|Home | About | Journals | Submit | Contact Us | Français|
The objective was to elucidate the inhibition requirements of the human Organic Cation/Carnitine Transporter (hOCTN2).
Twenty-seven drugs were screened initially for their potential to inhibit uptake of L-carnitine into a stably transfected hOCTN2-MDCK cell monolayer. A HipHop common features pharmacophore was developed and used to search a drug database. Fifty-three drugs, including some not predicted to be inhibitors, were selected and screened in vitro.
A common features pharmacophore was derived from initial screening data and consisted of three hydrophobic features and a positive ionizable feature. Among the 33 tested drugs that were predicted to map to the pharmacophore, 27 inhibited hOCTN2 in vitro (40% or less L-carnitine uptake from 2.5μM L-carnitine solution in presence of 500 μM drug, compared to L-carnitine uptake without drug present). Hence, the pharmacophore accurately prioritized compounds for testing. Ki measurements showed low micromolar inhibitors belonged to diverse therapeutic classes of drugs, including many not previously known to inhibit hOCTN2. Compounds were more likely to cause rhabdomyolysis if the Cmax/Ki ratio was higher than 0.0025.
A combined pharmacophore and in vitro approach found new, structurally diverse inhibitors for hOCTN2 that may possibly cause clinical significant toxicity such as rhabdomyolysis.
The human Organic Cation/Carnitine Transporter 2 (hOCTN2) is a high affinity cation/carnitine transporter widely expressed in human tissues, including skeletal muscle and kidney (1). hOCTN2 is also physiologically important for maintaining the homeostasis of the endogenous compound L-carnitine which is involved in intermediary metabolism(2). The primary role of hOCTN2 is to facilitate the transport of long-chain fatty acids into mitochondria for β-oxidation and subsequent energy production. The mutation of the hOCTN2 gene in humans is also known to cause primary systemic carnitine deficiency with clinical symptoms that include hypoketotic hypoglycemia, cardiomyopathy, and skeletal myopathy (3–4).
HMG-CoA reductase inhibitors, alone or in combination with other medications, are well known to be associated with rhabdomyolysis, a relatively rare condition resulting in a form of muscle weakness (5). Cerivastatin was withdrawn from the US market due to this side effect. The Food and Drug Administration (FDA) also recently issued an alert on the risk of rhabdomyolysis when simvastatin is used with amiodarone. However, the mechanism of muscle injury by HMG-CoA reductase inhibitors is still poorly defined. One possible mechanism involves the inhibition of L-carnitine uptake into muscle and/or inhibition of L-carnitine re-absorption in the kidney, both mediated by hOCTN2. An understanding of the inhibition requirements of hOCTN2 has potential to predict hOCTN2 inhibitors and potential utility in predicting drug-induced secondary carnitine deficiency.
Several reports have identified cationic or zwitterionic drugs as inhibitors and/or substrates of OCTN2 (6; 7–8). However, a systemic approach to study the molecular requirements of OCTN2 inhibition is lacking. In the current study, a pharmacophore was developed to identify the molecular features required for hOCTN2 inhibition. Twenty-seven drugs were initially screened for in vitro hOCTN2 inhibition. A common features pharmacophore was subsequently developed and then applied to search a database of 796 compounds. The pharmacophore, which consisted of three hydrophobic and a positive ionizable feature, identified potential hOCTN2 inhibitors from the database. Experimental testing was conducted on 53 additional compounds to further test the pharmacophore. Among 33 drugs that were predicted to be inhibitors and tested, 27 were observed to be active. Diverse therapeutic classes of drugs were found to be novel potent inhibitors of hOCTN2.
L-[3H]carnitine was purchased from American Radiolabeled Chemicals (St. Louis, MO). Fetal bovine serum, trypsin, and Dulbecco’s modified Eagle medium (DMEM) were purchased from Invitrogen Corporation (Carlsbad, CA). L-carnitine, all drugs, and other chemicals were obtained from Sigma Chemical (St. Louis, MO), Alexis Biochemicals (San Diego, CA), AK Scientific (Mountain View, CA), LKT Labs (St. Paul, MN), Spectrum Chemicals & Laboratory Products (Gardena, CA), Spectrum Pharmacy Products (Tucson, AZ), or TCI America (Portland, OR). Stably transfected hOCTN2-MDCK cells were kindly provided by Xin Ming and Dr. Dhiren R. Thakker from the University of North Carolina-Chapel Hill.
Stably transfected hOCTN2-MDCK cells were cultured at 37 °C, 90% relative humidity, and 5% CO2 atmosphere and fed every 2 days. Media was composed of DMEM supplemented with 10% FBS, 50 units/ml penicillin, and 50 μg/ml streptomycin. Cells were passaged after reaching 80% confluence.
hOCTN2-MDCK cells were seeded at a density of 0.7 million cells/cm2 in 12-well plates (Corning; Corning, NY). To enhance hOCTN2 expression, cells were treated with 10 mM sodium butyrate for 12–15 h at 37 °C prior to uptake or inhibition study.
Stably transfected hOCTN2-MDCK cells were characterized in terms of L-carnitine uptake. Uptake studies were performed at L-carnitine concentrations ranging from 0 to 200 μM and donor solutions were spiked with L-[3H] carnitine. Buffer consisted of either Hank’s balanced salts solution (HBSS) which contains 137 mM sodium chloride or a sodium-free, modified HBSS where sodium chloride was replaced with 137 mM tetraethylammonium chloride. Identical studies were conducted using sodium-containing buffer and sodium-free buffer, since hOCTN2-mediated uptake of L-carnitine is sodium-dependent. L-carnitine uptake was also performed in the presence of sodium using MDCK cells that were not transfected with hOCTN2.
At the end of the assay (10 min), active uptake was terminated by washing cells thrice with chilled sodium-free buffer. Cells were then lysed with 0.25 ml of 1 N NaOH for four hr. Cell lysate was neutralized with 0.25 ml of 1 N HCl and counted for associated radioactivity using an LS6500 liquid scintillation counter (Beckman Instruments, Inc., Fullerton, CA).
A passive transport model (eqn 1) was fitted to uptake data from sodium-free studies:
where J is L-carnitine flux, Pp is the passive L-carnitine permeability coefficient, and S is L-carnitine concentration. The Michaelis-Menten model with parallel passive permeability (eqn 2) was fitted to uptake data from sodium-containing studies:
where Jmax and Kt are the Michaelis-Menten constants for hOCTN2-mediated transport. Equations 1 and 2 were applied sequentially to sodium-free and sodium-containing uptake data to estimate Pp, Kt, and Jmax. The Pp estimate from sodium-free studies was applied to eqn 2.
Twenty-seven drugs were initially screened for hOCTN2 inhibition. These compounds were selected since they are drugs and reflect diverse structures that are either weakly basic or neutral, and are commercially available. Cis-inhibition studies of L-carnitine (2.5 μM with spiked L-[3H] carnitine) uptake were carried out using drug at a single concentration (500 μM) in sodium-containing buffer. At the end of the assay (10 min), active uptake was terminated by washing cells thrice with chilled sodium-free buffer. Cells were then lysed and quantified for radioactivity as described above. Results are expressed in terms of percent uptake of L-carnitine, compared to L-carnitine uptake without drug present.
The computational molecular modeling studies were carried out using Catalyst™ in Discovery Studio 2.0 or 2.1 (Accelrys, San Diego, CA) running on a Centrino Duo processor (Intel, Santa Clara, CA). Common features (HipHop) pharmacophore models attempt to describe the arrangement of key features that are important for biological activity and their generation has been widely described (9–10). Up to 255 molecule conformations were created for the following structures with the BEST conformer generation method, with the maximum energy threshold of 20 kcal/mol.
Pharmacophore common features were initially developed using verapamil and propantheline, which provided the highest inhibition from the inhibition screening studies. These common features served as the template molecules to which chlorpheniramine, diltiazem and imipramine were mapped (Supplemental Table I). Hence, the pharmacophore was based on the five most active compounds that were screened. Physostigmine and guanfacine were also used as less active compounds whose features were excluded from the pharmacophore, in order to highlight features of the five most active compounds. Physostigmine and guanfacine; chlorpheniramine, diltiazem and imipramine; and verapamil and propantheline were labeled principle 0, 1, and 2, respectively (Supplemental Table I).
The SCUT database of 796 compounds (656 drugs in the SCUT 2008 database plus additional drug metabolites and drugs of abuse) was created using structures in the MDL SDF format prior to conversion to a 3D Catalyst database after generating up to 100 molecule conformations with the FAST conformer generation method, with the maximum energy threshold of 20 kcal/mol.
The common features pharmacophore was then applied to screen the SCUT database using the FAST search method, as previously described (11–12). The quality of the molecule mapping to the pharmacophore was determined by the FitValue which is dependent on the proximity of a compound to the pharmacophore feature centroids and the weights assigned to each centroid, where a higher FitValue represents a better fit. An additional pharmacophore with shape restriction was developed and applied. This shape restriction involved the van der Waals shape of propantheline, which was the most active compound from the initial screen.
Pharmacophore database screening resulted in drugs that were predicted to strongly inhibit hOCTN2, as well as those predicted not to inhibit hOCTN2. Fifty-three drugs were subsequently selected and analyzed in this study. These drugs possessed a range of pharmacophore-predicted FitValues and were subjected to the hOCTN2 inhibition assay in order to test the pharmacophore. Inhibition studies were carried out as described above in the “hOCTN2 inhibition screening studies” section. A few compounds which were not soluble at 500 μM were tested at a lower concentration as noted.
Inhibitory Ki values were measured for 14 drugs. Verapamil and propantheline Ki values were measured since these compounds were the most potent from the inhibition screening studies and contributed to pharmacophore development. The other 12 drugs were selected primarily based on the pharmacophore results since they collectively exhibited a range of percent inhibition and FitValues. To measure Ki, inhibition studies were performed as described above, where a range of drug concentrations were applied to inhibit L-carnitine (2.5 μM spiked with L-[3H] carnitine) uptake. The following competitive inhibition model was applied:
where Ki is the competitive inhibition coefficient, I is the concentration of inhibitor, and S is the concentration of L-carnitine (i.e. 2.5 μM). In applying eqn 3, only Ki was estimated. The other three parameters (i.e. Jmax, Kt, and Pp) were estimated from L-carnitine uptake studies without inhibitor using eqn 1 and 2.
Vinblastine, the most potent inhibitor, was subjected to Lineweaver-Burk analysis where both vinblastine (0, 50, and 100μM) and L-carnitine (1, 2.5, and 5μM) concentrations were varied (13). Inhibition studies of L-carnitine by vinblastine were performed as described above. The competitive inhibition model (i.e. eqn 3) was fit to the data. The non-competitive inhibition model (i.e. eqn 4) was also fit to the data.
Percent uptake in presence of drug was expressed as mean ± S.E.M derived from three independent wells for each concentration of inhibitor, compared to L-carnitine uptake in absence of drug. Nonlinear curve fitting was performed using WinNonlin 4.1 (Pharsight, Mountain View, CA). Statistical analysis between Cmax/Ki ratio and rhabdomyolysis was performed using Pearson’s chi-square (χ2) test in MINITAB version 14.11.1 (Minitab Inc., State College, PA). The criterion of significance was p < 0.05.
To confirm the expression of functional carnitine transport activity in hOCTN2-MDCK cells, L-carnitine uptake was measured in the presence and absence of sodium. In Figure 1, the uptake of L-carnitine in the presence of sodium demonstrated saturable kinetics in the substrate range of 0–200 μM, while the uptake of L-carnitine in the absence of sodium demonstrated linear kinetics. In the absence of sodium, lower L-carnitine uptake was observed, in comparison to studies with sodium. Fitted Km and Vmax were 5.33 (±0.54)μM and 0.808 (±0.019) pmol/sec/cm2, respectively. The passive permeability of L-carnitine across hOCTN2-MDCK cells in the absence of sodium was a low 0.344 (±0.003) × 10−6 cm/sec. The uptake of L-carnitine into untransfected MDCK cells was also low and exhibited a passive permeability of 0.736 (±0.011) × 10−6 cm/sec.
Twenty-seven drugs were initially screened for inhibition of L-carnitine uptake into hOCTN2-MDCK cells. A wide range of inhibition potency was found, from 7.43 (± 0.19) percent uptake to 100 (± 3) percent uptake, compared to L-carnitine uptake without drug present (Table I). The five most potent inhibitors from this initial screening were propantheline, verapamil, chlorpheniramine, diltiazem and imipramine (i.e. bolded compounds in Table I).
A common features pharmacophore was developed using propantheline and verapamil as the template molecules, to which chlorpheniramine, imipramine, diltiazem were mapped. Physostigmine and guanfacine were applied as less active compounds whose features were excluded from the pharmacophore, in order to highlight features of the five most active compounds. Supplemental Table I lists molecules used in the HipHop pharmacophore and pharmacophore settings. The resulting HipHop pharmacophore consisted of three hydrophobic features and a positive ionizable (Figure 2A). The hydrophobic features radiate from a positive ionizable feature with distances of ~7 to 9.4 Å (Supplemental Figure 1); the five most potent screened inhibitors each mapped to the pharmacophore (Figure 2A, B). Physostigmine and guanfacine did not map to the pharmacophore. This approach was used in order to train the model to focus on the five most active compounds and exclude the features in physostigmine and guanfacine from this model.
The common features pharmacophore without shape restriction was used to search the database of 796 compounds and identified 136 molecules as potential inhibitors (Supplemental Table II). Four of the five inhibitors that were used to build the model were retrieved from the database as potential inhibitors (i.e. verapamil, chlorpheniramine, imipramine and diltiazem, but interestingly not propantheline). For each compound in the database that was identified as a potential inhibitor, a FitValue was computed. A higher FitValue anticipates greater inhibition. A second pharmacophore, denoted the pharmacophore with shape restriction, was made more restrictive by adding the van der Waals volume around propantheline (Figure 2C). This pharmacophore with shape restriction resulted in only 49 compounds as potential inhibitors and returned only imipramine from the five pharmacophore compounds (Supplemental Table III).
Fifty-three drugs were selected to test the pharmacophore via in vitro inhibition of L-carnitine uptake. These compounds were selected due to commercial availability and their wide range of predicted inhibition (i.e. range from anticipated potent hOCTN2 inhibitors to those that did not map to the pharmacophore). Table II lists the 32 compounds that were tested from the list of 136 compounds retrieved by the pharmacophore without shape restriction as potential inhibitors. These 32 compounds were selected as likely inhibitors, but with a wide range of FitValues, and were commercially available. Table III lists an additional 21 compounds that were tested. These compounds were not retrieved by the pharmacophore search of the database. Rather, these compounds were carnitine mimics, compounds previously reported to inhibit OCTN2, or drugs that cause rhabdomyolysis. Hence, including the 27 compounds from the initial hOCTN2 inhibition screening, a total of 80 compounds were tested for hOCTN2 inhibition in this study. Trifluoperazine, with a high pharmacophore FitValue of 3.408, was the most potent inhibitor found from all single concentration inhibition studies, with a percent uptake of 0.24 (±0.03) [i.e. over 99% inhibition].
Twelve compounds in Table II (i.e. over one third of those tested) were more potent than propantheline (percent uptake of 7.43±0.19 %), which was the most potent compound from the initial screening study of 27 compounds. Twenty-seven of the 32 drugs tested (84.4%) were found to be potent inhibitors, using 40% or lower percent uptake as the criterion to denote a potent inhibitor. These results reflect the capability of the pharmacophore to predict novel and more active hOCTN2 inhibitors. It should be noted that hOCTN2-mediated translocation can be either sodium-dependent or sodium-independent (14). Experiments here employed the substrate L-carnitine, whose translocation across hOCTN2 is sodium-dependent. Hence, observations here of hOCTN2 inhibition (regardless of mechanism of inhibition) do not necessarily reflect the sodium-independent pathway, as L-carnitine is not a probe for that pathway.
An additional 21 compounds were selected for in vitro testing without the use of the pharmacophore (Table III). Of these compounds, only emetine (which was not in the SCUT database) mapped to the pharmacophore. Emetine inhibited OCTN2 moderately (46.9% percent L-carnitine uptake). Among the remaining 20 compounds which did not map to the pharmacophore, only mirtazapine was found to be potent inhibitor. These results of compounds predicted not to map to the pharmacophore further support the utility of using the OCTN2 inhibition pharmacophore for selecting compounds for testing.
Of the 53 compounds tested in Table II and III, 33 compounds mapped to the pharmacophore, while 20 compounds did not map to the pharmacophore. Of the 33 compounds that mapped the pharmacophore, 27 inhibited hOCTN2. Of the 20 compounds that did not map to the pharmacophore, one inhibited hOCTN2. These results indicate that the pharmacophore had acceptable ability to screen for and select inhibitors over non-inhibitors.
Additionally, a pharmacophore with a van der Waals shape restriction around propantheline was also developed (Figure 3C). The purpose of adding a shape volume was to create a more restrictive features hypothesis in order to reduce the number of retrieved hits and to find molecules most similar to the shape and volume of propantheline. The pharmacophore with shape restriction identified only 49 molecules from the SCUT database (Supplemental Table III), which were a subset of the original dataset (Supplemental Table II). Of these 49, 17 molecules were tested in vitro (Supplemental Table III), and 14 drugs (93.7%) were found to be potent inhibitors based on our selection criteria.
Fourteen drugs were selected for Ki determination (Table IV). Verapamil and propantheline Ki values were measured since these compounds were the most potent from the screening studies and were used to develop the pharmacophore model. The other 12 drugs were selected since they collectively exhibited a range of percent inhibition values and FitValues. The hOCTN2 inhibition profile of vinblastine is shown in Figure 4. The hOCTN2 pharmacophore model predicted this drug as one of the most potent inhibitors of hOCTN2. Experimentally, the Ki of vinblastine was 4.85μM and was the most potent Ki value in this study. Vinblastine has not been previously been reported to be an OCTN2 inhibitor.
Figure 4 illustrates the Lineweaver-Burk plot of vinblastine inhibition of L-carnitine uptake. Vinblastine and L-carnitine concentrations were varied. The three lines show the linear fits to L-carnitine uptake data from experiments that employ vinblastine concentrations of 0 μM (filled circle), 50 μM (open circle) and 100 μM (filled triangle). Higher vinblastine concentration resulted in a greater slope in Figure 4. However, the y-intercept did not change and this is consistent with a competitive inhibition model. Additionally, AIC from nonlinear regression indicated the competitive inhibition model was better fitting than the non-competitive inhibition model.
An in silico common features pharmacophore modeling method and an in vitro cell culture approach have been combined to predict the molecular requirements of hOCTN2 inhibition and identify additional drugs as novel potent inhibitors of hOCTN2. Some of the most potent hOCTN2 inhibitors reflect diverse classes of molecules, both structurally and therapeutically, such as phenothiazine antipsychotics, atypical antipsychotics, selective estrogen receptor modulators, calcium channel blockers, anti-cancer compounds, and tricyclic antidepressants.
Considering the physiologically important role of hOCTN2 in carnitine transport as well as the widespread expression of hOCTN2 in human tissues (1), the high affinity inhibition of hOCTN2 of these therapeutic agents may have pharmacological implications. Secondary carnitine deficiency is known to be induced by long-term treatment of various drugs such as emetine (15) and cephaloridine (16). One suggested mechanism is the inhibition of hOCTN2-mediated L-carnitine uptake into various tissues, especially kidney, heart, and skeletal muscle, where hOCTN2 is highly expressed. Studies have shown that cephaloridine enhances renal excretion of carnitine, consistent with the inhibition of carnitine renal re-absorption by cephaloridine (17–18). In vitro Hela cell line studies have demonstrated that cephaloridine is a potent hOCTN2 inhibitor (IC50 = 0.79 ± 0.10 mM), as well as a substrate (16).
Since carnitine deficiency is linked with rhabdomyolysis (19–20), hOCTN2 inhibition might also be a possible contributor to drug-induced rhabdomyolysis since hOCTN2 inhibition might limit L-carnitine uptake. We have considered the possible association between clinical rhabdomyolysis and hOCTN2 inhibition by compiling a list of those compounds that were found to be the most potent inhibitors and least potent inhibitors in vitro, as well as compounds whose Ki values were determined (Table V). The Cmax/Ki ratio was also computed as a potential predictor for clinical rhabdomyolysis. Among the 12 compounds that were associated with rhabdomyolysis, 10 (83.3%) have a Cmax/Ki ratio higher than 0.0025. This ratio of 0.0025 separates the top half of Cmax/Ki values from the bottom half of Cmax/Ki values. In contrast, among 21 compounds that were not associated with rhabdomyolysis, only six (28.6%) have a Cmax/Ki ratio higher than 0.0025. Clinical rhabdomyolysis was associated with a Cmax/Ki value above 0.0025 (Pearson’s chi-square test p=0.00247).
HMG-CoA reductase inhibitors are well known to cause rhabdomyolysis. Cervistatin was removed from the market due to this side effect. FDA recently issued an alert for the risk of rhabdomyolysis when simvastatin is administered with amiodarone. In the present study, both cervistatin and atorvastatin were found to be inhibitors of hOCTN2. The Ki of cervistatin was 425 (±12) μM. Amiodarone inhibited hOCTN2 at a low micromolar concentration.
While clinical rhabdomyolysis was associated with a Cmax/Ki value above 0.0025, it should be emphasized that the use of Cmax/Ki as a predictor would result in false positives (e.g. rifampin) and false negatives (e.g. selegiline). If one were to extrapolate pharmacophore performance to rhabdomyolysis, the pharmacophore would exhibit false positives and false negatives. For example, succinylcholine, cerivastatin, levofloxacin, and selegiline did not map the pharmacophore, although are associated with rhabdomyolysis. However, the pharmacophore did successfully retrieve amiodarone, nelfinavir, thioridazine, nefazodone, and clozapine for in vitro testing, all of which yielded Cmax/Ki values above 0.0025 and are associated with rhabdomyolysis.
This observed association between rhabdomyolysis and Cmax/Ki is preliminary and perhaps should only be interpreted as an ongoing effort to leverage cell culture methods to screen for transporter-mediated drug side effects. While laboratory methods have been improved to anticipate metabolism-based drug-drug interactions (21), predictability for drug side effects and drug-drug interactions from transporter assays is not well developed. Further evidence is needed to better assess an association between rhabdomyolysis and Cmax/Ki. For example, direct association between drug plasma level and reduced tissue L-carnitine level may merit subsequent study. A rationale for a specific Cmax/Ki value that considers drug pharmacokinetics (e.g. tissue distribution, plasma protein binding) would also be advantageous, as a Cmax/Ki value of 0.0025 is applied here since this value separated the top half drugs from bottom half drugs in Table V.
A common features pharmacophore was developed and consisted of three hydrophobic features and a positive ionizable feature (Figure 2). These features are thus expected to be important for molecular interactions with hOCTN2. Of course, it is possible that in the resultant pharmacophore, not all of the pharmacophore features are important as they are just “common features” in the actives.
The pharmacophore models with and without shape restriction performed similarly and had very few false positives. However, the predictive power of the two pharmacophore models differed in terms of retrieving the five active drugs in the training set. While the pharmacophore without shape restriction retrieved four of the five drugs (it is uncertain why it failed with propantheline), the pharmacophore with the van der Waals shape restriction only retrieved imipramine. Moreover, the latter pharmacophore with shape restriction missed identifying several compounds that were subsequently found to be potent novel inhibitors, such as vinblastine and amlodipine, which were found with the pharmacophore without shape restriction. These results indicate pharmacophore with shape restriction is too restrictive in terms of shape, reliant on propantheline. Therefore, the pharmacophore without shape restriction was used to obtain FitValues for all database compounds, and is viewed here as the preferred pharmacophore to date for screening. However, the shape restricted pharmacophore still demonstrated a much higher percent active rate than the pharmacophore alone. This shape restricted pharmacophore approach might be useful if much larger databases were to be screened as this latter pharmacophore has the potential to retrieve hit lists approximately one third the size of the other pharmacophore with a higher level of true positives.
One report investigated the structural requirements of hOCTN2 inhibition using L-carnitine and cephaloridine as training compounds to derive a pharmacophore that was used to search the Merck Index (22). The pharmacophore included a constantly positively charged nitrogen atom and a carboxyl, nitrile or ester group connected by a 2–4-atom linker. The majority of the compounds retrieved were carnitine mimics. The pharmacophore search appeared to yield fewer, less chemically diverse, less potent, and much less drug-like compounds than reported here in the present study. Additionally, since hOCTN2-mediated carnitine uptake is sodium-dependent while hOCTN2-mediated cation uptake is sodium-independent, the recognition site of cations and carnitine may not significantly overlap.
Two missense mutations of OCTN2 (i.e. L352R, P478L) causing primary carnitine deficiency have been tested in vitro and result in complete loss of carnitine transport function. However, only the L352R mutant was associated with loss of organic cation transport function, whereas the P478L mutant had higher organic cation transport activity than the wild type OCTN2 (23). Hence, the superposition of L-carnitine with celphaloridine to derive the previously reported pharmacophore may be less applicable to drugs, since the pharmacophore for carnitine mimics and diverse cation/zwitterion drugs might be different overall. For example, verapamil was discussed in the report as a molecule that does not follow their pharmacophore since verapamil lacks a permanent positive charge but behaves as a potent inhibitor (22). Meanwhile, verapamil was used to derive our pharmacophore. It would therefore indicate that we are potentially modeling a unique, or a partially overlapping site in this study, compared to the previous published pharmacophore. The retrieval of diverse and potent inhibitors by our model also indicates it is well validated, in contrast to the previous model.
In summary, hOCTN2 inhibition molecular requirements were elucidated by the integration of computational modeling and in vitro testing. Twenty-seven drugs were screened initially for their potential to inhibit uptake of L-carnitine. A common features pharmacophore was also used to search a database of compounds. Fifty-three drugs, including some predicted not to be inhibitors, were subsequently selected and screened in vitro to test the pharmacophore. The pharmacophore performed exceptionally well predicting 27 out of 33 drugs as hOCTN2 inhibitors in vitro. Diverse therapeutic classes of drugs were also found to be low micromolar inhibitors, including many drugs not previously known to inhibit hOCTN2. This validation suggests that the pharmacophore consisting of 3 hydrophobic features ~7 to 9.4 Å from a positive ionizable feature captures important interactions between OCTN2 inhibitors and the transporter. The findings in this study may also have clinical implications for drug interactions mediated by compounds inhibiting hOCTN2.
This work was supported in part by National Institutes of Health grant DK67530. The authors kindly acknowledge Dr. Xin Ming and Dr. Dhiren R. Thakker (University of North Carolina-Chapel Hill) for providing the hOCTN2-MDCK cell line used in this study. The authors gratefully acknowledge Dr. Matthew D. Krasowski for his assistance in creating the SCUT 2008 database supplemented with metabolites and drugs of abuse. The authors also thank Accelrys, San Diego, CA for making Discovery Studio Catalyst available.