Modeling of complexes of GAPGraf with cognate GTPases
Efforts to crystallize the GAP domain of Graf, i.e. GAPGraf, in complex with either RhoA or Cdc42 were unsuccessful. In order to gain insights into the structural features of the GAPGraf-GTPase interactions, we used molecular modeling. As a starting point, we used the p50RhoGAP-RhoA and p50RhoGAP-Cdc42 crystal structures (1TX4 and 1GRN), in which we replaced p50RhoGAP with the crystallographic model of GAPGraf (1F7C). We then applied molecular dynamics to optimize the structure of the complexes, as described in the Methods section.
As expected, the simulated complexes retain most of the key stereochemical features of the p50RhoGAP complexes, especially within the switch regions of the GTPases. GAPGraf
does not undergo any significant conformational changes during the simulation on going from the isolated structure to one bound to either of the two GTPases. This was also expected based on the previous studies of p50RhoGAP in isolation and in complexes with cognate GTPases (Nassar et al., 1998
; Rittinger et al., 1997b
illustrates the nature of the contacts between the GAPGraf domain and the two GTPases in the respective models. In general terms, GAPGraf interacts with the GTPase via a concave large surface patch (997 Å2 and 1019 Å2 for RhoA and Cdc42 complexes, respectively) generated by the solvent exposed faces of helices α3, α7 and α8. The arginine finger, which includes a loop connecting α1 and α2, is located at the edge of this patch which binds an extensive surface on the nucleotide-binding face of the GTPase. The arginines-finger loop of GAPGraf, including the catalytic Arg220, is inserted in a canonical fashion allowing for a direct interaction of the latter with the catalytic Gln63 of RhoA (or Gln61 in Cdc42).
Figure 1 (A) The binding interface of RhoA- GAPGraf complex. The switch regions I (residues 34-42) and II (residues 62-79) are colored in red and blue, respectively. Mutation sites in the GAPGraf are marked in magenta for patch I, brown for patch II and dark green (more ...)
It is instructive to compare the surface that is engaged by GAPGraf
on RhoA, to that involved in the interaction with the DH (Dbl-homology) domain of a RhoA-specific nucleotide exchange factor (GEF), as seen, for example, in the complex with PDZRhoGEF (Derewenda et al., 2004
) (). The interface buried between RhoA and the DH domain is significantly larger (1514 Å2
) that the one engaged by GAPGraf
). Moreover, the DH domain, which is representative of the way that DH domains bind GTPases in other complexes (Kristelly et al., 2004
; Rossman et al., 2005; Snyder et al., 2002
) binds to a surface which is mostly (~70%) conserved between RhoA, Cdc42 and Rac1, but which also contains variable and highly variable residues accounting for selectivity. In contrast, ~90% of the surface engaged by the GAP domain is made up of highly conserved residues, with only ~10% moderately variable amino acids. This is consistent with, and to some degree rationalizes the observed tendency of GAP domains to exhibit a higher level of promiscuity in their in vitro
interactions with GTPases. Thus, in vivo
selectivity is probably primarily controlled by spatial targeting through other domains.
Figure 2 Conservation of residues within Rho family GTPases and involvement in the interaction with GAP and GEF regulators based on RhoA (GI:20379114), Rac1 (GI:8574038) and Cdc42 (GI:20379098). Binding interface of (A) RhoA with p50-RhoGAP (pdb:1TX4) (Rittinger (more ...)
Within the complex-forming surface on GAPGraf, we identified three specific patches which are important for binding and/or catalysis, as judged by the models of the complexes with RhoA and Cdc42. The first such epitope, i.e. patch I, includes the catalytic Arg220 located within the arginines-finger and two residues invariant among RhoGAPs which appear to be essential for the structural integrity of the arginines-finger, i.e. Lys262 and Arg266. The catalytic Arg220 is suitably positioned to interact with Gln63 (RhoA numbering), while the positively charged Lys262 and Arg266, located on the solvent exposed face of helix α3, interact with a negatively charged fragment of Switch II containing Glu64 and Asp65, which is highly conserved among RhoA, Rac1 and Cdc42.
Patch II includes several residues including Val224, Asn225 and the adjacent Glu253. These amino acids interact with a mostly negatively charged region of RhoA including Asp90, Glu93 and Glu97. Of these, only Glu93 is conserved in Rac1, while Glu93 and Glu97 (but not Asp90) are conserved in Cdc42. We previously showed that the cohesive interaction between Asp225 of GAPGraf
and Glu97 of Cdc42 provides a significant contribution to the observed RhoA/Cdc42 selectivity of GAPGraf
(Longenecker et al., 2000
), and so the entire patch II may play a role in selectivity in contrast to patch I.
Finally, a hydrophobic cradle (Patch III) formed by several solvent-exposed residues from helices 7 and 8, and in particular Val338 and Ile358, interacts with several conserved residues within the switch I region of RhoA/Cdc42, and specifically with the conserved Pro34 and Val36. Like patch I, this contact involves highly conserved amino acids and is less likely to be involved in specificity control.
Binding and enhancement of GTP hydrolysis on cognate GTPases by wild-type GAPGraf
It is well established that GAP domains bind target RhoGTPases only weakly when the latter are in the biologically inactive, GDP-bound state, and that the affinity increases dramatically for the transition state in which GTP transiently contains pentavalent phosphorus (Graham et al., 1999
). This transition state can be effectively mimicked by adding fluoride ions in the presence of Al3+
to a GDP-bound GTPase (Ahmadian et al., 1997a
; Hoffman et al., 1998
). In order to evaluate how wild type GAPGraf
binds to RhoA and Cdc42 in both the GTP-bound state of the GTPase and to the transition complex, we used GTPases loaded with a slowly hydrolysable GTP analogue – GMP-PNP, and GDP with AlFx, respectively. The thermodynamics of binding were monitored using isothermal titration calorimetry (ITC) (Jelesarov and Bosshard, 1999
). The results of ITC experiments are shown in . The association constants (Ka
) for both GMP-PNP-Mg2+
-bound RhoA and Cdc42 GTPases were 33.8 × 103
M and 26.1 × 103
M respectively. For the transition-state mimic these values increased 90 and 16 fold, respectively, suggesting preference for RhoA. This is in contrast to p50RhoGAP, where there is no preference in binding to two states of GTPase (Graham et al., 1999
). These results reaffirm that many GAP-GTPase interactions responsible for the specificity and strength of the functional complex are formed in the transition state (Fersht, 1990
Table 1 The thermodynamic and kinetic parameters of GAPGraf WT interaction with RhoA and Cdc42. The ITC measurements were performed using GMP-PNP-Mg2+ and GDP·Mg2+-loaded Cdc42 and RhoA in the presence of AlFx. Kinetic parameters of GAPGraf stimulated (more ...)
The interactions of GAPGraf
with both Cdc42 and RhoA are strongly endothermic and the data could be fitted to a single-site binding model with calculated stoichiometry of 1.0 ± 0.15%. Large, unfavorable positive enthalpy changes (ΔHa
) are in both cases compensated by positive (favorable) entropy changes (ΔSa
) which effectively drive the reactions. Unfavorable values of ΔHa
typically reflect the dehydration effect of polar groups associated with hydrogen bonding accompanying formation of a complex (Loladze et al., 2002
). The observed positive change in entropy suggests that the reaction is driven by the release of water molecules from the proteins' surfaces. To further elucidate the nature of the driving forces during the formation of the GAPGraf
-GTPase complex, we determined the change in heat capacity (ΔCp
) of association, which directly characterizes the nature of protein-protein interaction; a negative ΔCp
value indicates that a complex buries substantial hydrophobic surfaces. shows the temperature dependence of calorimetric ΔHa
plot for Cdc42- GAPGraf
and RhoA- GAPGraf
. The linear dependence of ΔHa
over the studied temperature range yields ΔCp
values of -2.14 and -1.05 kJ/mol·K for Cdc42 and RhoA, respectively. To assess, if the GTPase- GAPGraf
models are consistent with experimental results, we used an empirical relationship between the heat capacity and the change in the accessible surface area (ΔASA) (Murphy and Freire, 1992
), and we used the models to calculate theoretical ΔCp
values. We obtained -1.7 and -1.8 kJ/mol·K for Cdc42 and RhoA, respectively. These values are close to experimental data. The discrepancy between the measured and calculated ΔCp
values probably reflects unaccounted water molecules in the RhoA- GAPGraf
and Cdc42- GAPGraf
binding interface, as noted in similar studies Ras interactions with effectors (Rudolph et al., 2001
Figure 3 Representative calorimetric titrations of Cdc42 with GAPGraf. (A) panel shows raw heat data corrected for baseline drift obtained from 27 consecutive 7 μl injections of 1.72 mM I358A mutant at 260 second intervals into the sample cell (1.05 mL) (more ...)
In parallel to calorimetric titrations, we investigated the catalytic activity of wild-type GAPGraf
on Cdc42 and RhoA using the MESG/PNP assay. Reactions were carried out under single turnover conditions and the release of free phosphate group was monitored as a function of time (Zhang and Zheng, 1998
). This assay allowed us to determine both the KM
parameters for the hydrolysis reaction (, ). The KM
values of wild-type GAPGraf
-stimulated GTP hydrolysis for Cdc42 were 2.5 times lower and higher, respectively, than for RhoA, resulting in the overall catalytic efficiency six times higher on Cdc42 than on RhoA. This is consistent with the calorimetric studies which showed that the Cdc42 transition-state mimic is favored over the RhoA mimic as the binding partner.
Table 2 Table 2a. The thermodynamic and kinetic parameters of GAPGraf mutants interaction with Cdc42. The ITC measurements were performed in 25 mM HEPES, 1mM DTT, 5 mM MgCl2, 25 mM NaF, 1 mM AlCl3, pH 7.8 at 20°C using GMP-PNP-Mg2+ and GDP·Mg (more ...)
We have also re-evaluated the activity of GAPGraf against Rac1. As expected, the association constant for Rac1·GDP·Mg2+·AlFx was over twenty times lower than for RhoA and a hundred times lower than for Cdc42. The MESG/PNP assay revealed only marginal activity on Rac1 (data not shown)
Design of mutants probing structure/function relationships in GAPGraf
Based on the model of the RhoA(Cdc42)/GAPGraf complex, we designed fifteen point mutants to probe how the different epitopes on GAPGraf contribute to the interaction with the cognate GTPase and to catalysis.
Within patch I (i.e. the arginine finger and its adjacent loop) we targeted the strictly conserved Arg220, Lys262 and Arg266. Arg220 was mutated to Ala, Ser, Lys and Glu, to assess the effect of loss of side chain, retention of positive charge and charge reversal. Lys262 was mutated to Ala, while Arg266 was mutated to both Ala and Glu to assess the impact of loss of side chain and charge reversal. Within patch II we targeted Val224, Asn225 and the adjacent Glu253: Val224 was mutated to Glu, Asn225 to Ala and Glu253 to both Ala and Lys. Finally, within patch III we targeted both Val338, and Ile358 and we mutated Val338 to Leu and Ala, while Ile358 was mutated to Ala and to Glu.
Binding of cognate GTPases by mutant GAPGraf proteins and their catalytic properties
All mutants were assayed for their ability to bind the transition-state mimic of RhoA and Cdc42, and for their catalytic properties. In both series of experiments, using Cdc42 or RhoA as targets, we observe strong correlation between Ka, determined by calorimetry, and kcat/KM parameters, determined by MESG/PNP assay (). The mutants that conform to this paradigm exclude all those at the Arg220 position, as well as mutations of Lys262 and the R266E mutant. These data confirm that, in general terms, residues outside the arginine finger play a purely structural and non-catalytic function.
Figure 5 The correlation plot of Ka and kcat/KM parameters. A. Cdc42 B. RhoA. The Ka measurements were done in 25mM HEPES, 5mM MgCl2, 1mM DTT, 25mM NaF, 1mM AlCl3, pH 7.8 at 20°C, and kcat/KM determinations in 50mM Tris/HCl, GTP, 0,1mM MESG, 1U PNP, 5mM (more ...)
shows representative calorimetric data for the interactions of GAPGraf mutants with Cdc42. shows all determined ΔΔGa, defined as the ΔGa of the wild-type minus ΔGa of the GAPGraf mutant, for the interactions of all assayed GAPGraf mutants with both RhoA and Cdc42. For the majority of mutations, a decrease in the Gibbs association energy (ΔΔGa) is observed, with a maximum change of 8.9 kJ/mol. For some GAPGraf mutants, the decrease in association constant was so large, that accurate measurements of Ka required exceedingly high concentrations of the protein. In such cases, only the lower limit of Ka (below 1 × 104 M-1) could be estimated, corresponding to the reduction of Gibbs energy by at least 13 kJ/mol.
Comparison of the ΔΔGa values determined for the interaction of GAPGraf mutants with RhoA and Cdc42. Measurements were done in 25mM HEPES, 5mM MgCl2, 1mM DTT, 25mM NaF, 1mM AlCl3, pH 7.8 at 20°C.
lists all the numerical data of the calorimetric and catalytic assays. With one exception, all mutants showed a decrease in kcat/KM in comparison with wild-type GAPGraf. The results were similar for both RhoA and Cdc42 GTPases.
Arginine finger substitutions
Mutations of the catalytic Arg220 to Ser and Ala significantly reduced the Ka
values for the interaction with both RhoA and Cdc42, while mutants containing Lys and Glu at this position showed no detectable binding to either RhoA or Cdc42. In all above cases the catalytic activity was completely abolished. Thus, Arg220 in GAPGraf
plays a significant role in both binding the GTPase and in the protein's ability to stimulate the catalytic activity of Cdc42 and RhoA GTPases. While these data are consistent with experimental results reported for some RhoGAPs (Ahmed et al., 1994
; Hoffman et al., 1998
; Leonard et al., 1998
; Muller et al., 1997
), they are different from those reported for p50RhoGAP (Graham et al., 1999
). In the latter case the mutations of the catalytic Arg85 to Lys or Ala cause only moderate, 3-fold, decrease in the association constant with RhoA in the presence of AlFx
but result in a dramatic decrease in the catalytic activity on RhoA (Graham et al., 1999
The two conserved residues critical to the structural integrity of the arginine finger are Lys262 and Arg266. The crystal structure of the p50RhoGAP complex and our models suggest that Lys262 and Arg266 are involved in direct hydrogen bonds to Glu64 and Asp65 of RhoA, which is present in the G3 loop within the switch II region (). These residues stabilize the transition state conformation of G3 loop that bears the catalytic Gln63 (Gln61 in Cdc42). The Ka
values for the interaction of K262A/Q and R266E GAPGraf
mutants with Cdc42/RhoA were not measurable and, as expected, no activity was observed in kinetic MESG/PNP experiments (). In contrast, the R266A mutant causes only a 3- and 6-fold decrease of association constant for Cdc42 and RhoA, respectively. The effects on kcat
are of similar magnitude. These results suggest that the Lys262 side chain and its H-bonds to Glu64 and Asp65 of RhoA, as well as an H-bond to the main chain carbonyl of Tyr219, play a critical role in the formation of the catalytically competent interaction between the arginine finger and Switch II. Lys262 is equivalent to Arg903 in RasGAP334 and Lys122 in p50RhoGAP, and has been previously identified as the secondary arginine finger (Graham et al., 1999
). Loss of the side chain is as deleterious for Lys262 as charge reversal. In contrast, Arg266 appears to be less critical. The charge reversal mutant R266E is clearly disruptive, and must lead to severe electrostatic repulsion with Glu64 and Asp65, but the R266A mutant shows a milder effect, indicating that the side chain of Arg266 is electrostatically favorable, but not absolutely required for function.
Modeled complex of RhoA (electrostatic potential representation) and GAPGraf (main chain). A, B and C - structural details of Patch I, II and III, respectively. GAPGraf is shown in blue, RhoA in red and hydrogen bonds as green broken lines.
The role of residues within Patch II
The distinct, negatively charged cluster of Asp90, Glu93 and Glu97 is located on the solvent exposed face of helix 3 in RhoA. Interestingly, the distribution of charge on this surface is different in all three GTPases. In Rac1, Asp90 and Glu97 are replaced by alanines, and in Cdc42 Asp90 is replaced by a Ser. Our models show that this region comes into an intimate contact with several GAPGraf residues clustered on two adjacent loops: the loop containing the catalytic Arg220, i.e. between the N-terminal α-helix and the second α-helix, and the loop leading to helix 3. These residues are Val224, Asn225 and Glu253.
The V224K mutation was engineered to probe if a positively charged amino acid could confer a higher selectivity for RhoA. However, this mutant binds RhoA with 1.6-fold lower affinity, suggesting that unfavorable interactions are also created. A much larger drop in affinity, ~26-fold, was observed for Cdc42. This could be a result of the absence in Cdc42 of the negatively charged residue in the position analogous to Asp90, resulting in no new cohesive interactions. The binding affinities correlate well with the difference in kinetic parameters for both GTPases. The kcat/KM values for the GTP hydrolysis by RhoA in the presence of V224K mutant is only 2.3-fold lower, while for Cdc42, the effect is again more profound with kcat/KM about 12 times lower than determined for the wild-type.
The E253A/K mutants were tested to assess the consequences of side chain removal or charge reversal. Interestingly, both mutations affect the interaction with Cdc42 with 5.5- and 8-fold weaker binding, respectively, but with virtually no effect on the binding of RhoA. The effect on catalysis on Cdc42 is less significant with only about 2.7-fold decrease of kcat/KM.
The N225A was designed to assess the effect of loss of side chain. This mutation causes approximately 2-fold decrease in Ka for both RhoA and Cdc42. The effect on kcat/KM is negligible for RhoA but it produces much larger, 6.6-fold, decrease of catalytic efficiency in case of Cdc42.
The impact of mutations within Patch III
The V338A/L and I358A mutations have a moderate impact on GAPGraf ability to bind RhoA and Cdc42. Val338 is highly conserved among many RhoGAPs, although the position is often occupied by Ile. On the other hand, Ile358 is found in the variable C-terminal portion of the sequence which shows poor similarity to other GAPs, precluding credible sequence alignment. Both residues are in the center of a hydrophobic surface patch which packs between the Switch I and Switch II fragments of RhoA, and specifically Tyr66, Leu69, Val38 and Phe39 in RhoA (all these residues are conserved in Cdc42) The truncation of Ile358 to Ala resulted in 2.5-fold decrease in association constant and 4 to 5-fold reduced values of kcat/KM for the interaction with Cdc42 and RhoA. Insertion of negatively charged Glu in position of Ile358 is significantly more unfavorable, leading to about ~20-fold decrease in the Ka values for both GTPases and lack of catalytic activity. Substitution of Val338 with Ala results in a 6-fold decrease of the association constant for the interaction with both GTPases, and a moderate decrease in kcat/KM. A similar effect is observed for the V338L mutant.