Cryo-EM structures of TbAQP2 bound to either glycerol, melarsoprol or pentamidine.

a, Overall structure of the TbAQP2 tetramer viewed either from the extracellular surface or within the membrane plane. b, Structure of protomer A of TbAQP2 viewed as a cartoon (rainbow colouration) with glycerol (sticks) bound. The cryo-EM density for glycerol is shown as a grey surface. c, Cut-away views of the channel in each of the TbAQP2 structures showing bound substrates and drugs (spheres) with atoms coloured according to type: red, oxygen; yellow, sulphur; blue, nitrogen; purple, arsenic; carbon, grey (glycerol), orange (melarsoprol) or cyan (pentamidine). d, Cryo-EM densities (grey surface) for glycerol, melarsoprol and pentamidine in their respective structures. See Figures S5 and S6 for different views of the substrates and comparisons between densities.

Interactions between TbAQP2 and bound substrates and drugs.

a, Amino acid residues containing atoms ≤ 3.9 Å from substrate or drug model in the cryo-EM structures are depicted and coloured according to the chemistry of the side chain: hydrophobic, green, positively charged, blue; polar, orange. Interactions with a backbone carbonyl group is shown as CO (red) and potential hydrogen bonds are shown as red dashed lines. Residues highlighted in an additional colour (grey, pale blue, pink or dark orange) make contacts in two or less of the structures (as indicated on the figure), whilst those without additional highlighting make contacts in all three structures. b, Protomer A from each structure was aligned and the positions of the two glycerol molecules (grey), melarsoprol (orange) and pentamidine (cyan) are depicted.

Comparison between conserved glycerol-binding regions of TbAQP2 and PfAQP.

a, The structure of TbAQP2 is depicted (rainbow colouration) with residues shown that make hydrogen bonds (yellow dashed lines) to either glycerol (grey) or in the NPS/NSA motifs. CO, backbone carbonyl groups involved in hydrogen bond formation. b, The structure of PfAQP2 is depicted (rainbow colouration) with residues shown that make hydrogen bonds (yellow dashed lines) to either glycerol (yellow sticks) or water (red spheres), or in the NPS/NLA motifs. CO, backbone carbonyl groups involved in hydrogen bond formation. c, Structures of TbAQP2 (rainbow colouration) and PfAQP are superimposed to highlight structural conservation of the NPS and NSA/NLA motifs. d, Outline of channel cross-sections from aquaglyceroporins containing bound glycerol molecules (sticks): TbAQP2 (grey); PfAQP (yellow; PDB ID 3CO2); human AQP7 (magenta, PDB ID 6QZI); human AQP10 (green, PDB ID 6F7H); Escherichia coli GlpF (purple, PDB ID 1FX8).

Molecular dynamics simulations of pentamidine-bound TbAQP2.

a, Density of water molecules (blue) through the TbAQP2 central cavity from molecular dynamics. Bound pentamidine (sticks, green) abolishes the ability of TbAQP2 to transport water molecules. b, View of pentamidine (cyan, carbon; red, oxygen; blue, nitrogen; white, hydrogen) bound to TbAQP2. Cyan mesh shows the density of the molecule across the MD simulation. c, Upon application of a membrane potential, the pentamidine centre of mass (COM) moves along the z-dimension in relation to the COM of the channel, with three independent repeats shown in different shades of blue. The bottom graph is for the potential in the physiological direction (negative intracellular). d, Energy landscapes for pentamidine through the TbAQP2 central cavity as calculated using umbrella sampling. Separate calculations were made for monomeric WT TbAQP2 and TbAQP2 with I110W (IW) or L258Y/L264R (LY/LR) mutations. Each trace is built from 167 x 40 ns windows, with the histogram overlap and convergence plotted in Fig. S10a,b. The position of the membrane phosphates is shown as grey bars, and the structural binding pose is shown as a dotted line.

Comparison between drug-bound AQP2 structures.

Superposition of an AQP2 protomer from this work (cyan) with a protomer determined by Chen et al. 36 (grey), with the positions of drugs shown in stick representation: a, pentamidine; b, melarsoprol.

Alignment of aquaporin amino acid sequences.

Amino acid sequences of AQP2 and AQP3 from Trypanosoma brucei brucei (Tbb) are aligned with chimeric aquaporins from drug resistant strains (P1000, 427MR, R25) and a chimera from drug-resistant T. b. gambiense (TbgAQP2/3). Three identical sequences are shown of TbbAQP2 with residues making contact to either glycerol, pentamidine or melarsoprol highlighted (cyan, green or yellow, respectively). Red bars represent transmembrane regions. Residues highlighted in grey are conserved in all sequences. The conserved NSA/NPS motif in TbAQP2 and the analogous NPA/NPA motif in TbAQP3 are in purple font and the residues mutated for the MD simulation studies (I110W, L258Y and L264R) are in red. Sequence information was obtained from Munday et al. 12 except R2528 and TbgAQP2/3 (678-800) 15.

Purification of TbAQP2 and identification of the drug binding site by hotspot analysis.

a, SEC trace of affinity purified TbAQP2. b, Coomassie-stained SDS-PAGE gel of fractions during detergent solubilisation (Sol), Ni2+-affinity chromatography and from the SEC column. c, Cryo-EM electron micrograph showing the particle distribution of TbAQP2. d, A region of TbAQP2 was identified as being most likely to bind a ligand is shown (yellow surface) is mapped on to the structure of TbAQP2 determined here (rainbow colouration. e, as in d, but also containing pentamidine (cyan sticks). f, HOLE85 trace for our resolved TbAQP2 structure, compared to a previously released TbAQP2 structure (‘8yj7’) 36 and Plasmodium falciparum AQP24.

Flow chart of the cryo-EM image processing and structure determination.

Local resolution maps of TbAQP2 structures.

a-c, 2D class averages for each of the structures, FSC curves of the reconstructions with estimates for resolution determined using an FSC of 0.143 and local resolution estimations as calculated by Relion.

Depictions of the side chain and substrate densities.

Comparisons between substrate densities.

a, Comparison between melarsoprol and glycerol. b, comparison between pentamidine and glycerol.

Molecular dynamics simulations.

a, RMSDs for simulations of tetrameric WT TbAQP2. Trajectories were fitted on the Cα atoms and RMSDs were calculated for either Cα atoms for apo TbAQP2 (top), Cα atoms for pentamidine (pent) bound TbAQP2 (middle), or for the pent molecule itself (bottom). Five independent repeats are shown as blue traces. b, Percentage of time (occupancy) a given residue is within 3.9 Å of either pentamidine, water or glycerol during the MD simulations. c, Plotting the z-axis position of pentamidine when bound to TbAQP2 in a WT, IW, or LY/LR background. d, Pore radius profiles from simulations of apo monomeric WT, IW, or LY/LR TbAQP2. Analyses were run using the CHAP package81 with traces showing the mean (black line), standard deviation (dark grey), and range (light grey) for all frames over 5 x 800 ns of simulation. e, As panel a, but for Cα atoms from monomeric TbAQP2 in a WT background, I110W (IW) or L258Y/L264R (LY/LR) mutations.

Water in molecular dynamics simulations

a, hydrogen-bond analysis from gmx hbond between the pentamidine and channel. Across the 5 independent MD runs of the monomeric TbAQP2 MD simulations, 23 hydrogen-bonds are formed, but only 6 appear to be high occupancy. These are highlighted on the image along with their index number, and their quantification is shown below. b, hydrogen-bond distances from the bonds in panel a plotted as a frequency plot. The average bond length was 0.29±0.02 nm, suggesting moderate strength. c, snapshots from MD showing water progression through the TbAQP2 channel in the absence of pentamidine. Three snapshots are shown. Residues that form especially high contact with water molecules (values given in brackets) are highlighted in each snapshot. Water-water hydrogen bonds, as computed using VMD, are shown in yellow dashed lines. d, number of water molecules along the TbAQP2 channel ±pentamidine, as computed using gmx h2order on the monomeric TbAQP2 MD simulations. e, orientation of the waters in each of the slices along with channel, as computed using gmx h2order.

Molecular dynamics simulations.

a, Plot of the membrane potential (V) as applied via an electric field, and calculated using the Gromacs tool gmx potential. Shown are potentials produced by an electric field applied in the forward (-ve; solid line) or backward (+ve; dashed line) direction. For context, the densities of lipid phosphate atoms and pent atoms are shown as orange and green traces, as measured using gmx density. b, Histograms for the umbrella sampling simulations used to generate the landscapes in Fig. 4d. The data show considerable overlap between windows. c, Conversion plots for the PMF landscapes. Shown are landscapes calculated for increasing lengths of simulations (length in ns). After about 30 ns, the landscapes converged. d, Snapshots taken from post umbrella sampling windows for WT TbAQP2 at -1.35 nm and +2.7 nm.

Lysis assay of T. brucei cells.

a-d, Lysis assay with T. b. brucei wild-type and various AQP2 mutant cell lines treated with arsenical compounds: a and c, cymelarsan; b and d, phenylarsine oxide, a different arsenic-containing trypanocide that enters the parasite independently of TbAQP28,10,12 and is thus used as a control for transporter-related arsenic resistance versus resistance to arsenic per se. The cells were placed in a cuvette and treated with either compound at t = 10 min. All points shown are the average of triplicate determinations and SD. When error bars are not visible, they fall within the symbol. The slow decline with cymelarsan over time in AQP2-KO and the mutant cell lines is attributable to residual uptake of the compound through the TbAT1/P2 transporter86,87.

Cryo-EM data collection, refinement and validation statistics

Details of molecular dynamics simulations on TbAQP2.