The ATP-binding cassette proteins ABCB1 and ABCC1 as modulators of glucocorticoid action

Responses to hormones that act through nuclear receptors are controlled by modulating hormone concentrations not only in the circulation but also within target tissues. The role of enzymes that amplify or reduce local hormone concentrations is well established for glucocorticoid and other lipophilic hormones; moreover, transmembrane transporters have proven critical in determining tissue responses to thyroid hormones. However, there has been less consideration of the role of transmembrane transport for steroid hormones. ATP-binding cassette (ABC) proteins were first shown to influence the accumulation of glucocorticoids in cells almost three decades ago, but observations over the past 10 years suggest that differential transport propensities of both exogenous and endogenous glucocorticoids by ABCB1 and ABCC1 transporters provide a mechanism whereby different tissues are preferentially sensitive to different steroids. This Review summarizes this evidence and the new insights provided for the physiology and pharmacology of glucocorticoid action, including new approaches to glucocorticoid replacement. This Review discusses the ATP-binding cassette (ABC) proteins ABCB1 and ABCC1 and their preferential cellular export of cortisol and corticosterone, respectively. The article also explores the potential to select therapeutic glucocorticoids on the basis of their different tendencies for export to avoid harmful adverse effects. Humans have two circulating glucocorticoid hormones, cortisol and corticosterone, which diffuse into cells to become transcription factors when bound to their intracellular receptors. The availability of glucocorticoids to interact with their receptors depends not only on their plasma concentration but also on their intracellular concentration, which is modulated by intracellular enzymes and by transmembrane transporters. Glucocorticoids are susceptible to cellular export by membrane transporters from the ABC (ATP-binding cassette) transporter family: cortisol is a substrate for the ABCB1 transporter, and corticosterone for ABCC1. Tissues expressing ABCB1 (such as the brain) might be relatively sensitive to corticosterone over cortisol; those expressing ABCC1, such as adipose, might be more sensitive to cortisol. In future, therapeutic glucocorticoids could be selected on the basis of lower tendency to be exported from sites of efficacy and higher tendency for export from sites where harmful adverse effects occur. Humans have two circulating glucocorticoid hormones, cortisol and corticosterone, which diffuse into cells to become transcription factors when bound to their intracellular receptors. The availability of glucocorticoids to interact with their receptors depends not only on their plasma concentration but also on their intracellular concentration, which is modulated by intracellular enzymes and by transmembrane transporters. Glucocorticoids are susceptible to cellular export by membrane transporters from the ABC (ATP-binding cassette) transporter family: cortisol is a substrate for the ABCB1 transporter, and corticosterone for ABCC1. Tissues expressing ABCB1 (such as the brain) might be relatively sensitive to corticosterone over cortisol; those expressing ABCC1, such as adipose, might be more sensitive to cortisol. In future, therapeutic glucocorticoids could be selected on the basis of lower tendency to be exported from sites of efficacy and higher tendency for export from sites where harmful adverse effects occur.

Glucocorticoid hormones are vital for life; they confer diverse effects on multiple processes and systems. The adverse consequences of glucocorticoid excess are well demonstrated by the frank hypercortisolism of Cushing syndrome, but even subtle cortisol dysregulation has implications, contributing to cardiovascular disease, for example 1 . Over the past 30 years it has become clear that the concentration of glucocorticoid in the blood does not necessarily reflect that within tissues, as enzymes (such as 11β-hydroxysteroid dehydrogenase, which catalyses the interconversion of inert cortisone and active cortisol) and delivery mechanisms of corticosteroid binding protein can confer additional control over the absolute tissue levels 2,3 .
As lipophilic molecules, glucocorticoids can diffuse across cell membranes to interact with intracellular targets; however, they can also undergo active transmembrane transport. This process was first described for the ABCB1 transporter (of the ATP-binding cassette (ABC) protein family; also known as P-glycoprotein or multidrug resistance protein 1 (MDR1)), which exports cortisol and a variety of synthetic glucocorticoids from 'sanctuary sites' , including the brain 4,5 . Intriguingly, corticosterone is not readily exported by ABCB1, but we have discovered that the ABCC1 transporter (also known as multidrug resistance-associated protein 1 (MRP1)), found in tissues such as adipose, exports corticosterone but not cortisol 6 .
In this Review, we explore the implications of this tissue-specific glucocorticoid transport in the central control of the hypothalamic-pituitary-adrenal (HPA) axis, adipose tissue metabolism and pregnancy. We also consider whether the steroid specificity of ABCB1 and ABCC1 transport offers insights into the different roles of corticosterone and cortisol in humans and a potential opportunity for developing glucocorticoid therapies that are better targeted than those currently available to maximize efficacy and minimize toxicity.

The movement of lipophilic hormones
The 'free hormone hypothesis' determines that unbound lipophilic hormones move passively down a concentration gradient 7 and, indeed, steroids are taken up freely by cell types such as keratinocytes without the relevant membrane transporters 8 . Differences in steroid concentrations between tissues were previously attributed to differences in physicochemical properties, such as lipophilicity, until the discovery of the existence of specific thyroid hormone transporters challenged these traditional assumptions. In the case of triiodothyronine (T 3 ), which is highly lipophilic owing to the iodinated aromatic ring, the level of hormone available to receptors not only depends on hormone synthesis and peripheral enzymatic conversion, but also on transport into and out of cells, notably by the monocarboxylate 8 (MCT8) transporter 9 . The uptake of T 3 into neurons is critically impaired in the absence of MCT8, as occurs in the X-linked Allan-Herndon-Dudley syndrome of neurodevelopmental anomalies associated with abnormal thyroid function 10 .
The cellular uptake of glucocorticoids by membrane transporters has been demonstrated in Drosophila melanogaster, in which loss of the ecdysone importer membrane transporter produces a phenotype that is identical to that resulting from the loss of ecdysone or the ecdysone receptor 11 . Organic anion transporting polypeptide transporters mediate the uptake of glucocorticoids in rat liver ex vivo; however, this uptake has not been reproduced in humans 12, 13 . Furthermore, a saturable glucocorticoid uptake mechanism across the blood-brain barrier (BBB) and blood-cerebrospinal fluid barrier that was reported in mice was only discernible at supraphysiological concentrations, and so might not be physiologically relevant 14 .
Our increasing understanding of the importance of transporters for thyroid hormone function sets a biological precedent for a similar scenario for other lipophilic hormones; although the active cellular import of glucocorticoids in humans has not been shown, there is mounting evidence supporting the facilitated export of glucocorticoids from cells, particularly by two members of the ABC transporter family.
The ABC protein family As members of one of the most highly conserved protein superfamilies, ABC proteins shuttle toxins, xenobiotics and signalling molecules across eukaryotic and prokaryotic cell membranes. These proteins are classified into seven subfamilies according to their structural similarity and sequence homology, and have been actively researched for decades, particularly in relation to multidrug resistance. The evolution and relevance of this transporter superfamily in the context of cancer drug efflux has been well reviewed 15,16 ; however, of the over 50 human ABC proteins that have been identified, only ABCB1 and ABCC1 have recognized roles in glucocorticoid transport 17 .
The typical ABC transporter is a homodimer characterized by two transmembrane domains (TMDs) and two cytoplasmic nucleotide-binding domains (NBDs) 18 ( Fig. 1). Each TMD domain contains between six and ten transmembrane αhelices, depending on the specific transporter, and is involved in substrate recognition. The cytoplasmic NBDs contain conserved motifs for ATP binding and hydrolysis, including the ABC signature motif (or C-loop motif), Walker A motif (P-loop) and Walker B motif 17 . Together, these dimeric NBDs act to hydrolyse ATP and provide energy to drive transport against concentration gradients.
Several models have been proposed to explain the relationship between ATP hydrolysis and TMDmediated transport 19 , with most purporting that energy from ATP hydrolysis enables the TMDs to switch between inward-facing and outward-facing configurations (Fig. 1a). Individual ABC transporters are unidirectional: in eukaryotic cells, they are almost exclusively exporters, but both importers (of nutrients) and exporters (of toxins and cell wall substrates) exist in bacteria 20 . Consistent with this export function in eukaryotes, ABC transporters are typically found on the apical cell membrane on luminal surfaces to limit xenobiotic exposure 17 . Substrates range from ions to large proteins and there is a high degree of overlap between transporters, although the molecular basis for this overlap remains poorly documented.
ABCB1 and ABCC1 are steroid exporters ABCB1 and steroid export Initially named P-glycoprotein and later MDR1, ABCB1 has been extensively studied as the archetypal multidrug transporter, exporting a broad array of xenobiotics including antineoplastics, antimicrobials and antidepressants from cells (reviewed in 15,21,22 ). In humans, the ABCB1 gene, located on chromosome 7q21.12, encodes a protein of 1280 amino acids (141.5 kDa) with 12 membrane-spanning α-helices distributed among two TMDs 23 . The polyspecificity of ABC transporters is often purported to result from the plasticity of the drug-binding pocket, both in terms of side chain and backbone arrangements. Numerous attempts have been made over the years to determine the 3D structure of ABC proteins in an effort to understand their transport mechanisms and their substrate specificity; however, their size and hydrophobicity pose significant challenges 24 . Advances in the use of cryo-electron microscopy have enabled structural insights into substrate binding [25][26][27] . Reconstitution of the structure of human ABCB1 in complex with chemotherapeutic drugs has revealed the drug-binding cavity to be globular in shape, with interactions contributed by all 12 membrane-spanning α-helices 25 (Fig. 1b). Substrate-induced structural changes in NBD2 are thought to confer changes in ATPase activity, which determines transport action.
A putative steroid-binding site has been identified in human ABCB1, but this is based upon a homology model of only the NBDs 28 and is not definitive. However, physiological data do support selective ABCB1-mediated transport of steroids. In the 1960s, murine fibroblasts were observed exporting steroids in an energy-dependent and temperature-dependent

Key points
• Humans have two circulating glucocorticoid hormones, cortisol and corticosterone, which diffuse into cells to become transcription factors when bound to their intracellular receptors. • The availability of glucocorticoids to interact with their receptors depends not only on their plasma concentration but also on their intracellular concentration, which is modulated by intracellular enzymes and by transmembrane transporters. • Glucocorticoids are susceptible to cellular export by membrane transporters from the abC (aTP-binding cassette) transporter family: cortisol is a substrate for the abCb1 transporter, and corticosterone for abCC1. • Tissues expressing abCb1 (such as the brain) might be relatively sensitive to corticosterone over cortisol; those expressing abCC1, such as adipose, might be more sensitive to cortisol. • In future, therapeutic glucocorticoids could be selected on the basis of lower tendency to be exported from sites of efficacy and higher tendency for export from sites where harmful adverse effects occur.

α-Helices
Secondary protein structures formed by hydrogen bonding between the amine and carbonyl groups of amino acids located four residues apart, resulting in a helical structure with a tightly coiled central backbone, with side chains extending outwards.

Luminal surfaces
The lining surfaces of body channels, such as the intestines or blood vessels.

Polyspecificity
The capacity to bind multiple unrelated substrates.
NaTure revIewS | EndoCrInology manner, consistent with active transport 29 . Cortisol export was later (in 1992) demonstrated in a porcine renal tubular cell line (LLC-PK1) overexpressing human ABCB1 (reF. 30 ). Since then, several endogenous and synthetic steroids have been confirmed as ABCB1 substrates. Depending on the presence of hydroxyl groups at positions 11 and 17, steroids were stratified into three categories 31 . ABCB1-mediated efflux was highest for steroids with both hydroxyl groups (including dexamethasone, cortisol and prednisolone), lowest for those with neither (deoxycorticosterone and progesterone), and intermediate in those with one hydroxyl group (including corticosterone and aldosterone). A-ring planarity and 6α-methyl substitution were reported to enhance transport when compared with passive diffusion in the LLC-PK1 line, in keeping with the presence of a critical hydrophobic pocket in the steroid-binding region 32 . Methylprednisolone is the glucocorticoid most effectively exported by ABCB1, followed by prednisolone, betamethasone, prednisone, dexamethasone, cortisol and cortisone [31][32][33] . Aldosterone appears to be weakly transported, and there is no evidence that sex steroids or 11-deoxycorticosterone undergo ABCB1-mediated export 31 , although progesterone does bind avidly to ABCB1 with an inhibitory effect 34 . Corticosterone -the predominant glucocorticoid in rats and mice -was initially shown to be an ABCB1 substrate on the basis of efflux from murine macrophage-like cells 35 , and subsequent in vitro work in murine adrenocortical cells has demonstrated that pharmacological ABCB1 inhibition blocked the ability of these cells to secrete corticosterone 36 . However, this work is in contrast to previous in vitro work showing that corticosterone was not exported in the murine LMCAT fibroblast line 31,[37][38][39] . Studies of the human transporter have not shown corticosterone to be transported by ABCB1, so affinity might be species specific 4,40 . Importantly, studies in murine thymoma cells overexpressing Abcb1 in which corticosterone and cortisol transport was compared showed a lower efflux of corticosterone compared with cortisol 31 , indicating an overall preference of this transporter for cortisol.

ABCC1 and steroid export
First identified and cloned as MRP1, ABCC1 was also discovered in multidrug resistance studies where high levels of expression are poor prognostic indicators in certain malignancies [41][42][43] . Since then, ABCC1 has been shown to efflux a diverse range of conjugated xenobiotics and physiological organic anions 44 . Like ABCB1, ABCC1 demonstrates a polarized distribution in epithelial cells, but is located on the basolateral rather than apical membrane 45 . ABCC1 is encoded by the human ABCC1 gene on the short arm of chromosome 16 (16p13.11). Strikingly, ABCC1 and ABCB1 share only 23% sequence identity, and differ substantially in their structural and physiological functions. To date, the structure of only bovine ABCC1 has been determined by cryo-electron microscopy 45 . The 190-kDa ABCC1 protein has 17 trans membrane α-helices distributed among three TMDs (TMD0, TMD1 and TMD2) rather than the two TMDs observed in ABCB1 45 (Fig. 1c). The binding site between TMD1 and TMD2 is 'bipartite': it has a positively charged 'P pocket' , which forms hydrogen bonds with glutathione residues, and an 'H pocket' , which interacts with hydrophobic moieties. This bipartite binding domain explains why glutathione coupling facilitates the transport of a wide range of compounds 45 .
ABCC1 substrates tend to be organic anions, whereas those for ABCB1 tend to be weak cations 45 ; and ABCC1 uniquely exhibits affinity for phase ii hepatic metabolites. There are differences in substrate preference between human and other mammalian isoforms -for example, the glucuronide conjugate of 17β-oestradiol is a substrate only in humans 46 . It has also been shown in vitro, both in virally transfected mouse fibroblast LMCAT cells and subsequently in human adipocytes, that ABCC1 can export corticosterone and 11-deoxycorticosterone, but not cortisol, prednisolone or dexamethasone 6,39 . And whereas ABCB1 is thought to transport substrates partitioning through the bilipid cell membrane (the 'hydrophobic vacuum') 47

ABCB1 and ABCC1 expression in tissues
The mRNA expression profiles of human ABCB1 and ABCC1 in various tissues are summarized in Fig. 2. ABCB1 is highly expressed (both at the mRNA and protein levels) in the adrenal gland, but is also found at absorptive surfaces (for example, in the intestines), protective barriers (for example, testis, BBB and placenta) and in secretory tissues (for example, biliary canaliculi and renal tubule) 23 . ABCC1 is widely expressed in almost all cell types, with the highest levels in the thymus, parathyroid glands and skeletal muscle. It seems to be poorly expressed in the liver 48 and nervous system but, notably, is found in greater quantities than ABCB1 in adipose tissue and skeletal muscle 23,49,50 .  23 . Tissues are ranked in order of ABCB1 to ABCC1 ratio, such that those towards the top of the y axis have greater ABCB1 expression, and those at the bottom have higher ABCC1 expression.
A model for the consequences of this tissuespecific transporter expression on the intracellular concentrations of different glucocorticoids is outlined in Fig. 3. Combining in vitro studies from three different laboratories, glucocorticoids can be separated into three groups depending on their relative propensity to be exported by ABCB1 and ABCC1 (reFS. 6,31,39 ). According to this model, the intracellular concentrations of cortisol will be lower in tissues that predominantly express ABCB1 (including the central HPA axis negative feedback sites behind the BBB), and those of corticosterone will be lower in tissues that predominantly express ABCC1, such as adipose tissue. Experimental data from animal and human studies can be used to show how this differential export might modulate the physiology of the HPA axis, influence lipogenesis within adipocytes and alter glucocorticoid transfer across the placenta.

ABCB1 and ABCC1 and the HPA axis
Insights from murine models. Central control of the HPA axis depends on feedback from circulating glucocorticoids to the hypothalamus and pituitary but, to reach the brain, the glucocorticoids must traverse the tightly packed endothelium of the BBB, where ABCB1 is found 51 . Murine models have been used extensively to assess ABCB1dependent modulation of steroid concentrations within tissues, including the brain. Importantly, rodents have two ABCB1 isoforms: ABCB1A (also known as MDR1A or MDR3) and ABCB1B (also known as MDR1B or MDR1) 52,53 , which broadly share the characteristics of the human protein 53 . Indeed, Abcb1a-knockout mice accumulate 87 times more of the ABCB1 substrate ivermectin in the brain than do wild-type animals 54 , while ABCB1 inhibition with tariquidar increases cerebral retention of labelled verapamil on PET imaging and demonstrates the role of ABCB1 at the human BBB 55 .
Abcb1a-knockout mice exhibit enhanced retention of cortisol and dexamethasone in the brain 4,5,54,56 . As seen in vitro, results for corticosterone export in vivo vary, perhaps reflecting redundancy between the murine isoforms. One study found no difference in the levels of infused radiolabelled corticosterone in the brain between adrenalectomized Abcb1a-knockout and wild-type mice 4 . However, Abcb1ab double-knockout mice retained an excess of cortisol and corticosterone in the brain 57 ; this retention was greater for cortisol than for corticosterone, suggesting that, overall, ABCB1 activity in mice favours cortisol transport over corticosterone transport, as was also found in vitro. However, another group found the opposite effect: a retention of both glucocorticoids in Abcb1a-knockout mice, and cortisol retention alone in the Abcb1ab double-knockout mice [58][59][60] . The authors highlight methodological differences between the studies which might limit comparisons: for instance, in one study isotope radioactivity rather than intact steroid concentration was measured, and the use of labelled corticosterone in adrenally intact animals might have resulted in isotope dilution by endogenously secreted corticosterone.
From these findings we might predict that the HPA axis would be relatively suppressed by the accumulation of glucocorticoids in the brain if ABCB1 activity is reduced. Indeed, Abcb1a-knockout mice do show evidence of HPA axis suppression, with lower basal and stress-stimulated levels of corticosterone, adrenocorticotropic hormone (ACTH) and corticotrophin-releasing hormone than control animals, with the effect localized to the hypothalamic level 61 . Furthermore, mice treated with the ABCB1 inhibitor tariquidar show an attenuated corticosterone response to stressful stimuli 62 .

Insights from dogs and humans.
The ABCB1 protein is well conserved in larger, cortisol-dominant species, with a notable exception being in collie-derived dogs. Like Abcb1a-knockout mice 54 , these animals are exquisitely sensitive to ivermectin owing to a 4-bp deletion mutation (termed Mdr1-1Δ), for which 40-50% of this breed are homozygous 63,64 . This mutation results in a severely truncated protein (<10% of normal length), which is predicted to be non-functional. Anecdotally, collie dogs are reported to recover relatively slowly from illness 65 , and animals with the MDR1 −/− genotype show chronic suppression of the HPA axis, with lower basal cortisol levels and greater ACTH suppression in response to dexamethasone than their wild-type counterparts. It has been hypothesized that enhanced brain retention of cortisol (the dominant canine glucocorticoid) leads to this HPA axis suppression, and predisposes the animals to a form of relative corticosteroid insufficiency 65 . This hypothesis has been supported by a metabolomics study demonstrating lower urinary cortisol metabolites in MDR1 −/− dogs than in control dogs (reaching  66 . In a study in humans, the corticosterone to cortisol ratio in brain autopsy specimens was five times greater than the corresponding ratio in plasma from age-matched and sex-matched healthy controls 4 . Similarly, the ratio of corticosterone to cortisol in live subjects is five or six times higher in cerebrospinal fluid than in plasma 67 . Many drugs, including verapamil and cyclosporin A, inhibit ABCB1, but their experimental use to test ABCB1 physiology in humans is hampered by toxicity at levels that are too low to carry out meaningful studies of ABCB1 inhibition 68 .

ABCB1 and ABCC1 modulate the HPA axis.
These results are all consistent with the hypothesis that ABCB1 at the BBB exports cortisol and thereby modulates negative feedback of the HPA axis in cortisol-dominant species. The absence of ABCC1 in the brain and BBB is consistent with corticosterone being retained to a greater extent than cortisol in the brain. One additional complexity, however, is that the pituitary gland (which expresses both transporters) 69 lies outside the BBB but also contributes to the control of the HPA axis. We have demonstrated that administration of probenecid, an inhibitor of ABCC1, induces greater tonic negative feedback of the HPA axis in healthy subjects than placebo as judged by elevations in ACTH and cortisol during combined mineralocorticoid and glucocorticoid receptor antagonism 70 . This finding is consistent with ABCC1 also contributing to the export of corticosterone from the pituitary gland or other central feedback areas, and warrants further investigation in animal models.

ABCC1 transporters in adipose tissue
In contrast to the BBB, where ABCB1 is more abundant than ABCC1, the reverse is true in adipose tissue. Glucocorticoids within adipose tissue induce lipogenesis; in particular, they stimulate the accumulation of lipids in visceral tissue and the production of adipokines 71 . Abcc1-knockout mice infused with corticosterone and cortisol show an enhanced accumulation of corticosterone but not cortisol in adipose tissue, accompanied by the upregulation of both glucocorticoid-responsive and adipogenic genes 6 .
We have also demonstrated that human adipocytes preferentially accumulate cortisol over corticosterone, and that this accumulation is reversed in vitro after treatment with the ABCC1 inhibitors probenecid or MK-571 (reF. 6 ). This accumulation is also accompanied by activation of glucocorticoid-responsive and adipogenic genes (PER1, ADIPOQ, ATGL and HSL) and results in the increased accumulation of fatty acids in lipid droplets 6 . Moreover, during infusion of cortisol or corticosterone in vivo in patients with primary adrenal insufficiency, the induction of glucocorticoid-responsive gene expression (PER1 and LPL) in adipose tissue was greater in response to cortisol than to corticosterone (achieved at plasma levels of glucocorticoid that were equipotent for ACTH suppression) 6 . These results suggest that corticosterone could have a more favourable metabolic profile than cortisol in glucocorticoid replacement, particularly when ACTH suppression is a target.

ABCB1 and ABCC1 in the placenta
As the interface between the mother and the fetus in pregnancy, the placenta functions both as a nutritive source and a barrier, including to glucocorticoid transport. The fetus is unable to synthesize cortisol until the third trimester, and therefore depends on maternal cortisol; however, although maternal cortisol levels increase severalfold during pregnancy, this increase is not transferred to the fetus indiscriminately 72 . In early pregnancy, excessive glucocorticoids are detrimental to the fetus, so the placenta provides a glucocorticoid barrier 73 , but it confers a more facilitative role towards term for fetal organ maturation 74 .
The placental glucocorticoid barrier: 11β-hydroxysteroid dehydrogenase 2. The enzyme 11β-hydroxysteroid dehydrogenase 2 is viewed as the main component of the placental glucocorticoid barrier, converting active cortisol to inactive cortisone 75 . However, the results of a study in which 11β-hydroxysteroid dehydrogenase 2 was inhibited during ex vivo perfusion of human placentas collected on ice immediately after delivery suggested that the enzyme might contribute only part of the glucocorticoid barrier, as cortisol transfer was restricted even at maximal inhibition of 11β-hydroxysteroid dehydrogenase 2 (reF. 76 ). Further consideration of the role of other mechanisms that are operating at the placental barrier, such as transmembrane transport, is therefore warranted.
The placental glucocorticoid barrier: ABCB1 and ABCC1. ABCB1 is located within syncytiotrophoblasts at the apical brush-border membrane, in direct contact with maternal blood 77 . It is highly expressed in early pregnancy and decreases towards term, consistent with the physiological role suggested above 78 . As occurs in other tissues, glucocorticoids have been shown to upregulate the expression of ABCB1 in the placenta in the first trimester, which might enhance the barrier effect 79 . Data demonstrating low concentrations of ABCB1 substrates (antiretrovirals, for example) in the fetal circulation both at birth and in the ex vivo perfused placenta indicate that ABCB1-mediated export towards the maternal circulation is active in vivo 80 . ABCC1 is located on the fetal-facing placental surface and has been identified in cytotrophoblasts, syncytiotrophoblasts and the fetal endothelium 81 . This localization might be consistent with a role in transferring ABCC1 substrates such as folic acid to the fetus and, in contrast to ABCB1, ABCC1 is upregulated towards term 81,82 . Studies of other ABCC1 substrates using the inhibitors probenecid and MK-571 have not demonstrated a clear effect on cross-placental transfer, so cannot be extrapolated to corticosterone transport 83 . It has been shown that the ratio of circulating cortisol to corticosterone is higher in the maternal circulation (15:1) than in the umbilical vein (7:1) at term 84 , which might be accounted for by differing fetal versus maternal adrenal cortisol:corticosterone secretion rates or by the facilitated transport of maternal corticosterone by ABCC1 into the fetal circulation.

Syncytiotrophoblasts
Cells forming the outer layer of the placenta, and the major site of gas and nutrient exchange between mother and fetus.

Cytotrophoblasts
The inner stem cell layer of the placenta villicellular precursors to syncytiotrophoblasts.

Regulation of ABCB1
The mechanisms underpinning the regulation of the expression of ABCB1 have been reviewed thoroughly elsewhere [85][86][87] . The ABCB1 promoter contains a number of areas of interest, including binding sites for the tumour suppressor p53, heat shock proteins and adopted orphan receptors, including the pregnane X receptor (PXR) and constitutive androstane receptor (CAR), which bind a number of xenobiotic ligands 88 . Xenobiotics, inflammatory mediators and cellular stress (such as irradiation, heat shock, hypoxia) typically upregulate ABCB1 expression through common pathways involving nuclear factor κ-B (NF-κB) and Y-box binding protein 89,90 . This upregulation appears to be a protective response, and polymorphisms in NF-κB are linked with increasing colon cancer risk, potentially owing to enhanced cellular exposure to toxins 91 .
Glucocorticoids modulate the expression of ABCB1 mRNA and protein in rodents and humans. This modulation has been demonstrated across multiple tissues using dexamethasone, prednisolone, cortisol, methylprednisolone and some inhaled glucocorticoids 33,79,[92][93][94][95][96][97] . Although glucocorticoids predominantly induce the expression of ABCB1, this effect might be specific to some species or cell types, as there are also instances of ABCB1 expression being downregulated 98 . This glucocorticoid effect is inhibited in the presence of the glucocorticoid receptor blocker RU486, indicating that this effect is at least partly mediated via the glucocorticoid receptor but, as no consensus glucocorticoid response element has been found in the human ABCB1 promoter, it is assumed to be an indirect genomic effect. Dexamethasone-mediated upregulation of ABCB1 in retinal pigment epithelium was found to be abolished when the PXR receptor was silenced, implying that PXR (which does contain a consensus glucocorticoid response element) is either a co-regulator or a target of the glucocorticoid receptor 97,99,100 . This upregulation of expression raises concerns about increased drug efflux when glucocorticoids are used in combination with other ABCB1 substrates (as often occurs in chemotherapy protocols), and is theorized to be a cause of glucocorticoid resistance in conditions such as asthma 33 . However, this effect has also been exploited clinically -for example, methylprednisolone is used in the treatment of paraquat toxicity to increase excretion of the drug 101 .
The regulation of ABCB1 in inflammation is complex and potentially biphasic. Evidence from rodent studies indicates that, in the very early stages of inflammation, ABCB1 is functionally inhibited by lipopolysaccharides and inflammatory cytokines, despite mRNA levels remaining constant, perhaps owing to ABCB1 being trafficked away from the cell membrane. However, later in the evolution of inflammation, ABCB1 mRNA and protein levels are upregulated by the cytokines tumour necrosis factor and endothelin 1 converging on the NF-κB pathway 89 . Protein turnover at the cell surface under normal conditions is relatively slow (the half-life of ABCB1 is estimated at just over 24 h) 102 and there might be a role for post-translational and other mechanisms in modulating this turnover. Taken together, this evidence suggests that in times of increased physiological stress (for example, in response to illness or injury), ABCB1 can be upregulated both by stress-activated glucocorticoids and by signals released by cellular damage. This upregulation might result in positive feedback on cortisol production by further restricting glucocorticoid access to sites of higher negative feedback.

Regulation of ABCC1
Most research on factors affecting the expression levels of ABCC1 and its protein activity relates to cancer biology and chemotherapeutics, whilst physiological regulation has been poorly studied to date. Basal transcription of ABCC1 is stimulated by the SP1 transcription factor 103 which is, in turn, inhibited by the tumour suppressor protein p53 (reF. 104 ). It has not been clearly established whether PXR affects ABCC1 transcription 105,106 and, although early mapping of the ABCC1 promoter in a human leukaemic cell line did reveal a putative glucocorticoid response element, dexamethasone has not been shown to alter ABCC1 expression in the human placenta or in lymphocytes 94,[107][108][109] . Furthermore, we cannot clearly conclude whether ABCC1 is affected by acute inflammation in the same way that ABCB1 is, as both unchanged and increased mRNA expression have been found in response to mediators such as lipopolysaccharide, tumour necrosis factor, IL-1 and IL-6 (reFS. [110][111][112] ).
In vitro studies investigating the metabolic regulation of ABCC1 have focused on endothelium, and have demonstrated that expression of the transcript is downregulated in a hyperglycaemic environment 113 . Metformin, a drug commonly used in the treatment of type 2 diabetes mellitus, is known to reduce ABCC1 expression in a human hepatocellular carcinoma cell line through the AMP-activated protein kinase-hypoxia-inducible factor 1 pathway 114 .
Whilst limited, overall this evidence suggests that ABCC1 is regulated differently from ABCB1, and is predominantly responsive to metabolic and immunomodulatory signals rather than to mediators of acute stress or inflammation.

Adopted orphan receptors
An orphan receptor is a receptor whose ligand has not been identified; it can later be termed an 'adopted orphan receptor' when a ligand is discovered.

Box 1 | Multidrug resistance
multidrug resistance describes the ability of malignant cells to evade the actions of a broad range of chemotherapeutic agents. Tumours, which are initially very sensitive, can become resistant to multiple agents over the course of the disease, ultimately resulting in treatment failure and disease progression. There are several potential reasons for this, but the key mechanism is increased drug efflux out of malignant cells by membrane transporters, particularly those of the aTP-binding cassette (abC) family. Some tumours have innately high levels of transporter expression; others develop high levels after exposure to chemotherapy 151 .
abCb1 is the transporter that is most widely associated with multidrug resistance, particularly as alkylating agents, anthracyclines and vinca alkaloid drugs are all substrates 21 . abCC1 and abCG2 (abC subfamily G member 2, also known as breast cancer resistance protein) are also implicated in multidrug resistance.
as examples, survival rates from lung cancer, multiple myeloma and acute myeloid leukaemia are inversely associated with the levels of expression of abCb1 (reFS. [152][153][154] ). High levels of abCC1 expression are associated with poor outcomes in childhood neuroblastoma 155 , whilst overexpression of abCG2 is a negative prognostic factor in pancreatic ductal adenocarcinomas 156 .

Pathological dysregulation
There have been few studies of variations in ABC transporter expression beyond the descriptions in vari ous cancers, as mentioned (Box 1). A transcriptomic analysis utilizing single-cell RNA sequencing showed upregulation of ABCB1 in the adrenal cortex of patients with ACTH-dependent Cushing disease 36 . This upregulation probably reflects the effects of glucocorticoids on ABCB1 expression, but might contribute to pathogenicity by further enhancing the export of cortisol from the gland. Hypothesizing that steroid retention in adipocytes owing to low levels of ABCC1 could be a driving mechanism for obesity, we actually found that ABCC1 mRNA levels are upregulated in the adipose tissue (subcutaneous and visceral) of individuals with obesity compared with the levels in lean individuals, which might reduce glucocorticoid concentrations in adipocytes, although this reduction might only be true for corticosterone 6 .

Lessons from human genetics
Germline mutations in ABCB1. Human germline mutations in ABCB1 are rare. To our knowledge, there are only two publications of ABCB1 mutations: twin girls with recurrent reversible toxic encephalopathy alongside febrile illness 115 , and a 13-year old boy with ivermectin sensitivity 116 . In both cases, the mutations were identified by whole-exome sequencing and showed compound heterozygosity. The twin girls were found to have a nonsense mutation (p.Pro1182X) combined with a splice variant (c.2786+1G>T) and showed markedly enhanced central nervous system retention of 11 C-verapamil on PET imaging in comparison with their parents. Their symptoms were suspected to be caused by enhanced retention of inflammatory mediators within the brain during acute illness, and a mouse model showed that the cytokines tumour necrosis factor, IL-1, IL-6 and CC-chemokine ligand 2 (CCl-2) were retained in the brain 24 h after lipopolysaccharide injection in Abcb1abknockout animals, whereas they had completely cleared from the brain in wild-type animals. The investigators estimated from studies in lymphocytes derived from the sisters that only ~10% of functional ABCB1 protein was expressed. In the other case, the affected boy presented with severe adverse neurological effects after a single oral dose of ivermectin to treat scabies and was found to have inherited a nonsense mutation in ABCB1 from each parent (c.2380C>T and c.3053_3056delTTGA), both of which are predicted to result in the loss of the carboxyterminal nucleotide-binding domain. The boy and twin girls were otherwise healthy and growing normally.

Germline mutations in ABCC1.
Similarly, there is only one published mutation in ABCC1 of clinical significance: a heterozygous missense mutation (c.1769A>G) identified as causing familial sensorineural deafness 117 . ABCC1 has been found within the rodent cochlea, where it could be protective against neurotoxins 118 . This mutation is thought to disrupt hydrogen bonds, and thus stability between the helices of the TMDs in the proteins, but analysis of lymphoblastoid cell lines derived from affected family members showed loss of around 40-45% of ABCC1 mRNA expression when compared with those unaffected, suggesting an additional impairment in mRNA stability. Extrusion of SNARF-1, a known ABCC1 substrate, from lymphoblastoid cells as a measure of transport activity was subsequently shown to be slower 117 .
Polymorphisms in ABCB1 and ABCC1. With nonsense and frameshift mutations being rare, there have been attempts to correlate common polymorphisms with clinically relevant outcomes (reviewed in reF. 119 ).
Three ABCB1 variants are common in humans: c.2677G>A/T, c.3435C>T and c.1236C>T. The c.3435C>T allele is synonymous, but might affect mRNA stability 120 ; c.1236C>T is silent; and c.2677G>A/T results in an amino acid substitution (alanine to serine or threonine), which could potentially result in substrate changes. There is marked variation in the frequency of these polymorphisms across different races; for example, c.3435C>T is much less common in African populations (~80% of people from West Africa are homozygous for the C allele versus ~20% of individuals from western Europe) 120,121 . However, it has not been convincingly demonstrated that these variants affect substrate transport; for instance, levels of the ABCB1 substrate digoxin have been found to be increased, decreased and unchanged in the plasma of individuals with these polymorphisms. Subsequent attempts to correlate polymorphisms with response to chemotherapeutics, adverse drug effects, and resistance to anti-retroviral and anti-epileptic therapies have been similarly inconclusive [122][123][124] .
Studies of the HPA axis in individuals with ABCB1 variants have unfortunately been inadequately powered. No differences were found in the levels of evening cortisol and ACTH in 30 Japanese men with C/C, C/T or T/T c.3435 genotypes (the variant associated with potentially reduced transporter mRNA stability). However, another study, of 51 women, found lower levels of cortisol in the plasma, taken at 6 p.m., in individuals with one or two copies of the T allele compared with C/C controls; these lower levels reached significance only in the follicular menstrual phase, so an interaction with sex hormones was proposed 125,126 . In one candidate gene study of over 5,000 Japanese individuals, the c.2677G>A/T variant was highly associated with increased BMI, which could potentially reflect increased HPA axis activity, whilst in a study in 154 individuals with depression, the response of cortisol (but not ACTH) to corticotrophin-releasing hormone was lower in c.2677 TT homozygotes than in those with the major allele (GG) or heterozygotes (TG), which was taken to reflect reduced adrenal cortisol release 36,127 . However, neither the plasma level of cortisol nor BMI has been associated with any ABCB1 polymorphisms in larger cohorts.
Genetic studies have also been undertaken in patients taking exogenous steroids. In a cohort of 171 patients requiring long-term treatment with glucocorticoids for adrenal insufficiency, those patients with the c.3435 TT genotype had lower bone density than the those with the CC or CT genotype, suggesting greater systemic steroid absorption or enhanced bone retention 128 . There have been attempts to correlate glucocorticoid treatment outcomes in patients with rheumatoid arthritis, inflammatory bowel disease, immune thrombocytopenic

Compound heterozygosity
The presence of two different mutant alleles at a genetic locus.

Lymphoblastoid cell lines
immortalized cells that are derived from, and closely resemble, peripheral blood lymphocytes.

Synonymous
A silent genetic mutation in which a change in DNA sequence does not result in a change in the amino acid sequence of the protein produced. purpura and nephrotic syndrome with ABCB1 polymorphisms [129][130][131][132] . Most, but not all, results indicate a higher steroid response in those with the minor allele of the studied polymorphism, but studies are limited by sample size and a failure to control for multiple testing.

NaTure revIewS | EndoCrInology
Documented polymorphisms for ABCC1 are mostly rare and non-coding, and have not been assessed in the context of HPA axis activity or metabolism 133 . Three poly morphisms might predict the outcome of acute myeloid leukaemia, but any corresponding effect of these polymorphisms on transporter expression or function has so far not been established 134 .

Implications and future research
The observations that two ABC transporters influence the retention of glucocorticoids in tissues allow us to add membrane transporters to the list of factors that are involved in the metabolism of glucocorticoids at the pre-receptor level (Fig. 4). These observations provide insights into HPA axis physiology and how corticosterone and cortisol might carry out different functions in species that produce both of these steroids. These findings also provide opportunities for anti-inflammatory and physiological replacement steroid therapies that might better target tissues mediating efficacy while avoiding those mediating toxicity.

Revised glucocorticoid physiology
In rodents, the lack of steroid 17-hydroxylation means that corticosterone is necessarily the sole endogenous glucocorticoid 135 . In humans and other species in which both glucocorticoids circulate, it is common to consider them interchangeable. Indeed, cortisol and corticosterone share similar metabolic pathways (for example, susceptibility to metabolism by 11β-hydroxysteroid dehydrogenase enzymes) and affinities for the glucocorticoid and mineralocorticoid receptors [136][137][138][139] . However, corticosterone does exhibit differences from cortisol, including more rapid clearance from the circulation, and a greater response to ACTH, such that the corticosterone to cortisol ratio rises under stress [140][141][142] .
The findings outlined in this Review further demonstrate that cortisol and corticosterone are not interchangeable with respect to glucocorticoid action. Specifically, in tissues where ABCB1, but not ABCC1, is present, such as the brain, cortisol concentrations are constrained by export back into the circulation and corticosterone can play a disproportionate role. Conversely, in tissues such as adipose where ABCC1, but not ABCB1, is expressed, corticosterone is exported and the response to cortisol can be disproportionate (Fig. 5). This observation raises the concept of a distinctive role for corticosterone in mediating HPA axis negative feedback. In the stressed state, the ability to restrict high circulating levels of cortisol from accessing higher centres might prevent axis suppression and facilitate recovery, as demonstrated by the Mdr1-1Δ collie dogs, which lack this capacity 65 . It is recognized in other species that the ratio of cortisol to corticosterone and the peak levels of circulating glucocorticoids vary seasonally 143 , possibly in response to photoperiod length. If corticosterone is more accessible to negative feedback sites, and less peripherally anabolic than cortisol (in terms of effects on adipose tissue), then the energy-expending stress response might be restrained and access to vital adipose energy stores when food is scarce might be improved. Conversely, with a slower turnover than corticosterone in the circulation and adipose tissue than in other tissues such as the brain and liver 144 , cortisol might provide the option for medium-term adjustments, in comparison with the acute changes mediated by corticosterone.
Understanding the implications of the differential control and actions of cortisol and corticosterone in glucocorticoid physiology will require a detailed dissection of the dynamics of ligand availability for receptors within human target tissues in vivo. The increasing use of exome-wide sequencing in clinical as well as research settings might well identify further individuals or families with significant ABCB1 and ABCC1 mutations and offer new routes to addressing these key physiological issues.

Novel glucocorticoid therapies
A major limitation of current glucocorticoid therapies is their narrow therapeutic index. Despite extensive efforts, it has proved difficult to develop selective glucocorticoid receptor modulators with pharmacodynamic interactions that discriminate between efficacious and toxic gene transcriptions 145 . An alternative approach depends on the premise that efficacious and toxic effects are often mediated in different tissues, suggesting that the therapeutic index could be improved by modifying the pharmacokinetics of steroid drugs to 'target' them to the tissues These processes restrict access to the nuclear glucocorticoid receptor (GR) and/or the mineralocorticoid receptor (MR), which mediate the cellular response (5).

Therapeutic index
The margin between the desirable and undesirable effects of a drug; the narrower the margin, the more likely it is that adverse effects will occur at a therapeutic dose.

www.nature.com/nrendo
where efficacy is mediated while avoiding tissues where toxicity is mediated. Could this be achieved by using steroids with different affinities for the ABCB1 and ABCC1 transporters? When considering physiological replacement in patients with adrenal insufficiency, the challenges of this narrow therapeutic index are well documented, with adverse outcomes including, but not limited to, obesity, osteopenia and insulin resistance attributable to the steroid treatment regimen of these patients 146,147 . Such challenges are particularly evident in patients with congenital adrenal hyperplasia (CAH), in whom doses of glucocorticoid that achieve adequate adrenal androgen suppression are invariably associated with morbidity 146 . All glucocorticoids currently used to replace cortisol (hydrocortisone, prednisolone, dexamethasone and the active metabolites of pre-drugs cortisone and prednisone) are substrates for ABCB1, but not ABCC1. Although pharmacokinetic adjustments such as delayed release preparations might confer some benefits 148,149 , they cannot overcome the closeness of the dose-response relationship between efficacy and toxicity, and the prospect of choosing a glucocorticoid based on affinity for ABCC1 over ABCB1 is an intriguing therapeutic prospect.
As one such glucocorticoid, corticosterone is not currently available in an oral form, but our experimental work using intravenous corticosterone has provided proof-of-concept of the potential advantages of corticosterone in avoiding harmful metabolic effects mediated in adipose tissue. As described earlier, infused cortisol induced a greater response of glucocorticoid-responsive gene expression compared with infused corticosterone in the adipose tissue of patients with Addison disease 6 . In a similar study, 14 individuals with CAH also underwent ramped cortisol and corticosterone infusions; despite higher plasma levels of corticosterone than cortisol being achieved, the amount of insulin releaseda marker of glucocorticoid effect on adipose to induce insulin resistance -was greater in response to cortisol than to corticosterone 150 .
The potential for glucocorticoid therapies that avoid toxicity in metabolic tissues deserves further investigation but would require the generation of an oral corticosterone preparation for practical administration to patients.

Conclusions
We have collated evidence from cell, animal and human studies that the ABC transporters ABCB1 and ABCC1 differentially export cortisol, corticosterone and synthetic glucocorticoids from tissues and contribute to pre-receptor glucocorticoid regulation. Differing transporter expression profiles in the brain, placenta and adipose confer different tissue sensitivities to these steroids, which might be important for optimizing the responsiveness of the HPA axis, controlling fetal exposure to steroids throughout gestation, and optimizing adipose fuel metabolism. Although much is known about these transporters in the context of multidrug resistance, their physiological roles and regulation remain largely unexplored. The prospect of developing steroid therapies with transporter affinities that are tailored to give improved efficacy, without deleterious peripheral toxicity, offers new avenues for exploration for the management of inflammatory and endocrine diseases. Glucocorticoids are secreted from the adrenal cortex upon stimulation by signals from the hypothalamus and pituitary. They act peripherally on sites throughout the body, and feed back to the hypothalamus, pituitary and higher centres to maintain homeostasis. ABCB1 present at the blood-brain barrier might act to restrict the access of cortisol to feedback sites. Conversely, ABCC1, which is found without ABCB1 in adipose and skeletal muscle, exports corticosterone but not cortisol. The activity of the adrenal enzyme CYP17 (17-hydroxylase) determines the ratio of secreted cortisol to corticosterone.