Crystallographic study of FABP5 as an intracellular endocannabinoid transporter

Abstract

In addition to binding intracellular fatty acids, fatty-acid-binding proteins (FABPs) have recently been reported to also transport the endocannabinoids anandamide (AEA) and 2-arachidonoylglycerol (2-AG), arachidonic acid derivatives that function as neurotransmitters and mediate a diverse set of physiological and psychological processes. To understand how the endocannabinoids bind to FABPs, the crystal structures of FABP5 in complex with AEA, 2-AG and the inhibitor BMS-309403 were determined. These ligands are shown to interact primarily with the substrate-binding pocket
via
hydrophobic interactions as well as a common hydrogen bond to the Tyr131 residue. This work advances our understanding of FABP5–endocannabinoid interactions and may be useful for future efforts in the development of small-molecule inhibitors to raise endocannabinoid levels.

Full text

electronic reprint
Acta Crystallographica Section D
Biological
Crystallography
ISSN 1399-0047
Crystallographic study of FABP5 as an intracellular
endocannabinoid transporter
Benoˆıt Sanson, Tao Wang, Jing Sun, Liqun Wang, Martin Kaczocha, Iwao
Ojima, Dale Deutsch and Huilin Li
Acta Cryst.
(2014). D70, 290–298
Copyright c
International Union of Crystallography
Author(s) of this paper may load this reprint on their own web site or institutional repository provided that
this cover page is retained. Republication of this article or its storage in electronic databases other than as
specified above is not permitted without prior permission in writing from the IUCr.
For further information see http://journals.iucr.org/services/authorrights.html
Acta Crystallographica Section D: Biological Crystallography
welcomes the submission of
papers covering any aspect of structural biology, with a particular emphasis on the struc-
tures of biological macromolecules and the methods used to determine them. Reports
on new protein structures are particularly encouraged, as are structure–function papers
that could include crystallographic binding studies, or structural analysis of mutants or
other modified forms of a known protein structure. The key criterion is that such papers
should present new insights into biology, chemistry or structure. Papers on crystallo-
graphic methods should be oriented towards biological crystallography, and may include
new approaches to any aspect of structure determination or analysis. Papers on the crys-
tallization of biological molecules will be accepted providing that these focus on new
methods or other features that are of general importance or applicability.
Crystallography Journals Online is available from journals.iucr.org
Acta Cryst.
(2014). D70, 290–298 Sanson
et al.
·FAB P5

research papers
290 doi:10.1107/S1399004713026795 Acta Cryst. (2014). D70, 290–298
Acta Crystallographica Section D
Biological
Crystallography
ISSN 1399-0047
Crystallographic study of FABP5 as an intracellular
endocannabinoid transporter
Benoı
ˆt Sanson,
a
Tao Wang,
a
‡
Jing Sun,
b
Liqun Wang,
b
Martin
Kaczocha,
b
Iwao Ojima,
c,d
Dale
Deutsch
b,d
* and Huilin Li
a,b,d
*
a
Biosciences Department, Brookhaven National
Laboratory, Upton, NY 11973-5000, USA,
b
Department of Biochemistry and Cell Biology,
Stony Brook University, Stony Brook,
NY 11794-5213, USA,
c
Department of
Chemistry, Stony Brook University, Stony Brook,
NY 1794-3400, USA, and
d
Institute of Chemical
Biology and Drug Discovery, Stony Brook
University, Stony Brook, NY 11794-3400, USA
‡ Current address: School of Chemical Biology
and Biotechnology, Shenzhen Graduate School,
Peking University, Shenzhen 518055, People’s
Republic of China.
Correspondence e-mail:
dale.deutsch@stonybrook.edu, hli@bnl.gov
#2014 International Union of Crystallography
In addition to binding intracellular fatty acids, fatty-acid-
binding proteins (FABPs) have recently been reported to
also transport the endocannabinoids anandamide (AEA) and
2-arachidonoylglycerol (2-AG), arachidonic acid derivatives
that function as neurotransmitters and mediate a diverse set of
physiological and psychological processes. To understand how
the endocannabinoids bind to FABPs, the crystal structures of
FABP5 in complex with AEA, 2-AG and the inhibitor BMS-
309403 were determined. These ligands are shown to interact
primarily with the substrate-binding pocket via hydrophobic
interactions as well as a common hydrogen bond to the Tyr131
residue. This work advances our understanding of FABP5–
endocannabinoid interactions and may be useful for future
efforts in the development of small-molecule inhibitors to
raise endocannabinoid levels.
Received 10 May 2013
Accepted 30 September 2013
PDB references: mouse
FABP5, 4azn; 4azo; complex
with AEA, 4azp; complex
with 2-AG, 4azq; human
FABP5, complex with
BMS-309403, 4azm;
complex with AEA, 4azr
1. Introduction
Endocannabinoids are signaling lipids that activate cannabi-
noid receptors in the central nervous system and peripheral
tissues (Howlett et al., 2011). The best characterized endo-
cannabinoids, anandamide (AEA) and 2-arachidonoylglycerol
(2-AG), are ethanolamine and glycerol derivatives of arachi-
donic acid, respectively. In contrast to hydrophilic neuro-
transmitters, endocannabinoids are not stored in vesicles.
Instead, the magnitude and duration of endocannabinoid
signaling is regulated through ‘on demand’ biosynthesis and
prompt catabolism. AEA is principally hydrolyzed by fatty-
acid amide hydrolase (FAAH), while 2-AG is inactivated
by monoacylglycerol lipase (MAGL), ABHD6 and ABHD12
(Cravatt et al., 2001; Blankman et al., 2007; Deutsch & Chin,
1993).
Owing to their limited solubility, endocannabinoids require
carrier-assisted transport through the cellular cytoplasm.
Recently, we identified fatty-acid-binding proteins (FABPs)
as intracellular carriers that transport AEA from the plasma
membrane to intracellular FAAH for hydrolysis (Kaczocha
et al., 2009). FABPs are small (15 kDa) widely expressed
intracellular lipid-binding proteins (Furuhashi & Hotamisligil,
2008) and bind a variety of lipophilic ligands including fatty
acids, fatty-acid amides and xenobiotics (Velkov et al., 2005;
Chuang et al., 2008; Kaczocha et al., 2012).
It is well established that inhibition of FAAH or MAGL
potentiates endocannabinoid-mediated antinociceptive and
anti-inflammatory effects (Ahn et al., 2009; Long et al., 2009).
Inhibition of intracellular endocannabinoid carriers such as
FABPs may provide an alternative strategy to modulate
endocannabinoid inactivation. We have previously shown that
FABP knockdown or inhibition with the selective inhibitor
electronic reprint

BMS-309403 (Sulsky et al., 2007) reduces AEA inactivation in
cells (Kaczocha et al., 2009). FABP inhibitors augment endo-
cannabinoid levels and produce beneficial anti-inflammatory
and antinociceptive effects (Berger et al., 2012). As such, the
design of novel selective FABP inhibitors hinges upon
understanding the bonding interactions between current-
generation inhibitors and the FABP-binding pocket.
Ten isoforms of FABP have been identified in various
tissues, with FABP4 mainly present in adipocytes and FABP5
in epidermis (Furuhashi & Hotamisligil, 2008). At the
primary-sequence level, the conservation of FABP isoforms
varies from low (15%) to very high (70%). The FABPs are
conserved in three-dimensional structure: they form a ten-
stranded -barrel (Furuhashi & Hotamisligil, 2008; Hamilton,
2004). The -barrel is comprised of two orthogonal five-
stranded -sheets: -sheet 1 and -sheet 2. One side of the
-barrel is capped by a helix–loop–helix motif and the other
side by the amino-terminal peptide. Therefore, the structure
is also referred to as a -clamshell, with the two -sheets as
a pair of valves (Sacchettini et al., 1989; Hodsdon & Cistola,
1997; Jenkins et al., 2002; Richieri et al., 1999). Endogenous
fatty acids such as palmitic acid and oleic acid generally bind
to FABPs in a similar manner, with their carboxylates binding
to one or two conserved basic residues and their hydrocarbon
chains nesting in the largely hydrophobic chambers (Furu-
hashi & Hotamisligil, 2008; Hohoff et al., 1999). In this study,
we determined the crystal structures of mouse and human
FABP5 in complex with AEA or 2-AG and with the inhibitor
BMS-309403. These two proteins are highly conserved, with
80% (108/135 residues) sequence identity.
2. Experimental procedures
2.1. Cloning, expression and purification of human and
mouse FABP5
The FABP constructs were as described by Kaczocha et al.
(2012) and the purification of the FABPs was based on a
published procedure (Hohoff et al., 1999). Briefly, both human
and mouse FABP5 sequences were encoded into the pET-28a
vector (Novagen) fused with an N-terminal 6His tag. Proteins
were expressed in Escherichia coli BL21(DE3) cells using the
T7 expression system. Cells were grown in LB medium at 37C
with shaking at 250 rev min
1
.AtanOD
600
of 0.7, protein
expression was induced by adding IPTG to a concentration of
0.4 mM. After 20 h incubation at 20C, the cells were pelleted
by centrifugation at 5000gat 4C for 15 min and then resus-
pended in 30 ml ice-cold column buffer (250 mMNaCl,
20 mMTris pH 8.5). These cells were lysed by sonication on
ice, followed by 30 min centrifugation at 15 000gat 4C. The
supernatant was loaded onto an Ni–NTA column (Qiagen,
Valencia, California, USA). After mixing the supernatant with
Ni–NTA agarose at 4C for 10 min, the column was washed
with ten bed volumes of column buffer containing 20 mM
imidazole. The protein was then eluted with column buffer
containing 200 mMimidazole. The affinity-purified samples
were concentrated on a Spin-X UF cartridge (Corning,
England) and loaded onto a Sephacryl S-100 XK 16/70 column
equilibrated with 1PBS pH 8.5 using an A
¨KTAprime plus
system (GE Healthcare Life Sciences). The peak fractions
were collected and delipidated by incubation with Lipidex-
5000 (Sigma) at 37C for 1 h with occasional mixing.
For preparations where the 6His tag was removed, the
affinity-purified proteins were incubated with thrombin (GE
Healthcare Life Sciences) at 10 units per milligram of protein
at 4C overnight. The reaction was stopped by the addition
of phenylmethylsulfonyl fluoride to a final concentration of
1mM. The solution was loaded onto an Ni–NTA column and
the unbound fractions containing the 6His-tag-cleaved
FABP5 was collected and then concentrated. The concen-
trated sample was subjected to the same gel-filtration chro-
matography and delipidation procedure as described above.
All purified proteins were concentrated to approximately
10 mg ml
1
and then stored at 80C.
2.2. Crystallization and structural solution
The frozen purified mouse FABP5 was thawed on ice.
Crystals of the 6His-tagged mouse FABP5 grew in 20% PEG
3350, 100 mMbis-tris pH 5.5. Crystals of the tag-cleaved
mouse FABP5 grew in 50 mMsodium acetate pH 4.8, 5%
MPD, 25% PEG 3350, 200 mMNaCl. Some crystals were
soaked overnight in the mother liquor supplemented with
25% glycerol and saturated with AEA or 2-AG.
Frozen samples of the tag-free human FABP5 were thawed
on ice and incubated with saturating concentrations of AEA
or the inhibitor BMS-309403. The AEA–FABP5 complex was
co-crystallized in 25% PEG 3350, 0.1 MHEPES pH 7.5.
The inhibitor–FABP5 complex was co-crystallized in 1.32 M
sodium citrate, 0.1 MHEPES pH 7.0. The crystals were
cryoprotected in mother liquor containing 25 or 28% glycerol
in the case of the complexes with AEA and the inhibitor,
respectively.
X-ray diffraction data were collected on beamlines X25 and
X29 at the National Synchrotron Light Source and were
integrated and scaled using XDS (Kabsch, 2010). We set aside
5% of the reflections as the test set. The structures were solved
by molecular replacement using PDB entry 1b56 (Hohoff et
al., 1999) as a search model. Structure refinement was
performed using REFMAC (Murshudov et al., 2011). TLS
refinement was carried out as a final step, which generally
improved the R
work
and R
free
factors by 1% (Painter &
Merritt, 2006). We used two TLS groups per monomer in all
models. Simulated-annealing OMIT maps, composite OMIT
maps and F
o
F
o
difference maps were calculated using CNS
(Brunger, 2007). Radiation damage was assessed using
co-crystals of FABP5–BMS-309403 by calculating the F
o
F
o
difference maps between two successive data sets from the
same region of a single crystal. All ligand molecules were
built and their associated geometry files calculated using
PRODRG (Schu
¨ttelkopf & van Aalten, 2004). The models
were validated using MolProbity (Chen et al., 2010). Figures
were prepared using PyMOL (http://www.pymol.org). The
research papers
Acta Cryst. (2014). D70, 290–298 Sanson et al. FABP5 291
electronic reprint

atomic coordinates and structure factors have been deposited
in the PDB as entries 4azn, 4azo, 4azp, 4azq, 4azm and 4azr.
3. Results and discussion
3.1. Crystal structures of mouse FABP5 in complex with AEA
AEA is an arachidonic acid derivative containing four
double bonds, with a chemical composition of C
22
H
37
NO
2
and
a mass of 347 Da (Fig. 1). To determine by X-ray crystal-
lography how AEA binds to FABP5, we
overexpressed mouse FABP5 in E. coli
and purified the protein by Ni
2+
-affinity
column chromatography via an engi-
neered 6His tag at the amino-terminus
of the protein. We found that the His-
tag mediated crystal contact by inter-
action with -strand 4 of a neighboring
protein molecule, potentially interfering
with the FABP5 structure (Supplemen-
tary Fig. S1
1
; Table 1; PDB entry 4azn).
In subsequent work, we removed the
His tag by thrombin cleavage and then
delipidated the purified FABP by
treatment with Lipidex-5000 (see x2).
The resulting protein was concentrated
and stored at 80C.
The delipidated mouse FABP5 was
first crystallized in space group P6
4
22
and the apo FABP5 structure was
solved at 2.33 A
˚resolution (Table 1;
PDB entry 4azo). We then soaked the
crystals in saturated solutions of AEA.
The structure of the FABP5–AEA
complex was solved at 2.1 A
˚resolution
(Table 1; PDB entry 4azp; Fig. 2). The
protein structures were essentially the
same in the presence or absence of
AEA. The presence of AEA in the
substrate-soaked crystals was demon-
strated by the elongated electron
density inside the substrate-binding
pocket of the F
o
F
c
electron-density
map (Fig. 2b). The initial density for the
ligand was rather weak and contained
gaps. The AEA model was initially built
into the F
o
F
c
map. We also calculated
composite OMIT maps to confirm the
placement of AEA.
In the crystal structure, the hydroxyl
group of AEA forms a hydrogen bond
to Tyr131 and a second, water-mediated,
hydrogen bond to Arg129 (Fig. 2c). The
long lipophilic chain of AEA forms a
loop that nests in the largely hydro-
phobic substrate pocket, with the
nearest residues being Tyr22, Leu26,
Leu32, Ala36, Pro41, Val60, Ala78 and Val118, mostly from
the N-terminal region of the transporter. Interestingly,
a recent molecular-dynamics simulation of AEA complexed
with FABP7 predicted a similar mode of binding (Howlett et
al., 2011). AEA was shown to bind to FABP5 with an affinity
of 1.3 mM, approximately tenfold lower than its parent fatty
acid arachidonic acid (Kaczocha et al., 2012). Consistent with
research papers
292 Sanson et al. FABP5 Acta Cryst. (2014). D70, 290–298
Figure 1
The chemical structures of AEA, 2-AG and the FABP inhibitor BMS-309403.
Table 1
Statistics of data collection and structure refinement.
Values in parentheses are for the last resolution shell.
mFABP5
(His-tagged) mFABP5 mFABP5 mFABP5 hFABP5 hFABP5
Data collection
Ligand — — AEA 2-AG BMS AEA
Space group C2P6
4
22 P6
4
22 P6
4
22 P42
1
2C222
1
Unit-cell parameters
a(A
˚) 97.29 79.32 79.71 79.48 104.57 66.12
b(A
˚) 64.97 82.32 83.15 82.70 58.61 115.15
c(A
˚) 60.10 82.32 83.15 82.70 58.61 108.35
=() 90 9090909090
() 122.34 120 120 120 90 90
Oscillation () 1 1 1 111
Exposure (s) 1 2 1 1 1 3
Maximum resolution (A
˚) 2.50 2.33 2.10 2.00 2.75 2.95
Completeness (%) 93.7 (62.0) 99.2 (98.9) 99.4 (99.2) 99.6 (100) 99.9 (100) 99.9 (100)
R
merge
† (%) 5.2 (21.6) 4.0 (45.5) 6.8 (48.1) 8.1 (51.5) 6.2 (60.1) 11.2 (57.8)
Multiplicity 7.4 10.6 11.4 11.4 12.4 6.5
hI/(I)i28.4 (8.2) 33.1 (4.0) 22.0 (5.3) 18.5 (4.9) 35.6 (4.7) 16.1 (3.2)
Refinement
Resolution range (A
˚) 20–2.51 20–2.33 20–2.10 20–2.00 20–2.75 20–2.95
R
cryst
‡ (%) 21.44 22.44 19.95 21.63 20.50 21.91
R
free
‡ (%) 28.19 27.39 23.62 25.73 25.20 26.80
No. of protein atoms 2132 1060 1053 1079 2101 2108
No. of waters 82 27 84 68 44 32
No. of ligands 0 0 1 1 1 1
R.m.s.d., bond lengths (A
˚) 1.606 1.402 1.412 1.392 1.450 1.293
R.m.s.d., bond angles () 0.015 0.011 0.012 0.011 0.011 0.012
Average Bfactor (A
˚
2
) 51.6 56.4 40.2 44.1 55.2 38.5
Ramachandran quality
Favored (%) 95.9 99.2 99.2 98.5 97.4 95.9
Outliers (%) 0.0 0.0 0.0 0.0 0.4 0.0
TLS groups per chain 2 2 2 2 2 2
PDB code 4azn 4azo 4azp 4azq 4azm 4azr
†R
merge
=Phkl PijIiðhklÞhIðhklÞij=Phkl PiIiðhklÞ.‡R=Phkl jFobsjjFcalc j=Phkl jFobs jfor both R
cryst
and
R
free
.R
cryst
is calculated from the working data set, whereas R
free
is calculated from the test set (5% of the total
reflections).
1
Supporting information has been deposited in the IUCr electronic archive
(Reference: XB5074).
electronic reprint

this lower binding affinity, we found that the electron densities
at both ends of AEA are weak. We suggest that the binding at
the terminal regions of the AEA molecule in FABP5 may be
less specific and may have multiple conformations; therefore,
the model as built may represent an average of these
conformations.
Because AEA is longer than the most common endogenous
ligand, palmitic acid, we examined whether AEA binding
causes structural changes in FABP5 compared with that of
palmitic acid (Supplementary Fig. S2; PDB entries 4azp and
1b56; Hohoff et al., 1999). We found that AEA binds to
FABP5 in a manner similar to that of palmitic acid. However,
the extra length of the AEA hydrocarbon tail has pushed the
surrounding H1–H2 motif and the S3–S4 loop outwards by
1.5–2 A
˚. Since the H1–H2 cap and S3–S4 loops are generally
considered to line the substrate entrance, our observed
movement of these structural components by AEA is consis-
tent with such a substrate-entrance model.
3.2. Crystal structure of mouse FABP5 in complex with 2-AG
2-AG was the second endocannabinoid to be discovered
following AEA. 2-AG is a glycerol derivative of arachidonic
acid (Fig. 1). The structure of 2-AG complexed with mouse
FABP5 was solved at 2.0 A
˚resolution using essentially the
research papers
Acta Cryst. (2014). D70, 290–298 Sanson et al. FABP5 293
Figure 2
Interactions between AEA and mouse FABP5. (a) Cartoon view of the
FABP5–AEA complex crystal structure. The secondary-structural
elements, -helices H1 and H2 and -strands S1–S10, are labeled. The
red letters ‘N’ and ‘C’ denote the amino- and carboxyl-termini of the
protein, respectively. The bound AEA is shown as green sticks. (b)
Electron density of AEA in the binding pocket of FABP5. The simulated
OMIT map is contoured at the 2.5threshold and is shown as a blue
mesh. (c) Detailed interactions between AEA (green sticks) and FABP5
(contacting residues shown as yellow sticks).
Figure 3
Crystal structure of mouse FABP5 in complex with 2-AG. (a) Cartoon
view of the structure. The secondary-structural elements, -helices H1
and H2 and -strands S1–S10, are labeled. The bound 2-AG is shown as
green sticks. The red letters ‘N’ and ‘C’ denote the amino- and carboxyl-
termini of the protein, respectively. (b) Electron density of 2-AG in the
binding pocket of the FABP5–2-AG complex. The simulated OMIT map
is contoured at the 2.5threshold and is displayed as a blue mesh. (c)
Superposition of 2-AG with AEA in mFABP5. The two 2-AG hydroxyl
groups insert deeper into the substrate chamber. (d) Detailed interactions
between 2-AG and FABP5. Hydrogen bonds are shown as red dashes.
Note that the viewing direction of (d)is90rotated from that in (b)to
provide a different perspective. The FABP5 residues close to or in contact
with 2-AG are shown as yellow sticks.
electronic reprint

same procedure as described above (Fig. 3; Table 1; PDB entry
4azq). The electron density of 2-AG is stronger than that of
AEA except at the non-alkyl end of 2-AG, where it is slightly
weaker (Figs. 3band 2b). Compared with AEA, the hydro-
philic head group of 2-AG inserts deeper into the substrate
pocket (Fig. 3c). In the crystal structure, the lowered head
position of 2-AG enables it to form five hydrogen bonds to the
transporter (Fig. 3d), namely between the carbonyl O atom of
2-AG and Arg129, between the carbonyl O atom of 2-AG and
Tyr131, between Arg109 and one hydroxyl of 2-AG, which also
makes a hydrogen bond to Cys43, and between the second
hydroxyl group and Thr56. The looped lipophilic fragment of
2-AG is within 4 A
˚distance of FABP5 residues Phe19, Ala36,
Val60, Ala78, Ile107 and Val118.
The two endocannabinoids have similar chemical structures
(Fig. 1). Both are sparsely branched alkyl chains with nearly
identical lengths: the alkyl chain of AEA has 22 C atoms and
that of 2-AG has 23 C atoms. By comparing their binding
modes to FABP5, we note that Tyr131 and Arg129 of FABP5
make hydrogen bonds to the polar ends of both ligands and
that essentially the same residues of the transporter are
involved in hydrophobic contacts with the lipophilic arachi-
donyl chain (Figs. 2cand 3d). However, AEA forms fewer
hydrogen bonds than 2-AG, but AEA is in tighter hydro-
phobic contact with FABP5 than 2-AG. Their binding modes
are consistent with their similar binding affinity to FABP5,
which is 1.26 0.18 mMfor AEA and 1.45 0.21 mMfor
2-AG (Kaczocha et al., 2012). The experimental electron
densities for these endocannabinoids are weak at the ends and
are partially broken (Figs. 2band 3b), suggesting that both
AEA and 2-AG may adopt multiple conformations at their
respective termini in FABP5.
3.3. Crystal structure of human FABP5 in complex with
BMS-309403
BMS-309403 is a biphenyl azole that was initially developed
as a high-affinity inhibitor of the adipocyte FABP4 (2 nM;
Fig. 1) in an effort to develop anti-obesity drugs (Sulsky et al.,
2007). The crystal structure of the inhibitor in complex with
FABP4 has previously been reported (Sulsky et al., 2007).
We found that the compound also inhibits human FABP5,
although with a lower affinity of 890 nM(Kaczocha et al.,
2012). We have now determined the crystal structure of this
inhibitor in complex with human FABP5 (Table 1; PDB entry
4azm), with the expectation that the binding mode would yield
clues about how to improve the potency of the compound
towards FABP5. Interestingly, the FABP5–inhibitor complex
crystallized as a domain-swapped dimer (described further
below). An F
o
F
c
map calculated before ligand modeling
revealed a clear positive electron density in the substrate
pocket. Inspection of the F
o
F
c
map showed that the elec-
tron density for one monophenyl ring is weak, probably as a
result of rotational freedom (Fig. 4a). BMS-309403 contains
a terminal carboxylate that, like cysteine, is known to be
sensitive to radiation damage (Weik et al., 2000; Garman &
Weik, 2011). We collected two consecutive data sets from the
same region of a single crystal and calculated an F
o
F
c
difference map. Displayed at a 3level, the difference map
shows two negative electron-density peaks in the regions of
the inhibitor carboxylate and Cys120, respectively (Fig. 4b),
clearly indicating a loss of electrons owing to X-ray-induced
oxidation of these chemical groups. This experiment confirms
our modeling of the BMS-309403 inhibitor.
research papers
294 Sanson et al. FABP5 Acta Cryst. (2014). D70, 290–298
Figure 4
The binding mode of BMS-309403 in human FABP5. (a) The simulated-
annealing F
o
F
c
OMIT map calculated around the inhibitor is
contoured at the 3level and is displayed as a blue mesh. (b)AnF
o
F
c
difference density map calculated from two consecutive data sets
collected from the same region of a single crystal. The map is contoured
at the 3level and is shown as a red mesh. The loss of electrons at the
carboxylic acid moiety of BMS-309403 and, to a lesser extent, at the S
atom of Cys120 is caused by radiation damage. (c) The hydrogen-bonding
network that stabilizes the inhibitor in the FABP5 substrate pocket. The
hydrogen bonds are shown as red dashes. One water molecule is shown as
a red ball. (d) Superposition of BMS-309403 in complex with FABP5 (in
yellow) with the same inhibitor complexed to FABP4 (PDB entry 2nnq;
blue; Sulsky et al., 2007). Several residues surrounding the inhibitors are
shown as sticks. (e) Enlarged view of the BMS-309403 structure bound to
FABP5 (green) superimposed on the inhibitor bound to FABP4 (gray).
BMS-309403 forms hydrogen bonds to two conserved arginine residues in
both FABP4 (blue) and FABP5 (yellow). For clarity, hydrogen bonds are
not shown.
electronic reprint

research papers
Acta Cryst. (2014). D70, 290–298 Sanson et al. FABP5 295
Figure 5
Human FABP5 crystallized as a domain-swapped dimer. (a) Topology diagram of the crystallographic domain-swapped dimer. The diagram was
prepared in Pro-origami (Stivala et al., 2011). (b) Overview of the dimeric FABP5 crystal structure. The secondary structure is shown as a cartoon view.
The two FABP5 molecules are displayed in gold and cyan, respectively. (c) Close-up view of the induced -strands extending between the two monomers
in the domain-swapped FABP5 and the associated F
o
F
c
electron-density map. The map is contoured at the 3level and is shown as a green mesh. The
region shown corresponds to the region in the red box in (b).
The binding modes of the inhibitor in the two monomers of
the FABP5 dimer are very similar except that the inhibitor
electron density is stronger in one monomer than in the other.
This is reflected in their respective average Bfactors (55.9 and
70.2 A
˚
2
, respectively). In the crystal structure, one carboxyl O
atom of the inhibitor forms a hydrogen bond to Arg129 of the
transporter and another hydrogen bond to Tyr131 (Fig. 4c).
The other carboxyl O atom makes one direct and one water-
mediated hydrogen bond to Arg109. Several hydrophobic
contacts contribute to the binding; these include residues
Tyr22, Leu26, Ala39, Pro41, Ile54 and Ala78 that are within
4A
˚of the inhibitor. Phe19 appears to be particularly impor-
tant because its side chain is even closer at only 3.0 A
˚away
from the distal phenyl ring as well as from the ethyl moiety of
the inhibitor. The central pyrazole ring and the oxyacetate
group are the best defined moieties of the ligand. Indeed, the
oxyacetate group is involved in hydrogen bonding between
the inhibitor and the protein. The inhibitor forms a total of
four hydrogen bonds to Tyr131, Arg109 and Arg129.
The primary sequences of human FABP4 and FABP5 are
55% identical and 72% homologous and share essentially
the same structure. We superimposed the structures of the
respective proteins both complexed with the same inhibitor
BMS-309403 (Figs. 4dand 4e). We found that the four freely
rotatable benzene rings of the inhibitor are in different
conformations, apparently exploring the local landscapes of
the two proteins. In FABP4, the ethyl group of the inhibitor
is in van der Waals interaction with Ser53. This feature was
noted previously as a key factor in the high-affinity binding
(Sulsky et al., 2007). In the domain-swapped FABP5, the ethyl
group is next to but makes no direct interaction with Gly36.
We suggest that this binding difference may contribute to the
lowered affinity of the inhibitor for FABP5. Nevertheless, the
inhibitor shares binding features in both structures. Firstly, the
oxyacetate moiety of the inhibitor forms hydrogen bonds to
the three same residues in both structures: Arg109 (106 in
FABP4 residue numbering), Arg129 (126) and Tyr131 (128).
Secondly, hydrophobic interactions dominate the binding
mode in both structures.
3.4. Ligand-bound human FABP5 dimerizes via a
domain-swapping mechanism
The human FABP5 structure was previously solved as a
monomer either in its apo form or bound to exogenous lipid
by NMR or X-ray crystallography (Gutie
´rrez-Gonza
´lez et al.,
2002; Hohoff et al., 1999). However, human FABP5 forms a
domain-swapped dimer when complexed with BMS-309403
(Fig. 5; PDB entry 4azm). The crystals of the human FABP5–
inhibitor complex belonged to space group P42
1
2.
To examine whether the inhibitor was the cause of the
observed domain swapping, we also solved the crystal struc-
electronic reprint

ture of human FABP5 in complex with AEA (Table 1; PDB
entry 4azr). The human FABP5–AEA complex crystallized
in a different space group, C222
1
. We found that the human
FABP5–AEA complex in the new crystal form is also a
domain-swapped dimer, essentially the same as the dimer seen
in the presence of BMS-309403 (Fig. 6). However, the binding
mode of AEA in the domain-swapped human FABP5 dimer
is modified compared with that in the monomeric mouse
FABP5. Domain swapping of FABP5 makes the portal region
more open (Fig. 6a), enabling the AEA to penetrate deeper
into the substrate pocket (Fig. 6b) and resulting in an altered
hydrogen-bonding interaction at the AEA head region
(Fig. 6c). Because of the lower position of AEA, the hydroxyl
group at the tip can no longer form a hydrogen bond to
Tyr131. Instead, the hydrophilic AEA head region rotates by
180compared with the structure in monomeric mFABP5,
such that the hydroxyl now forms a hydrogen bond to Arg109.
The lowered AEA orients its carbonyl group towards Tyr131
and forms a hydrogen bond to the hydroxyl of Tyr131.
In the FABP5 dimer structures, the N-terminal half of the
first molecule (residues 1–59) forms a complete -barrel with
the C-terminal half of the second molecule (residues 60–134).
In so doing, the loop (Glu57–Thr62) connecting the S3 and
S4 -strands in the monomeric structure is converted into a
-strand, and together with S3
and S4 forms an unusually long
-strand that connects the two
domain-swapped monomers
(Figs. 5aand 5b). The electron
density in the connecting loop
region between the S3 and S4
-strands was impossible to
model without swapping the N-
and C-terminal subdomains of
each monomer. Furthermore, a
stretch of difference electron
density appears if residues 57–62
from both chains are omitted
during model building, clearly
indicating the continuity of the
-strands across the two copies of
the protein (Fig. 5c).
In the size-exclusion chroma-
tography profile, the bacterially
expressed and delipidated human
FABP5 existed predominantly as
a monomer and a dimer, with a
smaller amount being in a higher
oligomer state, possibly a
tetramer (Supplementary Fig.
S1). The presence or absence of
substrates or inhibitors does not
appear to affect the equilibrium
between monomers and dimers.
Fractions from the monomer
peak, when subjected to another
round of gel filtration, showed the
coexistence of monomers and
dimers. A similar gel-filtration
profile was also observed for the
purified and delipidated mouse
FABP5 (Supplementary Fig. 1).
This observation suggests that the
quaternary structure of FABP5
in solution may be in a dynamic
equilibrium between monomers
and dimers.
Domain swapping is a recog-
nized mechanism of protein
research papers
296 Sanson et al. FABP5 Acta Cryst. (2014). D70, 290–298
Figure 6
The AEA binding mode in domain-swapped human FABP5. (a) Superposition of hFABP5–AEA with
mFABP5–AEA. The H1–H2 cap in hFABP5 (cyan) moves left by >5 A
˚compared with that in the monomer
(gray), significantly opening up the portal region. (b) The AEA in the domain-swapped hFABP5 (orange)
enters 3.5 A
˚deeper into the substrate chamber compared with that in the mFABP5 monomer (gray). (c)
Stereoview of the interaction of AEA with hFABP5. The hydrocarbon chain of AEA is surrounded by
Met35 and Cys120 of one FABP5 (orange) and by Phe19, Tyr22, Met23 and Pro41 of the other FABP5
(cyan). The AEA hydroxyl forms a hydrogen bond to Arg109 as shown by the dashed red line (2.7 A
˚). The
AEA carbonyl forms a hydrogen bond to Tyr131 (2.4 A
˚).
electronic reprint

oligomerization (Bennett et al., 1995). The observed domain
exchange in human FABP5 requires the transient breakage of
several hydrogen bonds between S3 and S4 on one side and
subsequently between S1 and S10 on the opposite side of the
-clamshell of the monomeric (unswapped) FABP5. The
domain-swapped dimer may gain enthalpy because five addi-
tional hydrogen bonds form as the S3–S4 loop in the monomer
structure is converted to the -sheet configuration in the
domain-swapped structure (Fig. 5c). However, the enthalpy
gain is likely to be counterbalanced by a loss of entropy in the
solvent-exposed monomeric S3–S4 loop region. As a result,
the energy barrier associated with domain swapping in human
FABP5 may be low enough to be within the reach of ther-
modynamic fluctuation (Bennett et al., 1995). Domain swap-
ping in human FABP5 may suggest that the protein structure
is partially flexible and perhaps exhibits conformational
dynamics.
It is thought that the lipophilic ligands enter and exit the
deep and largely enclosed substrate cavity via the ‘portal’
region largely comprised of the capping helices H1 and H2
(Richieri et al., 1999; Sacchettini et al., 1989; Zhang et al., 1997;
Jenkins et al., 2002; Chen et al., 1998). Our observation of the
outwards movement of the portal region of FABP5 bound to
AEA compared with palmitic acid is indeed in agreement with
such a model (Supplementary Fig. 2). However, the portal
region alone may not be sufficient for admitting or releasing
the large and hydrophobic substrates. For example, in the
absence of steering forces the substrate failed to enter the
binding pocket of FABP in a molecular-dynamics simulation
(Friedman et al., 2006). In a steered molecular-dynamics
simulation in which the substrate was forced into the FABP
pocket, it was found that the structure in the -barrel region
underwent significant changes (Tsfadia et al., 2007). Therefore,
the potential -strand dynamics in human FABP5 may func-
tion in concert with the recognized portal region to facilitate
substrate binding. Such a possibility merits further investiga-
tion. However, engineering a disulfide bond into FABP5 may
be complicated by the presence of numerous cysteine residues
(six) in both the human and the mouse proteins.
4. Conclusions
Understanding endocannabinoid transport and signaling is
important for the development of small molecules that may
serve as potential analgesics. Our crystallographic studies of
FABP5 in complex with AEA and 2-AG provide structural
evidence for our previous cell-based conclusion that endo-
cannabinoids are transported intracellularly by FABP5. All
three ligands described in this report, the ligands AEA and
2-AG and the inhibitor BMS-309403, bind to FABP5 via
similar interactions that involve hydrogen bonding of their
polar regions mainly to Tyr131 and Arg129 of the transporter
and hydrophobic interactions of their lipophilic fragments
with the side chains lining the substrate pocket of FABP5. The
carrier substrate pocket contains multiple hydrophobic resi-
dues that appear to be sufficiently flexible to accommodate a
variety of fatty-acid-like molecules. While mouse FABP5 was
found only in the monomeric form, human FABP5 can exist as
a monomer as well as a domain-swapped dimer, suggesting a
higher degree of structural dynamics in the human protein.
Such dynamics in human FABP5 presents a challenge as well
as an opportunity for the development of the specific inhibi-
tors.
This work was partially supported by an SBU/BNL Seed
Grant from the Provost and by the Targeted Research
Opportunity Fusion Award from the Medical School at Stony
Brook University. DGD was supported by NIH DA016419
and DA026953. MK was supported by NIH DA032232.
Diffraction data for this study were collected on beamlines
X25 and X29 of the National Synchrotron Light Source.
Financial support comes principally from the Offices of
Biological and Environmental Research and of Basic Energy
Sciences of the US Department of Energy and from the
National Center for Research Resources (P41RR012408)
and the National Institute of General Medical Sciences
(P41GM103473) of the National Institutes of Health.
References
Ahn, K. et al. (2009). Chem. Biol. 16, 411–420.
Bennett, M. J., Schlunegger, M. P. & Eisenberg, D. (1995). Protein Sci.
4, 2455–2468.
Berger, W. T., Ralph, B. P., Kaczocha, M., Sun, J., Balius, T. E., Rizzo,
R. C., Haj-Dahmane, S., Ojima, I. & Deutsch, D. G. (2012). PLoS
One,7, e50968.
Blankman, J. L., Simon, G. M. & Cravatt, B. F. (2007). Chem. Biol. 14 ,
1347–1356.
Brunger, A. T. (2007). Nature Protoc. 2, 2728–2733.
Chen, V. B., Arendall, W. B., Headd, J. J., Keedy, D. A., Immormino,
R. M., Kapral, G. J., Murray, L. W., Richardson, J. S. & Richardson,
D. C. (2010). Acta Cryst. D66, 12–21.
Chen, X., Tordova, M., Gilliland, G. L., Wang, L., Li, Y., Yan, H. & Ji,
X. (1998). J. Mol. Biol. 278, 641–653.
Chuang, S., Velkov, T., Horne, J., Porter, C. J. & Scanlon, M. J. (2008).
J. Med. Chem. 51, 3755–3764.
Cravatt, B. F., Demarest, K., Patricelli, M. P., Bracey, M. H., Giang,
D. K., Martin, B. R. & Lichtman, A. H. (2001). Proc. Natl Acad. Sci.
USA,98, 9371–9376.
Deutsch, D. G. & Chin, S. A. (1993). Biochem. Pharmacol. 46,
791–796.
Friedman, R., Nachliel, E. & Gutman, M. (2006). Biophys. J. 90,
1535–1545.
Furuhashi, M. & Hotamisligil, G. S. (2008). Nature Rev. Drug Discov.
7, 489–503.
Garman, E. F. & Weik, M. (2011). J. Synchrotron Rad. 18, 313–317.
Gutie
´rrez-Gonza
´lez, L. H., Ludwig, C., Hohoff, C., Rademacher, M.,
Hanhoff, T., Ru
¨terjans, H., Spener, F. & Lu
¨cke, C. (2002). Biochem.
J. 364, 725–737.
Hamilton, J. A. (2004). Prog. Lipid Res. 43, 177–199.
Hodsdon, M. E. & Cistola, D. P. (1997). Biochemistry,36, 2278–2290.
Hohoff, C., Bo
¨rchers, T., Ru
¨stow, B., Spener, F. & van Tilbeurgh, H.
(1999). Biochemistry,38, 12229–12239.
Howlett, A. C., Reggio, P. H., Childers, S. R., Hampson, R. E., Ulloa,
N. M. & Deutsch, D. G. (2011). Br. J. Pharmacol. 163, 1329–1343.
Jenkins, A. E., Hockenberry, J. A., Nguyen, T. & Bernlohr, D. A.
(2002). Biochemistry,41, 2022–2027.
Kabsch, W. (2010). Acta Cryst. D66, 133–144.
Kaczocha, M., Glaser, S. T. & Deutsch, D. G. (2009). Proc. Natl Acad.
Sci. USA,106, 6375–6380.
Kaczocha, M., Vivieca, S., Sun, J., Glaser, S. T. & Deutsch, D. G.
(2012). J. Biol. Chem. 287, 3415–3424.
research papers
Acta Cryst. (2014). D70, 290–298 Sanson et al. FABP5 297
electronic reprint

Long, J. Z., Li, W., Booker, L., Burston, J. J., Kinsey, S. G., Schlosburg,
J. E., Pavo
´n, F. J., Serrano, A. M., Selley, D. E., Parsons, L. H.,
Lichtman, A. H. & Cravatt, B. F. (2009). Nature Chem. Biol. 5,
37–44.
Murshudov, G. N., Skuba
´k, P., Lebedev, A. A., Pannu, N. S., Steiner,
R. A., Nicholls, R. A., Winn, M. D., Long, F. & Vagin, A. A. (2011).
Acta Cryst. D67, 355–367.
Painter, J. & Merritt, E. A. (2006). Acta Cryst. D62, 439–450.
Richieri, G. V., Low, P. J., Ogata, R. T. & Kleinfeld, A. M. (1999).
Biochemistry,38, 5888–5895.
Sacchettini, J. C., Gordon, J. I. & Banaszak, L. J. (1989). J. Mol. Biol.
208, 327–339.
Schu
¨ttelkopf, A. W. & van Aalten, D. M. F. (2004). Acta Cryst. D60,
1355–1363.
Stivala, A., Wybrow, M., Wirth, A., Whisstock, J. C. & Stuckey, P. J.
(2011). Bioinformatics,27, 3315–3316.
Sulsky, R. et al. (2007). Bioorg. Med. Chem. Lett. 17, 3511–
3515.
Tsfadia, Y., Friedman, R., Kadmon, J., Selzer, A., Nachliel, E. &
Gutman, M. (2007). FEBS Lett. 581, 1243–1247.
Velkov, T., Chuang, S., Wielens, J., Sakellaris, H., Charman, W. N.,
Porter, C. J. & Scanlon, M. J. (2005). J. Biol. Chem. 280, 17769–
17776.
Weik, M., Ravelli, R. B. G., Kryger, G., McSweeney, S., Raves, M. L.,
Harel, M., Gros, P., Silman, I., Kroon, J. & Sussman, J. L. (2000).
Proc. Natl Acad. Sci. USA,97, 623–628.
Zhang, F., Lu
¨cke, C., Baier, L. J., Sacchettini, J. C. & Hamilton, J. A.
(1997). J. Biomol. NMR,9, 213–228.
research papers
298 Sanson et al. FABP5 Acta Cryst. (2014). D70, 290–298
electronic reprint