The assembly of a GTPase–kinase signalling complex by a bacterial catalytic scaffold

To identify new signalling pathways targeted by bacterial pathogens, we used a human growth hormone (hGH) secretion assay⁹ to measure the ability of type III and type IV effector proteins to regulate vesicle trafficking through the general secretory pathway (Fig. 1a, b). Consecutively, each bacterial effector was tagged with enhanced green fluorescent protein (eGFP) and assessed for localization at host organelles (Fig. 1b). We noted that several type III effectors encoded by the extracellular pathogen enterohaemorrhagic E. coli (EHEC) O157:H7 inhibited host trafficking events, whereas effectors secreted by Salmonella Typhimurium, Legionella pneumophila and Bartonella henslae showed little inhibitory functions, consistent with their intracellular life cycles (Fig. 1b). In particular, the EHEC type III effector EspG blocked exocytosis of hGH through an unknown molecular mechanism (Fig. 1b). eGFP-tagged EspG localized to the cis-Golgi apparatus, where it induced severe fragmentation of the organelle (Fig. 1c, d and Supplementary Fig. 1a, b). The Golgi disruption phenotype was observed when 10 nM recombinant EspG protein was microinjected into cells to mimic the protein concentration delivered by E. coli through the type III secretion apparatus¹⁰ (Fig. 1d). In addition, EspG disrupted the recycling endosome compartment in both transfection (Supplementary Fig. 1c, d) and microinjection experiments (data not shown). Previous genetic studies have implicated EspG^11,12 and related Shigella family members¹³ in microtubule depolymerization. However, microtubules were intact in EspG microinjected cells (Fig. 1d), consistent with previous reports showing that these effectors do not disrupt cytoskeletal architectures^14,15. Thus, EspG represents a new class of bacterial signalling effector that functionally regulates cargo trafficking from membrane organelles.

We used a yeast two-hybrid screen to identify host enzymes targeted by EspG. The screen resulted in 26 positive interactions with multiple overlapping complementary DNA clones expressing two ARF GTPase family isoforms (ARF1 and ARF6) and three p21-activated kinase family members (PAK1, PAK2 and PAK3) (Fig. 1e, f). ARF GTPases function within a broad range of organelle systems, where they organize vesicle transport machinery, phospholipids and signalling molecules at membrane microdomains^16,17, whereas the PAK family of serine/threonine kinases transduce Cdc42 and Rac1 GTPase signals that establish intracellular polarity¹⁸. Direct interactions between EspG and the GTPase domain of ARF family members (Fig. 1g) and the autoinhibitory domain (AID) of PAK kinases (Fig. 1h) were shown by in vitro binding studies using purified recombinant proteins. These findings establish two EspG substrates that are consistent with its regulatory function in host protein trafficking identified here and in bacterial infection studies conducted in vivo^19,20.

Next we crystallized EspG (residues 42–398) in complex with the GTPase domain of ARF6 (residues 13–175) and solved the structure to a resolution of 2.5 Å (Supplementary Table 1). EspG buries 602 Ų of ARF6 surface area and the complex interface is mediated by a collaboration of EspG loops (loops connecting β5 and β6, β8 and α6, and β12 and β13) that specifically engage the switch I loop of ARF6 and several residues lining the guanine-nucleotide-binding pocket (Fig. 2a and Supplementary Fig. 2a, b). The conformational state and amino-acid sequence of switch I are highly conserved between ARF family members, indicating that the EspG–ARF6 structure illustrates the nature of EspG’s interaction with several ARF isoforms (Supplementary Fig. 3a, b). The importance of conserved switch I residues for binding EspG were confirmed by mutational analysis on ARF6 (Supplementary Fig. 3c).

ARF6 is GTP bound in the crystal and adopts an active-state conformation nearly identical to that of ARF6_GTPγS (ref. ²¹; Supplementary Fig. 4a). Further structural analyses revealed that switch I is inaccessible to EspG when ARF6 adopts the GDP-bound conformation (Supplementary Fig. 4a). EspG selectively bound the GTP-loaded forms of ARF1 and ARF6 but did not recognize GDP–ARF complexes (Fig. 2b). Moreover, EspG interacted with ARF6_GTP in its full-length myristoylated form, which was isolated from membrane fractions (Supplementary Fig. 4b). Thus, EspG preferentially targets the active ARF_GTP signalling molecule.

COPI coat, vesicle complex adaptors and signalling enzymes primarily associate with switch 2 and the β2/3 interswitch of ARF_GTP (refs ^22–26; Supplementary Fig. 5a). Given the frequent occurrence of this binding mode, we were surprised to find that EspG is rotated away from these common binding elements and is positioned directly over the guanine-nucleotide-binding pocket (Fig. 2a, c). Surprisingly, however, EspG does not function as a guanine nucleotide exchange factor (Supplementary Fig. 5b, c) or a GTPase-activating protein (GAP) (Fig. 2d, cyan circles). Rather, EspG is appropriately positioned to hinder binding of ARF–GAP and its catalytic access to the γ-phosphate of GTP (Fig. 2c). EspG completely abolished the GAP-stimulated GTPase hydrolysis on lipid-anchored ARF1 (Fig. 2d, diamonds), in comparison with the fast ARF–GAP reaction (Fig. 2d, blue triangles). The inhibition of GAP by EspG relied on a direct interaction between EspG and ARF because the binding-deficient mutant EspG Glu 392 Arg (characterized in Supplementary Fig. 6) had no affect on GAP-stimulated hydrolysis (Fig. 2d, open circles).

GTP hydrolysis and exchange on ARF is required for proper membrane transport functions, suggesting that EspG inhibits Golgi trafficking by blocking its guanine nucleotide cycle^17,26. Several lines of evidence support this idea. First, EspG disrupted the Golgi complex with rapid inhibitory kinetics (Fig. 2e) and a phenotype similar to the fungal toxin brefeldin A (Supplementary Fig. 7a), a potent ARF1 GTPase inhibitor that also interferes with the guanine nucleotide cycle²⁷. Second, microinjection of dominant-negative ARF protein (ARFΔN13) caused a significant delay in Golgi disruption induced by EspG (Fig. 2e, open circles). Third, EspG Glu 392 Arg, a mutant that does not interact with host substrates (Supplementary Fig. 6), had no affect on Golgi morphology or trafficking function (Fig. 2e and Supplementary Fig. 7b, c). Finally, EspG co-localized with the ARF1 effector β-COP (ref. ²⁶) on Golgi membranes (Supplementary Fig. 1b). These combined structure and cellular studies provide a mechanism for bacterial regulation of membrane trafficking: EspG prevents vesicle transport by directly inhibiting ARF guanine nucleotide turnover on host membranes.

Having established the mechanism of ARF GTPase regulation, we next explored a second possible function of EspG: regulation of PAK family kinases. The EspG-binding site on PAK2 was defined to residues 121–136, a highly conserved sequence that encodes the Iα3-helix within the kinase AID (Supplementary Fig. 8). We crystallized EspG in complex with the PAK2 Iα3-helix fragment and solved the structure to a resolution of 2.8 Å (Supplementary Table 1). EspG recognized the initial turn of the Iα3-helix whereas the remainder of the peptide adopted an extended strand conformation that lies orthogonal to the EspG six-stranded β-sheet (Fig. 3a, b). EspG buries 684 Ų of the PAK2 surface area and the binding is primarily supported by a large hydrophobic interface and hydrogen bonding by residues Asn 212 and Asn 323 of EspG (Fig. 3b). This structural interface was confirmed by a series of in vitro binding studies and kinase assays using PAK2 and EspG mutant proteins (Supplementary Fig. 9).

To determine how EspG may regulate the kinase, we compared the peptide structure from EspG–PAK2 with the structure of Iα3-helix in the autoinhibited PAK1 homodimer²⁸ (Fig. 3c). In autoinhibited PAKs, the Iα3-helix is sandwiched between the kinase domain and the AID, where it has three autoinhibitory functions: (i) it folds onto the EF- and G-helices to block substrate binding; (ii) it positions the ‘kinase inhibitory’ loop across the enzyme catalytic cleft; and (iii) it stabilizes the AID three-helix bundle maintaining PAK homodimerization (Fig. 3c). Hence, EspG binding to the Iα3-helix suggested a mechanism for PAK activation. EspG induced a (7.6 ± 2.5)-fold (n = 11) increase in PAK2 activity, a profile that is comparable to PAK stimulation by GTPγS-loaded Cdc42 ((7.8 ± 3.4)-fold, n = 3) (Fig. 3d). Notably, EHEC type III effectors Map and EspF_U showed no PAK stimulatory activity, demonstrating the signalling specificity of EspG (Fig. 3e).

To gain further insight into the mechanism of kinase activation, we first examined the details of the EspG–PAK2 interface. EspG residue Asn 212 probably initiates the kinase reaction because this residue engages the surface-accessible residues Asp 131 and Ser 132 in the autoinhibited PAK homodimer (Fig. 3b and Supplementary Fig. 10). On initial recognition, EspG displaces the Iα3-helix by reorganizing its secondary structure and by displacing side chains that normally contact the autoinhibitory interface between the kinase domain and the AID (Fig. 3c). These data suggest a novel allosteric mechanism for PAK activation. To further confirm that PAK is stimulated by local perturbations in the environment surrounding Iα3, we mutated the hydrophobic PAK2 residues Val 123 and Phe 129 that stabilize the AID and the kinase domain, respectively. Both Val 123 Asp and Phe 129 Asp resulted in a constitutively active kinase with more than 60-fold enhancement of substrate phosphorylation (Fig. 3f). We note that the mechanism of EspG binding to PAK is structurally distinct from that of Cdc42 binding²⁸, indicating that the catalytic machinery of EspG is a unique bacterial invention.

The two EspG structures reported here are nearly identical, with a root mean squared deviation of 0.612 Å over 349 Cα atoms. As shown in Fig. 4a, ARF6 and PAK2 occupy distinct, non-overlapping binding sites on adjacent surfaces of EspG. Consistent with this view, EspG nucleates a trimeric complex between the kinase and GTPase in solution (Supplementary Fig. 11). This complex could also be reconstituted on Golgi mimetic liposomes (Fig. 4b). ARF1_GTP recruited EspG to the artificial membrane surface (Fig. 4b, lane 4), which in turn localized PAK2 to these sites (Fig. 4b, lane 6). Notably, PAK2 localization was strictly dependent on formation of the EspG–ARF1_GTP complex (Fig. 4b, lanes 6–8) and ARF1 tethering to the membrane (Fig. 4b, lanes 9 and 10). As predicted by these findings, EspG co-localized with ARF1at the Golgi (Fig. 4c). We further speculated that PAK would also be recruited to these sites. To test this hypothesis, an in vivo ‘activity’ probe was engineered by fusing the PAK2 Iα3-helix sequence (residues 121–136) to the carboxy terminus of the mCherry fluorophore. The PAK2 probe recognized cellular EspG and was targeted to the Golgi complex in 78 ± 5% of EspG-transfected cells (Fig. 4c). By comparison, mutant EspG Asn 212 Ala that lacked all kinase stimulatory activity (Supplementary Fig. 9) localized to the Golgi complex but did not recruit PAK2 to these sites (Fig. 4c). Together, our studies support the function of EspG as a catalytic scaffold that links GTPase inhibition with kinase signal transduction pathways at membrane organelles (Fig. 4d).

EspG belongs to large family of type III effectors secreted by diverse bacterial pathogens. Our studies show that EspG has structural homology with VirA (refs ^{14, 15}) from Shigella flexneri, suggesting that it too may function as an enzyme scaffold (root mean squared deviation, 3.1 Å; Z-score, 5.9) (Supplementary Fig. 12). However, a detailed structural comparison indicates that VirA is unlikely to target the same signalling pathways as EspG during Shigella pathogenesis (Supplementary Fig. 12). We provide mechanistic insights and structural evidence that EspG harbours two unique pathogenic activities, ARF GTPase inhibition and PAK stimulation. Moreover, EspG targets PAK to specific membrane surfaces through its association with ARFs. Froma strategic point of view, the assembly of an artificial enzyme complex enables bacteria to precisely control signalling events with little competition between endogenous regulatory systems. Thus, it is intriguing to speculate that EspG organizes a higher-order signalling network to effectively subvert key cellular processes including cell polarity, adhesion, receptor trafficking and protein secretion.