24, 25, 26). Overproduction of P-gp usually results in only modest resistance to MX (27, 28), implying that this drug is poorly transported by P-gp (28). The ABC transporter Mrp1 has a moderate and controversial effect in conferring resistance to MX, as shown in mouse cells lacking ABC transporters (28, 29) and cells transfected with MRP1 (30). Nevertheless, we recently showed that Mrp1 is not functional at the luminal side of the mouse BBB (15). The present study demonstrates that Abcg2 is a major factor controlling MX and PRA transport at the BBB of wt mice.

Vinca alkaloid drugs, such as VBL, are not transported by Abcg2, whereas MX and PRA are efficiently transported in vitro (6, 17, 19). We investigated Abcg2-mediated transport at the BBB by measuring the transport of MX, PRA, and VBL with and without GF120918, which inhibits P-gp and Abcg2 transport, and PSC833, which inhibits P-gp. The results indicate that VBL transport at the mouse BBB is not restricted by Abcg2. In contrast, the brain transport of PRA and MX is not sensitive to PSC833 but is significantly increased by GF120918 in wt mice. Thus, a mechanism unrelated to P-gp mediates the brain efflux of MX and PRA at the BBB. Although P-gp does not limit MX or PRA uptake at the BBB of wt mice, we perfused the brains of mutant mice lacking P-gp to clearly reveal this non-P-gp transport. The brain uptake of PRA was increased 2.1-fold and that of MX 3.0-fold in mdr1a(−/−) mice when the drugs were perfused together with GF120918 compared with their uptake by P-gp-deficient mice not perfused with this inhibitor. PSC833 also had no effect on the brain transport of MX or PRA at the BBB in P-gp-deficient mice, confirming in vivo that it does not overcome Abcg2 resistance. The inhibition of the brain transport of PRA and MX by GF120918 confirms the existence of a mechanism unrelated to P-gp, probably Abcg2-mediated transport, at the mouse BBB. We also inhibited the carrier-mediated transport of PRA and MX by use of a high concentration of unlabeled compound. PRA (30 μm) inhibited the efflux of radiolabeled PRA in mdr1a(−/−) mice, whereas 100 μm MX did not inhibit its own efflux. Higher concentrations of MX were not perfused because of their adverse effect on the BBB.

Surprisingly, GF120918 had a greater effect on the brain transport of PRA (0.67-fold) and MX (0.71-fold) in mdr1a(−/−) mice than on their transport in wt mice. The brain transport of MX was also lower in mice lacking P-gp than in the wt mice. We therefore postulate that Abcg2 is up-regulated in P-gp-deficient mice or that the mediated-drug efflux is more rapid in this mdr1a-deficient mouse strain. The first possibility was checked by measuring abcg2 mRNAs in the brain cortex and capillaries by quantitative real time RT-PCR. The brain capillaries of wt mice had significantly (∼700 times) more abcg2 mRNA than the brain cortex. A recent report also showed significantly more abcg2 mRNA in the capillary endothelial cells of pig brains than in the whole brain (9). This again points to the importance of Abcg2 in brain capillaries, as suggested by the MX and PRA in situ brain transport studies, given that Abcg2 tends to be concentrated in brain microvessel cells. One concern about the use of capillary microvessels could be the cell extract composition. Isolated brain capillaries contain approximately one-third pericytes and some astrocyte endfoot membranes (23). The endothelial cells, like pericytes, could contain the abcg2 mRNA measured in our experiments. However, pericytes from pig brains contain no abcg2 mRNA as measured by Northern blotting (9). Moreover, in situ brain perfusion measures only the transport at the luminal membrane of endothelial brain cells (16). These observations together suggest that Abcg2 is located at the luminal membrane of the brain microvessels. We also found that the brain capillaries of mdr1a(−/−) mice contain 3 times more abcg2 mRNA than wt capillaries. These data confirm our transport studies indicating that Abcg2 synthesis is increased in mdr1a(−/−) mouse brain capillaries. This abcg2 increase is probably correlated with the overproduction of the Abcg2 protein at the mouse BBB. The Abcg2 protein level is well correlated with resistance to MX and PRA in vitro (31). Although Abcg2 overproduction by drug selection in human or murine cells could be associated with a mutation at residue 482 that affects the specificity and transport function of the pump (6, 32), such alterations remain to be established in vivo for the increased Abcg2 in mdr1a(−/−) mice. However, the inhibitor GF120918 also reverses wt protein and Abcg2 with the 482 polymorphism (6). We have yet to determine whether a lack of mdr1a directly results in compensation involving Abcg2. The basal amount of Mrp1 is slightly elevated in cells from knockout mouse embryos lacking functional mdr1a/1b, but the basal amounts of Abcg2 in wt cells and cells lacking P-gp remain to be determined (28). There is a lack of functional mdr1a in two situations in vivo. One is in mdr1a(−/−) knockout mice that have increased mdr1b expression in the liver and the kidney but not in the brain or intestine, which normally contain little or no mdr1b (33). The other one is in CF-1 mutant mice lacking mdr1a(−/−), the mouse strain used in our study, where the amount of mdr1b is unaltered (34). Thus, the cause of the up-regulation remains to be clarified in mice lacking mdr1a, but it points to a tissue-specific compensation involving ABC transporters. Such compensation between drug transporters has yet to be explored in Abcg2(−/−) knockout mice. A recent study showed that there was more ABCG2 in the brain capillaries extracted from a human glioblastoma than in nonmalignant brain tissue (10).

In conclusion, this study demonstrates that Abcg2 is physiologically present and functional at the mouse BBB. The basal expression is sufficient to limit the brain uptake of MX and PRA, so that Abcg2 is thus a newly identified factor limiting the permeability of the brain to drugs and a potential source of drug-drug interactions. We also found that the concentration and transport activity of Abcg2 in vivo are increased by a lack of mdr1a at the mouse BBB. This highlights the care that must be taken with deficient mouse models, in which compensatory gene expression and protein synthesis may offset the lack of a specific gene.

We thank Dr. Owen Parkes for linguistic revisions.