GlpF, the glycerol facilitator protein of Escherichia coli, is an archetypal member of the aquaporin superfamily. To assess its structure, recombinant histidine‐tagged protein was overexpressed, solubilized in octylglucoside and purified to homogeneity. Negative stain electron microscopy of solubilized GlpF protein revealed a tetrameric structure of ∼80 Å side length. Scanning transmission electron microscopy yielded a mass of 170 kDa, corroborating the tetrameric nature of GlpF. Reconstitution of GlpF in the presence of lipids produced highly ordered two‐dimensional crystals, which diffracted electrons to 3.6 Å resolution. Cryoelectron microscopy provided a 3.7 Å projection map exhibiting a unit cell comprised of two tetramers. In projection, GlpF is similar to AQP1, the erythrocyte water channel. However, the major density minimum within each monomer is distinctly larger in GlpF than in AQP1.
The glycerol uptake facilitator of Escherichia coli (GlpF; Boos et al., 1990) is one of the few known diffusion facilitators in the inner membrane of this bacterium. Glycerol diffuses into the cell through GlpF and is phosphorylated by the glycerol kinase (GlpK), which prevents back‐diffusion. Besides glycerol transport, the diffusion of polyols and urea derivatives through GlpF has been reported (Maurel et al., 1994), but none of these substrates is transported in a phosphorylated state. In contrast, it remains unclear whether GlpF allows the passage of water.
GlpF is a member of the aquaporin protein superfamily (formerly known as the MIP family), which includes water channels (AQPs) and the facilitators of small solutes like glycerol (GLPs; Reizer et al., 1993; Park and Saier, 1996; Heymann and Engel, 1999). All of these channels are strictly selective for non‐ionic compounds, thus preventing the dissipation of the membrane potential. Sequence comparisons have revealed an internal homology between the C‐ and N‐terminal halves of all aquaporins, with highly conserved regions in the cytoplasmic loop B and the extracellular loop E (Heymann and Engel, 2000). According to the model of Jung et al. (1994), these two loops fold back into the membrane to form a channel akin to an hourglass. This region is believed to be the site of solute selectivity. Loop E of GlpF contains additional amino acids compared with AQP1. Other conserved residues have been found in the aquaporin superfamily by Froger et al. (1998) and were demonstrated by Lagree et al. (1999) to discriminate AQPs from GLPs. In addition, Heymann and Engel (2000) have identified a pair of residues that may determine the size of the pore. Tetrameric organization seems to be a general structural feature of aquaporins (Mitra et al., 1995; Walz et al., 1995; Li et al., 1997; Hasler et al., 1998; Daniels et al., 1999; Ringler et al., 1999). In contrast, analytical ultracentrifugation (Lagree et al., 1998) and freeze fracture electron microscopy (Bron et al., 1999) have indicated a monomeric organization for GlpF. Here we show by electron microscopy that purified GlpF solubilized in octyl‐β‐d‐glucopyranoside (OG) is in fact a tetrameric particle that assembles into highly ordered arrays suitable for solving its structure by electron crystallography.
Protein expression and single particle analysis
GlpF carrying 10 C‐terminal His residues was overexpressed in E. coli, solubilized and purified in OG as described by Borgnia et al. (1999). Analysis of purified and solubilized GlpF by SDS–PAGE revealed a prominent band at an apparent mol. wt of 30 kDa. GlpF purified in this way is functionally active (M.J. Borgnia, personal communication). Transmission electron microscopy (TEM) of negatively stained solubilized GlpF revealed two different particle populations (Figure 1A): smaller, often square‐shaped particles and larger complexes of about twice the size of the small ones (arrowheads). Tilting of the tetramers possibly related to the His tag (Ringler et al., 1999) may explain the heterogeneous appearance of the small particles. Nevertheless, 1294 single particles were selected automatically (Figure 1B), windowed and subjected to reference‐free alignment (Penczek et al., 1992). The resulting square‐shaped average projection had a side length of ∼80 Å and four weak peripheral densities (Figure 1B, large inset).
Purified GlpF particles were also freeze‐dried and imaged with a scanning transmission electron microscope (STEM) for mass measurements (Müller and Engel, 1998). The low‐dose dark‐field image in Figure 1C shows particles of different brightness and size, whose masses were calculated and sorted in a histogram after mass loss correction (Figure 1D; Müller et al., 1992). The mass histogram resulting from the analysis of 100 images (2073 particles in total) exhibits a broad asymmetric distribution with a single maximum at ∼190 kDa. Two independent Marquardt algorithms (see Methods) produced a three‐peak fit with Gauss profiles having a standard deviation of 111 kDa located at 170, 327 and 511 kDa. The experimental errors for these three peaks were estimated as 12, 20 and 31 kDa, respectively, taking the standard deviation, the number of particles within a peak and the calibration accuracy into account. A size of 170 kDa is compatible with a GlpF tetramer including the four His tags and their spacers (12 kDa), plus 60 kDa of OG accounting for two micelles (Walz et al., 1994). Thus, 327 kDa corresponds to an octamer and 511 kDa to a dodecamer. The abundance of tetrameric particles documented by the mass histogram is consistent with the results from negatively stained preparations (Figure 1A).
Solubilized GlpF was reproducibly crystallized using a continuous flow dialysing device (Jap et al., 1992) as described in Methods. Analysis in the TEM showed polycrystalline vesicles with diameters up to 40 μm and mostly monocrystalline double‐layered sheets with rectangular shapes and diameters up to 8 μm (Figure 2A). Some tubular structures were seen in many reconstitution experiments. However, their crystallinity was far inferior to that of the double‐layered sheets. The crystal quality of the latter was assessed by electron diffraction of frozen–hydrated samples. The diffraction pattern recorded with a 1k × 1k CCD camera shows spots beyond 3.6 Å (Figure 2B) but weak structure factors between 7 and 5 Å. The four (28,7) orders marked by arrowheads correspond to a resolution of 3.6 Å. Since the two layers were rotated with respect to each other, two lattices are indicated.
Electron microscopy and image treatment
Images of negatively stained crystals were recorded in a TEM and analysed by correlation averaging (Baumeister et al., 1982). The unit cell containing two tetrameric structures exhibited a P4 symmetry. Adjacent tetramers were differently stained similar to AQP1 crystals (Walz et al., 1994), indicating their up–down orientation (data not shown).
Images of frozen–hydrated crystals recorded at low dose revealed sharp spots when examined by optical diffraction out to a resolution of 7 Å. The eight best images were digitized and processed by the MRC program package (Henderson et al., 1986, 1990). The unit cell size was determined to 104 Å. The phase residuals of the merged data obtained after lattice unbending and transfer function correction indicated significant information up to a resolution of 3.7 Å (Table 1), yielding the projection map shown in Figure 3. In contrast to AQP1 crystals with P4212 symmetry, the GlpF crystals exhibit a P4 symmetry (Table 2), since oppositely oriented tetramers are rotated about their 4‐fold axes by different amounts. Thus, the striking similarity of the two oppositely handed tetramers shown in Figure 3 demonstrates the quality of the map.
Here we present the first structural analysis of GlpF, the archetypal member of the GlpF subcluster of the aquaporin superfamily (Heymann and Engel, 1999). Negative stain electron microscopy and mass measurements of freeze‐dried unstained preparations in the STEM demonstrate that OG‐solubilized GlpF exhibits the same tetrameric structure as other aquaporins studied previously: AQP1 (Smith and Agre, 1991; Walz et al., 1994), AqpZ (Ringler et al., 1999), MIP (Hasler et al., 1998) and TIP (Daniels et al., 1999). This result is in contrast with reports of monomeric GlpF and the hypothesis that glycerol facilitators are monomeric whereas water channels require tetramerization (Lagree et al., 1998; Bron et al., 1999). However, the apparent tetrameric nature of GlpF reported here is consistent with the model proposed by Voegele et al. (1993) in which GlpF interacts with the tetrameric glycerol kinase GlpF to stimulate glycerol phosphorylation.
In our hands, GlpF crystallized under similar conditions to AQP1 (Walz et al., 1995), but at pH 8.5 instead of pH 6 and at much lower MgCl2 concentration. The P4 unit cell houses eight GlpF monomers, has a side length of 104 Å and is thus 8% larger than the P4212 unit cell of AQP1 (96 Å; Walz et al., 1994). Electron diffraction experiments, yielding diffraction maxima to 3.6 Å, document the crystalline quality of these double‐layered sheets and suggest their suitability for structure determination at atomic resolution. This is further supported by the 3.7 Å projection map obtained by processing images of frozen–hydrated crystals recorded at liquid helium temperature.
It is interesting to compare the first high‐resolution projection structure of the archetypal member of the GLP subcluster with that of AQP1, the first aquaporin structurally analysed to high resolution (Figure 4; Cheng et al., 1997; Li et al., 1997; Walz et al., 1997; Mitsuoka et al., 1999). The density maxima marked in the AQP1 map (Figure 4B) appear to be slightly shifted and of different amplitude to the maxima in the GlpF map (Figure 4A). They are related to six tilted helices that surround a central structure produced by loops B and E (Mitsuoka et al., 1999). According to the hourglass model (Jung et al., 1994), these loops are close to the channel (marked by an X in Figure 4). Thus, differences between GlpF and AQP1 in this region are of particular interest. While the projection map of AQP1 shows a weak density at this position, GlpF seems to have a much larger hole with no inner structure discernible in the projection map.
The striking similarity of the GlpF tetramers of opposing handedness in Figure 3 suggests that that their 4‐fold axes were parallel to the optical axis of the microscope, similar to the situation of the AQP1 crystals. Hence, tilting of tetramers can be excluded as a major reason for the differences observed. Another possibility could be that the pore of GlpF is parallel to the 4‐fold axis of the tetramer, whereas the AQP1 channel is tilted. Nevertheless, it is more compelling to speculate that the apparently larger pore is required for the diffusion of glycerol and other small anionic solutes (Maurel et al., 1994). How such a pore prevents the passage of protons and other ions is a mystery still to be enlightened. The highly ordered two‐dimensional crystals presented here are a first essential step towards this goal.
A construct of GlpF with a His10 tag at the N‐terminus was overexpressed in E. coli, solubilized in OG and isolated by Ni‐chelation chromatography (Borgnia et al., 1999; Ringler et al., 1999). The solubilized protein was stable in 3% OG at 4°C for weeks and could be concentrated by centrifugation to 12 mg/ml without precipitation.
The GlpF was dialysed in a continuous flow dialysis machine (Jap et al., 1992) against a buffer containing 10 mM tricine pH 8.5, 5 mM MgCl2, 100 mM NaCl, 10 mM dithiothreitol and E. coli lipid (Avanti Polar lipids, Inc., USA) with a lipid to protein ratio (LPR) ranging from 0.6 to 1.4 (w/w) using the following temperature profile: 25°C for 12 h, a linear increase to 40°C over the next 12 h, 40°C for 24 h and a linear decrease to 25°C over 6 h. Protein concentrations between 1 and 3 mg/ml were used.
STEM mass determination.
Purified GlpF (50 μg/ml) was adsorbed to thin carbon films, washed extensively with quartz distilled water, freeze‐dried, and imaged with a Vacuum Generators HB5 STEM at doses of ∼3.3 e/Å2 (Müller and Engel, 1998). The IMPSYS software package was used to extract mass values of all particles discernible on the dark‐field images. These values were corrected for the experimentally measured dose‐associated mass loss (Müller et al., 1992), distributed in a histogram and approximated by Gaussian curve fitting with a Marquardt algorithm (Bevington, 1969). The mass values were also analysed using a commercial program (Igor pro, www.wavemetrics.com) for Histogram calculations and Gauss peak fitting by a Levenberg–Marquardt algorithm. The total experimental error was calculated as the standard error of the mean, plus 5% of the measured particle mass to account for the absolute calibration uncertainty.
Transmission electron microscopy.
For single particle analysis, the protein (50 μg/ml) sample was absorbed for 5 s onto glow discharged carbon film‐coated copper grids, washed three times in distilled water for 5 s and stained with 0.75% uranyl formate. Images were recorded under low‐dose conditions with a Hitachi H‐8000 TEM at 200 kV and nominal magnification of 50 000× on Kodak SO‐163 films.
For cryoelectron microscopy, grids were prepared by the back‐injection technique (Hirai et al., 1999) with 1% trehalose, quickly frozen in liquid ethane and transferred with a Gatan 626 cryo holder into a Hitachi H‐8000 TEM. Images were recorded at 90 K at 200 kV and 50 000× nominal magnification on Kodak SO‐163 film, using a homemade spot‐scan and low‐dose set‐up programmed on the Tietz CCD remote control system (Tietz Video & Imaging Processing System, Gauting, Germany). The dose per negative was 5 e/Å2. Electron diffraction patterns of vitrified crystals were acquired by the 1k × 1k Tietz CCD camera. Alternatively, images were recorded with a JEOL 3000SFF (MPI for Biophysics, Frankfurt) in spot‐scan mode and operated at 300 kV, 4.2 K and 70 000× nominal magnification. The dose for recording an image was 17 e/Å2.
Micrographs of negatively stained samples were digitized using a Leafscan‐45 scanner (Leaf Systems, Inc., Westborough, MA) at 4 Å/pixel at the specimen level. Single particles were picked automatically, windowed, aligned with a reference‐free alignment procedure (Penczek et al., 1992; www.wadsworth.org/spider_doc/spider/docs/reffreealign.html) and classified using the SPIDER software (Frank et al., 1996).
Cryo‐TEM negatives of two‐dimensional crystals were digitized using a Zeiss Phodis scanner (Carl Zeiss, Oberkochen, Germany) at 1 Å/pixel on the specimen level. Image processing was performed with the MRC software package (Henderson et al., 1986, 1990). In a first run, images were unbent three times using the Fourier‐filtered images themselves as a reference. The merged amplitudes and phases from eight images were then used with the program MAKETRAN to create two synthetic references, applying two different negative temperature factors. All images were now unbent using the reference with the smaller negative temperature factor. The unbent images were then refined by unbending them a second time, using the reference with the stronger negative temperature factor, this time allowing only very small (5 pixels/unit–cell) displacements of the units–cells.
To compare the GlpF and AQP1 monomers, the SEMPER software was used. First, the non‐crystalline symmetry in the GlpF crystals was applied (mirroring and rotational alignment of the two adjacent tetramers) to approach the P4212 symmetry used for AQP1. The monomers of GlpF and AQP1 were then interpolated to the same scale, aligned translationally and rotationally, and displayed with identical grey value level ranges and contours to aid comparison.
We thank Deryck Mills for the introduction and expert support with the Jeol 3000SFF microscope and Vinzenz Unger for advice on the use of the MAKETRAN routine. We are indebted to Yoshinori Fujiyoshi, Kaoru Mitsuoka and Thomas Walz for providing the AQP1 data, and Lorenz Hasler and Shirley Müller for their expert help. The work was supported by the Swiss National Foundation for Scientific Research (NF grant No. 4036‐44062 to A.E.), the M.E. Müller Foundation of Switzerland and the National Institutes of Health (to P.A.).
- Copyright © 2000 European Molecular Biology Organization