Structure and mechanism of the mammalian fructose transporter GLUT5.

The altered activity of the fructose transporter GLUT5, an isoform of the facilitated-diffusion glucose transporter family, has been linked to disorders such as type 2 diabetes and obesity. GLUT5 is also overexpressed in certain tumour cells, and inhibitors are potential drugs for these conditions. Here we describe the crystal structures of GLUT5 from Rattus norvegicus and Bos taurus in open outward- and open inward-facing conformations, respectively. GLUT5 has a major facilitator superfamily fold like other homologous monosaccharide transporters. On the basis of a comparison of the inward-facing structures of GLUT5 and human GLUT1, a ubiquitous glucose transporter, we show that a single point mutation is enough to switch the substrate-binding preference of GLUT5 from fructose to glucose. A comparison of the substrate-free structures of GLUT5 with occluded substrate-bound structures of Escherichia coli XylE suggests that, in addition to global rocker-switch-like re-orientation of the bundles, local asymmetric rearrangements of carboxy-terminal transmembrane bundle helices TM7 and TM10 underlie a 'gated-pore' transport mechanism in such monosaccharide transporters.

Introduction

GLUT transporters belong to the solute carrier 2 family (SLC2) and, so far, fourteen different isoforms GLUT1-14 have been identified[1,2]. Except for GLUT13, GLUT transporters are uniporters, which facilitate the diffusion of monosaccharides like glucose and fructose across the cell membrane in a concentration-dependent manner[1,2]. Each GLUT transporter shows a distinct pattern of tissue distribution, gene regulation, substrate preference and kinetic properties[1,2]. For example, GLUT1 is distributed in a wide range of tissues, including the blood-brain barrier and is essential for glucose transport into the brain[3], whereas GLUT4 is mostly localized to skeletal muscles and adipose tissue, and is the major insulin-stimulated glucose transporter[1,4]. GLUT5 is the only member specific to fructose[5,6] and together with GLUT2, which transports fructose in addition to glucose, they make up the major fructose transporters in the body [7,8]. GLUT5 is primarily expressed in the small intestine, but lower levels are also expressed in brain, adipose tissue, kidney, testes and skeletal muscle [5,9,10]. GLUT activity is also associated with various diseases[1,11]. For instance, increased GLUT5 expression has been linked to several metabolic disorders[12,13], and several types of cancers like breast cancer[14], because of the higher energy demand of cancer cells stimulating sugar uptake, the so-called Warburg effect[15]. GLUT transporters belong to the larger major facilitator superfamily (MFS), whose members have a fold consisting of two symmetrical six transmembrane helix (TM) bundles[16,17]. Within the MFS they belong to a subfamily of sugar porter transporters[18,19], whose members are found in all domains of life and are important targets for industrial and biomedical applications[20]. Recently, an open inward-facing structure of human GLUT1 was reported with a bound sugar from a detergent head-group in the substrate-binding site, and compared to previous structures of the related Escherichia coli D-xylose:H+ symporter XylE in the outward- and inward-occluded conformations, suggesting a “rocker-switch” type transport mechanism [21-23]. However, as little is known about the molecular basis of substrate binding and release in GLUT transporters, their alternating-access mechanism is yet to be fully understood.

Open outward and inward GLUT5 structures

Rat and bovine GLUT5 (rGLUT5 and bGLUT5) that share ~81% sequence identity to human GLUT5 were selected and optimised for structural studies using fluorescence-based screening methods (Methods). rGLUT5 was crystallized in complex with an Fv antibody fragment (rGLUT5-Fv, Methods). The rGLUT5-Fv and bGLUT5 structures were solved by molecular replacement (MR) and refined against data extending up to 3.3 Å and 3.2/4.0 Å (anisotropic data), respectively (Extended Data Table 1 and 2, Extended Data Fig 1, and Methods). The GLUT5 structure shows the typical MFS fold, plus five additional helices on the intracellular side, one at the C-terminus (ICH5) and the other four, ICH1-4, located between the N- and C-terminal TM bundles (Fig. 1). bGLUT5 crystallized in an open inward-facing conformation (Fig. 1) and although human GLUT1 (hGLUT1) and bGLUT5 share only 43% sequence identity, their inward-facing structures superimpose well, with an r.m.s.d. of 1.12Å for 364 pairs of Cα atoms (Methods and Extended Data Fig. 2a). The rGLUT5-Fv structure shows an open outward-facing conformation, which is a state that has not been observed previously in any of the related sugar porter structures[22-25] (Fig. 1). The open outward-facing conformation is possibly stabilized by the Fv fragment, which binds to the ICHs (Extended Data Fig. 3).

Extended Data Table 1

Crystallographic data collection and refinement statistics.

	rat GLUT5-Fv	bovine GLUT5 #
Data collection
Space group	P21	P212121
Cell dimensions
a, b, c (Å)	76.78, 151.54, 106.40	74.61, 112.15, 139.57
α, β, γ (°)	90.00, 97.25, 90.00	90,00, 90.00, 90.00
Resolution (Å)	50-3.27 (3.39-3.27)	100 - 3.00 (3.11 - 3.01)*
Rsym or Rmerge	21.5 (>100)	10.06 (>100)
I/σ/	11.6 (1.5)	11.21 (0.86)
Completeness (%)	100 (100.0)	99.4 (93.9)
Redundancy	13.0 (13.0)	10.2 (8.6)
Refinement
Resolution (Å)	50-3.27 (3.5-3.27)	33.3-3.2 (3.31 - 3.20)
No. reflections (Rfree set)	38017 (3355)	13346 (1331)
Rwork/ Rfree	24.2/28.8 (35.4/37.5)	23.6/25.8 (32.8/36.5)
No. atoms
Protein	10657	3382
B-factors
Protein	157.0	149.2
R.m.s deviations
Bond lengths (Å)	0.004	0.003
Bond angles (°)	0.97	0.90
Ramachandran plot (%)
Favored	98.0	94.5
Outliers	0	0

Data were obtained by scaling together two datasets collected on the same crystal

Highest resolution shell used in the final refinement is shown in parenthesis.

Extended Data Table 2

Completeness of bovine GLUT5 data per resolution shell after correction for anisotropy

Completeness of bovine GLUT5 data per resolution shell after correction for anisotropy

Resolution range (Å)	Completeness (%)
100.0 - 8.10	95.5
8.10 - 6.43	100.0
6.43 - 5.62	99.9
5.62 - 5.10	99.9
5.10 - 4.74	100.0
4.74 - 4.46	99.7
4.46 - 4.24	99.6
4.24 - 4.05	100.0
4.05 - 3.90	78.4
3.90 - 3.76	53.4
3.76 - 3.64	40.6
3.64 - 3.54	33.4
3.54 - 3.45	26.3
3.45 - 3.36	21.2
3.36 - 3.29	15.4
3.29 - 3.22	13.1
3.22 - 3.15	9.4
3.15 - 3.09	6.0
3.09 - 3.04	3.5
3.04 - 3.00	1.7

Extended Data Fig. 1

Anisotropy descriptors of bGLUT5 data reported by the UCLA-MBI Diffraction Anisotropy Server and 2Fo-Fc electron density maps for the bovine and rat GLUT5 structures

a. Degree of anisotropy of bGLUT5 data, resolution limits for the 3 principle axes (left), and panel illustrating steps along correction of bGLUT5 data for anisotropy (right). b Representative portions of the electron density map (1.5σ) for bGLUT5 overall model (left) and a close-up of the substrate binding site (right); residues highlighted are numbered based on rGLUT5 for sake of clarity c. Electron density (1.0σ) for rGLUT5 showing one of the inter-bundle salt-bridge clusters that form in the open outward-facing conformation.

Fig. 1

Structures of rat GLUT5 in the open outward-facing conformation and bovine GLUT5 in the open inward-facing conformation

a. Ribbon representation of open outward-facing rat GLUT5 (left) and open inward-facing bovine GLUT5 (right) structures, viewed in the plane of the membrane. TMs 1 and 4 and TMs 2, 3, 5 and 6 in the N-terminal TM bundle are colored in blue and light-blue, respectively. TMs 7 and 10 and TMs 8, 9, 11 and 12 in the C-terminal TM bundle are colored in red and yellow-brown, respectively. The intracellular domain helices ICH1 to ICH5 are shown in grey. b. Slab through the surface electrostatic potential of the open outward- (left) and open inward-facing (right) GLUT5 structures, as viewed within the plane of membrane, which highlight the accessibility of the sugar to the central cavity (shown as a dotted ellipse). c. Ribbon diagrams of GLUT5 viewed from the cytoplasm in the open outward- (left) and inward-facing (right) conformations.

Extended Data Fig. 2

Superimposition of open inward-facing bGLUT5 and hGLUT1 structures, and comparison of the substrate-binding site in bGLUT5 and inward-facing XylE

a. Ribbon representation of inward-facing bGLUT5 (colored as in Fig. 1a) and inward-facing hGLUT1 (light grey) structures, as viewed in the plane of the membrane. The D-glucopyranoside moiety of the detergent molecule bound to GLUT1 (n-nonyl-β-D-glucopyranoside (β-NG)) is shown as sticks. Density for ICH5 at the C-terminus is missing in both hGLUT1 and bGLUT5 inward-facing structures and highlighted with the dotted ellipse. The beginning of TM1 kinks further outwards in the bGLUT5 structure compared to hGLUT1 and residues 1 to 18 could not be built. The r.m.s.d. (root mean square deviation) after superposition of the two structures is 1.12Å for 364 pairs of Cα atoms (see Methods). b The substrate-binding in the inward-facing bGLUT5 structure (coloured as in Fig. 1), is very similar to that seen in inward-facing XylE (4JA4) structure (shown in light-grey). Only non-conserved residues and the equivalent glutamine to Q166 are labeled for XylE.

Extended Data Fig. 3

Structure of the rat GLUT5-Fv complex

a. Cartoon representation of the complex between rGLUT5 (grey) and 4D111Fv (heavy-chain variable region (VH) is in blue; light-chain variable region (VL) is in red). 4D111Fv binds to the cytoplasmic domain of GLUT5, including ICH2 (residues 226, 230, 234), the loop between ICH2 and ICH3 (residues 238, 240, 241), and ICH3 (residue 243), with ~848 Å2 of buried surface area at the interface. b. Packing of the rat GLUT5-Fv complex molecules in the crystal. The unit cell is represented as green lines.

Central fructose-binding site of GLUT5

The GLUT5 substrate-binding site is closely related to those of hGLUT1 and E. coli XylE[21,22] (Fig. 2a and Extended Data Fig. 2b). Many of the residues lining the central cavity are conserved between GLUT5 and hGLUT1, and include Ile169, Ile173, Gln166, Gln287, Gln288, Asn324 and Trp419 (Fig. 2a and Extended Data Fig. 4). In GLUT5, Trp419 is the only tryptophan positioned in the substrate-binding site (Fig. 2a and Extended Data Fig. 5a), and it is essential for transport[26]. Consistent with rGLUT5 transport activity (Extended Data Fig. 6a), strong quenching of tryptophan fluorescence could be observed with the addition of D-fructose, but not with the addition of L-fructose or known GLUT1 substrates like D-glucose, D-galactose or D-mannose (Extended Data Fig. 5). Using this assay, the affinity of rGLUT5 for D-fructose was measured to have a Kd around 6 - 9 mM (Fig. 2b and Extended Data Fig. 6b), which is similar to that reported for human GLUT5[5,8]. Throughout this study tryptophan fluorescence quenching of rGLUT5 was henceforth used to assess substrate binding.

Fig. 2

The fructose-binding site of GLUT5

a. The substrate-binding site in the inward-facing bGLUT5 structure (left panel; colored as in Fig. 1) is very similar to the inward-facing hGLUT1 structure (right panel; light-grey). To facilitate comparison to rGLUT5, bGLUT5 residues are labeled with rGLUT5 numbering. For hGLUT1 only Q161 and all other residues that are different in bGLUT5 are labeled. The D-glucopyranoside moiety of bound n-nonyl-β-D-glucopyranoside in hGLUT1 is shown as sticks. b. D-fructose binding to GLUT5 as measured by tryptophan (Trp) fluorescence quenching (excitation 295 nm; emission 338 nm) after addition of increasing concentrations of D-fructose to WT (black squares) and to WT protein that had been previously incubated with the GLUT inhibitor HgCl2 (open circles). c. Trp fluorescence quenching for purified substrate-binding site mutants after addition of 40 mM D-fructose (non-filled bars) relative to WT (filled bar). d. Trp fluorescence quenching after addition of either 40 mM D-fructose or D-glucose to purified WT (black-filled bar) or Q166E (non-filled bars); pre-incubation with the inhibitor HgCl2 is indicated. e. Trp fluorescence quenching after addition of increasing concentrations of D-glucose to either purified Q166E (black squares), Q166E previously incubated with HgCl2 (open circles) or WT (open triangles). In all experiments errors bars, s.e.m.; n = 3.

Extended Data Fig. 4

Sequence alignment of rat GLUT5 (rGLUT5), bovine GLUT5 (bGLUT5), human GLUT5 and GLUT7 (hGLUT5, hGLUT7), human GLUT1-4 (hGLUT1-4), Saccharomyces cerevisiae (HXT7), Plasmodium falciparum (PfHT1), Arabidopsis thaliana (GlcT) and E. coli XylE

Secondary structure elements of rat GLUT5 are indicated above the alignment, and coloured as in Fig. 1a. Strictly conserved residues are highlighted in black-filled boxes, and highly conserved residues are shaded in grey. Green boxes highlight central cavity residues that are specific to GLUT5 and red boxes highlight those that are conserved among GLUTs. Purples boxes highlight residues forming the salt-bridges between cytosolic TM segments. A blue box (TM5) highlights Gln166, whose mutation to glutamic acid, as present in GLUT7, weakens D-fructose binding but supports strong D-glucose binding in rGLUT5. The brown box (TM8) highlights Glu336 that is conserved across all the GLUTs and replaced with glutamic acid in XylE. Red bars underneath the alignment indicate the sugar porter (SP) family motifs[18,19]. Note that because bGLUT5 and hGLUT5 have an additional amino acid at position 8 their numbering differs from rGLUT5 by 1 amino acid. For clarity, bGLUT5 residues are labeled using rGLUT5 numbering.

Extended Data Fig. 5

D-fructose binding monitored by tryptophan fluorescence quenching

a. Cartoon representation of the outward-facing rGLUT5 structure, as viewed from the plane of the membrane with the coloring as shown in Fig. 1a. Atoms in all tryptophan residues are shown as spheres and tryptophan W419, whose fluorescence is quenched by substrate, is labeled. b. Emission fluorescence spectra for purified deglycosylated rGLUT5 wildtype like mutant N50Y (referred to as “WT”), shown in the range of 320-360 nm with an excitation wavelength of 295 nm after the addition of 40 mM D-fructose (top), and 40 mM L-fructose (bottom). Emission fluorescence spectra for purified WT protein that had been previously incubated with the inhibitor HgCl2 is also shown for D-fructose (middle). c. Tryptophan fluorescence quenching (excitation 295 nm; emission 338 nm) after incubation of purified rGLUT5 N50Y with either 40 mM D-fructose (black-filled bar) or L-fructose, D-glucose, D-mannose, D-xylose or D-galactose as labeled (open bars). Tryptophan fluorescence quenching for purified WT protein that had been previously incubated with the inhibitor HgCl2 is also shown for D-fructose (non-filled bar) d. As in c., rGLUT5 with a single tryptophan residue (W419), which contains the following mutations: N50Y, W70F, W191F, W239F, W265F, W275F, W338F and W370F. No tryptophan quenching was observed for D-fructose (5 mM HgCl2), L-fructose, D-glucose or D-galactose. In all experiments errors bars, s.e.m.; n = 3.

Extended Data Fig. 6

Substrate-specificity in GLUT5

a. Time-dependent uptake of D-[14C]-fructose by rGLUT5 wildtype (open squares and triangles) and the deglycosylated mutant N50Y (filled squares and triangles) in proteoliposomes incubated with or without the inhibitor HgCl2 as labeled. Non-specific uptake was estimated with 0.1 mM L-[14C]-Glucose for wildtype (filled circles) and the N50Y mutant (open circles). In all experiments errors bars represent a spread of duplicates. Inset shows SDS-PAGE analysis of the purified rat GLUT5 wildtype and the deglycosylated N50Y mutant. b. Tryptophan fluorescence quenching (excitation 295 nm; emission 338 nm), after incubation of purified rat GLUT5 mutant (N50Y, W70F, W191F, W239F, W265F, W275F, W338F, W370F) that contains one single tryptophan residue, W419, with increasing concentrations of D-fructose (filled squares) and to the protein previously incubated with the inhibitor mercury chloride (open circles). c. Slab through the surface of the outward-facing rGLUT5 structure as viewed in the plane of membrane. The structure of substrate-bound XylE structure was further superimposed onto rGLUT5 and is shown here as a grey ribbon. In XylE, Trp392 (Trp388 in hGLUT1) is located at the bottom of the cavity (spheres; magneta) and coordinates D-xylose (stick form; yellow). In GLUT5, the equivalent residue is an alanine, making the cavity deeper. d. D-fructose binding as measured by tryptophan fluorescence quenching (excitation 295 nm; emission 338 nm) after incubation with 40 mM D-fructose for WT (open bar), and TM7 mutations of Ile295 (interacts with TM10 residues) and Tyr296 and Tyr297 residues. Equivalently located tyrosine residues in XylE occlude the sugar-binding site from the outside[22]. Fluorescence quenching for the mutants are displayed as a percentage of total WT binding. In all experiments errors bars, s.e.m.; n = 3.

Gln166 is the only conserved residue in the substrate-binding site that is positioned differently between the hGLUT1 and GLUT5 structures (Fig. 2a). In hGLUT1 the equivalent glutamine (Gln161) points away, rather than towards, the central cavity as it has likely re-orientated to accommodate the acyl chain of the bound detergent in the sugar-binding site (Fig. 2a). In XylE the equivalent glutamine (Gln168) is orientated similarly as in GLUT5 (Extended Data Fig. 2b). Other residues that line the binding site, but are not conserved between GLUT5 and hGLUT1 are Tyr31 (Phe26 in hGLUT1), His386 (Phe379), Ala395 (Trp388), His418 (Asn411) and Ser391 (Gly384) (Fig. 2a and Extended Data Fig. 4). Single alanine mutants of each of these residues show strongly reduced D-fructose binding (Fig. 2c). The alanine mutants that show the weakest D-fructose binding are Tyr31 (TM1), Gln287 (TM7), His386 (TM10) and His418 (TM11). Except for Tyr31, these residues belong to the C-terminal TM bundle, indicating an asymmetrical binding mode of the sugar between the N- and C-terminal TM bundles in GLUT5, as seen in the sugar-bound hGLUT1 and XylE structures[21,22]. The substrate-binding cavity is deeper in GLUT5, because the tryptophan residue found at the bottom of the cavity in hGLUT1 (Trp388) and XylE (Trp392) is replaced with alanine (Ala395) in GLUT5 (Fig. 2a and Extended Data Fig. 6c). Trp388 in hGLUT1 is critical for inhibition by cytochalasin B[26], whereas GLUT5 is insensitive to this inhibitor[5].

GLUT7 is the closest isoform to GLUT5 and transports both D-fructose and D-glucose[27]. Among the substrate-binding site residues of GLUT5 almost all are identical to those in GLUT7 (Extended Data Fig. 4). The most notable difference is that Gln166 in GLUT5 is replaced in GLUT7 with glutamate (Extended Data Fig. 4). In rGLUT5, the substitution of Gln166 (TM5) with glutamate clearly weakens D-fructose binding (Fig. 2c and 2d). Furthermore, the Q166E mutant now shows robust binding to D-glucose with an apparent affinity (Kd) of ~ 4 mM (Fig. 2d and Fig. 2e). The introduced carboxylate is likely to play a similar role to Glu380 (TM10) in hGLUT1 (Fig. 2a), which is essential for D-glucose transport[28].

Salt-bridges stabilize the outward conformation

A common feature observed in recent structures of MFS transporters are inter-TM bundle salt-bridges that break and form near the central cavity during transport[16,29,30]. In H+-coupled transporters, it is thought that the protonation state of these salt-bridges is coupled to substrate-binding and further to structural transitions[16,29,30]. In GLUT5, no salt-bridges are observed near the central cavity in either conformation. Instead, inter-TM bundle salt-bridges are observed only in the outward-facing conformation and far from the central cavity, linking the cytoplasmic ends of TMs 3, 4, 5 in the N-terminal TM bundle to those of TMs 9, 10 and 11 in the C-terminal TM bundle (Fig. 3a). Specifically, Glu151 (TM4) forms a salt-bridge to both Arg97 (TM3) and Arg407 (TM11), and similarly Glu400 (TM10) forms a salt-bridge to both Arg158 (TM5) and Arg340 (TM9). Due to their strict conservation these residues make up the well-described sugar porter signatures[18,19] and are related by a pseudo two-fold symmetry axis running through the center of the transporter and perpendicular to the membrane plane (Fig. 1a and Fig. 3b and Extended Data Fig. 4). In the inward-facing conformation the inter-TM bundle salt-bridge pairs Glu400-Arg158 and Glu151-Arg407 are separated by some ~7 and 13 Å, respectively (Fig. 3a). In the inward-facing conformation, no inter-TM bundle salt-bridges are formed on the extracellular side, indicating that in the absence of substrate GLUT proteins may favor the outward-facing conformation. Consistently, the substitution of Glu400, Arg407 and Glu336 equivalent residues in GLUT4 to neutral amino acids arrests the transporter in an inward-facing conformation[31]. Moreover, in the outward-facing conformation Glu336 (TM8) in the C-terminal TM bundle forms a salt-bridge to Arg340, which is connected to the inter-TM bundle Glu400-Arg158 salt-bridge (Fig. 3a). The equivalent glutamate to Glu336 in hGLUT1 (Glu329) was substituted to glutamine to stabilize the inward-facing hGLUT1 structure[21]. The salt-bridges formed between inter-TM bundles seem to be coupled also to substrate-binding in GLUT5, because substituting Glu151 and Glu400 with alanine shows significantly reduced D-fructose binding (Fig. 3c).

Fig. 3

Inter-TM salt-bridges form between bundle cytoplasmic ends in the outward-facing conformation

a. Cartoon representation of GLUT5 as viewed from the cytoplasm in the outward- (left) and inward-facing (right) conformations. ICHs are not shown for clarity. TMs are colored as in Fig. 1a., and residues forming salt-bridges are shown as sticks. To facilitate comparison to rGLUT5, bGLUT5 residues are labeled with rGLUT5 numbering. Breakage of conserved salt-bridges as seen here for GLUT5 has also been predicted for XylE [23] b. Superimposition of the N- and C-terminal 6-TM bundles. Strictly conserved and pseudo symmetry related charged residues forming the salt-bridges are labeled and are shown as sticks. c. D-fructose binding as measured by Trp fluorescence quenching after incubation with 40 mM D-fructose for purified WT GLUT5 (black-filled bar) and single alanine mutations of key acidic residues E400 and E151 that form inter-bundle salt-bridges, and G389, which is located in the hinge point of TM 10 critical for TM10 conformational change (non-filled bars). Trp fluorescence quenching by D-fructose for these mutants is displayed as a percentage of WT binding. In all experiments errors bars, s.e.m.; n = 3.

The ICH domain is below this cytoplasmic salt-bridge network, and positioned similarly, with respect to the N- and C-terminal TM bundles, in both outward-open GLUT5 and outward-occluded XylE structures[22] (Extended Data Fig. 7a). ICH1, ICH2 and ICH3 are linked together by salt-bridges, as previously shown for XylE and GLUT1[21,22]. In GLUT5, ICH5 lacks charged residues that could interact with ICH1-3 (Extended Data Fig. 7b), as it does in the outward-occluded XylE structure[22]. Rather, the N-terminal helices ICH1-3 interact with the C-terminal bundle via a salt-bridge formed between Glu252 in ICH3 and Arg407 in TM11; thus linking the ICH domain to a TM involved in the inter-bundle salt-bridge network. In inward-facing GLUT5, these interactions are broken (Extended Data Fig. 7c) and, as observed in the inward-facing hGLUT1 and XylE structures[21,23,24], ICH5 could not be built (Extended Data Fig. 2a). Therefore, the role of the ICH domain might be to provide additional stabilization of the outward-facing conformation, as suggested previously[21].

Extended Data Fig. 7

The intracellular helical domain (ICH)

a. Cytoplasmic view of the ICH domain after superposition of the open, outward-facing rGLUT5-Fv (grey) and outward-facing occluded E. coli XylE (teal) (4GBY) structures. b. In the outward-facing GLUT5 structure ICH1-3 are linked together by several salt-bridges (side chains are labeled and shown as sticks in yellow). In contrast, no polar interactions are formed between ICH5 and either ICH1-3 or cytoplasmic ends of N-terminal TM bundle helices. A salt-bridge forms (dotted line in magenta), however, between Glu225 in ICH3 and Arg407 in TM 11, which also forms part of the inter-bundle salt-bridge network (side chains are labeled and shown as sticks in cyan). c. In the inward-facing GLUT5 structure, this inter-bundle salt-bridge network is not formed, because the cytoplasmic ends of the N- and C-terminal bundle have moved apart; consistently, the ICH domain functional role is proposed to act as a scaffold domain that further helps to stabilize the outward-facing conformation[21].

TMs 7 and 10 form substrate-induced gates

In GLUT5, the N- and C-terminal TM bundles undergo a small rotation of ~15° between the open outward- and inward-facing conformations (Fig. 1). As observed in other MFS transporter structures[16], cavity-closing contacts in GLUT5 form mostly between TMs 1 and 7 on the outside, and between TMs 4 and 10 on the inside (Fig. 1c and Fig. 4a). Among these helices, however, TMs 7 and 10 in the C-terminal TM bundle seem to play the most prominent roles in occluding the substrate-binding site from the outside and inside, respectively (Fig. 4b). Comparisons of the substrate-free open GLUT5 structures with the substrate-occluded XylE structures[22,23] highlight the central role of TM7 and TM10 in gating (Extended Data Fig. 8). Inverted-symmetry related TMs 7 and 10 make up highly conserved sugar transporter signatures[2], and they also form a large fraction of the substrate-binding site in hGLUT1[21] and XylE[21,22], in agreement with previous functional data[28,32,33]. Between the outward-open GLUT5 conformation and outward-occluded XylE conformation[22], the extracellular half of TM7 shows the largest shift towards the substrate-binding site (Extended Data Fig. 8a). Similarly, from the inward-occluded XylE conformation[23] to the inward-open GLUT5 conformation, the intracellular half of TM10 shows the largest shift away from the substrate-binding site (Extended Data Fig. 8b and c); this TM10 movement has also been described previously for XylE [23,24] . In GLUT5, the observed conformational changes in TMs 7 and 10 occur at hinge points that contain glycine residues (Fig. 4b and 4c and Extended Data Fig. 4). Tyr382 in TM10 interacts tightly with TM7 residues Ileu295 and Val292 in the outward-facing state and these interactions do not take place in the inward-facing conformation (Fig. 4c). The interactions between TMs 7 and 10 seem important to coordinate their conformational changes during transport, because mutation of the Ile295 equivalent residue in hGLUT5 to valine or alanine abolishes D-fructose transport[34]. Consistently, similar interface-perturbing mutations of Ile295 significantly reduce D-fructose binding in rGLUT5 (Extended Data Fig. 6d). Further, in hGLUT1 the equivalent residue to Val292 (Ile287) was substituted to every other amino acid, and each mutant shows markedly different D-glucose affinities and transport kinetics[35]. Therefore, it suggests that the interactions between TM7 and TM10 are important and tuned with respect to substrate affinity and transport kinetics.

Fig. 4

Substrate-induced gates are predominantly formed by TMs 7 and 10 in the C-terminal bundle

a. Superposition of GLUT5 open outward- and inward-facing (*) structures, as viewed from the extracellular (left) and intracellular (right) side of the membrane. TMs are colored as in Fig. 1a, except inward-facing TMs 1* and 4* and TMs 7* and 10* that are coloured in orange and cyan, respectively. ICHs have been removed for clarity. b. Superimposition of the GLUT5 open outward- and inward-facing structures as viewed in the plane of the membrane. For clarity, TMs 5, 5*, 8, and 8* are not shown. Cavity-closing contacts are mostly formed by TMs 1* and 7* on the extracellular side in the inward-facing conformation and by TMs 4 and 10 on the intracellular side in the outward-facing conformation. These TMs are the first TMs in each of the four 3-TM repeats of the MFS fold[16,37]. D-xylose, as it is in the occluded-outward-facing XylE structure (4GBY), is shown in stick-form. With the inward movement of TM7 conserved tyrosine residues are likely to occlude the substrate from exiting, as seen for the equivalently located tyrosine residues in the substrate-occluded XylE structure[22] and as supported by D-fructose binding data (Extended Data Fig. 6d). The opening movement of TM10 to enable cytosolic substrate release has been described previously for XylE [23,24] and other unrelated MFS transporters[38,39] c. Interactions between hydrophobic residues between TM7 and TM10 in the outward-facing conformation (left) are lost in the inward-facing conformation (right). To facilitate comparison to rGLUT5, bGLUT5 residues are labeled with rGLUT5 numbering.

Extended Data Fig. 8

Access to the central cavity and substrate-binding site is gated by TM7 on the outside and TM10 on the inside

a. Superposition of outward-facing open GLUT5 and outward-facing occluded E. coli XylE (4GBY) structures. The TM numbering for outward-facing occluded XylE has an additional asterisk “*”. The inward-facing GLUT5 structure is colored as in Fig. 1a and that of XylE in grey. The bound D-xylose in stick form in green. The r.m.s.d. is 1.38Å for 290 pairs of Cα atoms (see Methods). b. Superposition of inward-open GLUT5 and inward-occluded E. coli XylE structure (4JA3) with coloring and annotation as described in a. The r.m.s.d. is 1.80Å for 274 pairs of Cα atoms (see Methods). The bound D-xylose in 4GBY is represented in stick form in green. The ICH domain is not shown for clarity. c. Superposition of inward-facing open GLUT5 and inward-facing open XylE (4JA4) structures as viewed from the cytoplasmic side with coloring and annotation as described in a. The ICH domain is not shown for clarity. The r.m.s.d. is 1.70Å for 273 pairs of Cα atoms (see Methods).

Conclusions

Symmetrical substrate binding and rigid-body movements of the N- and C-terminal TM bundles around a centrally located substrate-binding site form the structural basis of the “rocker-switch” mechanism in MFS transporters[16,36]. However, here we have described asymmetrical rearrangements in the MFS subfamily of sugar porters, consistent with the asymmetrical binding mode of the sugar reported previously in XylE and hGLUT1 structures[21,22]. Therefore, we conclude that transport in GLUT5 is not only controlled by the rocker-switch type movement[21] of the N- and C-terminal TM bundles, but also by a gated-pore mechanism, in the form of local, gating movements by TMs 7 and 10 in the C-terminal TM bundle that are coupled to substrate binding and release (Fig. 5 and Supplementary Videos 1 - 3). TM10 is also part of the inter-TM bundle salt-bridge network, indicating how local gating and global rocker-switch type movements are coupled. Importantly, a deeper understanding of GLUT5 structure and transport mechanism, as described here, should facilitate the structure-based design of novel inhibitors with therapeutic potential.

Fig. 5

Alternating-access transport mechanism in GLUT5

Schematic representation of the “rocker-switch” type movement of the N- and C-terminal TM bundles and of the local, gating conformational changes of TMs 7 and 10 supporting a “gated-pore” type transport mechanism in GLUT5.

METHODS

Rattus norvegicus GLUT5 cloned full-length sequence (UniProt accession number: P43427); N50Y deglycosylation mutation is underlined and additional C-terminal residues retained after TEV cleavage are shown in italic (see next section for cloning details).

MEKEDQEKTGKLTLVLALATFLAAFGSSFQYGYNVAAVNSPSEFMQQFY

Bos taurus GLUT5 cloned sequence containing 1 to 473 out of 501 residues (UniProt accession number: P58353); N51A deglycosylation mutation is underlined and additional C-terminal residues retained after TEV cleavage are shown in italic (see next section for cloning details).

MEPQDPVKREGRLTPVIVLATLIAAFGSSFQYGYNVAAINSPSEFMKDFY

Fv light chain variable region (VL); additional C-terminal residues, retained after TEV cleavage, are shown in italics

ELDIVLTQSPLSLPVSLGDQASISCRSSQSIVHSNGNTYLEWYLQKPGQSPKLLIYKVSNR FSGVPDRFSGSGSGTDFTLKISRVEAEDLGVYYCFQGSHVPYTFGGGTKLEIK

Fv heavy chain variable region (VH); additional C-terminal residues retained after TEV cleavage, are shown in italics

LEVNLVESGGGLVQPGGSRKLSCAASGFTFSSFGMHWVRQAPEKGLEWVAHISSGSRTI DYADTVKGRFTISRDNPKNTLFLQMTSLRSEDTAIYYCARGNGYYDALDYWGQGTSVT VSSAKTTPPSV

Target identification using fluorescence-based screening methods

GLUT5 homologues were cloned into the GAL1 inducible TEV cleavable GFP-His8 2μ vector pDDGFP2, transformed into the S. cerevisiae strain FGY217 (MATa, ura3–52, lys2Δ201, and pep4Δ)[40] and overexpressed as described previously[41,42]. In brief, 10 ml S. cerevisae FGY217 cell cultures in –URA media and 0.1% glucose were grown at 30°C and expression was induced by the addition of final 2% (v/v) D-galactose at an OD600 of 0.6. After ~22 hrs, cells were harvested, resuspended in 1 × PBS buffer and overexpression levels assessed by whole-cell GFP fluorescence[41,42]. Fusions with detectable expression levels were re-grown in larger 2-L culture volumes and membranes subsequently isolated. The monodispersity of expressed fusions were screened in crude dodecyl-β-D-maltopyranoside (DDM), decyl-β-D-maltopyranoside (DM), nonyl-β-D-maltopyranoside (NM) and n-dodecyl-N,N-dimthylamine-N-oxide (LDAO) solubilized membranes by fluorescence-detection size exclusion chromatography (FSEC)[43] as described previously[41]. Out of the GLUT5 homologues screened, rat and bovine GLUT5 showed the sharpest monodispersity profiles and was the most stable after purification in detergent[44].

Large-scale production and purification of rat and bovine GLUT5

For rat GLUT5, cells were harvested from 10 L S. cerevisiae cultures, resuspended in buffer containing 50 mM Tris-HCl pH 7.6, 1 mM EDTA, 0.6 M sorbitol, and lysed by mechanical disruption as described previously[42]. Membranes were isolated by ultracentrifugation at 195,000 g for 3 h, homogenized in 20 mM Tris-HCl pH 7.5, 0.3 M sucrose, 0.1 mM CaCl2, frozen in liquid nitrogen and stored at −80 °C. Rat GLUT5 membranes were solubilized with 1% DDM in equilibration buffer (EB) consisting of 1 × PBS, 10 mM imidazole, 150 mM NaCl, 10 % glycerol. After ultracentrifugation at 195,000 g for 45 min, the supernatant was incubated with 10 ml of Ni2+-nitrilotriacetate affinity resin (Ni-NTA; Qiagen) for 2 h at 4 °C. The resin was washed with 100 ml of EB containing 0.05% DDM and 30 mM imidazole, and the protein eluted in 25 ml of EB containing 0.05% DDM and 250 mM imidazole. The eluted protein was incubated with TEV-His6 protease to remove the GFP-His8 tag and dialyzed against 20 mM Tris-HCl pH 7.5, 150 mM NaCl, 10% glycerol, 0.03% DDM. Dialyzed sample was then loaded onto a 5 ml HisTrap column (GE Healthcare) equilibrated in dialysis buffer, and the flow-through was collected and concentrated. The protein was further purified by size-exclusion chromatography (SEC) using a Superdex 200 column (GE Healthcare) in buffer consisting of 20 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.02 % DDM. All rat GLUT5 mutants were generated with a standard PCR-based strategy and were sub-cloned, overexpressed and purified as described for wildtype.

Protein complexes were prepared by incubation of the HisTrap-purified rat GLUT5 with the SEC-purified Fv fragment (supplemented with 0.02% (w/v) DDM) at a molar ratio of 1:1.5 for 1 h on ice. The complex was subjected to SEC (Superdex 200, GE Healthcare). The SEC step was repeated twice to ensure reproducibility of crystals. Peak fractions containing rat GLUT5-Fv complex were concentrated to ~10 mg/ml by ultrafiltration (Millipore, MWCO 50 kDa), and immediately used for crystallization experiments.

For bovine GLUT5, membranes were isolated from 40 L S. cerevisiae cultures. After membrane solubilization and ultracentrifugation (as described for rat GLUT5), the supernatant was supplemented with 55 mM Imidazole, and incubated with 30 ml of Ni-NTA resin for 2 h at 4 °C. The resin was washed with 600 ml of EB supplemented with 55 mM imidazole and 0.03% DDM, and the protein was eluted using 125 ml of EB containing 250 mM imidazole and 0.03% DDM. After cleavage of GFP-His8 cleavage by TEV-His6 protease the material was re-loaded onto the same 30 ml Ni-NTA resin column (no imidazole), the flow-through was concentrated and passed through a 1 ml Ni-NTA column for further clean up. The untagged protein was further purified by SEC in buffer containing 10 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 0.09 % undecyl-β-maltopyranoside (UDMβ), and then concentrated to ~4 mg/ml for crystallization experiments.

Generation of single-chain Fv (scFv) fragments

Animal experiments conformed to the guidelines outlined in the Guide for the Care and Use of Laboratory Animals of Japan and were approved by the University of Tokyo Animal Care Committee (approval no. RAC07101). Rat GLUT5-specific scFv fragment were generated essentially as previously described[45]. In brief, purified GLUT5 was reconstituted into liposomes containing chicken egg yolk phosphatidylcholine (egg PC; Avanti) and adjuvant lipid A (Sigma-Aldrich) by detergent removal method. Small unilamellar proteoliposomes were prepared by sonication. MRL/lpr mice were immunized with 0.1 mg of the proteoliposome antigen three times at two-week intervals. Immunized mice were sacrificed, and RNA in their splenocytes was isolated and converted into cDNA via reverse-transcription PCR. The VL and VH repertoire was assembled via an 18 amino acid flexible linker and cloned into the phage-display vector pComb3XSS. Biotinylated proteoliposomes were prepared by reconstituting GLUT5 with a mixture of egg PC and 1,2-dipalmitoyl-sn-glycero-3-phosphoethanolamine-N-(cap biotinyl) (16:0 biotinyl Cap-PE; Avanti), and used as binding targets for scFv-phage selection. The targets were immobilized onto streptavidin-coated paramagnetic beads (Dynabeads) or streptavidin-coated microplates (Nunc). After four rounds of biopanning, proteoliposome-targeted enzyme-linked immunosorbent assays (“liposome ELISAs”) were performed on periplasmic extracts of individual colonies. Positive clones were collected and evaluated using a Biacore T100 (GE Healthcare).

Expression and purification of Fv fragment

Antibody fragments in the scFv format are undesirable for use as crystallization chaperones because they are able to intermolecularly form domain-swapped dimers, and dimer-monomer equilibrium may increase structural heterogeneity. Therefore, we used Fv fragments for crystallization trials. The Fv fragment were expressed in Brevibacillus choshinensis as a secreted His6-tagged protein and purified from culture medium. The cells were grown at 30 °C with 200 r.p.m. in 2SY medium (soypton 40 g/l, yeast extract 5 g/l, glucose 20 g/l, CalCl2 0.15 g/l) supplemented with 50 mg/l neomycin. The cells expressing the VL and VH were initially grown separately as overnight precultures. The precultures were then combined and diluted in 2SY medium to give OD600 of 0.02 of each strain and grown for further 65-70 h. The cells were removed by centrifugation at 6,000 g for 15 min. The recovered culture supernatant was adjusted to a final ammonium sulfate concentration of 60% saturation, and the precipitate was pelleted, dissolved in TBS buffer (10 mM Tris-HCl, pH 7.5, 150 mM NaCl), and dialyzed overnight against the same buffer. Dialyzed sample was mixed with Ni-NTA resin equilibrated with Fv1 buffer (10 mM Tris-HCl, pH 7.5, 150 mM NaCl, 20 mM imidazole). Bound proteins were eluted with Fv2 buffer (10 mM Tris-HCl, pH 7.5, 150 mM NaCl, 250 mM imidazole), mixed with TEV-His6 and dialyzed overnight against TBS buffer. Cleaved His6 tag and TEV-His6 were removed by binding to a HisTrap column equilibrated with Fv1 buffer. The tag-free Fv fragment was concentrated and loaded onto a HiLoad16/60 Superdex 75 column (GE Healthcare) equilibrated with TBS buffer. Peak fractions were pooled, concentrated, flash frozen in liquid nitrogen, and stored at −80°C.

Transport activity of reconstituted GLUT5

The proteoliposome D-fructose uptake assay was modified from that previously described for human GLUT1[46]. In brief, purified rat GLUT5 was reconstituted by the freeze–thaw/extrusion method. Crude lipids were extracted from bovine liver and sonicated to make unilamellar liposomes. 500 μL of a mixture containing ~10 μg of purified GLUT5 and 20 mg of liposomes in 10 mM TrisSO4 (pH 7.5) was flash frozen and thawed at room temperature. Large, unilamellar proteoliposomes were prepared by extrusion (LiposoFast, Avestin; membrane pore size, 400 nm). For each time point, 10 μl proteoliposomes (0.4 mg lipid; 0.2 μg GLUT5) was added to 10 μl transport buffer containing 10 mM TrisSO4 and 2 mM MgSO4 (pH 7.5) with or without the addition of 0.5 mM HgCl2. Time course of 0.1 mM [14C]-D-fructose transport was measured at 25 °C at indicated time intervals and stopped by the addition of cold buffer containing 10 mM TrisSO4, 2 mM MgSO4 (pH 7.5) and 0.5 mM HgCl2 and immediately filtered. Non-specific uptake was estimated with 0.1 mM [14C]-L-glucose. The radioactivity corresponding to the internalized substrate was measured by scintillation counting. Each experiment was performed in triplicate and data-points shown indicate average values of two technical replicates.

Substrate specificity

Unless otherwise stated the rat GLUT5 deglycosylation mutant (N50Y) used for structure determination is referred to as “wild-type like protein (WT)”. Only rat GLUT5 WT and mutants showing a monodisperse peak in DDM during gel filtration were assessed for their ability to bind sugar by tryptophan fluorescence. In all experiments, either purified rat GLUT5 WT or mutants were diluted to 0.06 mg/ml in purification buffer containing 150 mM NaCl2, 20 mM Tris pH 7.5, 0.03% DDM. D-fructose, L-fructose, D-glucose, D-mannose, D-xylose and D-galactose stocks, that were freshly prepared in purification buffer, were added to the diluted protein to reach a final concentration of 40 mM. After each sugar addition the sample was incubated at room temperature for 2 minutes before measuring tryptophan fluorescence. Measurements were performed using a Cary Eclipse Fluorescence Spectrophotometer (Agilent Technologies), with an excitation wavelength of 295 nm and the emission spectra recorded in the range of 300-400nm with a 5 nm excitation slit in a High Precision Cell Quartz Suprasil 10 × 2mm cuvette (Hellma Analytics). The emission peak of 338 nm was taken as an average of the four wavelengths 337.03, 337.96, 339.06 and 340 nm. Each experiment was carried out in triplicate in the absence and presence of the well-known GLUT inhibitor HgCl2 (5 mM)[47]. To compare mutant binding to D-fructose, nonspecific tryptophan fluorescence quenching of 2.5% (as measured with 40 mM L-fructose) was first subtracted from both wildtype and mutant quenching levels prior to calculating the final percentage of quenching relative to wildtype.

Binding analysis

D-fructose or D-glucose stocks, that were freshly prepared in purification buffer, were sequentially added to the diluted protein to a final concentration of 40 mM. After each sugar addition the sample was incubated at room temperature for 2 minutes prior to measuring. Tryptophan fluorescence measurements were performed as above. Each experiment was carried out in triplicate in the absence and presence of the known GLUT inhibitor HgCl2 (5 mM). The rat GLUT5 curves were fitted by non-linear regression using the software Prism. The bovine GLUT5 construct used for structural studies binds D-fructose with similar affinity as rat GLUT5 with a Kd of 5.5 ± 0.6 mM.

Crystallization

Fv fragment

The structure of the Fv fragment was used as a search model for the molecular replacement. Crystals of the Fv fragment appeared in the well buffer at 20°C containing 0.1 M CAPS-NaOH (pH 10.5), 1.6 M (NH4)2SO4, 0.1 M Li2SO4, 6.5 mM n-nonyl-β-D-glucoside.

Rat GLUT5

Crystals of rGLUT5-Fv complex (in DDM) used for structural determination were grown at 20°C by hanging drop vapor diffusion. A 400-μl reservoir containing 33-35 % PEG 400, 0.12 M CaCl2, 0.1 M Tris-HCl (pH 8.0) was equilibrated against a 2-μl drop containing a 1:1 mixture of the complex and reservoir solution. After 2 to 3 weeks of growth, crystals were dehydrated by stepwise equilibration of the drops against 400 μl reservoirs containing increasing concentration of PEG 400, in steps of 5% and up to a final concentration of 70%. Crystals were then flash-frozen and stored in liquid nitrogen.

Bovine GLUT5

Crystals were grown over the course of 2 to 3 weeks at 4 °C in 2 μl drops (500 μl well) consisting of the well solution 31 to 35% PEG 300, 0.125 M HEPES pH 8.0, 0.125 M NaCl, 0.125 M LiSO4, mixed first with 70 mM HEGA-10 (4:1) and then the protein solution (1:1), and placed over the well solution diluted with water (4:1) before sealing. For freezing in liquid nitrogen, crystals were soaked for one minute in the mother liquor supplemented with ~ 15% PEG 300.

Data collection, structure determination and analysis

For the Fv fragment, diffraction data were collected at 100 K at SPring-8 beamline BL41XU (Japan) and processed using HKL2000 packages[48] and the CCP4 suite[49]. The crystal belonged to the space group C2 with two molecules per asymmetric unit. The structure was determined by molecular replacement with Molrep50 using the Fv portion of an antibody structure (PDB code: 1IGC) as a search model. Refinement were performed till the Rfree value decreased to ~22% with REFMAC5. For rat GLUT5-Fv complex, diffraction data were collected at 100 K at SPring-8 beamline BL41XU (Japan) and processed using HKL2000 packages[48] and the CCP4 suite [49]. The dataset used for structure determination and refinement was generated by the combination of 4 datasets from 4 independent crystals. The space group was determined to be P21, with two rat GLUT5-Fv complexes per asymmetric unit. The structure was determined by molecular replacement with Molrep[50] using two search models (polyalanine of the transmembrane region in the outward-facing conformation of FucP (PDB code: 3O7Q), and the separately determined 1.5 Å structure of the Fv fragment used in this study). Refinement were performed with PHENIX[51] followed by manual examination and rebuilding of the refined coordinates using COOT[52]. Recently determined structures of XylE (PDB code: 4GC0) and human GLUT1 (PDB code: 4PYP) helped with the modeling. The 6 N-terminal residues (Met1-Gln6), 22 C-terminal residues (Ser481-Gln502), and 22 residues (Asn39-Asn60) in molecule A and 12 residues (Met45-Arg56) in molecule B in TM2 are not included in the structure as they did not have interpretable densities. For bovine GLUT5, data were collected on frozen crystals at beamline I02, Diamond Light Source (UK). The dataset used for structure determination and refinement was generated using HKL2000 by processing and scaling together two datasets collected on two different parts of the same crystal, and by correcting for anisotropy (UCLA server, http://services.mbi.ucla.edu/anisoscale/) (Extended Data Table 1 and 2 Extended Data Figure 1). The structure was solved by molecular replacement using the N-terminal and C-terminal bundles of rat GLUT5 as independent search models in PHASER, and refined using REFMAC5 and BUSTER against data up to 3.0 Å resolution after anisotropy correction[53], with rounds of re-building in COOT[52]. Final refinement was performed using PHENIX[51] at 3.2 Å resolution (Extended Data Figure 1). Structural alignments were performed using the align command of PYMOL software using C-alpha coordinates.

Homology modeling

Models for the outward-occluded and inward-occluded conformations were based on corresponding XylE crystal structures (4GBY and 4JA3, respectively) and were produced using Modeller 9v12[54]. Multiple alignment of GLUT5 homologs (Homo sapiens, Canis lupus, Rattus norvegicus, Felis catus, Gallus gallus, Anolis carolinensis) and xylE sequences were combined with structural data from the XylE structures and the GLUT5 structures using expresso mode of t-coffee[55]. The multiple alignment in combination with helical restraints corresponding to the TMs of bovine GLUT5 structure was used as an input for Modeller. For each conformation, 20 models were generated with loop optimization. The DOPE scoring function was used to select the final model. Outward-facing bovine GLUT5 was modeled with the same protocol based on the rat outward-facing GLUT5 structure and alignment of GLUT5 sequences. The four conformations (outward-open, outward-occluded, inward-occluded and inward-open) were morphed and rendered using PyMOL (http://www.pymol.org/).