Both in vivo studies, involving the use of mutant mouse models, and in vitro cell culture experiments have established scavenger receptor class B, type I (SR-B1) as a key a mediator for uptake of β-carotene from the intestinal lumen into the enterocyte [11,12,13]. Van Bennekum et al. studied cholesterol and β-carotene uptake by WT and SR-B1 knockout mice and concluded that SR-B1 is required for β-carotene absorption, at least for mice consuming a high fat diet [11]. These authors further showed that both SR-B1 and the plasma membrane fatty acid transporter CD36 can facilitate absorption of dietary cholesterol, but van Bennekum et al. were unable to establish whether SR-B1 acts essentially in vivo in facilitating this cholesterol uptake [11]. No evidence was obtained that CD36 acts in facilitating β-carotene absorption. SR-B1 expression in transfected COS-7 cells [11] and in intestinal Caco-2 cells [12] was found to confer on these cells the ability to take up β-carotene from mixed bile salt micelles, phospholipid small unilamellar vesicles, and triglyceride emulsions, thus, providing further evidence for a role for SR-B1 in β-carotene absorption. In addition to facilitating mucosal uptake of the proretinoid carotenoid β-carotene, SR-B1 also acts in facilitating uptake into the enterocyte of the non-proretinoid carotenoids lycopene and lutein [14,15].

Retinol uptake into cultured intestinal Caco-2 cells has been reported by During and Harrison to occur via both a saturable process, when retinol was provided at concentrations below 10 µM, and a nonsaturable process at higher retinol concentrations [12]. Expression of SR-B1 is not required for retinol uptake into Caco-2 cells, since knockdown of SR-B1 expression with small inhibitory RNAs (siRNAs) failed to influence retinol uptake by the cells [12]. Interestingly, inhibition of expression of the cholesterol efflux transporter ATP-binding cassette, sub-family A, member 1 (ABCA1) in Caco-2 cells through use of either siRNAs or the drug glyburide, which inhibits transport activity of ATP-binding cassette family members including ABCA1, diminished retinol efflux from the basolateral surface of the polarized cultures of Caco-2 cells [12]. These findings led to the proposal that retinol efflux from enterocytes is probably partially facilitated by the basolateral cholesterol transporter ABCA1. However, later studies by Reboul et al., making use of both Caco-2 cells and ABCA1-deficient mice, failed to confirm this earlier observation [16]. Although Reboul et al. were able to convincingly demonstrate a role for ABCA1 in facilitating absorption of both α- and γ-tocopherol, their data provide no support for the idea that ABCA1 contributes to retinoid efflux, in nascent chylomicrons, from enterocytes [16].

The enzymes involved in converting proretinoid carotenoids to retinoid were long a matter of heated research controversy [5,6]. From studies undertaken in the late 1950s and early 1960s, it was clear that enzymes existed within mammalian tissues that are able to cleave β-carotene, either symmetrically at its central 15,15′ carbon-carbon double bond, forming two molecules of retinaldehyde, or asymmetrically at other carbon-carbon double bonds, forming two products of unequal chain length (see Figure 2 below). Largely owing to technical difficulties in purifying and cloning these enzymes, from the 1970s through the 1990s, there was considerable debate as to whether only the central cleavage reaction contributed to retinoid formation or whether asymmetric cleavage also gave rise to quantitatively significant retinoid formation. This controversy was resolved in the last decade with the cloning and study of cDNA for two gene products, Bcmo1 and Bcmo2, which encode enzymes able to catalyze central and asymmetric β-carotene cleavage, respectively. cDNAs for Bcmo1 have now been cloned for the chicken [17], mouse [18,19,20], rat [21], and human [22,23], as well as for the fruit fly [24] and zebrafish [25]. Kiefer et al. reported cloning cDNAs for Bcmo2 from the fruit fly, zebrafish, and the mouse and human [26]; whereas, Hu et al. reported cloning the ferret cDNA [27]. It is now established that only the Bcmo1 gene product, which encodes an enzyme able to catalyze the central cleavage of β-carotene, is significant within the body for mediating retinoid formation from dietary proretinoid carotenoids [28,29].

Two structurally related proteins, β-carotene-15,15′-monooxygenase (BCMO1), encoded by Bcmo1, and β-carotene-9′,10′-monooxygenase (BCMO2), encoded by Bcmo2, are the sole mammalian enzymes known to cleave carotenoids. BCMO1 and BCMO2 are highly homologous to each other, with mouse BCMO1 and BCMO2 sharing 39% sequence identity. Each has a high degree of sequence homology to the retinal protein RPE65, which catalyzes the isomerization of all-trans-retinoid to the 11-cis-isomer for use in the visual cycle (see Section 5.3 below for more details). The older literature, based on the study of only partially purified enzyme, referred to BCMO1 as β-carotene-15,15′-dioxygenase, rather than as a monooxygenase [3,4,5,6]. However, a reevaluation of the reaction mechanism for this enzyme, employing recombinant protein, indicated that it acts through a monooxygenase, rather than a dioxygenase mechanism [30]. Biochemical data obtained by different laboratories regarding the properties and expression of mammalian BCMO1 are in general agreement [17,18,19,20,21,22,23]. BCMO1 is a soluble, Fe²⁺-containing, 63 kDa protein. It is expressed in the small intestine (at higher levels more proximal to the stomach), liver, kidney, lungs, skin, testis, the retinal pigment epithelium within the eye, and in a number of embryonic tissues.

The biochemical properties and expression of BCMO2 have been less extensively studied than those of BCMO1. Like BCMO1, BCMO2 contains Fe²⁺ and has a cytoplasmic localization within the cell [26,27]. Both recombinant mouse and ferret BCMO2 catalyze cleavage of β-carotene at its 9′,10′ carbon-carbon double bond, and both will catalyze lycopene cleavage at its 9′,10′ carbon-carbon double bond [26,27]. In the mouse, Bcmo2 is expressed in small intestine, liver, kidney, spleen, brain, and heart [26]. A similar tissue distribution has been reported for the ferret [27].

The gene for Bcmo1 and the regulation of its expression have been studied by a number of laboratories. The mouse and human genes for Bcmo1 have both been shown to contain functional peroxisome proliferator-activated receptor (PPAR) response elements (PPREs) [31,32]. The PPRE present in the mouse gene is located within 60 bp upstream of the start site, and deletion or mutation of this element reduces promoter activity to its basal level [31]. Electrophoretic mobility shift assays established that PPARγ specifically binds this site, and administration of the PPARα/γ agonist WY14643 stimulated promoter activity in reporter assays, as well as increased BCMO1 protein expression in livers of mice administered the agonist. Using similar experimental approaches, Gong et al. established that the human Bcmo1 gene contains a functional PPRE in its proximal promoter [32]. Moreover, these investigators demonstrated that the proximal promoter of the human Bcmo1 gene also contains a functional myocyte enhancer factor 2 (MEF2) binding element and reported data suggesting that MEF2C, one member of the MEF2 transcription factor family, and PPARγ interact synergistically to transactivate Bcmo1 expression.

Other transcription factors may also play important roles in regulating Bcmo1 expression, especially in intestine. Takitani et al. reported that intestinal Bcmo1 mRNA expression was markedly increased in rats fed a retinoid-deficient diet and that expression was suppressed upon feeding of either all-trans- or 9-cis-retinoic acid [21]. However, although Takitani et al. demonstrated actions of retinoic acid in modulating intestinal Bcmo1 expression, they couldn’t explain the molecular basis for their observation. Further insight was provided by subsequent work from Seino et al. [33] and Lobo et al. [13]. Seino et al. reported that the intestine-specific transcription factor Isx (intestine specific homeobox) plays a key role in regulating expression of Bcmo1 in the mouse intestine [33]. These investigators, who generated and studied Isx-deficient mice through knock-in of LacZ, convincingly established that mRNA levels for both Bcmo1 and SR-B1 are greatly increased in the intestines of Isx-knockout mice. They further showed that severe vitamin A-deficiency markedly decreased Isx expression and that this was accompanied by an increase in Bcmo1 expression in both duodenum and jejunum. Based on their data, Seino et al. suggested that Isx participates in the maintenance of retinoid metabolism by regulating Bcmo1 expression in intestine. Lobo et al. carried this idea further by showing that retinoic acid, acting through retinoic acid receptors, induces Isx expression [13]. This effect of retinoic acid on Isx expression resulted in repression of both the Bcmo1 and SR-B1 genes. Through study of BCMO1-deficient mice, Lobo et al. were also able to establish that increased SR-B1 expression and systemic β-carotene accumulation could be prevented through administration of dietary retinoid, which induced Isx expression, resulting in a downregulation of SR-B1 expression and β-carotene uptake and systemic accumulation. Thus, the work of Lobo et al. established the existence of a diet-responsive regulatory network that controls β-carotene absorption and retinoid production through negative feedback regulation of Isx [13].

A number of single nucleotide polymorphisms (SNPs) in the human Bcmo1 gene have been identified. One of these, a T170M missense mutation, results in a 90% reduction in enzyme activity when analyzed in vitro using purified recombinant enzymes [34]. A person identified as being heterozygous for the T170M mutation was reported in an earlier study to possess very high levels of serum β-carotene (14.8 µM), even though consuming only a typical Western diet not supplemented with β-carotene. Two common SNPs in the Bcmo1 gene, R267S and A379V, alter β-carotene uptake in female volunteers receiving a pharmacologic dose (120 mg) of β-carotene [35]. Both variant alleles, when studied as recombinant proteins, showed reduced catalytic activity towards β-carotene. Carriers of both the 379V and 267S + 379V variant alleles displayed reduced ability to convert β-carotene, as indicated through reduced retinyl palmitate:β-carotene ratios in the triglyceride-rich lipoprotein fraction and increased fasting plasma β-carotene concentrations. Collectively, these data from SNP studies provide a compelling molecular explanation for why some individuals may be better able than others to take up and/or convert proretinoid carotenoids to retinoids.

The structure and regulation of the Bcmo2 gene have been much less extensively studied than is the case for Bcmo1. Mutations in the Bcmo2 gene have been identified, and these influence carotenoid accumulation in chickens [36], cows [37] and sheep [38]. In domestic chickens, yellow skin coloration was found to be caused by a regulatory mutation that prevents expression of the Bcmo2 gene in skin, allowing for accumulation of yellow carotenoids [36]. In cows, a mutation giving rise to a premature stop codon in the Bcmo2 gene results in increased β-carotene concentrations in both serum and milk [37]. In sheep, a nonsense mutation in the Bcmo2 gene is strongly associated with high levels of carotenoids deposited in fat, resulting in a yellow fat phenotype [38].

Newly absorbed dietary retinol within the enterocyte must be esterified prior to its packaging as retinyl ester in nascent chylomicrons. The older literature had indicated that the intestine possesses two distinct enzyme activities able to synthesize retinyl esters from retinol [3,4,5,6]. One of these, lecithin:retinol acyltransferase (LRAT), catalyzes the transesterification of retinol employing a fatty acyl group present in the A1 position of a membrane phosphotidyl choline molecule. The other, acyl-CoA:retinol acyltransferase (ARAT), catalyzes the fatty acyl-CoA-dependent esterification of retinol. Studies by O’Byrne et al. of LRAT-deficient mice established, for mice receiving a physiologic dose of retinol (6 µg), that LRAT-catalyzed retinol esterification accounted for approximately 90% of intestinal retinyl ester formation [39]. Chylomicrons isolated from the dosed LRAT-deficient mice contained some retinyl ester, which was presumably synthesized by an intestinal ARAT, and relatively high levels of the free alcohol retinol, which were not observed in chylomicrons obtained from WT mice. Subsequent studies by Wongsiriroj et al. established that the enzyme diacylglycerol acyltransferase 1 (DGAT1), which catalyzes triglyceride synthesis from diacylglycerol and fatty acyl-CoA, acts as a physiologically significant ARAT in the mouse intestine [40]. Normally, for a physiological dose of retinol, DGAT1 accounts for approximately 10% of the retinol esterified in the intestine. However, the contribution that DGAT1 makes to intestinal retinol esterification becomes considerably greater upon administration of a large pharmacologic dose of retinol (1000 µg) [40]. Under physiologic conditions, LRAT and DGAT1 account for all retinol esterification within the enterocyte since no retinyl esters could be detected in chylomicrons isolated from LRAT/DGAT1-double knockout mice given a physiologic dose of retinol. However, when LRAT/DGAT1-double knockout mice were administered a pharmacologic dose of retinol, some retinyl ester could be detected in chylomicrons isolated from these mice, indicating the existence of other minor ARAT activities that can become active in response to excessive retinol intake [40]. In summary, LRAT accounts for the great majority of retinyl ester formed in the enterocyte upon consumption of normal dietary levels of retinoid; DGAT1, an intestinal ARAT, accounts for the remaining esterification activity.

Retinoids are very insoluble in water and consequently within the aqueous environment of the body they are usually found bound to specific retinoid-binding proteins (see Table 1 below for a listing of retinoid-binding proteins). In the adult, CRBPII is reported to be expressed solely in the intestinal mucosa and is proposed to facilitate optimal retinol absorption from the diet [41]. Within the enterocyte, CRBPII represents 0.4–1.0% of the total cytosolic protein [42]. To investigate the physiological role of CRBPII, Li and colleagues generated and studied CRBPII-deficient mice [43]. When maintained on a retinoid-enriched diet, the knockout mice were found to have reduced (by 40%) hepatic retinoid stores, but the mutant mice grew and reproduced normally. However, when maternal dietary retinoid levels were reduced to marginal levels during the latter half of gestation, a 100% mortality rate, within 24 hours after birth, was observed for these litters [43]. These studies convincingly demonstrate that CRBPII acts to ensure adequate delivery of retinol to the developing fetus when dietary retinoid is limiting. Subsequent investigations making use of CRBPII-deficient mice bred into the LRAT-deficient background (lacking both CrbpII and Lrat) established that CRBPII metabolically channels retinol to LRAT for retinyl ester synthesis [40]. However, it could not be demonstrated experimentally that CRBPII directly prevents retinol from being acted upon in vivo by intestinal DGAT1 or other intestinal ARAT activities, as had been proposed in the older literature [4,5,6].

When the retinyl ester-containing chylomicron remnant arrives at the liver, it passes into the space of Disse (located between the endothelium and the hepatocyte) in a process referred to as sieving [57]. Only remnants of appropriate size can pass through, while larger particles, including whole chylomicrons, are excluded. Once inside, the remnant is taken up exclusively by hepatocytes by one of two possible receptor-mediated pathways. The topic of receptor-mediated remnant uptake has been extensively and well reviewed by Cooper, and the reader is referred to this work for more detail [58]. Cooper recounts that one receptor-mediated pathway involves direct uptake by the low density lipoprotein (LDL) receptor, which has a high affinity for the apoE-rich chylomicron remnant particles, and internalization via endocytosis. If the LDL receptor is absent, down-regulated or saturated, the remnants may be sequestered in the space of Disse by binding to heparin sulfate proteoglycans (HSPGs), mediated by apoE. The remnants may also be sequestered through binding to hepatic lipase, which is enhanced by the presence of apoB. The remnants may eventually be transferred to LDL receptors as they become available or, if the remnants acquire enough apoE, transferred to an alternative receptor, the LDL receptor-related protein (LRP). LpL, acquired by the remnants during their formation, can facilitate uptake of remnants by LRP. It has not yet been established the extent to which each receptor contributes to chylomicron remnant-retinyl ester removal, but it likely depends on the metabolic state of the animal, which in turn can influence the amount of apoE being secreted.

The importance of HSPGs in chylomicron remnant uptake was demonstrated by Zeng et al. who showed that remnant binding to the hepatocyte is dependent on both the expression of HSPG core proteins and the functionality of HSPG heparin sulfate chains [59]. Remnant binding in HepG2 cells was significantly decreased by antisense oligonucleotide knockdown of HSPG, antibodies to heparin sulfate, heparinase treatment and by various disruptions of the heparin sulfate chains, including inhibition of glycosylation. More recently, it’s been shown in vivo that the primary HSPG core protein involved in hepatic clearance of remnants is syndecan-1, which facilitates lipoprotein binding in the space of Disse [60].

Though less understood, evidence for a third hepatic remnant uptake pathway is emerging. Studies in normal and apoE-deficient mice fed low- or high-fat diets showed that plasma clearance of chylomicron remnants is delayed with non-lipolyzed particles and is inhibited by lactoferrin, which blocks LRP [61]. When lipolysis was restored with the addition of hepatic lipase, no difference in uptake was observed between the two diet groups or in the presence and absence of apoE. This provided evidence for an apoE-independent pathway that functions through LRP. This idea was further investigated by Out et al., who proposed a mechanism involving interactions between remnant phospholipids and a receptor on the surface of hepatocytes, SR-BI [62]. In SR-BI-deficient mice, triglyceride-rich chylomicron-like particles associate significantly less with isolated hepatocytes compared to WT mice, with a concomitant delay in postprandial triglyceride clearance [62]. Adenovirus-mediated hepatic overexpression of SR-BI significantly decreased serum cholesterol, phospholipids and TG [63]. While a role for SR-BI in chylomicron remnant metabolism in the liver is likely, it remains to be determined if and how it may interact with the other receptor systems.

Upon entry into the hepatocyte, the retinyl ester is associated with early endosomes and undergoes rapid hydrolysis [64]. The hydrolysis of retinyl ester is carried out by a number of enzymes referred to in the literature as REHs, carboxylesterases and/or lipases.

The first well-characterized REH was the bile salt-dependent CEL (see above for more details regarding CEL actions in the intestine), which is a potent bile salt-dependent REH in vitro [65,66]. Hepatic CEL was found to have very close similarity to pancreatic CEL in terms of mRNA sequence, enzymatic activity and antibody recognition [67]. Like pancreatic CEL, hepatic CEL is also a secreted enzyme, so it was first hypothesized to be involved in retinyl ester hydrolysis in the space of Disse [68]. However, it was shown that lack of CEL expression does not affect uptake of dietary chylomicron remnant-retinyl ester in the liver, and furthermore, the livers of CEL-deficient mice display similar REH activity compared to wild type livers [69]. The presence of REH activity in the liver distinct from CEL was confirmed by the discovery of neutral, bile acid-independent REH activity in rat liver homogenates that localized to the microsomal fraction, consistent with a role in retinyl ester hydrolysis in the plasma membrane and/or endosome, and was not cross-reactive with antibodies to the pancreatic enzyme [70]. It was later found that this activity was stimulated by the presence of apo-CRBPI at physiological concentrations, suggesting apo-CRBPI may be a regulator of retinyl ester hydrolysis in vivo [71]. Acidic bile salt-independent REH activity has also been observed in rat liver plasma membrane and endosomal fractions [72]. It has been postulated that the neutral REH acts on chylomicron remnant-retinyl ester at the cell surface and when it enters the early endosome, and as the pH drops in the late endosome, retinyl ester hydrolysis is continued by the acidic REH [73].

Another group of enzymes proposed to be important to retinyl ester hydrolysis in the liver is the carboxylesterases. Although over 30 liver carboxylesterases with various oxyester substrates have been identified, most of these enzyme activities are thought to arise as the gene products of five major loci in linkage group V and are referred to as ES-2, ES-3, ES-4, ES-10 and ES-15 [74]. Of these five enzymes, ES-2, ES-4 and ES-10 have been shown specifically to possess REH activity in the liver in vitro [75,76]. ES-4 functions primarily as a thioesterase to catalyze the hydrolysis of long-chain acyl Co-A [75], but it has also been shown to hydrolyze retinyl palmitate [77]. ES-2 and ES-10 have been shown to function as neutral bile salt-independent REHs. Sun et al. purified REH activity from rat liver microsomal fractions that reacted with antibodies against ES-2, showed substrate preference for retinyl palmitate and had a pH optimum of 7 [76]. Similarly, they were able to purify REH activity that corresponded to the protein size and amino acid sequence of ES-10 and also demonstrated immunoreactivity to antibodies directed against ES-10. More recently, ES-22 has been reported to be a hepatic REH. Schreiber et al. identified ES-22 as a hepatocyte-expressed esterase that localizes to the ER [78]. These investigators showed that ES-22 specifically hydrolyzes retinyl palmitate, but not triolein or cholesteryl oleate. Additionally, overexpression of ES-22 inhibited the accumulation of retinyl esters in COS-7 cells.

A number of in vitro studies have also identified some well characterized lipases that possess REH activity, including hepatic lipase [79], LpL [49,50], PTL [9,10], and hormone sensitive lipase (HSL) [80]. Mello et al. have shown that, of these, only hepatic lipase and LpL are expressed in the liver (mRNA levels of PTL and HSL in all liver cell types were very low) [81]. Hepatic lipase is exclusively expressed in hepatocytes and is the only secreted lipase expressed in these cells [81]. It is secreted by the hepatocyte into the space of Disse, where it can function in the binding and uptake of chylomicron remnants and possibly in the hydrolysis of chylomicron remnant retinyl ester, though there is currently no direct evidence for the latter possibility. LpL is expressed at very low levels in hepatocytes and HSCs, but its expression is induced 32-fold in activated HSCs [81]. Thus, Mello et al. propose that it is unlikely LpL functions in the hydrolysis of newly absorbed hepatic chylomicron remnant-retinyl ester, but rather may have a role in the hydrolysis of lipid droplet retinyl esters in activated HSCs [81]. How this might occur is unclear, given that LpL is a secreted enzyme, and there is no evidence for it acting intracellularly. A summary of proposed REHs in hepatocytes can be found in Table 2.

The physiological significance of the many lipid hydrolases and carboxylesterases in the liver remains to be indentified. However, this uncertainty is true for many other tissues in the body as well; in vivo retinyl ester hydrolase activity has only been determined for a number of these enzymes. For example, it is clear that PTL and PLRP2 are physiologically significant retinyl ester hydrolases in the intestine. As discussed in further detail below, LpL-catalyzed retinyl ester hydrolysis is important for facilitating postprandial retinol uptake into adipose tissue, heart and skeletal muscle. And similarly, it is known that HSL catalyzes retinyl ester hydrolysis in adipocytes to allow for retinol mobilization into the circulation. With these exceptions, the identities of most other physiologically significant retinyl ester hydrolases remain obscure. It is likely that several other enzymes will be shown to have physiologically signifcant and distinct roles in catalyzing this reaction. With respect to the liver, we speculate that this may include hepatic lipase in hepatocytes and HSL in HSCs. Much work remains to elucidate the roles of these many enzymes in the various tissues that function in the uptake and metabolism of retinoid in the body.

Early studies suggested that this transfer is mediated by retinol-binding protein (RBP). Blomhoff et al. reported that antibodies against RBP completely blocked the transfer of retinol from hepatocytes to HSCs [82]. Following loading of hepatocytes in vivo by intravenous injection of [³H]retinyl ester-labeled chylomicrons, livers were perfused in situ with either a control buffer or a buffer containing an excess of antibodies against RBP. In both groups, the radioactivity recovered 10 min after loading was significantly higher in hepatocytes than in HSCs; after 40 min of perfusion, the control group showed significantly higher ³H-cpms associated with the HSCs, while the HSCs from the livers perfused with the RBP antibodies still had less ³H-cpms than hepatocytes. A similar study by Senoo et al. showed that antibodies to RBP blocked its transfer from HepG2 cells to rat HSCs, when the cells were co-cultured [83]. This study also showed that RBP could be internalized by cultured HSCs when the cells were incubated with human RBP. RBP was first observed to be associated with vesicles close to the cell surface and membrane and was later found deeper in the cytoplasm within endosomes. The findings that RBP is able to be bound and internalized by HSCs and that antibodies to RBP inhibited transfer of retinol into HSCs was taken to suggest that RBP mediates this transfer.

However, the generation of RBP-deficient mice allowed this hypothesis to be tested directly, and the results are not in agreement with these earlier proposals. Quadro et al. showed that, in 3- and 13‑week-old mice, there is no statistical difference in hepatic total retinol levels between WT and RBP-deficient mice; thus, the absence of RBP does not impair hepatic uptake of retinol [47]. It was later demonstrated that there is no quantitative or qualitative differences in the lipid droplets present in liver sections from RBP-deficient compared to WT mice, suggesting that retinol transfer to HSCs for storage is unaffected by the absence of RBP [84]. These studies provide strong evidence that RBP does not act in an essential manner in the transfer of retinol from hepatocytes to HSCs.

The involvement of CRBPI in the transfer process has been suggested by data from Ghyselinck et al. [85], whose studies of CRBPI-deficient mice suggest that CRBPI may mediate the transfer of retinol from hepatocytes to HSCs [85]. CRBPI-deficiency was found to result in significantly lower hepatic retinyl ester levels, with mutant mice having a 50% reduction in levels of retinyl palmitate, the main retinyl ester form found in the liver. Light microscopy also revealed that the HSCs of CRBPI-deficient mice are characterized by a reduction in both the number and size of lipid droplets. This suggests a decrease in retinyl ester synthesis that likely occurs due to impaired delivery of retinol to the retinyl ester-synthesizing enzyme LRAT, which is highly expressed in HSCs [86]. These studies clearly indicate that CRBPI is necessary for assuring efficient retinol esterification and storage in vivo. Future studies are needed to confirm how the presence and absence of CRBPI affects total retinol levels specifically in hepatocytes and HSCs.

The most distinguishing feature of the HSC is the presence of numerous retinyl ester-containing lipid droplets in the cytoplasm, which have recently been proposed to be specialized organelles for retinoid storage [86]. This idea arises from numerous studies showing the unique retinoid content of these droplets, their responsiveness to dietary retinoid status, their dependence on the synthesis of retinyl ester, and their loss in different types of hepatic disease. Taken together, these studies have demonstrated a strong regulatory role of retinoids in HSC lipid droplet physiology. This will be discussed below in more detail.

After transfer from the hepatocyte to the HSC, retinol is esterified back to retinyl ester for storage in HSC lipid droplets. Once in the HSC, retinol is bound by CRBPI and transferred to LRAT for esterification [87]. The binding of retinol to CRBPI is necessary for its transport in the aqueous environment of the cytosol and has been proposed to prevent acyl CoA-dependent enzymes in the liver from catalyzing retinyl ester formation [87]. Both LRAT and CRBPI are enriched in HSCs [5,6,86], and both proteins are needed to assure optimal HSC accumulation of retinoid stores.

LRAT is the only known enzyme capable of esterifying retinol in the liver in vivo. Earlier in vitro studies suggested that an unidentified ARAT(s) may work in conjunction with LRAT to catalyze retinyl ester formation [88]. As mentioned above, DGAT1 has been shown to participate in retinol esterification in vitro [39,89,90] and in the intestine and skin in vivo [40,91]. Thus, DGAT1 became a candidate for in vivo esterification of retinol in the liver, as well. However, the generation and study of LRAT-deficient mice fail to support both possibilities. O’Byrne et al. show that the livers of LRAT-deficient mice have undetectable levels of retinyl ester and are completely void of lipid droplets [39]. These studies not only proved that LRAT is the only retinol esterifying enzyme present in the liver, but also that LRAT and/or its product retinyl ester is necessary for HSC lipid droplet formation. Around the same time, Liu and Gudas reported that disruption of the Lrat gene makes mice more susceptible to retinoid-deficiency [92]. After maintenance on a retinoid-deficient diet for six weeks, LRAT-deficient mice had significantly lower serum retinol levels than WT mice and undetectable levels of retinol in a number of tissues studied, including the liver. Thus, Liu and Gudas proposed that LRAT-deficient mice may serve as a readily inducible, and hence useful, model to study retinoid-deficiency [92].

Moriwaki et al. reported the lipid composition of lipid droplets isolated from primary rat HSCs [93]. In rats maintained on a control diet, the mean percent lipid composition consisted approximately of 40% retinoid lipid and 60% non-retinoid lipid, broken down as follows: 39.5% retinyl ester, 31.7% triglyceride, 15.4% cholesteryl ester, 6.3% phospholipid, 4.7% cholesterol and 2.4% free fatty acids. The retinyl esters present in these droplets possess only long chain fatty acyl moieties, with retinyl palmitate followed in abundance by retinyl stearate, retinyl oleate and retinyl linoleate [94]. In the same study, the rats were maintained on four additional experimental diets containing either low or high retinol and low or high triglyceride. The investigators found that HSC lipid droplet retinyl esters, as well as other lipids, are decreased in response to a low retinol diet, and both retinoid and non‑retinoid lipids are elevated in response to a high retinol diet; however, neither retinoid nor non‑retinoid content is affected by low or high fat diets. This data was taken to indicate that the lipid composition of HSC lipid droplets is strongly influenced by dietary retinoid status, but not by dietary triglyceride intake.

Following acute liver injury, a wound healing response is initiated, which ultimately returns the liver to its healthy state [54,55,56]. However, when the liver is confronted with chronic injury, the result is often hepatic fibrosis, which is the formation of excess fibrous connective tissue as a reparative process to contain the site of injury. The most common causes of hepatic fibrosis include chronic hepatitis B and C infection, alcoholism, non-alcoholic fatty liver disease (NAFLD), and bile duct obstruction. Further progression of disease will lead to cirrhosis and eventually hepatocellular carcinoma. Seminal work by Leo and Lieber established that with this progression of liver disease in human subjects is the progressive depletion of hepatic retinoid [95]. These investigators found that there is a near 5-fold decrease in total hepatic retinol levels with the development of alcoholic hepatitis and another approximate 4-fold decrease with the development of cirrhosis. It remains unclear whether this loss of hepatic retinoid content is a causative agent or simply a consequence of disease.

As a result of injury, HSCs undergo a process of activation in which they transition from a quiescent state to a myofibroblastic phenotype [96,97]. Activated HSCs are characterized by increased synthesis of extracellular matrix and thereby become fibrogenic and proliferative [96,97]. There are currently several in vivo models of HSC activation, including carbon tetrachloride injection and bile duct ligation, and one commonly studied in vitro model, the culture of primary HSCs on plastic. It has been demonstrated that the changes in gene expression that accompany HSC activation and the loss of HSC retinyl ester lipid droplets are regulated differently in the in vitro model and the in vivo models [98]. Thus, the study of HSC activation can only be studied in the context of the disease model employed.

Two hypotheses have been proposed to account for the loss of retinyl ester lipid droplets in hepatic disease. One theory is that HSC activation results in a “toxic burst” of transcriptionally active retinoids from the lipid droplet stores and, acting through the retinoic acid receptors and the retinoid X receptors (all of which are expressed in HSCs), lead to the altered gene expression patterns seen with activation [99,100]. The other proposes that the retinoid stores in these lipid droplets play a protective role, such that their presence buffers against hepatic insult [101]. When these stores are absent, an injury to the liver will result in fibrosis and hepatic disease. This second hypothesis seems unlikely considering LRAT-deficient mice are not predisposed to developing spontaneous hepatic fibrosis [39]. However, more definitive studies need to be conducted before the linkage between the loss of HSC retinoid stores and hepatic disease can be determined.

As mentioned above, HSC lipid droplet retinyl ester stores must be mobilized in times of dietary retinoid-insufficiency to supply peripheral tissues with retinoid needed for maintaining various essential biological functions. Mobilization requires that the retinyl ester first be hydrolyzed back to retinol. It is currently not known which lipases are involved in the hydrolysis of HSC retinyl ester, but there are several candidate enzymes based on their REH activity and/or expression in HSCs (see Table 2). As discussed above, ES-2, ES-4 and ES-10 are three hepatic carboxylesterases shown to have REH activity [75,76]. Mello et al. show that ES-4 and ES-10 are expressed in HSCs, albeit at very low levels compared to hepatocytes [81]. Their REH activity and expression in HSCs make ES-4 and ES-10 strong candidates for HSC lipid droplet retinyl ester hydrolases. Another candidate for HSC retinyl ester hydrolysis is adipose triglyceride lipase (ATGL). ATGL is the key enzyme for triglyceride hydrolysis in adipocytes [102], and its lipase activity in vivo is dependent on coactivation by CGI‑58 [103]. Earlier studies suggested ATGL does not have REH activity, but these studies were done in the absence of CGI-58 [102]. Mello et al. found that ATGL is expressed highly in HSCs and propose that, in the presence of CGI-58, will have REH activity [81]. However, this possibility has not yet been demonstrated directly.

Upon hydrolysis of HSC lipid droplet retinyl ester, retinol is thought to be transferred back to the hepatocyte, where it is bound by its specific transport protein, RBP. RBP is a 21 kDa protein with a single binding site for one molecule of all-trans-retinol, and the hepatocyte is the major (though not exclusive) site of RBP synthesis in the body [104]. Once bound by RBP, the retinol-RBP complex enters the bloodstream for transport to peripheral tissues. Secretion of RBP by the hepatocyte is highly regulated by the retinoid status of the animal, such that RBP secretion is blocked in times of dietary deficiency and restored upon retinol-repletion [105].

The importance of RBP in maintaining retinoid homeostasis in the body was demonstrated by Quadro et al. through generation of RBP-deficient mice [47]. These investigators show that, in 3- and 13-week-old mice, there is no statistical difference in hepatic total retinol levels between WT and RBP-deficient mice; thus, the absence of RBP does not impair hepatic uptake and accumulation of retinol. However, when these investigators measured hepatic and serum retinol levels in 5-month-old mice, they found that hepatic retinol and retinyl ester levels were significantly higher in RBP-deficient compared to WT mice and, additionally, serum total retinol levels were significantly lower. These data indicate that RBP-deficient mice can acquire hepatic retinol stores, but these cannot be mobilized. It was later shown using transgenic mice that express human RBP that extrahepatically synthesized RBP cannot be taken up by hepatocytes and cannot function in the mobilization of hepatic retinol stores [84]. Only RBP synthesized in the liver can mobilize hepatic retinol stores.

In the blood, RBP is found in a 1:1 protein-protein complex with a 55 kDa serum protein, transthyretin (TTR) [104]. The retinol-RBP-TTR complex is the predominant transport molecule through which retinol is delivered to peripheral tissues in the fasting circulation. In the absence of TTR, plasma retinol and RBP levels are only approximately 5% of those that are observed in matched WT mice [104]. It has been established that the association of RBP with TTR prevents filtration of the relatively small RBP molecule through the kidney glomeruli [106]. Studies by van Bennekum et al. employing TTR-deficient mice have demonstrated the importance of TTR in maintaining normal levels of retinol and RBP in the circulating plasma. Subsequent studies established that, when a physiologic dose of human retinol-RBP was injected intravenously into TTR-deficient mice, it was cleared more rapidly from the plasma and accumulated more rapidly in the kidneys of TTR-deficient than WT mice [106]. These data were taken by the authors to indicate that the reduced levels of retinol and RBP in the circulations of TTR-deficient mice arise, at least in part, due to increased filtration of the retinol-RBP complex. Other studies of TTR-deficient mice showed that the livers of these mutants have similar levels of retinol and retinyl ester compared to WT mice, but 60% higher levels of RBP protein [107]. These data indicated that TTR does not have a role in hepatic uptake or storage of dietary retinol, but suggested it may play a role in hepatic secretion of RBP. However, van Bennekum et al. were unable to establish directly that the absence of TTR affects RBP secretion from hepatocytes [106]. These investigators showed that cultured primary hepatocytes isolated from TTR-deficient mice accumulated RBP in their culture media to the same degree as hepatocytes from WT mice; thus, RBP was being secreted from hepatocytes at the same rate in the presence and absence of TTR.

White adipose tissue is a significant site of extra-hepatic retinoid storage, accounting for ~15% of tissue retinoid stores in the adult rat [146]. Adipocytes and the stromal-vascular fraction of adipose have been shown to express key mediators of retinoid metabolism, supporting the notion that retinoid signaling is an important mediator of adipose physiology [147]. Within adult adipose tissue, RBP is predominantly expressed in adipocytes, with only weak expression in stromal-vascular cells [146]. Conversely, CRBPI was found to have relatively higher expression in the stromal-vascular cells versus adipocytes, though in vitro experiments show that retinoic acid can upregulate CRBPI expression in adipocytes [146,148]. CRBPIII has been shown to be expressed in adipose tissue homogenates [149] and in differentiated 3T3-L1 adipocytes in vitro [150]. However, through the use of a specific CrbpIII-LacZ mouse, expression of this gene appears to be restricted to the vascular endothelial cells of adipose tissue [151]. The mRNA transcripts for retinoic acid receptor-α, -β, and -γ have all been reported to be expressed in adipose tissue [152,153]. However, work in differentiated 3T3-L1 adipocytes reveals only a low level of RAR-α at the protein level [154]. Similarly, mRNA for the three major retinoid X receptor isoforms is also expressed in adipose/3T3-L1 adipocytes [153]. The synthesis of retinyl ester in adipose tissue was presumed to be mediated by LRAT, but this is certainly not the case, since LRAT-deficient mice have increased levels of adipose retinyl ester stores. Given the importance of LRAT in the synthesis of retinyl ester in the liver and nearly all other tissues, adipose levels of retinyl ester would be expected to be lower than WT levels, yet a greater than 3-fold increase in retinyl ester was observed [39,40,91]. This observation suggests that there is an unidentified enzyme expressed within adipocytes, the cellular site of retinoid storage within adipose tissue [146], which can synthesize retinyl ester.

There are several studies that indicate an important role for retinoid signaling in adipose tissue physiology, including the observation that retinoic acid can inhibit adipogenesis [147,155], CRBPI-deficient mice show increased adiposity [156], and CRBPIII-deficient mice have altered lipid metabolism and decreased adiposity [150]. Although CRBPIII was originally identified as a cellular retinol-binding proteins based on the ability of the recombinant protein to bind retinol, a true physiological ligand for CRBPIII has not been identified. It seems possible, even likely, that the actions of CRBPIII in lipid metabolism may involve other ligands than retinol.

In addition to research studying the importance of retinoid signaling on adipose tissue biology, a novel role for adipocyte-derived RBP in metabolic disease has been proposed. As discussed above, the sole established physiologic role for RBP is to mobilize hepatic retinoid stores and transport retinol in the circulation [47]. Moreover, RBP-deficient mice and RBP-deficient humans have a similar relatively mild phenotype. However, work from Kahn and colleagues shows that RBP-deficient mice display enhanced insulin sensitivity [143]. This observation, coupled with data from a number of follow-up studies, suggests that adipose-derived RBP may be a significant contributor to metabolic disease.

In 2005, Kahn and colleagues published a convincing series of studies leading to the conclusion that RBP is an important mediator of insulin resistance in obesity and type 2 diabetes [143]. It was previously known that mice with an adipose-specific deletion of the glucose transporter, GLUT4, were insulin resistant [157]. Yang et al. demonstrated that these mice over-express RBP in adipose tissue and that circulating levels of RBP were elevated in adipose-specific GLUT4-deficient mice. RBP expression in the liver was unchanged, suggesting that the increase in serum RBP was adipose-derived. Treatment of adipose-GLUT4-deficient mice with the PPAR-gamma agonist, rosiglitazone, was associated with improved insulin sensitivity and glucose tolerance. This improvement was paralleled by a normalization of circulating RBP levels. Interestingly, increased circulating RBP levels were found to be a common feature of multiple mouse models of insulin resistance, and serum RBP levels were also increased in human subjects with diabetes. Further evidence suggesting a role for RBP in modulating insulin sensitivity came from experimental manipulations of circulating RBP levels. Mice over-expressing human RBP in skeletal muscle have increased serum RBP levels and were found to be insulin resistant, an observation mirrored by chronic treatment with purified RBP protein. Conversely, RBP-deficient mice have enhanced insulin sensitivity, as do mice with lowered circulating RBP levels following fenretinide treatment. Taken together these studies clearly indicate that increased circulating RBP is associated with insulin resistance, whereas lowered serum RBP is associated with enhanced insulin sensitivity. The publication of these studies led to the suggestion that RBP is a novel adipokine and a potential therapeutic target in human type II diabetes [158,159,160].

Follow-up studies into the correlation between serum RBP levels and altered insulin sensitivity in humans have not resulted in consensus on this point. Several studies have confirmed the observation linking insulin resistance to high circulating levels of RBP. Indeed this concept has been expanded to suggest that the ratio of retinol to RBP more strongly correlates with insulin sensitivity [161,162,163,164,165,166]. However, other groups have been unable to show an association between high serum RBP levels and insulin resistance [167,168,169,170,171,172]. It is apparent that further study is required to definitively establish a link between elevated serum RBP levels and insulin resistance in humans.

The studies mentioned above suggest that RBP is an important contributor to metabolic disease. What remains unclear is whether this effect is mediated through retinoid signaling or a hitherto unknown mechanism of RBP action. Studies into the molecular mechanisms of RBP-induced insulin resistance have been recently reviewed elsewhere [173,174]. There are a growing number of studies that indicate an intersection between energy balance, metabolic disease, and retinoid signaling. These developments have opened up an exciting area of retinoid biology and brought renewed attention to this essential nutrient.

Another expanding and important area of retinoid biology is the importance of retinoid signaling in the heart. While it has been well established that retinoic acid is an important signaling molecule during heart development in the embryo, there is a growing body of evidence that suggests this molecule is also an important factor in heart disease and cardiac remodeling [130]. In several animal models of heart disease, retinoic acid supplementation has been shown to have beneficial effects on tissue pathology. Experiments employing rat cardiomyocytes in vitro have shown that retinoic acid can attenuate cardiomyocyte hypertrophy induced by endothelin-1, phenylephrine, and angiotensin-II [175,176,177,178]. Following the promising results of these in vitro studies, experiments in whole animals have borne out the hypothesis that retinoic acid may suppress cardiac hypertrophy in vivo. Retinoic acid treatment has proven effective in attenuating cardiac remodeling in a rat pressure overload model (by aortic banding) and a rat model of myocardial infarction [179,180]. Experimental myocardial infarction is also associated with an increase in heart retinoid metabolism, with a significant mobilization of retinoid from the liver to the heart observed in infarcted versus control animals [181]. In contrast to the beneficial effects of retinoic acid in a disease setting, there is evidence to suggest that retinoid-deficiency may have adverse effects on the heart. Retinoid-deficiency in female rats has been associated with markers of cardiac remodeling and ventricular dysfunction [182]. How these animal studies relate to heart disease in humans remains largely untested. Unfortunately, in the context of acute promyelocytic leukemia, retinoic acid treatment has been associated with cardiac dysfunction in at least one case report [183]. Continuing research aimed at verifying the therapeutic potential of retinoic acid treatment in heart disease is warranted.

There has been a very substantial body of research carried out in the last decade aimed at defining better retinoid metabolism and actions within the eye. Notably, this literature has established many linkages between genetic eye disease and genes involved in retinoid metabolism and transport. This had long been anticipated but has only been established through groundbreaking research carried out in the last 10 to 15 years. In this review, we will only focus on a few key advances for understanding retinoid metabolism in the eye, and we will not focus on relationships of these processes with disease. A number of extensive and authoritative reviews on this topic have been published recently [145,146] and consequently the reader is referred to these reviews for more detail.

The central question regarding retinoid metabolism in vision that was left unanswered from the early work of Wald and others [184,185] concerned the mechanism through which all-trans-retinoid is converted to the energetically less stable 11-cis-retinoid that is required for vision. Based on work commencing in the mid-1980s, Rando and colleagues and others proposed the existence of an isomerohydrolase that converts all-trans-retinyl ester to 11-cis-retinol [186,187,188]. It was proposed that the isomerohydrolase utilizes energy liberated from the hydrolysis of the ester bond to drive steroisomerization from the all-trans- to the energetically less favorable 11-cis-configuration. Recently, it has been demonstrated that the abundant retinal pigment epithelium protein (RPE) RPE65 is the isomerohydrolase [189,190,191]. RPE65-deficient mice have high levels of all-trans-retinyl esters in their eyes and no detectable 11-cis-retinoids [192]. These mice are blind because they are unable to generate 11-cis-retinoid needed for vision. As mentioned above in Section 2.2, RPE65 structurally resembles BCMO1 and BCMO2 and, like these two other enzymes, contains a conserved Fe²⁺ binding site that is required for catalytic activity [18]. RPE65 (originally referred to as p63) had been identified in the earlier literature as a cell surface receptor for RBP, but this has proven to be incorrect [6,192].

Another important advance in understanding retinoid metabolism in the eye has been the elucidation of the mechanism through which all-trans-retinoid produced upon photoexcitation is removed from the photoreceptor outer segments for transport back to the RPE and reconversion to 11-cis-retinoid. This process has been identified to require a retina-specific ATP binding cassette transporter referred to as either ABCR or ABCA4, which localizes to the disc membranes present in rod and cone outer seqments [193,194,195]. ABCR-deficient mice show a much slower rate of removal of all-trans-retinal from the retina following exposure to light [196,197]. This is accompanied by an increase in retinal-phosphotidylethanolamine adducts, which have been identified as the substrates for ABCR-catalyzed removal of all-trans-retinal from the disc membranes to the cytoplasmic space [198,199].

Another important advancement in understanding retinoid metabolism in the eye has been the identification of retinol dehydrogenase 8 (RDH8; also referred to as photoreceptor RDH or prRDH) as one enzyme within the outer segments of the retina that catalyzes the reduction of all-trans-retinal to all-trans-retinol [200,201,202]. RDH8-deficient mice show a slower rate, by several orders of magnitude, of clearance for all-trans-retinal from the retina following exposure of dark adapted animals to a flash of light [200,201,202,203]. However, the kinetics of rhodopsin regeneration for RDH8-deficient mice is not different than that of WT mice, suggesting that other dehydrogenases are also involved in all-trans-retinal reduction to all-trans-retinol within the photoreceptors. Subsequent studies by Palczewski and colleagues identified RDH12 as another photoreceptor dehydrogenase that is contributes to all-trans-retinal reduction, although RDH12 too was observed to be dispensible [203].

In summary, the last decade has been a very exciting one for the study of retinoid metabolism in the eye. Much has been learned and the eye remains the organ where retinoid metabolism is best and most deeply understood.