Water Permeation Across Biological Membranes: Mechanism and Dynamics of Aquaporin-1 and GlpF

See allHide authors and affiliations

Science  14 Dec 2001:
Vol. 294, Issue 5550, pp. 2353-2357
DOI: 10.1126/science.1066115


“Real time” molecular dynamics simulations of water permeation through human aquaporin-1 (AQP1) and the bacterial glycerol facilitator GlpF are presented. We obtained time-resolved, atomic-resolution models of the permeation mechanism across these highly selective membrane channels. Both proteins act as two-stage filters: Conserved fingerprint [asparagine-proline-alanine (NPA)] motifs form a selectivity-determining region; a second (aromatic/arginine) region is proposed to function as a proton filter. Hydrophobic regions near the NPA motifs are rate-limiting water barriers. In AQP1, a fine-tuned water dipole rotation during passage is essential for water selectivity. In GlpF, a glycerol-mediated “induced fit” gating motion is proposed to generate selectivity for glycerol over water.

Aquaglyceroporins constitute a large family of integral membrane proteins that facilitate highly efficient and specific passive permeation of water and other small uncharged solutes across biological membranes (1,2). Osmotic water regulation is essential for all life forms, and aquaglyceroporins are found throughout nature, with nearly 300 proteins identified and sequenced so far. In humans, more than 10 different aquaporins with specialized functionality are expressed in tissues as diverse as kidney, red blood cells, and brain. Malfunctions of these proteins cause a wide range of diseases, including nephrogenic diabetes insipidus, congenital cataract, and impaired hearing (1, 3, 4).

The human water channel aquaporin-1 (AQP1) (Fig. 1) (5) permeates water molecules across the membrane at a rate of 3 × 109s−1 per channel (6–8), with an activation energy nearly as low as the one associated with the self-diffusion rate in bulk water (8). The homologous bacterial glycerol facilitator GlpF is selective for glycerol and other linear alcohols (9, 10) and shows lower water permeability (10, 11) despite a wider pore. The low activation energies allow one to study entire water-permeation events through both proteins by molecular dynamics (MD) simulations in “real time,” without the need to accelerate the process by additional driving forces.

Figure 1

System setup for AQP1 and GlpF simulations. (A) Top view and (B) side view. All MD simulations were carried out with full electrostatics in a periodic simulation box containing the aquaglyceroporin tetramer (blue, cyan, orange, magenta), embedded within a POPE lipid bilayer (yellow head groups and green tails) surrounded by water (red, white). The total system consists of about 101,000 atoms (19).

The structural models of human AQP1 (12–14) and the atomic structure of GlpF from Escherichia coli (15) have confirmed and extended the early sequence-based “hourglass” model (16): The walls of the pore are formed by six transmembrane helices, 1 through 6, connected by five loops, A through E; the pore center is formed by the two highly conserved fingerprint asparagine-proline-alanine (NPA) motifs contained in the B and E loops, which fold back into the protein. The C-terminal halves of these two loops form two short helices that together form a seventh, kinked transmembrane helix. Despite a wealth of experimental data, major issues need to be resolved at the atomic level: How is this extremely high rate achieved while maintaining strict selectivity? How are ions, and particularly protons, excluded, even though they are known to be conducted well by hydrogen-bonded water chains (17,18)? What is the exact pathway of water molecules through the channel? How are the known structural differences between AQP1 and GlpF reflected in the permeation mechanism? An especially intriguing question is how GlpF facilitates permeation of (larger) glycerol molecules while hindering passage of (smaller) water molecules.

We have carried out extensive MD simulations (19) of AQP1 and GlpF in their biologically active form (Fig. 1): as tetramers (20, 21), embedded in a fully solvated bilayer membrane. Each of the simulations covered 10 ns. For AQP1, 205 water molecules visited the four pore regions, and 16 full-permeation events were observed, in good agreement with the experimental rate. In both AQP1 and GlpF, no permeation was observed for the central cavity, as has been expected from experiments (16,22, 23).

Figures 2 and 3 focus on the permeation mechanism. By averaging over the large number of partial permeation events, accurate profiles could be obtained. At the intracellular side (bottom), the pore is relatively wide (see also the red curves in Fig. 3, right), and the protein interacts only weakly with the water molecules (Fig. 2, green curve). The main interaction sites are formed by the backbone carbonyl groups of the residues preceding the first NPA motif, Gly72and Ala73, and the His74 side chain. Farther up the pathway, the first strong interaction site with water is formed, with both asparagines of the NPA motifs on one side of the pore, and the hydrophobic side chains of Phe24, Val176, and Ile191 on the other. At the extracellular side of the NPA motifs, nearly symmetric to the intracellular side, the carbonyl groups of residues Ile191, Gly190, and Cys189 interact with the water molecules in the pore.

Figure 2

Overlaid snapshots from a trajectory of a water molecule passing through AQP1 (left and middle panels, surface and ribbon representation of the same protein structure) and hydrogen bond energies per water molecule (right). The permeation event shown on the left lasted 3.3 ns. Those residues that interact with water molecules are colored yellow in the surface representation and are labeled in the ribbon representation, most notably the NPA region (between Asn76, Asn192, and Phe24) and the ar/R region around Arg195. In the surface representation, the front part of the channel has been removed to enable a view at the pore. The right panel shows hydrogen bond energies per water molecule along the pore axis, averaged over all observed (partial) permeation events (205 and 252 for AQP1 and GlpF, respectively). Hydrogen bond energies were calculated from donor-acceptor distances as described (38). The rendered protein model was made with BOBSCRIPT (39) and Raster3D (40).

Unexpectedly, 8 Å above the NPA motif, a second main interaction site is found, formed by the aromatic side chains of Phe56 and His180, and the positively charged Arg195(ar/R). Within this ar/R region, which also forms the narrowest part of the pore (Fig. 3, red curves), the hydrophobic Phe56 side chain orients the water molecules such as to enforce strong hydrogen bonds to Arg195 and His180. These residues are conserved in the water-selective aquaglyceroporins (24). Farther up the pore, at the extracellular side, the A and C loops interact with water mainly through Lys36 and Ser123, respectively.

Figure 3

Water dipole orientation (left) and water statistics (right) for AQP1 (top) and GlpF (bottom). Orientation and strength of the water dipole moment were calculated along the pore axis for both proteins, averaged over all (partial) permeation events observed in the simulations. The arrowheads represent the positive ends of the dipoles. The protein structures are colored according to the local electrostatic potential [from negative (red) to positive (blue)], as calculated with DELPHI (41). The arrow sizes and colors (from green to yellow) indicate the size of the average dipole moment per water molecule. The right panel shows relative free energiesG PMF(z) (kJ/mol) of water molecules (black curves), the entropic contribution TS toG(z) (green curves) due to restrained rotational freedom of the water molecules (42), pore radiir(z) (Å) (red curves), and the degreec(z) of correlated motion of adjacent water molecules along the pore axes z (blue curves). For GlpF, pore dimensions for the first (solid red curve) and last (dashed curve) nanosecond of the simulation are shown separately. The rendered protein models were made with BOBSCRIPT (39) and Raster3D (40).

Contiguous hydrogen-bonded water chains through the channels, a prerequisite for proton conductance, are not observed in any of the MD snapshots for either AQP1 or GlpF. The red curves in Fig. 2 show that for both proteins, such chains are most frequently interrupted within the ar/R region around the conserved (24) arginine (Arg195 in AQP1 and Arg206 in GlpF). This is also the region where water-water hydrogen bonds are weakest (blue curves). Together with the electrostatic repulsion by the positively charged arginine, this characterizes the ar/R region as the main filter for protons and other positive ions, including hydronium ions.

Contrary to what has been expected from more qualitative models (15), our simulations reveal the energetic cost of the disruption of water shells (Fig. 2, blue curves) in the ar/R region to be fully compensated by polar water-protein interactions (green curves). A similar compensation occurs within the NPA regions. Such disruption is thus unlikely to be rate-limiting. Instead, as can also be seen from the free-energy patterns in Fig. 3 (black curves), the simulations identify the hydrophobic zones adjacent to the NPA regions, where such compensation cannot occur, as the rate-limiting barriers of both AQP1 and GlpF.

Focusing on the different specificities of the two channels, the qualitative structure-based model of the fully conserved NPA motifs as important selectivity-determining regions (12) and the relevance of the hydrophobic residue facing the NPA motifs (24–26) are confirmed and explained. The simulations show that in GlpF, the NPA region interacts less strongly with passing water molecules than in AQP1 (Fig. 2, green curves). In AQP1, the hydrophobic Phe24 side chain intrudes more into the pore and thus forces passing water molecules to interact more strongly with the NPA loops than in GlpF (where this Phe is replaced by Leu21 and thereby acts as a size-exclusion selectivity filter, hindering the passage of larger molecules such as glycerol). Indeed, preliminary results from a simulation of the Leu21→Phe mutant of GlpF confirm this finding, showing a reduced water permeability.

Strong translational and orientational control of the water molecules, involving a fine-tuned dipole inversion, is observed in the pore region, particularly for AQP1 (Fig. 3, left). In both proteins, the water dipoles align with the helix macrodipoles (27) caused by the B and E helices, as speculated for AQP1 (12). The center of the rotation lies near the NPA motifs, where both N2-termini of the helices meet. As suggested by Berendsen (28), only small molecules with a large dipole moment are likely to be able to follow this inversion fast enough. This effect, therefore, is crucial for selectivity. A similar, although less pronounced, pattern is seen for GlpF. Here, this orientation pattern is broken within the ar/R region, in contrast to AQP1, where the dipoles remain oriented perpendicular to the membrane.

The combined enthalpic (mainly hydrogen bonds, Fig. 2) and entropic (Fig. 3) barriers computed for single molecules are much larger than the effective rate-determining free-energy barrierG PMF (Fig. 3, right). This discrepancy suggests that the observed high water-permeation rate is achieved through the highly collective motion of water molecules, which effectively lowers the activation energy. Such strong correlation is actually seen in the simulations (Fig. 3, blue curve). This multiparticle effect renders a straightforward (single-particle) rate estimate from the Arrhenius activation energy, e.g., using Kramers' theory, problematic. Indeed, considerable mismatch is seen between measured rates and corresponding activation energies (11).

For GlpF, the water permeability is found to decrease in the course of the simulation from an initial rate (during the time span from 1 to 3 ns) of 2.5 × 109 s−1 per monomer to a value of 1.0 × 109 s−1 in the last 2 ns. This decrease, which apparently continues at the end of the simulation, is compatible with the experimental rate of 0.5 × 109 s−1 (11). Also seen is a simultaneous narrowing motion of the pore, localized mainly at the NPA region (Fig. 3, right, red curves) which, however, does not affect the features of the other profiles. We suggest that this motion, which is not seen for AQP1, is a response to the removal of the crystallographic glycerol molecules from the starting structure and is the cause of the observed decrease. This observation challenges the current view (15) that the measured low water permeability is primarily caused by disruption of water shells and, instead, suggests “induced fit” gating motions, triggered by glycerol passage, as the main water blockage mechanism of GlpF. Additionally, it confirms the the role of the NPA region as the selectivity filter.

Full functional permeation events have been simulated and analyzed at the atomic level for the prototypic transmembrane water channel AQP1. Biomolecular permeation rates have been computed from first principles, i.e., without relying on the validity of rate theories. The essential features of aquaglyceroporins are multistage filters and task sharing (Fig. 4). For AQP1 and GlpF, the NPA region acts mainly as a size-exclusion selectivity filter, whereas the ar/R region predominantly provides selectivity against ions and protons.

Figure 4

Schematic summary of the permeation mechanism for AQP1 and GlpF, as extracted from our simulations. In both proteins the ar/R region is the narrowest part of the pore (even more so in AQP1) and forms the upper filter (red). Here, water-water hydrogen bonds are weakened to the largest extent (light blue) as compared with bulk water (darker blue), and therefore, together with the positive charge of the conserved Arg195 and Arg206, renders this region a proton filter. This effect is more pronounced in GlpF than in AQP1. In GlpF, the strongest protein-water hydrogen bonding site (green) is also located in this region, at Arg206, due to water repulsion by the aromatic Trp48 and Phe200 (gray ellipses). Farther down, the second stage of the filter (orange) is located at the conserved NPA motifs and is mainly a size-exclusion selectivity filter. Additionally, the water repulsion due to the hydrophobic Phe24 group enhances the water-asparagine interactions at the NPA motif and renders this region the strongest protein-water interaction site of AQP1 (green). In both proteins, the dipoles of water molecules passing the pore (red arrows) undergo a rotation during passage. For both AQP1 and GlpF, two rings of exclusively hydrophobic residues, directly adjacent to the NPA regions, form the rate-limiting free-energy barriers (yellow) for water molecules.

  • * To whom correspondence should be addressed. E-mail: hgrubmu{at}


Stay Connected to Science

Navigate This Article