Structure-based discovery of prescription drugs that interact with the norepinephrine transporter, NET.
ABSTRACT The norepinephrine transporter (NET) transports norepinephrine from the synapse into presynaptic neurons, where norepinephrine regulates signaling pathways associated with cardiovascular effects and behavioral traits via binding to various receptors (e.g., β2-adrenergic receptor). NET is a known target for a variety of prescription drugs, including antidepressants and psychostimulants, and may mediate off-target effects of other prescription drugs. Here, we identify prescription drugs that bind NET, using virtual ligand screening followed by experimental validation of predicted ligands. We began by constructing a comparative structural model of NET based on its alignment to the atomic structure of a prokaryotic NET homolog, the leucine transporter LeuT. The modeled binding site was validated by confirming that known NET ligands can be docked favorably compared to nonbinding molecules. We then computationally screened 6,436 drugs from the Kyoto Encyclopedia of Genes and Genomes (KEGG DRUG) against the NET model. Ten of the 18 high-scoring drugs tested experimentally were found to be NET inhibitors; five of these were chemically novel ligands of NET. These results may rationalize the efficacy of several sympathetic (tuaminoheptane) and antidepressant (tranylcypromine) drugs, as well as side effects of diabetes (phenformin) and Alzheimer's (talsaclidine) drugs. The observations highlight the utility of virtual screening against a comparative model, even when the target shares less than 30% sequence identity with its template structure and no known ligands in the primary binding site.
- SourceAvailable from: Lei Xie[Show abstract] [Hide abstract]
ABSTRACT: Genome-Wide Association Studies (GWAS), whole genome sequencing, and high-throughput omics techniques have generated vast amounts of genotypic and molecular phenotypic data. However, these data have not yet been fully explored to improve the effectiveness and efficiency of drug discovery, which continues along a one-drug-one-target-one-disease paradigm. As a partial consequence, both the cost to launch a new drug and the attrition rate are increasing. Systems pharmacology and pharmacogenomics are emerging to exploit the available data and potentially reverse this trend, but, as we argue here, more is needed. To understand the impact of genetic, epigenetic, and environmental factors on drug action, we must study the structural energetics and dynamics of molecular interactions in the context of the whole human genome and interactome. Such an approach requires an integrative modeling framework for drug action that leverages advances in data-driven statistical modeling and mechanism-based multiscale modeling and transforms heterogeneous data from GWAS, high-throughput sequencing, structural genomics, functional genomics, and chemical genomics into unified knowledge. This is not a small task, but, as reviewed here, progress is being made towards the final goal of personalized medicines for the treatment of complex diseases.PLoS Computational Biology 05/2014; 10(5):e1003554. · 4.83 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: Aflatoxin, a naturally occurring mycotoxin, produced by Aspergillus parasiticus infects peanuts and many other legumes. Among 23 enzymatic reactions involved in aflatoxin biosynthesis, only 15 are identified so far. Versicolorin B synthase (vbs) is the key enzyme involved in aflatoxin biosynthesis. It converts hydroxy versiconal acetate (VHA) to versicolorin B (VERB), which is finally converted into aflatoxin in a series of reactions. We selected two naturally available compounds, allicin and ajoene present in onion and garlic, to analyse their inhibitory effect on aflatoxin biosynthesis. Additionally, we virtually derived a new compound called ajocin (allicin + ajoene) to study their synergistic effect in preventing aflatoxin biosynthesis. We aim to compare the inhibitory action of these compounds against 15 known enzyme targets through docking simulations using Autodock. The best inhibitory action is reported by ajocin against the key enzyme, vbs. In silico docking studies thus confirm the enhanced inhibitory property of ajocin and reduce the risk of aflatoxin (carcinogen) exposure to human.
- [Show abstract] [Hide abstract]
ABSTRACT: Discovery of new inhibitors of the plasmalemmal monoamine transporters (MATs) continues to provide pharmacotherapeutic options for depression, addiction, attention deficit disorders, psychosis, narcolepsy and Parkinson's disease. The recent windfall of high-resolution MAT structural information afforded by x-ray crystallography has enabled the use of in silico methods in MAT drug discovery. Here, virtual screening (VS) of the PubChem small molecule structural database using the S1 (primary substrate) ligand pocket of a serotonin transporter computational homology model yielded 19 prominent "hit" compounds. In vitro pharmacology of these VS hits revealed four structurally unique MAT substrate uptake inhibitors with high nanomolar affinity at one or more of the three MATs. In vivo characterization of three of these hits revealed significant activity in a mouse model of acute depression, at doses that did not elicit untoward locomotor effects. This constitutes the first report of MAT inhibitor discovery using the primary substrate pocket as a VS tool. Novel-scaffold MAT inhibitors offer hope of new medications that lack the many classic adverse effects of existing antidepressant drugs.ACS Chemical Neuroscience 07/2014; · 4.21 Impact Factor
Structure-based discovery of prescription drugs that
interact with the norepinephrine transporter, NET
Avner Schlessingera,b,c,1, Ethan Geiera, Hao Fana,b,c, John J. Irwinb,c, Brian K. Shoichetb,c,
Kathleen M. Giacominia, and Andrej Salia,b,c,1
aDepartment of Bioengineering and Therapeutic Sciences,
Biosciences, University of California, San Francisco, CA 94158
bDepartment of Pharmaceutical Chemistry, and
cCalifornia Institute for Quantitative
Edited by Barry Honig, Columbia University, Howard Hughes Medical Institute, New York, NY, and approved July 20, 2011 (received for review April 15, 2011)
The norepinephrine transporter (NET) transports norepinephrine
from the synapse into presynaptic neurons, where norepinephrine
regulates signaling pathways associated with cardiovascular
effects and behavioral traits via binding to various receptors
(e.g., β2-adrenergic receptor). NET is a known target for a variety
of prescription drugs, including antidepressants and psychosti-
mulants, and may mediate off-target effects of other prescription
drugs. Here, we identify prescription drugs that bind NET, using
virtual ligand screening followed by experimental validation of
predicted ligands. We began by constructing a comparative struc-
tural model of NET based on its alignment to the atomic structure
of a prokaryotic NET homolog, the leucine transporter LeuT. The
modeled binding site was validated by confirming that known
NET ligands can be docked favorably compared to nonbinding
molecules. We then computationally screened 6,436 drugs from
the Kyoto Encyclopedia of Genes and Genomes (KEGG DRUG)
against the NET model. Ten of the 18 high-scoring drugs tested ex-
perimentally were found to be NET inhibitors; five of these were
chemically novel ligands of NET. These results may rationalize the
efficacy of several sympathetic (tuaminoheptane) and antidepres-
sant (tranylcypromine) drugs, as well as side effects of diabetes
(phenformin) and Alzheimer’s (talsaclidine) drugs. The observa-
tions highlightthe utility ofvirtualscreeningagainsta comparative
model, even when the target shares less than 30% sequence
identity with its template structure and no known ligands in the
primary binding site.
that channel neurotransmitters, amino acids, and osmolytes into
the cell (1). These transporters regulate a variety of biological
activities such as synaptic transmission, neurotransmitter recy-
cling, metabolism, and fluid homeostasis. The norepinephrine
or noradrenaline transporter (NET, SLC6A2) is a monoamine
transporter mostly expressed in the peripheral and central ner-
vous systems (CNS). NET recycles neurotransmitters, primarily
norepinephrine, but also serotonin and dopamine, from synaptic
spaces into presynaptic neurons (2, 3). Mutations in NET have
been associated with a variety of behavioral disorders, such as
attention-deficit hyperactivity disorder (ADHD) and panic disor-
der, as well as severe orthostatic hypotension (4). Low levels
of the NET mRNA and protein have been found in brains of
suicidal patients with major depression (4).
NET is a target of drugs treating a variety of mood and beha-
vioral disorders, such asdepression, anxiety, andADHD(4).Many
of these drugs inhibit the uptake of norepinephrine into the pre-
synaptic cells through NET (4). These drugs therefore increase the
availability of norepinephrine for binding to postsynaptic receptors
that regulate adrenergic neurotransmission. NET inhibitors can be
specific. For example, the ADHD drug Atomoxetine (Strattera®)
is a norepinephrine reuptake inhibitor (NRI) that is highly selec-
tive for NET. NET inhibitors can also bind multiple targets,
increasing their efficacy as well as their potential patient popula-
tion. For instance, the antidepressant Venlafaxine (Effexor®) is a
serotonin-NRI (SNRI) that targets both NET and the serotonin
he solute carriers 6 (SLC6) or the sodium:neurotransmitter
symporters is a family of ion-dependent transporter proteins
transporter (SERT, SLC6A4) (4). Because of the lack of high-
resolution structural information, most drug discovery efforts
targeting NETand other SLC6 transporters, including SERTand
dopamine transporter (DAT, SLC6A3), haverelied on quantitative
structure-activity relationship (QSAR) approaches and pharmaco-
phore modeling (5).
As for other SLC6 family members, NET is predicted to have
one domain containing 12 transmembrane helices (2). No struc-
tures of human SLC6 members have been determined at atomic
resolution; however, the leucine transporter LeuT from the bac-
terium Aquifex aeolicus has been determined by X-ray crystallo-
graphy in different conformations, with various ligands, including
substrates and inhibitors (6). Additional crystallographic and
model structures of related transporters (7–11) revealed that the
SLC6 family members transport ligands across the cell membrane
via the “alternating access” transport mechanism (12, 13). The
structures of LeuT in complex with different amino acid ligands
suggested a competitive inhibition mechanism (14). In particular,
for a substrate to be transported efficiently, it needs to bind to
the binding site on the transporter surface as well as fit within the
binding cavityof the“occluded” transporter state(14); in contrast,
a competitive inhibitor binds to the binding site, but is too large to
ported through the transporter. In addition to the primary binding
site (S1 site), the family members may have additional binding
sites such as the clomipramine-binding site on LeuT (6, 15).
Here, we identify approved and “natural” drugs that inhibit
NET, using comparative protein structure modeling and virtual
ligand screening against the primary binding site, followed by
experimental testing of predicted ligands. The ability of these re-
sults to explain efficacy and side effects that may result from NET
inhibition are considered, as is the potential use of our approach
for discovering ligands of other membrane transporters.
Model of NET Binding Site and Its Assessment. NET was modeled
based on the structure of the leucine transporter LeuT from
Aquifex aeolicus (Fig. 1A and Figs. S1 and S2). The comparative
model contains the whole transmembrane domain, including
the 12 transmembrane helices, and the residues comprising the
primary binding site (Figs. S1 and S2). The binding site model
was assessed by its ability to discriminate between known ligands
and likely nonbinders (“decoys”) using two measures (Fig. 1B and
Table S1) (16, 17).
Author contributions: A. Schlessinger, E.G., J.J.I., B.K.S., K.M.G., and A. Sali designed
research; A. Schlessinger, E.G., and H.F. performed research; A. Schlessinger, E.G., H.F.,
J.J.I., B.K.S., K.M.G., and A. Sali contributed new reagents/analytic tools; A. Schlessinger,
E.G., H.F., J.J.I., B.K.S., K.M.G., and A. Sali analyzed data; and A. Schlessinger and A. Sali
wrote the paper.
The authors declare no conflict of interest.
This article is a PNAS Direct Submission.
1To whomcorrespondencemaybe addressed.E-mail:email@example.com firstname.lastname@example.org.
This article contains supporting information online at www.pnas.org/lookup/suppl/
www.pnas.org/cgi/doi/10.1073/pnas.1106030108PNAS Early Edition
1 of 6
First, we calculated the enrichment factor (EFn)—the fraction
pared to their fraction in the entire docking database that includes
both ligands and decoys (Eq. S1). In particular, we computed the
early-stage enrichment factor at 1% of the database (EF1) as well
as the late-stage enrichment factor at 20% of the database (EF20)
(TableS1)(16–18).Our refinedmodel obtainedthehighest enrich-
ment scores among all models (EF1of 13.5 and EF20of 3.2;
Table S1), even though the enrichment value was not considered
during its refinement. These enrichment factor values are consis-
tent with a relatively accurately modeled binding site (16).
Second, we calculated the logarithm of the area under the en-
richment curve as a measure for docking accuracy (logAUC;
Eq. S2, Fig. 1B, and Table S1) (16–18). A random selection of
compounds from the mixture of actual ligands and decoys yields
the logAUC of 14.5. Again, our refined model received the high-
est enrichment score among all models (logAUC of 37.6;
Table S1). As for the enrichment factor values above, such a
logAUC value suggests that the model is suitable for selecting
ligands for experimental testing (16).
The significance of the observed enrichment factors and
logAUC values can also be appreciated by comparing them with
the corresponding values for docking of the NET ligand to the
LeuT template structure. As expected from the dissimilar binding
profiles of the primary binding sites in NETand LeuT (i.e., mono-
amine neurotransmitters and leucine, respectively), the enrich-
ment of NET ligands yielded by the template (logAUC of 14.7
and EFmaxof 1.6) was comparable to that of random selection
(Fig. 1B and Table S1).
Mode of NET-Ligand Interaction. The modeled binding site of NET
is small, thus limiting the size of the ligands and their binding
modes (Fig. 1 and Figs. S2 and S3). Moreover, the binding site
consists of several hydrophobic residues that may contribute to
an increased binding affinity via van der Waals interactions with
the ligand and the hydrophobic effect. For example, Ala145 and
Val148 are predicted to allow for a hydrophobic effect with the
aryl portion of norepinephrine (Figs. 1 and 2). Previous studies of
SERT showed that its residues at the corresponding positions
(i.e., Ala169 and Ile172) also allow for a similar hydrophobic
effect with its natural substrate 5-HT (19).
In addition, several polar interactions occur in the model of the
NET-ligand complexes (Figs. 1A and 2). The modeled Phe72
forms π-cation interactions with norepinephrine via its benzyl
side chain, and a hydrogen bond via the amide oxygen of its main
chain; oxygen atom of Ala145 makes a hydrogen bond with the
catechol group of norepinephrine (Fig. 1A); the side chain of
Asp75 makes ionic interactions with the amine group of norepi-
nephrine (Fig. 1A). The corresponding aspartate residues in the
other two human SLC6 monoamine transporters [Asp79 in DAT
(20) and Asp98 in SERT (19)] are also predicted to make similar
critical interactions with their ligands. The model of SERT
suggests that this aspartate forms an ionic interaction with the
sodium ion “Na1”, adopting a conformation that cannot be ruled
out for NET (19).
Virtual Screening of Drugs. The refined model of NETwas compu-
tationally screened against a library of known drugs from the
Kyoto Encyclopedia of Genes and Genomes (KEGG DRUG)
database (21), filtered for molecules that are suitable for ligand
docking (SI Text). The filtered set included 6,436 molecules,
including drugs that are marketed in Europe, Japan, and the
United States, as well as natural drugs and traditional Chinese
medicines. Interestingly, several compounds that were ranked
highly in our screen have been shown previously to bind NET. For
example, the hypotension drug metaraminol (no. 1 hit) (22) and
the nasal decongestants ephedrine (no. 2) (23) are well-charac-
terized NET ligands. In our model, these molecules and other
top hits have similar binding modes to that of norepinephrine
(Fig. S4), interacting via key NETresidues conserved among the
monoamine transporters. For example, the norephedrine amine
group forms ionic interactions with the side chain of Asp75, and
its aromatic ring interacts with the aromatic rings of Phe72,
Tyr152, and Phe317. The correct identification of drugs known to
interact with NET increases our confidence in the NET binding
site model and therefore also in newly predicted ligands, similarly
to the correct identification of known substrates (above).
The 200 highest-ranked drugs (the top 3.1% of the drug
library) were analyzed manually for the similarities of their pre-
dicted poses to those in structurally defined complexes, frequent
scaffolds, and common pharmacological function (Dataset S1)
(24, 25). For example, the drug tuaminoheptane was ranked no.
18. The pose of tuaminoheptane was similar to the pose of known
ligands, such as norepinephrine (Fig. 2E). Moreover, tuaminohep-
tane is chemically similar to other top hits, such as octodrine, hep-
taminol, and milacemide (hits nos. 29, 86, and 122, respectively;
Dataset S1), further suggesting that molecules belonging to this
chemotypemay interact withNET. Finally, tuaminoheptane,which
is typically used as a nasal decongestant, is a stimulant of the
sympathetic nervous system, similarly toseveral other known drugs
targeting NET, such as pseudoephedrine and ephedrine.
For experimental testing, we excluded drugs that have been
already shown to bind NET (SI Text). The 18 drugs selected for
testing were classified into two groups.
The first “high-confidence” group included five molecules
(Tables S2 and S3) that have predicted binding modes nearly
identical to that of the natural substrate norepinephrine, interact-
ing with almost all key NETresidues (Figs. 1 and 2). These mo-
lecules are chemically similar to known NET ligands (Table S2),
sharing key pharmacophoric characteristics: They consist of an
aromatic ring and a positively charged amine group (5) (Fig. 2F).
For example, adrenalone, which is a topical nasal decongestant,
hemostatic, and vasoconstrictor, is a keton form of the natural
substrate epinephrine (Table S2). Adrenalone is chemically simi-
lar to known NET ligands [Tanimoto coefficient (Tc) of 0.40
and 0.61 for the EFCP4 and Daylight fingerprints, respectively
(SI Text); Table S2]. Additionally, similarly to norepinephrine,
adrenalone contains an aromatic ring that forms hydrophobic in-
teractions with Phe72, Tyr152, and Phe317, an amine group that
forms an ionic bond with Asp75, and a hydroxyl group that forms
a hydrogen bond with Ala145.
% of ranked database
% of known ligands found
dicted structure of the NET–norepinephrine complex.
Norepinephrine is colored in orange, with oxygen, ni-
trogen, and hydrogen atoms in red, blue, and white,
respectively. Sodium ions are visualized as purple
spheres. NET’s transmembrane helices are depicted
as white ribbons. Key residues are displayed as sticks;
the three hydrogen bonds between norepinephrine
and NET (involving residues Ala145, Phe72, and
Asp75) are shown as dotted gray lines. (B) Enrichment
plots for various structures: the refined NET model
(blue), random selection (red), the initial NET model
(green), and the LeuT template structure (orange).
Validation of modeling and docking. (A) Pre-
2 of 6
www.pnas.org/cgi/doi/10.1073/pnas.1106030108Schlessinger et al.
The second “medium-confidence” group included 13 drugs
(Tables S2 and S3). Their predicted binding modes are dissimilar
from the binding of the natural substrates. The majority of these
drugs have only one of the key pharmacophoric characteristics of
the known NETsubstrates—a polar group that interacts with the
key residue Asp75 (Figs. 1 and 2). For example, tuaminoheptane is
chemically dissimilar from known NET ligands [Tc of 0.23 (0.18);
Table S2]. Tuaminoheptane has an amine group that makes ionic
interactions with Asp72 but lacks an aromatic ring to make inter-
actions with the hydrophobic side of the binding site; alternatively,
the tuaminoheptane heptane group allowsfora hydrophobic inter-
action with Val148 and Tyr152 (Fig. 2E).
Experimental Characterization of Predicted NET Inhibitors. The 18
drugs selected above were tested experimentally for their ability
to inhibit [3H]norepinephrine uptake by NET in human embryo-
nic kidney (HEK) cells; desipramine, a known NET inhibitor,
served as a positive control (Fig. 3 and Table 1). [3H]norepi-
nephrine uptake was measured in the presence of 10 and 100 μM
drug concentrations: Although many drugs are therapeutically
dosed to plasma levels of approximately 10 μM, they can poten-
tially reach concentrations as high as 100 μM or greater in the
peripheral tissues. In addition, for many CNS-acting drugs and
endogenous compounds, the values for the inhibition constants
(Ki) at their corresponding receptor target sites have been re-
ported at approximately 10 μM, whereas maximal receptor
inhibition can be achieved at 100 μM for many compounds (26).
The potencies observed at 10-μM concentration ranged from
75% inhibition (tranylcypromine) to no inhibition (e.g., metfor-
min); the potencies at 100-μM concentrations ranged from 90%
(levonordefrin) to no inhibition (e.g., 6-mercaptopurine).
All high-confidence hits exhibited a significant level of inhibi-
tion: Substrate uptake by NETwas reduced to 25% (14%), 68%
(10%), 58% (20%), 99% (27%), and 70 (28%) at 10-μM (100-μM)
concentrations of tranylcypromine, levonordefrin, norfenefrine,
adrenalone, and octopamine, respectively (Fig. 3 and Table 1); the
half maximal inhibitory concentration (IC50) value of tranylcypro-
mine was 8.7 μM (Fig. 4 and Table 1). Supporting these results,
levonordefrin interactionwithNETwas alsofoundtobedocumen-
ted in the DrugBank database (27).
Several medium-confidence hits also exhibited a significant
level of inhibition: Substrate uptake by NET was reduced to
87% (44%) and 87% (79%) at 10-μM (100-μM) concentrations
Predicted binding modes of the known substrate nor-
epinephrine (A), and four ligands discovered in the
docking screen (B–E). Residues making polar interac-
tions with the ligand are illustrated with sticks; car-
bon atoms are colored in white, nitrogen atoms in
blue, and oxygen atoms in red; hydrogen bonds
are represented by dotted gray lines. The predicted
pose of the known ligand is shown in orange sticks
in A, and in green lines in B–E). The compounds
are adrenalone (B), tranylcypromine (C), phenformin
(D), and tuaminoheptane (E). The proposed key inter-
actions between NET and norepinephrine are high-
lighted. (F) Protein–ligand hydrogen bonds are
represented by dashed lines. Hydrophobic effectis re-
presented by green lines. Interactions involving π
electrons are represented by dotted green lines.
Predicted binding modes for NET ligands.
identified inhibitors in NET stable transfected HEK (HEK-NET) cells. HEK-NET
cells were incubated with 200 nM unlabeled NE (50 nM radiolabeled [3H]
norepinephrine) and either 10 μM (blue) or 100 μM (red) of inhibitor for
3 min. Drugs that are chemically dissimilar to known NET ligands (Table S2)
are annotated in bold font.
Uptake experiments. Inhibition of [3H]norepinephrine uptake by the
Table 1. Uptake experiments for predicted ligands of NET
Name*Substrate uptake†, %IC50
58 ± 1 (20 ± 3)
68 ± 2 (10 ± 7)
70 ± 2 (28 ± 9)
25 ± 1 (14 ± 3)
99 ± 6 (27 ± 16)
102 ± 5 (97 ± 10)
27 ± 7 (22 ± 3)
109 ± 9 (111 ± 6)
97 ± 12 (102 ± 10)
144 ± 46 (112 ± 20)
102 ± 8 (106 ± 5)
103 ± 5 (30 ± 3)
90 ± 9 (90 ± 4)
87 ± 5 (79 ± 1)
52 ± 7 (62 ± 13)
87 ± 8 (44 ± 11)
115 ± 14 (111 ± 29)
111 ± 10 (104 ± 11)
*“Name” is the generic name of the molecule.
†“Substrate uptake” gives the percent of [3H]norepinephrine uptake in the
presence of 10 μM (100 μM) drug concentrations relative to that uptake in
the absence of drugs (Materials and Methods and SI Text).
‡“IC50” is the estimated half maximal inhibitory concentration value based on
measurement of inhibition of substrate uptake at different concentrations
(Materials and Methods and SI Text)
§N/A, not applicable.
Schlessinger et al. PNAS Early Edition
3 of 6
of guanabenz and tolazolin, respectively (Fig. 3 and Table 1). In
addition, two of the medium-confidence hits were shown to be
potent inhibitors of NET: The uptake of substrate by NET for
the antidiabetic drug phenformin and the nasal decongestant
tuaminoheptane at concentrations of 10 μM were, respectively,
52 and 27% relative to that of uninhibited NET (Fig. 3 and
Table 1). Tuaminoheptane inhibition of NET is further supported
by its inhibition of norepinephrine transport into bovine chro-
maffin cells (28); Furthermore, substrate uptake by NET for the
anti-Alzheimer’s disease drug talsaclidine at concentration of
100 μM was reduced to 30% relative to that of uninhibited
NET. Although the inhibition by talsaclidine was substantial only
at relatively high concentrations, the relationship between the
pharmacodynamic and pharmacokinetic properties of this drug
suggest that this inhibition might also occur in vivo at clinical
concentrations (Ifu;max∕IC50≈ 0.1; SI Text).
Polypharmacology is a phenomenon in which a drug binds multi-
ple rather than a single target with significant affinity (29, 30).
The effect of polypharmacology on therapy can be positive
(effective therapy) and/or negative (side effects). Positive and/or
negative effects can be caused by binding to the same or different
subsets of targets; binding to some targets may have no effect.
Polypharmacology is observed in the treatment of various disor-
ders in the nervous system, often causing severe side effects. For
example, the antischizophrenia drug clozapine binds a variety of
serotonergic, dopaminergic, muscarinic, adrenergic, and other
receptors that are likely to contribute to its high efficacy; simul-
taneously, at least some of these interactions also cause serious
side effects, including a higher risk of agranulocytosis, seizure,
weight gain, and diabetes (30, 31).
Three key findings emerge from our study. First, several sympa-
thetic and antidepressant drugs that target a variety of receptors
(e.g., α1-adrenergic receptor) and enzymes [e.g., monoamine oxi-
dase (MAO)] (Table S3), are also inhibitors of NET (Figs. 3 and
4; Table 1 and Table S2), thus likely contributing to the efficacy
of these drugs and further suggesting the substantial level of poly-
pharmacology of neuroactive drugs. Second, drugs used for the
treatment of diseases that are seeminglyunrelated to NET function
(e.g., diabetes) also inhibit NET activity (Fig. 3; Table 1 and
Table S2). This result may explain some side effects of these drugs
(e.g., tachycardia) and suggests that side effects of chemically re-
lated drugs, used to treat other diseases (e.g., proguanil for malar-
ia), may also be mediated via NET activity. Third, our virtual
screening against a comparative model of NET, a membrane pro-
tein sharing only 27% sequence identity with its template structure
discriminated known ligands from nonligands (Figs. 2 and 3;
Table S1). This suggests that the modeling and docking approach
mightbe useful foridentifying unknown interactions between drugs
and other membrane transporters and for designing new reagents
and leads. We take each of the three key findings in turn.
Explaining Efficacy of Drugs. Our uptake experiments show that
several drugs inhibit substrate transport by NET (Fig. 3 and
Table 1), suggesting that they might achieve their efficacy in part
because of this “off-target” activity. These drugs can be classified
into three groups as follows.
The firstgroup includes sympatheticdrugs thattypicallyactivate
the sympathetic nervous system via several mechanisms. They
can directly activate postsynaptic adrenergic receptors, inhibit the
breakdown of neurotransmitters by enzymes, block the reuptake
into the presynaptic cells by transporters, and/or stimulate produc-
tionand release ofcatecholamines (32). Virtual screening revealed
that various activators (i.e., guanabenz, tolazolin, octopamine,
levonordefrin, and norfenefrine) of the sympathetic system, which
have well-characterized targets (Table S3), also inhibit substrate
uptake by NET (Table 1). Inhibition of norepinephrine reuptake
by NET may be a therapeutically enhancing effect of these drugs.
The second group includes sympathetic drugs with unknown
mechanism, such as adrenalone. Previous studies showed that
adrenalone inhibits the activity of dopamine β-oxidase (33). Be-
cause this enzyme catalyzes the conversion of dopamine to nor-
epinephrine, adrenalone is expected to attenuate the adrenergic
signal. Here, we show that adrenalone inhibits substrate transport
by NET (Fig. 3), thereby amplifying the adrenergic signal. Be-
cause the pharmacological effects of adrenalone are related to
effects that are associated with stronger adrenergic signal (i.e.,
nasal decongestion and vasoconstriction), it is conceivable that
the increase in norepinephrine concentration via NET inhibition
is the major molecular mechanism of adrenalone efficacy.
The third group includes antidepressants with known targets.
MAO inhibitors (MAOIs) prevent the breakdown of monoamine
neurotransmitters, such as norepinephrine and dopamine, by
inhibiting MAO, an enzyme that is responsible for their degrada-
tion. Therefore, the availability of these neurotransmitters for
activating transmission of signals controlling the mood and beha-
vior is increased. Our finding that the MAOI tranylcypromine is
an inhibitor of NET (Figs. 3 and 4; Table 1) has the following three
clinical implications: (i) polypharmacology may contribute to the
efficacy of tranylcypromine; (ii) tranylcypromine scaffold might be
used for indications that are typically treated by NET reuptake
inhibitors and not by MAOIs, such as methylphenidate (Ritalin)
for ADHD; (iii) orthostatic hypotension, a condition that is linked
to malfunctions in NET (34) and is also a side effect of MAOIs
(35), might be explained by NET binding to tranylcypromine.
Explaining Side Effects. As illustrated by tranylcypromine, side ef-
fects are more likely to occur when a drug displays polypharma-
cology. Our experiments show that two additional drugs (i.e.,
phenformin and talsaclidine) inhibit substrate transport by NET
(Fig. 3 and Table 1). These drugs cause severe side effects that
are similar to those triggered by NET inhibitors, suggesting that
NET inhibition by these compounds also occurs in vivo.
Talsaclidine is a muscarinic M1 receptor agonist that was under
development for the treatment of Alzheimer’s disease. Talsacli-
dine failed in clinical trials because it triggered dose-limiting side
effects (36). For example, in some patients talsaclidine caused
symptoms including tachycardia, high blood pressure, nausea,
diarrhea, excessive sweating, and palpitation. Although some
of thesesymptoms are likely tobe triggered by binding toa variety
of targets, including the M2 and M3 receptors, the pharmaco-
dynamic properties of talsaclidine indicate that its inhibition
of NETuptake is clinically relevant and might also occur in vivo
(Table 1 and SI Text) (37). These results further support the
hypothesis that talsaclidine effects can be explained by stimula-
tion of both the adrenergic and cholinergic systems (36).
Phenformin is a biguanide drug that is typically used for the
treatment of type II diabetes. It lowers the blood glucose concen-
tration by a variety of mechanisms, such as decreasing the absorp-
tion of glucose in the intestines (38). The drug is still in use
in many countries, including Brazil, China, Greece, Poland, Por-
tugal, and Uruguay. However, phenformin was withdrawn from
clinical use in United States in 1978 because it was found to be
associated with fatal side effects (39, 40). For some patients,
phenformin triggers anorexia and high blood pressure by an
unknown mechanism (39). Therefore, phenformin inhibition of
substrate uptake by NETcan be one cause for these side effects.
Furthermore, other drugs that are chemically similar to phenfor-
min, containing biguanide and aromatic ring groups, could trigger
related side effects via NET inhibition. For example, the prophy-
lactic antimalarial drug chloroguanide (proguanil), an inhibitor of
dihydrofolate reductase that is highly similar to phenformin,
causes anorexia in 5% of the patients (http://www.drugs.com/sfx/
atovaquone-proguanil-side-effects.html). Conversely, metformin,
4 of 6
www.pnas.org/cgi/doi/10.1073/pnas.1106030108Schlessinger et al.
another antidiabetic biguanide drug that can trigger lactic acidosis
(27), does not cause high blood pressure and anorexia, which is
consistent with our observation that it does not affect norepi-
nephrine uptake by NET (Fig. 3 and Table 1). Finally, four other
phenformin-like drugs (e.g., the antihypertensive drug moroxy-
dine) (Table S4), whose pharmacological effect (i.e., side effects,
efficacy,orboth)might beexplainedby NET binding,were ranked
highly by our virtual screening (Dataset S1 and Table S3). One of
these predicted drugs (i.e., guanabenz) was experimentally vali-
dated and was shown to inhibit substrate uptake by NET (Fig. 3).
Virtual Screening Against NET and Other Transporters. NET ligands
can inhibit transport by NET via (i) competitive binding resulting
in either competitive substrates or competitive inhibitors (e.g.,
ephedrine) or (ii) noncompetitive binding to a NETallosteric site
(e.g., venlafaxine) that reduces NET’s affinity for the natural sub-
model is whether the model approximates an active or inhibited
conformation of NET. As a modeling template, we used the struc-
ture of LeuT in complex with its natural substrate leucine, which
was proposed to represent an active, outward-occluded conforma-
tion of LeuT (14). Furthermore, recent studies revealed that differ-
ent inhibitors stabilize unique LeuTconformations (41–43). Taken
together, these studies suggest that virtual screening against a single
rigid model of NET may not be able to identify all NET ligands.
Nevertheless, the docking screen was able to distinguish between
some ligands and nonligands. Five of the 10 experimentally verified
NET ligands (guanabenz, tolazolin, talsaclidine, phenformin, and
tuaminoheptane) are chemically novel (Table S2). The other five
identified ligands are chemically similar to known substrates and
bind NET in the micromolar range, as expected for substrates.
In contrast, large NET inhibitors, such as some SNRIs and tricyclic
antidepressants, do not fit into the binding site of our model and
thus were not identified by virtual screening. Therefore, to identify
NET interactions with large inhibitors, future studies should screen
against NET models that are based on the outward-open confor-
mation structure of LeuT (14). Furthermore, to identify new classes
of NET ligands, including both substrates and inhibitors, virtual
screening against large sets of lead-like molecules (18) should be
combined with the chemical similarity ensemble approach (SEA)
(31); SEA can relate proteins based on the chemical similarity
among their ligands, thus building cross-target similarity maps that
are useful in identifying protein–drug interactions (29).
Membrane transporters are key proteins that control the up-
take and efflux of various solutes such as amino acids, sugars,
and drugs. They can be drug targets themselves or they can reg-
ulate the absorption, distribution, metabolism, and elimination
of drugs (37). For example, the organic cation transporter 1
(SLC22A1 or OCT1) regulates the cellular concentration of
the antidiabetic drug metformin (44, 45). Because of the difficulty
of determining atomic structures of membrane proteins, struc-
tures of transporters, particularly from human, are not well
represented in the Protein Data Bank (PDB) (46). However,
in the past few years, structures of several prokaryotic and eukar-
yotic membrane transporters have been determined at atomic
resolution (47, 48). Despite relatively low sequence similarity
between these structures and their human homologs with un-
known structures, some of these structures can serve as templates
for constructing useful models (47).
If the template structures have similar ligand-binding profiles
to those of their homologs with unknown structures, they can be
used for virtual screening to identify ligands for these homologs
(17). For example, the recent crystal structure of the sodium
galactose transporter from Vibrio parahaemolyticus (vSGLT) (49)
can serve to identify ligands of the human glucose transporters of
the SLC5 family, which are emerging antidiabetic targets (50).
Conversely, if a template has a dissimilar ligand-binding profile
from that of its homolog, virtual screening against the template
clearly cannot be used to identify homolog’s ligands (17), as is
exemplified by the LeuT template and its NET homolog (Fig. 2B
and Table S1), binding leucine and monoamines, respectively.
Nontrivially, however, virtual screening against a refined NET
model based on the LeuT template did identify some known
and unknown ligands (Figs. 2–4).
The number of membrane protein structures is expected to
grow substantially (51, 52). The combined computational and
experimental approach presented in this study is generally applic-
able to the characterization of transporter structures and their
interactions with ligands, including drugs. Thus, it is useful for
identifying unknown drug-transporter interactions as well as
designing reagents with optimized selectivities.
Materials and Methods
Comparative Model Construction. NET was modeled using MODELLER-10v8
based on the X-ray structure of LeuT from Aquifex aeolicus (PDB ID code
2A65) (53) (SI Text). The alignment between the NET and LeuT sequences was
based on a comprehensive comparison of the SLC6 family members (54) (SI
Text). The sequence identity between the modeled fraction of NET and LeuT
is 27%, covering 77% of NET (Table S1 and Fig. S5. The primary binding site
of the initial NET model (“initial model”) was refined by repacking the side
chains on a fixed backbone using Scwrl4 (55) (“refined model,” available in
the “Norepinephrine transporter (NET, SLC6A2)” model dataset in ModBase
or upon request) (SI Text). Individual side chains (i.e., Asp75, Phe72, Tyr152,
Phe317, and Ser419) were refined consecutively by Scwrl4, with or without
ligand constraints (SI Text); Ser419 conformation was refined manually using
PyMOL (56). We also tested a NET model (“ModBase”) that was based on an
automatically computed NET–LeuTalignment downloaded from the ModBase
database of annotated comparative protein structure models (Table S1) (57).
Virtual Screening and Ligand Docking. Virtual screening against the NET model
was performed using a semiautomatic docking procedure (16, 17, 24, 58),
relying on DOCK 3.5.54 (59, 60) (SI Text).
Binding Site Assessment. Theaccuracyofvirtualscreeningwas estimatedbythe
enrichment for the known ligands among the top-scoring decoy compounds
(61, 62), generated by the Directory of Useful Decoys protocol (16, 18) (SI Text).
Cell Lines. Stable transfected HEK293 cells were created by transfecting
pcDNA5/FRT (Invitrogen) vector containing the full-length human NET
cDNA (HEK-NET) and the empty vector (HEK-EV) using Lipofectamine 2000
(Invitrogen) per manufacturer’s instructions. Transfected HEK293 cells were
maintained in DMEM-H21 containing 10% FBS, 100 units∕mL penicillin,
100 μg∕mL streptomycin, and 200 μg∕mL hygromycin B at 37 °C in a humidi-
fied atmosphere containing 5% CO2.
Inhibition of Norepinephrine Uptake. Uptake studies were performed as de-
scribed previously (63). Briefly, cells were seeded at a density of 2 × 105cells
per well in poly-D-lysine-coated 24-well (BD Falcon) plates and were grown
to 80–90% confluence. Stable transfected HEK293 cells were rinsed with pre-
warmed Hanks’ balanced salt solution (HBSS) and then incubated in 0.3 mL of
prewarmed buffer containing 200 nM unlabeled norepinephrine and 50 nM
tranylcypromine on [3H]norepinephrine uptake under the same conditions
as in Fig. 3. Data are expressed as the percent [3H]norepinephrine uptake
in the absence of inhibitor, and are the mean ? standard error of the mean
of three independent experiments.
Uptake kinetic experiment. Concentration-dependent effect of
Schlessinger et al.PNAS Early Edition
5 of 6
radiolabeled [3H]norepinephrine (Perkin Elmer) in the presence and absence
of 10 and 100 μM test compound at 37°C for 3 min. The reaction was termi-
nated bywashing cells twicewith1.0mLofice-coldHBSS,followed byaddition
of700mLlysisbuffer (0.1%SDSwt∕vol,0.1 NNaOH). Intracellular radioactivity
was determined by scintillation counting and normalized to per well protein
content as measured by using the BCA protein assay (Pierce). Concen-
tration-dependent inhibition for tranylcypromine (Fig. 4) was measured using
the same conditions as for the single-point measurements. Cells were incu-
bated with 0.025, 0.1, 0.4, 1.6, 6.25, 25.0, and 100 μM drug concentrations.
ACKNOWLEDGMENTS. We thank Sook Wah Yee, Amber Dahlin, and Keren
Lasker for helpful discussions, as well as Ursula Pieper, Ben Webb, and Elina
Tjioe for technical assistance and maintenance of the computational
resources required for this study. The project was supported by grants from
National Institutes of Health (R01 GM54762 to A. Sali, U54 GM094625 and
U01 GM61390 to A. Sali and K.M.G., U54 GM093342 and P01 GM71790 to
A. Sali and B.K.S., and F32 GM088991 to A. Schlessinger). We also acknowl-
edge funding for computing hardware from Hewlett Packard, IBM, NetApps,
Intel, Ron Conway, and Mike Homer.
1. Chen NH, Reith ME, Quick MW (2004) Synaptic uptake and beyond: The sodium-
and chloride-dependent neurotransmitter transporter family SLC6. Pflugers Arch
2. Pacholczyk T, Blakely RD, Amara SG (1991) Expression cloning of a cocaine- and anti-
depressant-sensitive human noradrenaline transporter. Nature 350:350–354.
3. Rudnick G, Clark J (1993) From synapse to vesicle: The reuptake and storage of
biogenic amine neurotransmitters. Biochim Biophys Acta 1144:249–263.
4. Hahn MK, Blakely RD (2007) The functional impact of SLC6 transporter genetic
variation. Annu Rev Pharmacol Toxicol 47:401–441.
5. Andersen J, Kristensen AS, Bang-Andersen B, Stromgaard K (2009) Recent advances
in the understanding of the interaction of antidepressant drugs with serotonin and
norepinephrine transporters. Chem Commun (Camb) 3677–3692.
6. Nyola A, et al. (2010) Substrate and drug binding sites in LeuT. Curr Opin Struct Biol
7. Kanner BI, Zomot E (2008) Sodium-coupled neurotransmitter transporters. Chem Rev
8. Krishnamurthy H, Piscitelli CL, Gouaux E (2009) Unlocking the molecular secrets of
sodium-coupled transporters. Nature 459:347–355.
9. Abramson J, Wright EM (2009) Structure and function of Na(+)-symporters with
inverted repeats. Curr Opin Struct Biol 19:425–432.
10. Forrest LR, et al. (2008) Mechanism for alternating access in neurotransmitter trans-
porters. Proc Natl Acad Sci USA 105:10338–10343.
11. Shi L, Quick M, Zhao Y, Weinstein H, Javitch JA (2008) The mechanism of a neuro-
transmitter:sodium symporter—Inward release of Na+ and substrate is triggered by
substrate in a second binding site. Mol Cell 30:667–677.
12. JardetzkyO (1966)Simple allosteric model for membranepumps.Nature 211:969–970.
13. Guan L, Kaback HR (2006) Lessons from lactose permease. Annu Rev Biophys Biomol
14. Singh SK, Piscitelli CL, Yamashita A, Gouaux E (2008) A competitive inhibitor traps
LeuT in an open-to-out conformation. Science 322:1655–1661.
15. Piscitelli CL, Krishnamurthy H, Gouaux E (2010) Neurotransmitter/sodium symporter
orthologue LeuT has a single high-affinity substrate site. Nature 468:1129–1132.
16. Huang N, Shoichet BK, Irwin JJ (2006) Benchmarking sets for molecular docking.
J Med Chem 49:6789–6801.
17. Fan H, et al. (2009) Molecular docking screens using comparative models of proteins.
J Chem Inf Model 49:2512–2527.
18. Irwin JJ, Shoichet BK (2005) ZINC—A free database of commercially available com-
pounds for virtual screening. J Chem Inf Model 45:177–182.
19. Celik L, et al. (2008) Binding of serotonin to the human serotonin transporter.
Molecular modeling and experimental validation. J Am Chem Soc 130:3853–3865.
20. Beuming T, et al. (2008) The binding sites for cocaine and dopamine in the dopamine
transporter overlap. Nat Neurosci 11:780–789.
21. OkudaS, et al. (2008) KEGG Atlas mapping for global analysisof metabolic pathways.
Nucleic Acids Res 36:W423–W426.
22. Shore PA, Busfield D, Alpers HS (1964) Binding and release of metaraminol: Mechan-
ism of norepinephrine depletion by alpha-methyl-m-tyrosine and related agents.
J Pharmacol Exp Ther 146:194–199.
23. Rothman RB, et al. (2003) In vitro characterization of ephedrine-related stereoi-
somers at biogenic amine transporters and the receptorome reveals selective actions
as norepinephrine transporter substrates. J Pharmacol Exp Ther 307:138–145.
24. Shoichet BK (2004) Virtual screening of chemical libraries. Nature 432:862–865.
25. Kuntz ID (1992) Structure-based strategies for drug design and discovery. Science
26. Nikolaev VO, Hoffmann C, Bunemann M, Lohse MJ, Vilardaga JP (2006) Molecular
basis of partial agonism at the neurotransmitter alpha2A-adrenergic receptor and
Gi-protein heterotrimer. J Biol Chem 281:24506–24511.
27. Wishart DS, et al. (2006) DrugBank: A comprehensive resource for in silico drug
discovery and exploration. Nucleic Acids Res 34(Database issue):D668–D672.
28. Delicado EG, Fideu MD, Miras-Portugal MT, Pourrias B, Aunis D (1990) Effect of
tuamine, heptaminol and two analogues on uptake and release of catecholamines
in cultured chromaffin cells. Biochem Pharmacol 40:821–825.
29. Keiser MJ, et al. (2009) Predicting new molecular targets for known drugs. Nature
30. Roth BL, Sheffler DJ, Kroeze WK (2004) Magic shotguns versus magic bullets:
Selectively non-selective drugs for mood disorders and schizophrenia. Nat Rev Drug
31. Keiser MJ, et al. (2007) Relating protein pharmacology by ligand chemistry. Nat Bio-
32. Leonard BE (2003) Fundamentals of Psychopharmacology (John Wiley & Sons, Chi-
chester, UK), 3rd Ed, p 536.
33. Goldstein M, Contrera JF (1961) Inhibition of dopamine beta oxidase by adrenalone.
34. Hahn MK, Robertson D, Blakely RD (2003) A mutation in the human norepinephrine
transporter gene (SLC6A2) associated with orthostatic intolerance disrupts surface
expression of mutant and wild-type transporters. J Neurosci 23:4470–4478.
35. Kaludercic N, Carpi A, Menabo R, Di Lisa F, Paolocci N (2010) Monoamine oxidases
(MAO) in the pathogenesis of heart failure and ischemia/reperfusion injury. Biochim
Biophys Acta 1813:1323–1332.
36. Adamus WS, Leonard JP, Troger W (1995) Phase I clinical trials with WAL 2014, a new
muscarinic agonist for the treatment of Alzheimer’s disease. Life Sci 56:883–890.
37. Giacomini KM, et al. (2010) Membrane transporters in drug development. Nat Rev
Drug Discov 9:215–236.
38. Grodsky GM, Karam JH, Pavlatos FC, Forsham PH (1963) Reduction by phenformin of
excessive insulin levels after glucose loading in obese and diabetic subjects. Metabo-
39. Mc KJ, Kuwayti K, Rado PP (1959) Clinical experience with DBI (phenformin) in the
management of diabetes. Can Med Assoc J 80:773–778.
40. Wise PH, et al. (1976) Phenformin and lactic acidosis. Br Med J 1:70–72.
41. Claxton DP, et al. (2010) Ion/substrate-dependent conformational dynamics of a
bacterial homolog of neurotransmitter:sodium symporters. Nat Struct Mol Biol
42. Zhao Y, et al. (2010) Single-molecule dynamics of gating in a neurotransmitter trans-
porter homologue. Nature 465:188–193.
43. Nyola A, et al. (2010) Substrate and drug binding sites in LeuT. Curr Opin Struct Biol
44. Shu Y, et al. (2007) Effect of genetic variation in the organic cation transporter 1
(OCT1) on metformin action. J Clin Invest 117:1422–1431.
45. Shu Y, et al. (2003) Evolutionary conservation predicts function of variants of the
human organic cation transporter, OCT1. Proc Natl Acad Sci USA 100:5902–5907.
46. Kelly L, et al. (2009) A survey of integral alpha-helical membrane proteins. J Struct
Funct Genomics 10:269–280.
47. Schlessinger A, et al. (2010) Comparison of human solute carriers. Protein Sci
48. He X, et al. (2010) Structure of a cation-bound multidrug and toxic compound
extrusion transporter. Nature 467:991–994.
49. Faham S, et al. (2008) The crystal structure of a sodium galactose transporter reveals
mechanistic insights into Na+/sugar symport. Science 321:810–814.
50. Wright EM, Hirayama BA, Loo DF (2007) Active sugar transport in health and disease.
J Intern Med 261:32–43.
51. Stroud RM, et al. (2009) 2007 Annual progress report synopsis of the Center for
Structures of Membrane Proteins. J Struct Funct Genomics 10:193–208.
52. Love J, et al. (2010) The New York Consortium on Membrane Protein Structure
(NYCOMPS): A high-throughput platform for structural genomics of integral mem-
brane proteins. J Struct Funct Genomics 11:191–199.
53. Yamashita A, Singh SK, Kawate T, Jin Y, Gouaux E (2005) Crystal structure of a
bacterial homologue of Na+/Cl−-dependent neurotransmitter transporters. Nature
54. Beuming T, Shi L, Javitch JA, Weinstein H (2006) A comprehensive structure-based
alignmentof prokaryoticand eukaryotic neurotransmitter/Na+ symporters(NSS) aids
in the use of the LeuTstructure to probe NSS structure and function. Mol Pharmacol
55. Krivov GG, Shapovalov MV, Dunbrack RL, Jr (2009) Improved prediction of protein
side-chain conformations with SCWRL4. Proteins 77:778–795.
56. DeLano WL (2002) The PyMOL Molecular Graphics System (DeLano Scientific, San
57. Pieper U, et al. (2011) ModBase, a database of annotated comparative protein
structure models, and associated resources. Nucleic Acids Res 39:D465–D474.
58. Irwin JJ, et al. (2009) Automated docking screens: A feasibility study. J Med Chem
59. Lorber DM, Shoichet BK (2005) Hierarchical docking of databases of multiple ligand
conformations. Curr Top Med Chem 5:739–749.
60. Mysinger MM, Shoichet BK (2010) Rapid context-dependent ligand desolvation in
molecular docking. J Chem Inf Model 50:1561–1573.
61. Cavasotto CN, et al. (2008) Discovery of novel chemotypes to a G-protein-coupled
receptor through ligand-steered homology modeling and structure-based virtual
screening. J Med Chem 51:581–588.
62. Evers A, Gohlke H, Klebe G (2003) Ligand-supported homology modelling of protein
binding-sites using knowledge-based potentials. J Mol Biol 334:327–345.
63. Chen Y, Zhang S, Sorani M, Giacomini KM (2007) Transport of paraquat by human
organic cation transporters and multidrug and toxic compound extrusion family.
J Pharmacol Exp Ther 322:695–700.
6 of 6
www.pnas.org/cgi/doi/10.1073/pnas.1106030108Schlessinger et al.