Structure–function studies of the G-domain from human gem, a novel small G-protein

1. Introduction

The Ras superfamily of small G-proteins is both large and diverse. Several subfamilies have been categorized including the Rad,Gem/Kir (RGK) subfamily. This group now includes four members: Rad, Gem, Rem, and Rem2. These molecules were initially identified independently as transcriptionally regulated G-proteins [1–3] but their physiological function has remained challenging to define.

Within the Ras superfamily, the RGK subfamily is structurally distinguished by unique N and C-terminal sequence motifs, and within the G domain, by changes in normally highly conserved amino acids that are involved in nucleotide binding and GTP hydrolysis. Significant differences in Gem relative to Ras include the absence of an equivalent to T35, which stabilizes Mg-GDP/GTP binding, and a loss of the G3 motif, DXXG, which contributes to nucleotide binding and GTPase catalysis. To date, the biochemical functions identified for the N and C terminal extensions include binding 14-3-3 and binding Ca²⁺/calmodulin (CaM) [4]. These unique features raise questions regarding the subfamily’s enzymatic activity, directly related to its function as a molecular switch.

Recent cell biologic and biochemical characterization of different subfamily members has forged progress in revealing functional information (reviewed in [4]). In particular, two independent effector pathways have been defined for Gem, namely regulation of voltage-dependent calcium channel (VDCC) signaling and cytoskeletal organization as mediated through ROK (also known as ROCK/Rhokinase). In order to better understand the action of these molecules, we have initiated a structure–function study of human Gem. We chose first to characterize the three-dimensional structure and biochemical activity of Gem’s G-domain as a starting point, thereby providing essential basic biochemical information. Thus, we report here a 2.4 Å crystallographic structural determination and biochemical studies. On the basis of this data, we have also engineered several mutations that have been tested in a cell-based functional context.

2. Materials and methods

Subcloning, expression, purification, crystallization, data collection, and structure determination are described in detail in supplementary material available online. Gem G-domain was bacterially overexpressed, purified with the help of a removable histidine tag, and subsequently crystallized. Structure determination was by single anomalous diffraction using a selenomethionine derivative.

The reversed-phase HPLC nucleotide identification assay was performed as described [5]. GTP hydrolysis assays were performed as described [6]. Gem-dependent ROK-mediated cytoskeletal rearrangements were assayed using transient co-transfection of N1E-115 neuroblastoma cells as described [7]. Co-immunoprecipitation of Gem and Flag-tagged CaVβ2A or 1B was performed using lysates from transiently transfected COS7 cells precipitated with anti-FLAG antibodies and blotted with anti-Gem polyclonal antibodies.

3. Results and discussion

Using our purified Gem G-domain prepared with a non-hydrolyzable GTP analog, diffracting crystals were obtained. Data measured on a home X-ray source enabled molecular replacement calculations. The solution gave reasonable electron density maps in regions of the central β-sheet but poor density in other locations. Rounds of model building improved the maps and indicated regions of significant divergence from the starting model. In order to validate our structure and obtain independent phasing, selenomethionine protein was prepared and a single wavelength anomalous diffraction experiment performed. The experimentally phased map clearly confirmed and completed regions of our model (Table 1). Moreover, it unambiguously confirmed that in place of the expected GTP analog, GDP was found. This finding raised questions regarding the conditions of nucleotide exchange, binding and hydrolysis for the Gem G-domain. Further biochemical characterization described below leads us to conclude that our attempt to load Gem G-domain with β,γ-imido-GTP (GMPPNP) in preparation for crystallization was not successful.

The refined structure comprises residues 72–243, with breaks at residues 99–102 and 137–138. The G-domain takes the canonical Ras fold with the central β-sheet of six strands and five surrounding α-helices. A superposition with Ras-GDP indicates a RMS deviation of 1.1 Å for 146 Cα atoms. There are notable differences between the Ras family of structures and Gem. The most outstanding structural variation is the region of switch I (Fig. 1A). In contrast to Ras, whose switch I covers the bound nucleotide, in Gem the loop emerging from α1 helix changes trajectory and runs closer to the surface of the protein, completely leaving the nucleotide pocket exposed. The ‘‘phosphate binding’’ loop then ‘‘returns’’ to the canonical fold with a shortened (versus Ras) β2 strand. Inspection of the structure based sequence alignment (Fig. 2) shows that this sequence region linking α1 and β2, which includes G-1, has no similarity with Ras. Moreover, even within the RGK family there is little sequence conservation with varying lengths for this loop, Gem having the longest.

Another important but subtler difference in the G-domain fold between the prototypical Ras and Gem is found in the end of β3 and the loop linking it with α2. This sequence region comprises switch II. Gem’s fold only approximates the Ras trajectory, is further from the bound nucleotide, its α2 takes a somewhat different orientation and the packing between the extended chain emerging from β3 and α2 appears quite robust. A molecular lynchpin is created by the stacking of W133 and H143 (from α2), seeming to rigidify this structural module (Fig. 1C). W133 is perfectly conserved among RGK family members. In Ras, this switch II is known to encompass the γ-phosphate of the nucleotide and to be quite flexible. Sequence comparisons for this region indicate a remarkable divergence compared to Ras. The conserved G3 motif, DXXG, in Ras is replaced by DXWEX at the comparable position in RGK proteins.

Many nucleotide interactions between the G-domains and the bound nucleotide are similar (Fig. 1B and D). However, there are pointed exceptions. For example, N135, equivalent to Ras Q61, a residue critical for GTP hydrolysis as it stablilizes the nucleophilic water (see [8] for a summary of G-protein catalytic mechanisms), would be far from the γ-phosphate and its bridging O unless significant movement was induced. Such flexibility seems unlikely given the conformation of Gem’s switch II. Nevertheless, Q84 may be positioned to replace it since it is within 4 Å of the Mg²⁺ coordination shell . This residue superimposes well with Ras G12. Other important interactions are missing in Gem. In Ras, K147 and F28 interact with the guanine base by van der Waals interactions. In Gem, these are absent. Moreover, the equivalent of catalytically important Ras T35 is not positioned due to the change in switch I conformation. One interaction that does appear in Gem and is reminiscent of G_α is T90 hydrogen-bonding with the nucleotide α-phosphate. Nonetheless, Gem maintains fewer interactions with nucleotide in comparison to Ras.

A significant structural difference between Gem and the Ras family lies in the burial of the nucleotide. This difference is readily apparent from Fig. 1D. About 90 out of 600 Ų from the ligand is more solvent accessible in the Gem structure. Furthermore, the apparent electrostatic potential of Gem appears to be significantly more electronegative apposite the nucleotide phosphates and in particular where a γ-phosphate might fit than Ras (Fig. 1D). The substitution of E134 for glycine clearly causes the increased electronegativity (Fig. 1B). Together, these features underscore the uniqueness of Gem’s G-domain. In contrast to Ras, comparison of our Gem G-domain-GDP structure with the just published Rad G-domain-GDP structure [9] reveals that the superposed structures align well (0.6 Å for 144 Cα atoms). The notable structural features described above are generally shared, although switch I is not seen at all in Rad while switch II is also more disordered than in the Gem structure (supplemental Fig. 1).

Our structure solution provoked several questions: first, does Gem bind GTP?; second, if so, does it induce a conformational switch?; and third, if Gem indeed binds GTP, can it hydrolyze it? In our attempt to answer the fundamental question, i.e. that of GTP binding, we performed a comprehensive screen for conditions to empty the G-domain of nucleotide. Using an established reversed-phase HPLC assay to identify the nucleotide content of Gem, we defined conditions to remove the Mg²⁺:GDP. Protein was incubated at RT with immobilized alkaline phosphatase in the absence of Mg²⁺ (having been removed previously by desalting) and in the presence of the non-hydrolyzable GMPPNP [10]. The nucleotide state was examined by HPLC after desalting to remove the excess GMPPNP and alkaline phosphatase reaction products. After 2 h, no nucleotide was detectable. This empty Gem G-domain was then incubated with 10 mM Mg²⁺ and two-fold molar excess of GDP, GTP, or GMPPNP in independent experiments, run over a gel filtration column equilibrated with Mg²⁺ to remove unbound nucleotide and then analyzed for nucleotide bound. Wild type (WT) Gem and three substitution mutants in residues surrounding the nucleotide binding site (S89N, E134A and Q84A) were assayed (Fig. 3A). GDP was found bound to Gem while no detectable trace was found for GTP or GMPPNP. Thus, GDP could be exchanged i.e. loaded and unloaded onto Gem whereas GTP or its analog could not. This finding argues for a very marked preference of Gem for GDP versus GTP, with stable binding and slow off-rates. E134A and Q84A behaved similarly to WT, while S89N did not bind any nucleotide. This is in contrast to the homologous Ras S17N mutant that maintains a GDP bound state and acts as a dominant negative as a result of non-productive complexes with upstream activators. The GDP preference of WT Gem is entirely consistent with our crystallographic results that despite protocols to exchange the nucleotide and an extended multi-day purification, GDP was found homogeneously bound to the protein. Perhaps in a similar vein, Rem2, a RGK member, has a GDP dissociation rate markedly slower than for GTP [11]. In contrast, we have recently expressed and purified the Rad G-domain under the same conditions as for Gem. Rad was found to have a significant GTP-bound fraction (data not shown). This finding may be related to a more mobile switch II for Rad since, for example, the equivalent of E134 and N135 are not visualized in that structure but further structural studies will be necessary to test this possibility.

In parallel with our characterization of Gem’s nucleotide binding, we examined the ability of Gem to hydrolyze GTP. In enzymological assays measuring single turnover or steady-state γ-phosphate release, our isolated recombinant protein, whether loaded with GDP or empty of nucleotide produced no evidence of GTPase activity (Fig. 3B and C). The assays were executed under standard conditions for small G-proteins wherein Ras was utilized as the positive control. Further, we tested not only the Gem G-domain but the bacterially expressed full length protein alone and with Ca²⁺-CaM (supplemental Fig. 2). Neither of these forms showed hydrolytic activity.

We next asked what might be the functional consequences of perturbing six different residues that based on the structure seem to play unique roles in Gem function. Our functional test was dual-pronged: one, we examined binding of the full length Gem to the VDCC β subunit as assessed by co-immunoprecipitation in transfected cells and, two, we probed cytoskeletal reorganization, mediated by ROK with a cell morphology assay. Four mutations (S89N, T90A, E134A, N135A) surrounding the nucleotide binding region, which have been described above, were engineered. Q84, located in proximity of the GDP terminal phosphate was mutated to alanine. An additional mutation (D105K), designed to possibly lock switch I’s open conformation by formation of a novel salt bridge, was also tested. The results, summarized in Fig. 4, show that T90A mutant had reduced functional activity in both assays, but the S89N, E134A, N135A, and D105K mutants had differential effects relative to the two functional assays. The indicated mutants coprecipitated less with CaVβ2a or CaVβ1b as compared to WT Gem, while maintaining ROK-dependent activity. These data are consistent with and add to previous observations suggesting that nucleotide binding is necessary for Gem inhibition of VDCC activity but is not as stringently required for inhibition of ROK activity [12]. By contrast, Q84A interacted normally with CaVβ2a and CaVβ1b but was inactive in the ROK-dependent assay. Because S89N does not bind nucleotide, yet maintains ROK-mediated cytoskeletal effects, these data demonstrate that Gem-dependent ROK function is independent of nucleotide binding. In summary, several Gem loss of function mutants segregate with respect to Gem-dependent VDCC and ROK activities.

In conclusion, the Gem G-domain structure displays dramatically divergent structural features, particularly in the two switch regions. The protein also significantly prefers binding GDP, an unusual characteristic for G-proteins. This preference may be based in part on the electrostatics of the structure. It remains possible that the Gem N and C-terminal extensions provide exchange functionality for binding GTP, although we have no evidence for this. Perhaps other proteins might provide this activity. Reports describing functional differences dependent on nucleotide state were not performed with purified components [13]. Further investigation of Gem’s structure–function correlates promises to provide insight into this intriguing G-protein’s molecular mechanism.