To understand the impact of the isoprenyl moiety of RAC1 on GDI1 binding, we compared the biochemical properties of geranylgeranylated RAC1 (RAC1^GG) from insect cells and non-isoprenylated RAC1 full-length (RAC1^FL) from Escherichia coli. Previous mass spectrometric analysis and liposome sedimentation of intact RAC1^GG, compared to RAC1^FL, have revealed a fully modified RAC1^GG by geranylgeranylation [34]. Unless otherwise stated, the unmodified RAC1 purified from E. coli is designated as RAC1^FL, the modified RAC1 purified from insect cells as RAC1^GG, and the unmutated (wild-type) RAC1 as RAC1^WT in cell-based experiments.

We determined the GDI association rates with both RAC1^GG and RAC1^FL using a stopped-flow fluorometric assay. Figure 1A shows a rapid decrease in fluorescence after mixing GDI1 with the RAC1 proteins, which is directly related to the association reaction between the RAC1-GDI1 pairs. Observed rate constants (k_obs) obtained by a single exponential fitting increased linearly as a function of the GDI1 concentrations (Figure 1B), and yielded similar association rate constants (k_on) for both RAC1^GG and RAC1^FL. The dissociation of the GDI1 from mdGDP-bound RAC1 proteins was measured in a displacement experiment. Observed single exponential fluorescence increase yielded respective dissociation rate constants (k_off), which differ only 3-fold (Figure 1C). Notably, the RAC1^GG-GDI1 interaction showed a biphasic behavior (double exponential kinetics); particularly for the off-rate, an initial rapid increase in fluorescence was followed by a slow plateau phase, which can be attributed to the GG moiety of RAC1^GG. Calculated dissociation constants (K_d) from the ratio of the k_off and k_on values (Figure 1D) unexpectedly revealed only a 7-fold higher affinity for RAC1^GG vs. RAC1^FL.

Instead of the simple exponential decrease seen with GAP, there is an initial rapid increase in fluorescence followed by a decrease at a rate similar to that seen at high GAP concentrations. The first phase of the reaction is concentration-dependent, whereas the second is less obviously so. This suggests that an initial association between the proteins leads to an increase in fluorescence, which then decreases on mant-GTP hydrolysis (and consequent dissociation of the proteins) to a level below that of the starting level, in agreement with the observation that the fluorescence of Ras‚mant-GDP is lower than that of Ras‚mant-GTP. This interpretation means that the mechanism cannot be treated as a very rapid initial equilibration followed by a relatively slow cleavage step, which is the underlying assumption in the analysis of the data obtained with GAP.

This result clearly contradicts existing models which have suggested that the isoprenyl moiety of the RHO GTPases contributes to orders of magnitude higher binding affinity for GDI as compared to unmodified RHO GTPases [41,42]. The studies have determined K_d values of 0.4 nM and 5 pM for the interaction of GDI1 with prenylated RAC1 and RHOA, respectively, as compared to the 180 nanomolar K_d value determined in our study. The reason for the extraordinarily large differences between the binding affinities obviously lies in the proteins and the type of fluorescence reporter groups used in each case. On the one hand, we and Tnimov et al. applied native GDI1 while Newcombe et al. [41] used a coumarin-labeled GDI1 at position cysteine 79, which is actually buried and located in the back side of the C-terminal geranylgeranyl binding domain (GGBD); this modification at this position can drastically alter the confirmation and consequently its biochemical properties of GDI1. On the other hand, native RAC1^GG purified from insect cells was used by Newcombe et al. as well as in this study, Tnimov et al. [42] used a cell-free modification approach of RHOA^FL and purified geranylgeranyl transferase I and nitrobenzoxadiazole (NBD)-labeled geranyl-pyrophosphate [1]. In this way, the nature of the isoprenyl moiety and the absence of further posttranslational modifications by specific proteolytic removal of the terminal three residues, and carboxylmethylation of prenylted cysteine residue, are largely different from native RHOA^GG.

The above findings prompted us to unambiguously challenge the paradigm of RAC1 regulation by GDI1. To this end, we investigated the interactions of the three GDI paralogs, GDI1, GDI2 and GDI3, with various non-isoprenylated members of the RHO GTPase family. GDI3 was purified as an N-terminal truncated variant lacking the amphipathic helix (amino acids or aa 1-20). Kinetics of association of the GDIs (2 µM) with 0.2 µM mdGDP-bound RHO GTPase was monitored under the same conditions as described above for RAC1. Calculated k_obs values for each measurement were plotted as bar charts in Figure 2A, which clearly show that all three GDIs associated with RAC1, RAC3, RHOG and RHOA under the experimental conditions but apparently not with RAC2, CDC42, TC10, TCL, RHOB, RHOC, RHOD and RIF. GDI2 and GDI3 showed significantly faster association with RAC3 as compared to GDI1. Most remarkably, unlike GDI1 and GDI3, GDI2 was able to bind RAC2. These results clearly indicate for the very first time that GDIs can discriminate between the RHO GTPases by interacting with some but not all the RHO proteins.

To examine binding properties, the respective association rate constants (k_on) and the dissociation rate constants (k_off) were determined for the interaction of GDI1 with RAC1, RAC3, RHOG and RHOA, and GDI2 with RAC2 under conditions described above (Figure 1A–D). All kinetic parameters along with calculated dissociation constants (K_d) are summarized in Figure 2B. The data are very similar for the GDI1 interaction with RAC1, RAC3, RHOG and RHOA with K_d values between 0.9 and 3.2 µM. However, GDI2 interactions exhibited similar rate constants for RAC1 and RAC3, which were significantly different from that of GDI1 (Figure 2B). GDI2 exhibited more than 10-fold faster k_on values and up to 4-fold slower k_off values. GDI2-RAC2 interaction was characterized by a much slower rate of association as compared to RAC1 and RAC3, resulting in a binding affinity of 18.8 µM under the given experimental conditions (Figure 2B).

The biochemical characterization together with structural studies has shown that the RAC paralogs exhibit different properties concerning ligand– and protein–protein interactions [43]. Whereas RAC1 and RAC3 behave almost identically, RAC2 revealed a 25-fold lower nucleotide affinity because of a decreased nucleotide association rate, a slightly higher PAK1 (p21 activated kinase-1, a downstream effector for RAC1 and CDC42) binding affinity, and a significant increase in GEF-catalyzed nucleotide dissociation. These aberrant properties most likely are the consequence of different conformational flexibilities in the switch I region [44].

A comparison of available structures of the RAC1-GDI1 and RAC2-GDI2 complexes revealed a rather high sequence similarity [35,36]. An inspection of full interaction matrix revealed very few amino acid deviations within the RAC G domains (Y/F89) and GDI paralogs (A/P/G31, E/K/R53, A/T/V54). However, Y98 in RAC1 and RAC3 undergoes contacts with H23 and V25 of GDI1 which may not be achieved by F89 in RAC2, that may only contact V25 but not S23. Among the deviations in GDIs, the loop containing A/G31 in GDI1 and GDI3 is in close vicinity of D65/R66 of RAC paralogs, which may, in the case of P31 in RAC2, adopt a different orientation and thereby influence RAC2-GDI2 binding. Residues in SWBD, such as E53 and A54 (GDI1 numbering), appear to be critical for RHO GTPase-GDI interaction. The double mutation of L55/L56 to serines in GDI1 has been shown to drastically decrease its affinity for RAC1 [45].

A sequence analysis of HVRs of GDI1 associating RHO GTPases (‘binders’) versus those with no observed GDI1 association (‘non-binders’) showed clear differences in both numbers and relative positions of positively charged residues (Figure 3A). To examine the impact of HVR on the RHO GTPase-GDI1 interaction, we measured the kinetics of GDI1 association with different HVR variants of RAC1 and RAC2. Remarkably, a loss of RAC1 association was observed with a C-terminal truncated variant lacking HVR-CAAX (RAC1^ΔC10) as well as KRKRK-to-EEEEE (RAC1^5xE; charge-reversal variant) and KRKRK-to-QQKRA (RAC1-to-RAC2 or RAC1^RAC2 variant). In contrast, a gain of GDI association with RAC2 was observed with QQKRA-to-KRKRK (RAC2-to-RAC1 or RAC2^RAC1 variant; Figure 3B). These findings were verified by fluorescence polarization experiments and obtained data summarized in Figure 3C revealed that (1) RAC1^ΔC10 yet bound GDI1 with a 26-fold lower affinity as compared to RAC1, (2) RAC1^5xE, binding to GDI1 was yet observed with a very low affinity while this was not possible for RAC1^RAC2, and (3) RAC2^RAC1 did, in contrast to RAC2, bind GDI1 with an almost similar affinity as determined for RAC1. Taken together aided HVR alteration can completely abolish GDI1 association with RAC1 and revert GDI1 association with RAC2.

The results clearly demonstrate the critical role of the polybasic motif of RAC1 in determining GDI1 binding. It seems that both an increase of overall positive charge in HVR of RAC2 and the distance of the basic residues from the geranylgeranyl site strongly reinforce GDI1 binding affinity. We hypothesize that the GDI1 selectively binds RAC1 polybasic motif to pull the GG moiety from the plasma membrane and direct it into the hydrophobic cavity of its GGBD.

The relative position and the order of the basic residues in HVR seem to contribute to the formation of an electrostatic network (Figure 2E) that may significantly stabilize GDI1 interaction with, for example, RAC1 and RAC3 but not RAC2. Synthetic peptides containing the polybasic motifs of RAC1 (aa 178–188), but not RAC2 (178–188), have been shown to inhibit NADPH oxidase activity in a RAC1-dependent system, and interfere with the translocation of RAC1 proteins to the plasma membrane [47]. While the geranylgeranyl moiety mediates membrane anchorage, the polybasic motif of RAC1 interacts with plasma membrane phosphoinositides and stabilizes its proper orientation [48].

Considering the amino acid sequence identity of the G domain on the one hand (Figure 2C), and the sequence similarities among the hypervariable regions on the other (Figure 3A), it is striking that seven out twelve RHO family GTPases do not interact with GDI1 (Figure 2A). For example, the HVRs of RHOA versus RHOC and RAC3 versus CDC42 look very similar, and yet GDI1 binds one but not the other, under the same experimental condition in this study. An in-depth analysis of the RAC1 and RAC2 variants revealed that the HVR polybasic motif dictates GDI1 binding (Figure 3B,C). The number of positively charged residues appears not to be a binding determining factor since the HVR polybasic motif of the nonbinder RHOC has a higher positive net charge as compared to the binder RHOA (Figure 3A). This is also true if comparing the binder RAC3 with the nonbinder CDC42, which exhibits a much larger amino acid variability with their HVRs. Thus, we assume at this stage that, not the number of positive charges but rather the position of the basic residues relative to the C-terminal cysteine determines the bilateral binding selectivity of the HVR polybasic motif by the negatively charged residues of both GGBD and NTA (Figure 2D,E). This may be the reason for TC10 and RHOD not to interact with GDI1 related to the distance of the polybasic motif to the C-terminal Cysteine (Figure 3A). Notably, Gosser et al. has reported a binding affinity of 1.6 nM between unmodified CDC42 and GDI1, which significantly impaired upon N-terminal deletions of GDI [49]. This value is three orders of magnitude lower than the K_d values we have obtained from kinetic and equilibrium measurements (Figure 1). Gosser et al. have used in addition to mGDP-bound CDC42 also a fluorescein-conjugated GDI at position cysteine 79, which is actually buried and located in the back side of the GGBD; this modification at this position can drastically alter the confirmation and consequently int biochemical properties of GDI1. Moreover, residues next to the positively charged residues within the HVR seem to play a role, too. Glutamines in RAC2 and RIF seem to be deleterious for the interaction with GDI1. Glutamate 181 in CDC42 may exert electrostatic repulsive effects on the GDI binding (Figure 3A). Serine 185 is a phosphorylation site on CDC42, regulating its translocation to the cytosol by favoring its interaction with GDI1 [50]. Future studies will shed light on these issues.

GDI1 function appears to be driven and controlled by electrostatic forces, that attract the polybasic motif of RAC1. To examine this mechanism, we generated different deletion and charge reversal variants of GDI1 and measured both their binding capabilities to RAC1^GG and RAC1^FL, as well as their functional properties to displace RAC1^GG from PIP-enriched synthetic liposomes. GDI1^E121K, which apparently does not contact RAC1 HVR but the switch 2 region (Figure 2E), was used as a control. Kinetic analysis showed that most GDI variants are disabled in associating with RAC1 (Figure 4A). Substitutions of D140, E163 and E164 for lysines or deletion of the N-terminal and very C-terminal amino acids significantly impaired GDI1 binding to isoprenylated and non-isoprenylated RAC1, as compared to GDI1^WT. The most drastic effects were observed with 25 and 58 amino acids deleted at the N-terminus (∆N25 and ∆N58) on the one hand, and double (E163K and E164K or 2E > 2K) and triple (E140K, E163K and E164K or 3E > 3K) mutations, on the other, which did not bind to the RAC proteins under the experimental conditions. Fluorescence polarization measurements verified that most GDI1 variants were yet able to bind RAC1^FL, but with up to 145-fold lower binding affinities compared to GDI1^WT (Figure 4B). RAC1 binding was completely abolished in the case of ∆N58 and 3E > 3K variants. GDI1^ΔN58 not only lacks the very N-terminal acidic resides, that are integral elements of the electrostatic pincer function, but also the switch binding domain (SWBD), which forms multiple contacts with the RAC1 switch regions (Figure 2C,D). GDI1^3E>3K most likely creates intermolecular charge repulsion towards positively changed HVR.

To analyze the function of the GDI variants in extracting RAC1^GG from the liposomes, we performed a liposome sedimentation assay established previously [34]. Therefore, we mixed PIP-enriched liposomes (200 µm in diameter) with GDP-bound RAC1^GG and isolated liposome-bound RAC1^GG from the pellet fraction (Figure 4D, first lane) after sedimentation. Next, 1 µM of RAC1^GG•GDP bound to liposomes was mixed with 2 µM of GDI1 WT and its variants (2 µM, respectively) to measure their ability to displace RAC1^GG from the liposomes. Figure 4D shows that GDI1^WT quantitatively displaced RAC1^GG from the liposomes. In contrast, the majority of the GDI1 variants revealed a significant reduction in their activities, consistently with the kinetic and equilibrium measurements (Figure 4A–C). Particularly, GDI1^ΔN58 and GDI1^3E>3K were completely disabled in binding and extracting RAC1^GG from the liposomes, strongly supporting the notion that GDI1 supplies an electrostatic pincer to grasp RAC1 and pull it out from the plasma membrane. Moreover, GDI1^E121K and GDI1^2E>2K remained partially associated with liposomes and were sedimented in the pellet fraction (Figure 4D). The fact that these GDI1 variants were able to bind to RAC1^GG on the liposomes but could not extract it from the liposomes strongly suggests that an electrostatics-guided binding and extraction mechanism is impaired unilaterally. We think that binding of the GDI1 GGBD with the RAC1 HVR takes it away from its membrane association and release additional basic residues on HVR for the interaction with the negatively charged residues of the GDI1 NTA (the so-called electrostatic pincer; Figure 5). Loss of the GGBD-HVR interaction at this step obviously disabled the GDI1 variants (GDI1^E121K and GDI1^2E>2K), that is associated with RAC1^GG on the liposomes, to extract RAC1^GG from the membrane.