compounds for inhibition; C3361 was synthesized by BOC Sciences
and the MVV009085-homologue was supplied by Mcule. Transport
was initiated by the addition of 2 μl of [^14 C]d-glucose at 40 μM final
concentration. The reaction was stopped by the addition of 1 ml of
10 mM Tris-MgSO 4 buffer and followed by rapid filtering through a
0.22-μm filter (Millipore). The on-filter collected proteoliposomes
were washed with 6 ml of buffer containing 10 mM Tris-HCl 7.5 and
2 mM MgSO 4 , transferred to scintillation vials and emulsified in 5 ml
of Ultima Gold scintillation liquid (Perkin Elmer) before scintillation
counting (TRI-CARB 4810TR 110 V; Perkin Elmer).
The recorded radioactivity of empty liposomes at 4% (v/v) DMSO
was subtracted from each tested condition. Half-maximal inhibitory
concentration (IC 50 ) values were obtained by fitting one-phase decay
nonlinear regression by GraphPad Prism 7.0. All transport results are
represented as mean values (n = 3) with their corresponding standard
errors.
Crystallization and structure determination of PfHT1
Crystals of PfHT1 in complex with d-glucose were grown at 4 °C using
the hanging-drop vapour-diffusion method. Purified PfHT1 protein at
8 mg/ml was added d-glucose to a final concentration of 50 mM. One
microlitre of this solution was mixed 1:1 with reservoir solution consist-
ing of 0.1 M MES pH 6.5, 0.1 M MgCl 2 , 26–30% (w/v) PEG 300 and 0.2%
(w/v) n-nonyl-β-d-glucopyranoside (NG, Anatrace). Crystals appeared
within 1 week in 26% PEG 300 and were dehydrated by equilibration of
the drops against 500 μl reservoir solution containing increasing con-
centrations of PEG 300 in steps of 2% (w/v) up to a final concentration
of 32% (w/v). Crystals were subsequently collected and flash-frozen in
and stored under liquid nitrogen.
X-ray diffraction data from PfHT1 crystals were collected at 100K at
the European Synchrotron Radiation Facility (ESRF) at the beamlines
ID30A-3 and ID23-1. Two datasets from different crystals were indexed,
integrated and scaled together using XDS^36 before merging using Aim-
less^37. Initial phases of PfHT1 were obtained by molecular replacement
using phenix.mr_rosetta^38 ,^39 and the outward-facing occluded structure
of human GLUT3 as an input search model (PDB 4ZW9). There are four
PfHT1 molecules in the asymmetric unit. Structure refinement was car-
ried out using Phenix.refine^40 ,^41 and auto BUSTER^42 with local NCS (non-
crystallographic symmetry), one TLS (translation–libration–screw
rotation) group per chain and external constraints to human GLUT3,
interspersed with manual model building in Coot^43. The Ramachandran
statistics are 92.6% favoured, 6.91% allowed and 0.93% outliers. Other
data collection and refinement statistics are presented in Extended
Data Table 2. Structural alignments were performed using the align
command of PyMol software (http://www.pymol.org/) using Cα coor-
dinates.
Molecular dynamics simulations
The starting models used for molecular dynamics simulations were
human GLUT3 (PDB 4ZW9) and chain C of PfHT1. The cytosolic loops
of PfHT1 connecting TM5 to TM6 and TM9 to TM10, as well as ICH5,
were modelled using MODELLER^44 version 9.21 before simulations. Six
simulation systems were constructed (three for PfHT1 and three for
GLUT3), each of which consisted of the protein embedded in a POPC
bilayer. To do this, six lipid configurations were generated using the
CHARMM-GUI membrane builder^45 , in which the protein (and ligands
if applicable) was embedded. These systems were then solvated in
150 mM NaCl. Details of each simulation replica can be found in Sup-
plementary Table 1.
All systems underwent energy minimization using steepest descent.
Equilibration molecular dynamics was then performed for a total of 375
ps, gradually relaxing positional restraints on protein, POPC lipids and
ligands, when relevant. The duration of each production molecular
dynamics simulation can be found in Supplementary Table 1. Simu-
lations were carried out under periodic boundary conditions and
production molecular dynamics was carried out using a 2-fs time steps.
The temperature and pressure were maintained at 303.15K and 1 bar
using the Berendsen thermostat and barostat^46 , respectively. Pressure
coupling was performed using semi-isotropic coupling with a time
constant of 5 ps and compressibility of 4.5 × 10−5 bar−1. Temperature cou-
pling was performed using three separate groups for protein, lipids and
solvent. Hydrogen bonds were constrained using the linear constraint
solver (LINCS)^47. Electrostatic interactions were modelled with a 1.2-nm
cutoff, with a switching function between 1.0 and 1.2 nm. Long-range
electrostatics were calculated using particle mesh Ewald (PME)^48. All-
atom molecular dynamics simulations were performed using Gromacs
2018.1. Interactions were modelled with the CHARMM36m (protein,
lipids and ions) and the TIP3P (water) forcefields^49.Analysis of simulations and protein morphing
Analysis of the molecular dynamics simulations were performed using
the gromacs analysis tools gmx rmsf and gmx pairdist, for root mean
square fluctuation and gating-residue distance calculations. Gate dis-
tance was determined by measuring the centre of mass of residues that
remained closest around the extracellular gate during the simulations:
residues Val44 to Ile50 and Asn311 to Ser317 for PfHT1, and residues
Thr28 to Pro34 and Asn286 to Ser292 for human GLUT3. The gate dis-
tances plotted are the mean between the three replicas are indicated
by darkened lines (Fig. 3e). Python scripts were written to parse and
plot relevant data^50. Figure generation was performed using PyMol
(https://pymol.org/2/).
Morphing between structural states for movie generation was per-
formed using PyMol. From the PfHT1 structure presented here, chain C
was used to generate models in the following resolved conformations:
outward-open (GLUT5, PDB 4YB9), outward-occluded (GLUT3, PDB
4ZW9), inward-occluded (XylE, PDB 4JA3) and inward-open (GLUT5,
PDB 4YBQ). Structural alignments of the proteins were carried out
in PyMol, and subsequent model generation using MODELLER. The
N termini of these respective models were superimposed, morphed
between states and the video was made using PyMol.Sugar-porter principal component analysis
Principal component analysis (PCA) is a statistical technique to reveal
dominant patterns^51. Diagonalization of the covariance matrix of a
system of variables renders the major axes of statistical variance or
principal components, thus mapping complex multidimensional data
into a few coordinates, which contain the major trends that explain
the statistical variation. For the sugar-porter structures, a set of near-
intact structures sharing 30% homology with PfHT1—that is, eukaryotic
GLUT structures (GLUT1, GLUT3 and GLUT5) along with E. coli XylE
transporter (16 PDB codes in total: PfHT1, 6RW3; XylE, 4GBZ, 4GBY,
6N3I, 4JA3 and 4JA4; GLUT1, 4PYP, 5EQI, 5EQH and 5EQG; GLUT3, 5C65,
4ZWC, 4ZWB and 4ZW9; and GLUT5, 4YB9 and 4YBQ)—were aligned
to extract the common structural fold, mostly formed by conserved
helices (353 residues) (Extended Data Fig. 9a). Missing residues in some
of the structures were rebuilt with MODELLER, making sure that the
positions of the corresponding structural elements were strictly kept
for the core alignment. The structural ensemble was aligned to the
structure of GLUT5 in an open outward-facing conformation (root
mean square deviation (r.m.s.d.) of 2.7 ± 1.2 Å) and used to compute
the covariance matrix versus this reference; that is, the mean-square
deviations in atomic coordinates from their mean position (diagonal
elements) and the correlations between their pairwise fluctuations (off-
diagonal elements). The covariance matrix was diagonalized to obtain
a set of eigenvectors or principal components, ordered according to
their eigenvalues with decreasing variance from those representing
the largest-scale motions up to the smallest fluctuations in atomic
coordinates. Within this framework, any structure i is characterized by
its scalar product projections onto the conformational space defined
by the major components, PCk (k = 1,2...n):