Fatty Acid-binding Proteins (FABPs) Are Intracellular Carriers for Δ9-Tetrahydrocannabinol (THC) and Cannabidiol (CBD)*

Background: Δ9-Tetrahydrocannabinol (THC) and cannabidiol (CBD) modulate endocannabinoid tone in vivo through unknown mechanisms. Results: THC and CBD bind to fatty acid-binding proteins (FABPs) and reduce endocannabinoid metabolism. Neither THC nor CBD inhibit human fatty acid amide hydrolase activity. Conclusion: FABPs are intracellular transporters of THC and CBD. Significance: These findings identify a new mechanism by which phytocannabinoids influence endocannabinoid signaling. Δ9-Tetrahydrocannabinol (THC) and cannabidiol (CBD) occur naturally in marijuana (Cannabis) and may be formulated, individually or in combination in pharmaceuticals such as Marinol or Sativex. Although it is known that these hydrophobic compounds can be transported in blood by albumin or lipoproteins, the intracellular carrier has not been identified. Recent reports suggest that CBD and THC elevate the levels of the endocannabinoid anandamide (AEA) when administered to humans, suggesting that phytocannabinoids target cellular proteins involved in endocannabinoid clearance. Fatty acid-binding proteins (FABPs) are intracellular proteins that mediate AEA transport to its catabolic enzyme fatty acid amide hydrolase (FAAH). By computational analysis and ligand displacement assays, we show that at least three human FABPs bind THC and CBD and demonstrate that THC and CBD inhibit the cellular uptake and catabolism of AEA by targeting FABPs. Furthermore, we show that in contrast to rodent FAAH, CBD does not inhibit the enzymatic actions of human FAAH, and thus FAAH inhibition cannot account for the observed increase in circulating AEA in humans following CBD consumption. Using computational molecular docking and site-directed mutagenesis we identify key residues within the active site of FAAH that confer the species-specific sensitivity to inhibition by CBD. Competition for FABPs may in part or wholly explain the increased circulating levels of endocannabinoids reported after consumption of cannabinoids. These data shed light on the mechanism of action of CBD in modulating the endocannabinoid tone in vivo and may explain, in part, its reported efficacy toward epilepsy and other neurological disorders.


Chemicals
AEA was from Cayman Chemical (Ann Arbor, MI), THC, CBD, and fatty acid-free BSA were from Sigma, [ 14 C]AEA (arachidonoyl- [1-14 C]ethanolamide, 60 mCi/mmol) was provided by the Drug Supply Program at the National Institute on Drug Abuse.

Cell Culture and Transfections
Wild-type and FABP5 shRNA-expressing HeLa cells (12) were grown in DMEM supplemented with 10% fetal bovine serum, 100 units/ml of penicillin/streptomycin, 2 mM L-glutamine, and 1 mM sodium pyruvate. Transfections of human FAAH were carried out using the GenJet Plus reagent (SignaGen, Rockville, MD) according to the manufacturer's instructions.

In Vitro Binding of Ligands to FABPs
Binding analysis of THC and CBD to FABPs was carried out as described (14).
Simulation Setups and Molecular Dynamics-MD simulations were adapted from the methodology carried out in Ref. 27. Each MD trajectory (2 ns) was performed using NAMD 2.9 (28) employing charmrun to assign multiple processors (ϳ10) on the Seawulf Cluster at Stony Brook University. Preparation of each the initial MD coordinates started with docked complexes, which were subsequently stripped of crystallographic waters and prepared using the AMBER8 suite (leap, antechamber) (29). AMBER8 was used to assign force field parameters and solvate each complex in a periodic box containing ϳ5000 waters (ϳ60 Å 3 ). Force fields were assigned independently using AMBER8: FF99SB (30) for the protein, GAFF for the ligand, and TIP3P (31) for the water. Prior to running each simulation, a nine-step equilibration of three minimization steps (2000 steps of steepest decent) and seven molecular dynamics runs (50 ps each) as described in Ref. 32 were performed to equilibrate the system before the production run (2 ns) (32).
MD simulations were carried out in the NPT ensemble using Langevin dynamics at a constant temperature of 310 K and constant pressure of 1.01325 bar (33). All MD simulations employed a 1-fs time step that required the use of SHAKE (34). Additional key parameters included particle mesh Ewald (35) to compute long-range electrostatics (1.5 Å grid spacing) and a 10-Å cutoff (8 Å smoothing switch) for non-bonded interactions. Each MD frame is composed of 1000 fp (1 ps) and was run for a total of 2 ns.
MMGBSA Calculations-Free energies of binding (⌬G bind Calculated) were carried out using the MM-GBSA method (36,37). All MM-GBSA calculation were carried out post production. For each complex, AMBER8 (p-traj) was employed to strip all explicit water from each of the 2-ns MD trajectories. A single point calculation was carried out using AMBER8 (sander) for each ps (2000 ps in 2 ns) with the following parameters: AMBER radii (mbondi2), Dielectric constants (1 and 78.5), and GB model (igb ϭ 5) (38).
Following Equation 1, the MM-GBSA free energy of binding (⌬G bind ) was calculated using the non-bonded intermolecular protein-ligand interactions (⌬G coul and ⌬G vdw ), the change in hydration energy (⌬G hyd ), and the temperature multiplied by the change in solute vibrational, rotational, and translational entropy (T⌬S).
Specifically, ⌬G coul and ⌬G vdw were calculated from the change in energy for all nonbonded intermolecular protein-ligand interactions (Coulombic and van deer Waals) as the receptor and ligand go from a free state to a bound state (complex) shown in Equation 2.
The ⌬⌬G hyd was calculated from the ⌬G hyd ϭ G polar ϩ G nonpolar for each species (complex, receptor, and ligand) as shown in Equation 3.

Enzyme Assays
FAAH activity in homogenates was determined as described previously (11).

AEA Uptake
Inhibition of AEA uptake into cells was performed as described (11). HeLa cells were preincubated with 20 M THC, CBD, or vehicle (0.1% ethanol in DMEM supplemented with 0.15% BSA) for 15 min followed by incubation with 100 nM [ 14 C]AEA for 5 min. AEA uptake was quantified as previously described (11).

Immunohistochemistry
Brain sections were blocked in 1% serum or 1% nonfat dry milk in PBS-T (0.3% Trition X-100 in PBS) and then incubated with anti-human/mouse serum albumin primary antibody (MAB1455 from R&D Systems used at 1:20) and anti-macrophage/microglia Iba1 (rabbit polyclonal 019-19741 from WakoUSA used at 1:1000) overnight at 4°C. Incubation with Alexa Fluor 488 and Alexa Fluor 555-conjugated secondary antibodies at 1:1000 was performed at room temperature for 1 h, followed by washing with PBS and mounting using Fluoromount-G with DAPI (Southern Biotech). 4 -8 brain sections were imaged per biological replicate.

Statistics
Results represent mean Ϯ S.E. of at least three independent experiments performed in triplicate. Statistical significance was determined using two-tailed unpaired t tests against controls or by one-way analysis of variance with post-hoc Dunnett's test compared with vehicle controls (significance set at p Ͻ 0.05) on the raw data, as indicated. Analysis was performed using GraphPad (version 6.0, Prism).

RESULTS
Computational methods were initially employed to determine the likelihood of FABPs being carriers for cannabinoids. THC and CBD displayed energy scores consistent with appreciable binding to FABP3, FABP5, and FABP7, the main FABPs that are expressed in the brain. This analysis showed tighter binding of CBD to FABP3, FABP5, and FABP7 (⌬G binding ϭ Ϫ23. 64  Ϫ15.54 Ϯ 3.06 kcal/mol) ( Table 1). THC and CBD were accommodated in the pockets of the FABPs (Fig. 1). Conspicuously, THC binds deeper into the pocket of FABP7 than FABP3 and FABP5. Noticeably, the binding pose is unique for CBD in FABP5 (Fig. 1).
In vitro binding studies of THC and CBD, using a fluorescent NBD-stearate displacement assay, indicated that these two cannabinoids bind to FABP3, FABP5, and FABP7 with low micromolar affinities ( Table 2). In addition, for FABP5 there was agreement with the in silico results showing that CBD bound more tightly than did THC, however, a significant difference in binding affinities was not observed for FABP3 and FABP7. Interestingly, THC and CBD bind the FABPs as well or nearly as well as the endocannabinoids and a variety of other compounds including various transport inhibitors as discussed below. However, they do not bind as well as many of the fatty acids including oleic acid that binds with nearly 50-fold higher affinity for some of the FABPs (see Table 2).
Because albumin is a known carrier of the cannabinoids in the blood and was proposed to function as an intracellular endocannabinoid-binding protein (9, 15), we examined whether it could function in a similar capacity as an intracellular transporter in the brain by initially examining its expression pattern in the mouse brain. One group has previously reported that microglial cells synthesize albumin (41). Contrary to this report, we did not find any albumin expression in microglia ( Fig. 2). In the normal mouse brain albumin expression did not co-localize with the macrophage/microglial marker Iba1, but was rather detected in perivascular areas (Fig. 2, arrows). Very low levels of expression could be detected in the cytoplasm of neurons in the hippocampus after extensive incubation with the primary antibody (not shown). These results indicate that albumin is likely not a major intracellular transporter of the cannabinoids in the brain.
We next determined whether FABPs serve as intracellular carriers for THC and CBD. HeLa cells are a suitable cell type to examine interactions of exogenous cannabinoids with FABPs and their effects upon endocannabinoid transport because these cells conveniently express a single FABP subtype, FABP5 (12,14), AEA uptake is coupled to its breakdown by FAAH (42,43). Therefore, HeLa cells, which lack FAAH (44), were transfected with human FAAH and AEA uptake was subsequently examined.
Both THC and CBD inhibited AEA uptake, with CBD being slightly more efficacious (Fig. 3A). The greater potency of CBD as an inhibitor of AEA uptake mirrored its higher affinity (lower K i ) for FABP5 ( Table 2). As expected, FABP5 knockdown inhibited AEA uptake (Fig. 3A) (14). Importantly, THC and CBD did not reduce AEA uptake in cells bearing a knockdown of FABP5  3A). Furthermore, THC and CBD did not reduce the proportion of intracellular AEA that is hydrolyzed following uptake, suggesting that the cannabinoids block the delivery of AEA to FAAH but do not affect AEA hydrolysis by FAAH (Fig. 3B). In support of this, we found that CBD and THC did not inhibit AEA hydrolysis by human FAAH in cell homogenates (Fig. 3C). This result was surprising in light of previous studies demonstrating robust inhibition of rodent FAAH by CBD (17,(45)(46)(47). Therefore, we examined whether CBD inhibits rat and mouse FAAH. In agreement with previous studies, CBD inhibited both rat and mouse but not human FAAH (Fig. 3D), suggesting a species specificity that may underlie these effects (14,44).
To further investigate this species specificity, we employed site-directed mutagenesis to interconvert key residues within the active site of human FAAH to the homologous residues found in rat FAAH. Previous work has identified six residues that differ between the active sites of human and rat FAAH (F192L, Y194F, T377A, N435S, V491I, and M495V) (Fig. 4) (48). Of these, Phe 192 and Met 495 were identified as likely candidates to alter interactions with CBD based on the crystal structures of human and rat FAAHs with other bound ligands (23). These two residues appeared to restrict the entry of CBD into the active site of human FAAH. Therefore, we mutated these residues in human FAAH to generate F192L, M495V, as well as FAAH bearing the double mutant. Interestingly, the F192L and M495V mutants did not significantly alter FAAH inhibition by CBD. In contrast, human FAAH carrying mutations at both sites was robustly inhibited by CBD, with a similar inhibitor sensitivity profile as seen in rat FAAH (Fig. 5, A and B).
Computational methods were subsequently employed to characterize the interactions of CBD within the active site of FAAH. CBD was individually docked to human FAAH, rat FAAH, or the F192L/M495V mutant human FAAH, and predicted free energy scores were obtained (⌬G binding ϭ Ϫ26.55 kcal/mol, Ϫ36.84 kcal/mol, and Ϫ30.46 kcal/mol, respectively). Consistent with the observed biochemical data, CBD was predicted to bind human FAAH with a lower affinity than either rat FAAH or the human FAAH F192L/M495V mutant.
The membrane access channel of FAAH is a cavity of hydrophobic and aromatic residues that connect the catalytic core of the protein to the membrane-associated region, thought to allow lipid substrates to directly access the site of catalysis from B, THC appears sterically crowded in the binding pocket of FABP3 with water-mediated hydrogen bonding seen with TYR128, a key amino acid in the binding of fatty acids. THC instead adopts a direct hydrogen bond with Ala 75 (2.6 Å). CBD remains in hydrogen bonding distance to Tyr 128 (5.0 Å), whereas making closer hydrogen bonds through its 1Ј-OH with Thr 53 (2.4 Å), Gln 95 (2.2 Å), and Glu 72 (3.9 Å). Water-mediated hydrogen bonding with Tyr 128 , similar to THC, is critical for binding. C, THC shows a long-range water mediated hydrogen bonding with Tyr 130 (5.3 Å), a key amino acid residue critical to FA binding. The Ile 106 and Val 117 residues in FABP5 provide less steric hindrance compared with Leu 104 and Leu 115 of FABP3, allowing THC to penetrate further into the binding pocket. CBD adopts a unique orientation in FABP5, compared with similar binding poses in FABP3 and FABP7, blocking Tyr 129 by making a strong hydrogen bond with Arg 128 (2.1 Å). The 1Ј-OH of CBD forms hydrogen bonding with Thr 55 . D, THC forms a strong hydrogen bonding directly with Tyr 128 (2.6 Å), a key amino acid residue critical to fatty acid binding, whereas providing a weaker interaction with Arg 126 (4.1 Å). The Phe 104 and Met 115 residues in FABP7 provide even less steric hindrance compared with similar residues in FABP3 and FABP5, allowing THC to penetrate even further into the binding pocket, which is now without water. CBD adopts a similar binding mode as in FABP3, providing long range interactions between Thr 53 (2.8 Å) and Tyr 128 (5.3 Å), whereas additionally providing hydrogen bonding with Ala 75 (2.3 Å) and Glu 72 (2.1 Å) via the 1Ј-OH.  (49,50). We performed in silico docking of CBD to FAAH to investigate the preferred binding pose of this cannabinoid. Like other FAAH inhibitors, CBD was predicted to reside within the membrane access channel cavity, coming within 2.8 Å of Ser 241 of the catalytic core (Fig. 6A). AEA was subsequently docked to rat FAAH for comparison and the lowest energy pose was found to be in a highly similar location to that of CBD (Fig. 6B), further suggesting the possibility of CBD acting as a competitive inhibitor of FAAH. Inspection of the predicted geometry of CBD within the active site reveals a twisted conformation of the compound (Fig.  7). Although THC and CBD are isomers, the pyran ring of THC confers added rigidity to the molecule. The reduced structural flexibility of THC should not allow for the same twisted conformation as observed with CBD and likely accounts for the inability of THC to inhibit FAAH activity from either species. Superimposed images of CBD in rat (red), human (blue), or F192L/ FIGURE 2. Albumin is not expressed by microglia in the normal mouse brain. Hippocampal brain sections were obtained from C57Bl6 mice and incubated with anti-albumin and anti-Iba1 antibodies. The sections were incubated with the primary antibody overnight at 4°C. Images were visualized with Alexa Fluor 488-and Alexa Fluor 555-conjugated secondary antibodies. Nuclei were labeled with DAPI. n ϭ 4. Scale bars, 500 m (upper panels, low magnification images) and 100 m (lower panels, high magnification images). Arrows point to perivascular albumin staining. APRIL 3, 2015 • VOLUME 290 • NUMBER 14

JOURNAL OF BIOLOGICAL CHEMISTRY 8715
M495V mutant (purple) FAAHs reveal similar modes of binding (Fig. 7). CBD penetrates ϳ1 Å less deeply into the active site of human FAAH and has a less favorable predicted free energy of binding, which may account for it being less potent as an inhibitor in human FAAH in comparison to rat FAAH in our enzyme assays.

DISCUSSION
In the present study, we identified FABPs as the first intracellular carriers for phytocannabinoids and showed that THC and CBD modulate AEA uptake by competing with AEA for binding to FABPs. Therefore, THC and CBD may potentiate AEA signaling by reducing its catabolism. Unexpectedly, in  contrast to previous reports demonstrating robust inhibition of rodent FAAH by CBD, our findings clearly indicate that FAAH does not serve as a target for CBD in humans. Therefore, the recent observation linking CBD administration to elevated serum AEA levels in humans is unlikely to be attributable to FAAH inhibition, and points to modulation of AEA transport proteins such as FABPs (see schematic in Fig. 8).
In silico modeling demonstrated that THC and CBD interact with FABP3, FABP5, and FABP7 and that both ligands reside within the FABP binding pocket in a manner similar to that shown for AEA and 2-AG by x-ray crystallography (51). In vitro experiments confirmed that these cannabinoids bind the FABPs, all with a low micromolar affinity (1.04 -3.14 M). Interestingly, these affinities are only slightly lower than those observed with SBFI26, a truxillic acid compound recently identified as a specific inhibitor of the FABPs (K i for FABP3, FABP5, and FABP7 ϭ 3.86 Ϯ 0.70, 0.93 Ϯ 0.08, and 0.38 Ϯ 0.04, respectively) (16). However, unlike SBFI26, THC and CBD showed pronounced affinity for FABP3. These two cannabinoids also bind with similar affinities as endocannabinoids AEA and 2-AG ( Table 2). The relative similarity in binding affinity between the phytocannabinoids and endocannabinoids suggests that competition for these binding sites may exist. Our results show that the FABPs carry anandamide and that it can be displaced by CBD and THC in milieu of tighter binding of non-esterified fatty acids that, we surmise, are not saturating.
Endocannabinoid signaling is believed to be controlled through a balance of on demand synthesis and prompt catabo-  lism (52)(53)(54)(55). AEA is inactivated through hydrolysis by intracellular FAAH (56,57). We chose to employ HeLa cells because they do not express FAAH, and this permitted us to use cells transfected with human, rat, or mouse FAAH. The uptake of AEA is unique compared with other neurotransmitter systems in that AEA does not require a transmembrane transporter. FAAH enzyme activity maintains an inward concentration gradient that drives uptake and this process requires FABPs as carriers (11,13,14).
Conveniently, HeLa cells only express a single FABP subtype, FABP5 (12). Our results indicate that CBD and THC inhibit AEA inactivation by targeting FABPs. This is supported by the observations that both phytocannabinoids were without effect in cells bearing a knockdown of FABPs. Furthermore, we did not observe inhibition of human FAAH by either of these cannabinoids.
Previous reports have shown that CBD, but not THC, inhibits mouse and rat FAAH activity (17,(45)(46)(47), which we have confirmed here. Unexpectedly, we are the first to report that CBD does not inhibit human FAAH, even at high concentrations. We explored the species-specific differences and found two key residues within the active site of human FAAH that render it resistant to inhibition by CBD.
In addition to exogenous cannabinoids, FABPs bind to endogenous N-acylethanolamines related to AEA (12). Therefore, it is not surprising that previous studies have demonstrated that N-acylethanolamines that do not themselves activate cannabinoid receptors, potentiate endocannabinoid signaling (58 -60). A recent study reported that CBD alleviates the psychotic symptoms of schizophrenia, possibly in part by elevating levels of AEA and N-acylethanolamines in humans (17). Another report showed that administration of a single oral dose of 20 mg of THC to 30 healthy volunteers resulted in higher circulating concentrations of anandamide, 2-AG, palmitoyl ethanolamide, and oleoylethanolamide (61). Therefore, it is tempting to speculate that phytocannabinoids modulate the endocannabinoid tone by targeting FABPs in humans (Fig. 8). Indeed, inhibition of FABPs was recently shown to elevate brain levels of AEA in mice (16,62). This proposed mechanism of AEA modulation in humans is highly dependent on the pharmacological concentrations of cannabinoids in the brain following consumption. To our knowledge there is little data on this front, however, rats administered a low dose (1 mg/kg) of THC or CBD rapidly attained micromolar brain and plasma concentrations of the parent compound and/or their metabolites (63). The reported brain cannabinoid concentrations are above the K i for the FABPs, indicating the feasibility of a FABPdependent mechanism of action in modulating endocannabinoid levels.
Following ingestion, THC is biotransformed primarily to the psychoactive metabolite 11-OH-THC, which is further oxidized to the non-psychoactive 11-nor-9-carboxy-THC (THC-COOH). This inactivation pathway primarily occurs by the cytochrome P-450s in the liver, however, extrahepatic metabolism in other tissues such as the brain, intestines, and lungs may also partially contribute to THC metabolism (64). Phase II metabolism involves glucuronidation of THC-COOH to form soluble adducts that are excreted in urine and feces (10,64). Phase I metabolism of CBD is similar to that of THC, with primary oxidation at the C-9 position to the alcohol and carboxylic acid, however, little data are available describing its phase II metabolism (64,65). CBD has been reported to inhibit THC metabolite formation in vitro and some groups have found that co-administration of THC and CBD partially inhibits the formation of select THC metabolites in human subjects (66 -68). Metabolic reactions such as phase I hydroxylation by the cytochrome P-450s and phase II glucuronidation occur at the endoplasmic reticulum and, similar to endocannabinoids, phytocannabinoids likely require a carrier for translocation from the cell membrane to the endoplasmic reticulum. In light of our current study, we postulate that FABPs may play a role in trafficking phytocannabinoids through the cytosol to their respective intracellular enzymes, although it is possible that other identified carriers such as HSP70 may also be involved (15).
Cannabinoids have shown promise in treating several neurological disorders, including addiction, anxiety, schizophrenia, and epilepsy, as well as other conditions such as glaucoma, pain, and nausea (4, 5, 17, 69 -71). Sativex, an oromucosal spray containing ϳ1:1 ratio of THC:CBD, has shown promise in treating neuropathic pain and spasticity in patients with multiple sclerosis (26,72,73). The efficacy of CBD in the treatment of these various medical conditions suggests that both cannabinoid receptor-dependent and -independent mechanisms of action are at play in conferring such therapeutic benefits. Further studies characterizing the in vivo effects of phytocannabinoids are warranted, and the FABPs may emerge as targets of great therapeutic potential in the future of cannabinoid-based pharmacotherapies.