A common assembly module in injectisome and flagellar type III secretion sorting platforms

Translocating proteins across the double membrane of Gram-negative bacteria, type III secretion systems (T3SS) occur in two evolutionarily related forms: injectisomes, delivering virulence factors into host cells, and the flagellar system, secreting the polymeric filament used for motility. While both systems share related elements of a cytoplasmic sorting platform that facilitates the hierarchical secretion of protein substrates, its assembly and regulation remain unclear. Here we describe a module mediating the assembly of the sorting platform in both secretion systems, and elucidate the structural basis for segregation of homologous components among these divergent T3SS subtypes sharing a common cytoplasmic milieu. These results provide a foundation for the subtype-specific assembly of T3SS sorting platforms and will support further mechanistic analysis and anti-virulence drug design.

T ype III secretion systems (T3SS) allow the transport of protein substrates directly across the double membrane of Gram-negative bacteria. There are two evolutionarily related, yet functionally distinct subtypes of T3SS: 'injectisomes', which deliver effector proteins into the cytoplasm of eukaryotic host cells 1 , and the flagellar apparatus, which secretes the polymeric filament used for motility 2 . Despite their functional divergence, injectisomes and the flagella share a common core of homologous gene products and possess ultrastructural similarities 3 . For example, both systems share related elements of a 'sorting platform' that facilitates the hierarchical secretion of protein substrates 4 .
Proteomic analyses have identified the major components of the sorting platform for the Salmonella typhimurium SPI-1 injectisome: the AAA þ ATPase InvC, its regulator OrgB and the proteins SpaO and OrgA 4 . While SpaO has been shown to be necessary for formation of the sorting platform 4 , little is known about its molecular structure. In Yersinia, the SpaO homologue is expressed as a full-length protein as well as a carboxy-terminal fragment translated from an internal translation start site 5 ; this carboxy-terminal fragment dimerizes and can interact with the full-length protein. The crystal structure of the Yersinia carboxyterminal dimer is similar to that of its Pseudomonas 6 and flagellar 7 homologues, together characterizing a structural class known as the surface presentation of antigens (SPOA) domain. Whether other domains within SpaO possess a similar structure, and how these structures correlate with function remains unknown.
In the flagellar apparatus, the SpaO homologues FliM and FliN form a robust, stable ring (the 'C-ring') at the cytoplasmic face of the basal body 8 . Electron microscopic analyses have similarly localized the SpaO homologue to the cytoplasmic face of the Shigella injectisome 9 , and recent cryoelectron tomographic studies in the same organism identified SpaO homologuedependent 'pods' of density beneath the injectisome 10 .
In contrast to the flagellar C-ring, this sub-injectisome structure is less robust 10 , and fluorescence microscopic analysis of the Yersinia SpaO homologue show that there is dynamic exchange between cytoplasmic-and injectisome-associated forms 11 . How SpaO and its homologues interact with other elements of the T3SS has yet to be shown at high resolution, and how homologous flagellar and injectisome components are properly segregated to their cognate secretion systems remains an open question.
Here we show that a novel, heterotypic interaction between SPOA domains serves as a scaffold for sorting platform assembly in both injectisome and flagellar T3SS. Solution nuclear magnetic resonance (NMR) data support the crystallographic model, and structure-guided mutagenesis shows that this interaction is necessary for formation of the SpaO-OrgB-InvC complex, the proper localization of SpaO to the bacterial inner membrane and T3SS function. Structures of the flagellar SpaO-OrgB homologues FliM, FliN and FliH reveal a mechanism for the proper segregation of homologous sorting platform components among T3SS subtypes sharing a common cytoplasmic milieu. Together, these structures define a common module utilized in sorting platform assembly and provide insight into the subtype-specific assembly of T3SS.

Results
SpaO contains two bona fide SPOA domains. To dissect the structural basis for the sorting platform assembly, we determined the structures of individual domains of S. typhimurium SpaO and then characterized their interactions with other sorting platform components. Preliminary bioinformatic analyses suggested the presence of two putative SPOA domains in the carboxy-terminal half of SpaO, which we denote SPOA1 and SPOA2 (Fig. 1a). We first determined the structure of the SPOA2 homodimer to 1.35 Å resolution (Fig. 1b, 5,6 : like two left hands grasping one another, an antiparallel beta-sheet 'palm' of each protomer is grasped by the 'fingers' of the other, with a 'thumb' protruding from the top of the palm and strands from each protomer forming an antiparallel beta sheet on the 'floor' of the assembly (Fig. 1b,c). The Salmonella SPOA2 homodimer superposes on its Yersinia and Pseudomonas homologues with 2.24 and 3.05 Å root mean squared deviation (r.m.s.d.), respectively ( Supplementary Fig. 1b).
While SPOA1 alone was insoluble, constructs containing both SPOA1 and SPOA2 (residues 140-297) were stable and soluble. SpaO(140-297) was analysed by solution NMR ( Supplementary  Fig. 2a), and chemical shift deviation (CSD) analysis of backbone amide resonances suggested a secondary-structure pattern similar to that predicted by bioinformatic analyses: two SPOA domains connected by a flexible linker ( Supplementary Fig. 2b). We hypothesized that SPOA2 interacts with and stabilizes SPOA1; consistent with this hypothesis, a SPOA1 construct (145-213) could be co-refolded with SPOA2. This complex crystalized, and its structure was determined to 2.9 Å resolution (Fig. 1b,c; Table 1; Supplementary Fig. 3). SPOA1 and SPOA2 form a distinct, heterotypic SPOA-SPOA interaction with an overall topology similar to that of the SPOA2 homodimer. The SPOA1 backbone follows that of the prototypical SPOA fold, retaining the antiparallel beta-sheet floor and fingers-to-palm architecture (Fig. 1b,c). In both SPOA1,2 and the SPOA2 homodimer, interacting protomers each bury about 1,800 Å 2 against their binding partner. SPOA1,2 and the SPOA2 homodimer superpose with 2.47 Å r.m.s.d. (Supplementary Fig. 3b), and the conformation of SPOA2 in association with SPOA1 is grossly similar to that seen in the homodimer, superposing with an r.m.s.d. of 1.67 Å (Supplementary Fig. 3c).
Further supporting the hypothesis that SPOA1 and SPOA2 interact in solution, a post hoc analysis of the three dimensional 15 N-edited nuclear Ö verhauser enhanced spectroscopy-(3D NOESY-HSQC) for SpaO(140-297) revealed long-range amide proton correlations between SPOA1 and SPOA2 ( Supplementary  Fig. 4). Given the o20 residue linker connecting SPOA1 and SPOA2, they would experience a low millimolar-range relative concentration and would likely interact in an intramolecular fashion (Fig. 2a). However, at high local SpaO concentrations (for example, in association with the T3SS), intermolecular heterotypic SPOA interactions might explain the apparent oligomeric nature of the sorting platform (Fig. 2a). Indeed, a similar model of intermolecular domain swapping was recently suggested for the ring-forming injectisome protein PrgK 12 .
Hypothetical SpaO oligomerization driven by intermolecular heterotypic SPOA interactions would be dependent on the covalent linkage of SPOA1 and SPOA2. Thus, we tested whether genomic deletion of the SpaO amino-terminal domain and SPOA1 can be complemented in trans, as assayed by Salmonella T3SS function in vitro. When grown under T3SS-stimulating conditions, the culture supernatant of S. typhimurium has a stereotyped protein composition, consisting of both flagellar and injectisome secretion substrates (Fig. 2b, secreted proteins are annotated as per Mizusaki et al. 13 ). Deletion of spaO results specifically in the loss of injectisome-dependent secretory products from the culture supernatant, and deletion of spaO codons 1-203 phenocopies spaO deletion, indicating that the SpaO amino-terminal domain(s) and/or SPOA1 are necessary for T3SS function (Fig. 2c). Because SpaO(1-219) is able to complement the deletion of spaO codons 1-203 (Fig. 2c, red asterisks), the covalent linkage of SPOA1 and SPOA2 is not  (Fig. 2c).
SpaO SPOA1,2 is a scaffold for interaction with OrgB-InvC. Double-hexahistidine-tagged SpaO is able to co-affinity purify the sorting platform components OrgB and InvC when co-expressed in Escherichia coli (Fig. 3a). Formation of the SpaO-OrgB-InvC termary complex is OrgB dependent, as SpaO alone is insufficient to co-affinity purify InvC (Fig. 3a). We hypothesized that SPOA1,2 might serve as a scaffold for the interaction of SpaO with OrgB-InvC. Indeed, SPOA1,2 is sufficient to co-affinity purify OrgB-InvC (Fig. 3a). This construct contains a Val203 GTGto-Val203 GTT mutation to prevent the duplicitous translation of SPOA2 from its cryptic internal translation start site, demonstrating that the SPOA2 homodimer is dispensable for SpaO-OrgB-InvC complex formation. OrgB and its homologues are predicted to share a common amino-terminal organization: a disordered region followed by a coiled coil. In the flagellar system, the unstructured region at the amino terminus of the OrgB homologue FliH is necessary for its interaction with the SpaO homologues FliM and FliN 14 . We solublized the pre-coiled-coil region of OrgB (residues 1-30) by genetic fusion to T4 lysozyme and found that it bound to SpaO SPOA1,2. Herein, we will refer to the SPOA1,2-binding region at the amino terminus of OrgB and its homologues as the adaptor peptide of the ATPase regulator (APAR).
The SPOA1,2-OrgB(APAR)::lysozyme complex was crystalized and its structure solved to 2.0 Å resolution ( Fig. 3b; Table 1; Supplementary Figs 5 and 6). The OrgB APAR forms a lariat-like structure, contacting the thumb of SPOA2 and fingers of SPOA1 (Fig. 3b, (Fig. 4). Compared with apo-SpaO, the largest CSDs of backbone amide resonances in SpaO-OrgB map on the crystal structure to residues involved in the interface, which are highly conserved across both the Salmonella/Shigella and Yersinia/ Pseudomonas clades ( Fig. 4; Supplementary Fig. 7). In the crystal, these residues form the docking site for OrgB residues Ile17, Leu18 and Ile19 (Fig. 5a). The OrgB surface area buried by these three residues (360 Å 2 ) accounts for approximately one-third of the APAR's total buried area. Here, the APAR shows noteworthy sequence homology: immediately following a conserved glycine (Gly16, pseudo-lariat apex) is a string of aliphatic and basic amino acids in each homologue (Fig. 5a).

The SPOA1,2-APAR interaction is necessary for T3SS function.
To test whether the SPOA1,2-APAR interaction per se is necessary for T3SS function, we constructed an OrgB triple mutant (I17D,L18D,I19D) to disrupt its interaction with SpaO. As predicted, SpaO failed to co-affinity purify OrgB(I17D, L18D,I19D)-InvC when co-expressed in E. coli (Fig. 5b), and the aspartate triple mutation completely abolished T3SS in vivo (Fig. 5c). Fluorescence microscopic analyses of the Yersinia SpaO homologue have shown it to localize in discrete perimembranous punctae 11 . Might the SPOA1,2-APAR interaction function to localize SpaO to the bacterial inner membrane? In an otherwise wild-type genomic background, an EGFP::3 Â FLAG::SpaO fusion exhibits punctate, perimembranous localization, consistent with its recruitment to injectisome basal body channels (Fig. 5d). Deletion of orgB disrupts proper SpaO localization, producing a more diffuse, cytoplasmic pattern, and the asparate triple mutation was sufficient to phenocopy the orgB deletion mutant (Fig. 5d). Together, these data suggest that the SpaO(SPOA1,2)-APAR assembly is necessary for the proper localization of SpaO to discrete perimembranous puncta, and that this arrangement is required for T3SS function.
A divergent SPOA1,2-APAR assembly in the flagellar T3SS. The flagellar C-ring is primarily composed of three proteins: FliM, FliN and FliG 8 . The SpaO homologues FliM and FliN are predicted to contain one SPOA domain each, which we designate as SPOA1 and SPOA2, respectively. Paralleling the injectisome, FliN is known to interact with the OrgB homologue FliH 14 . The evolutionary relationship between injectisomes and flagella creates a practical conundrum: how are homologous T3SS components segregated to their corresponding secretion systems within a common cytoplasmic milieu? To qualitatively assess the subtype specificity of SPOA-APAR interactions, we co-affinity purified a panel of Salmonella SPOA domains with hexahistidinetagged APAR::lysozyme fusions (Fig. 6). Indeed, the OrgB and FliH APARs robustly co-affinity purify their cognate SPOA1,2 but not that of the other T3SS subtype (Fig. 6, red asterisks). Neither SpaO nor FliM-FliN are pulled down by the APAR from a second pathogenic T3SS found in S. typhimurium (SPI-2 SsaK). Consistent with the observation that the OrgB APAR interacts with surfaces on both SPOA1 and SPOA2, the OrgB and FliH APARs more robustly pull down their cognate SPOA1,2 than homodimeric SPOA2 (Fig. 6).
We hypothesized that divergence of the SPOA1,2-APAR assembly architecture contributes to proper component segregation among T3SS subtypes, and sought to structurally  Fig. 8a) and some swarming motility 15 . We crystallized the SPOA of FliM (residues 245-334) fused to FliN(5-137), and its structure was solved to 2.56 Å ( Table 2; Supplementary Fig. 8). Architecturally, the FliM(SPOA1)-FliN(SPOA2) interaction is similar to that of SpaO ( Supplementary Fig. 8e To elucidate the mechanism of FliH-specific assembly with FliM-FliN, we co-crystalized the FliM(SPOA)::FliN fusion with a FliH(1-18)::lysozyme fusion ( Fig. 7; Table 2; Supplementary  Fig. 9). As with its injectisome counterparts, the FliM-FliN SPOA1,2 did not undergo large conformational changes upon APAR binding (Supplementary Fig. 9c, 1.11 Å r.m.s.d.); however, the binding mode for the FliH APAR is radically different. In contrast to the OrgB pseudo-lariat, the FliH APAR adopts a near-linear conformation along the 'top' of FliM-FliN (Fig. 7a). As observed in the SpaO-OrgB assembly, the FliH APAR makes extensive contact with both SPOA1 and SPOA2 (Fig. 7a), supporting the observation that the FliH APAR interacts more strongly with the FliM-FliN heterodimer than the FliN homodimer (Fig. 6). The FliH-FliM-FliN assembly is characterized by the burial of several highly conserved hydrophobic FliH side chains. Two tryptophan side chains form an aromatic clamp, which binds hydrophobic pockets on opposite faces of the FliN thumb (Fig. 7b). These residues are critical for flagellar function 14 and are highly conserved (Fig. 7c). Similarly, the bulky side chain of FliH Leu15 fills a hydrophobic pocket on the thumb of FliM ( Supplementary Fig. 10). The binding interfaces for these three residues are formed by both FliM and FliN and are highly conserved across species (Supplementary Fig. 10). This structure presents a conserved model for FliH-FliM-FliN interaction, which is distinct from that of SpaO-OrgB.

Discussion
We present here a series of structures that yield critical mechanistic insights into T3SS sorting platform assembly across multiple species and secretion subtypes. The existence of heterotypic SPOA interactions provides a structural explanation for the observed B1:3 stoichiometry of SPOA1 to SPOA2 in SpaO homologues 5 . While two of these SPOA2 domains could be accounted for by a homodimer interacting with full-length SpaO, the conformation of the third SPOA2 (located in the full-length protein) was unclear. Previous reports had proposed the existence of an alternate autostabilizing conformation for the third SPOA2 (ref. 5). We show here that the third SPOA2 can be stabilized by a SPOA1-SPOA2 interaction.
Similar to SpaO and its injectisome homologues, the ratio of FliM to FliN in situ is estimated to be 1:3 (ref. 8). In the context of our FliM-FliN structure, this suggests a model for FliM-FliN interaction similar to that of SpaO. FliM(SPOA1) would engage FliN(SPOA2) in a heterotypic SPOA-SPOA interaction, and additional homodimeric FliN would interact with FliM-FliN in an as of yet undetermined fashion (analogous to the SpaO SPOA2 homodimer interaction with full-length SpaO). However, reports of FliN tetramerization and FliM:FliN ratios between 1:3 and 1:4 suggests that more complicated higher-order structures may be used by the flagellar apparatus 7 . It should also be noted that while previous investigations of the flagellar T3SS have focused on the interaction between FliH and FliN specifically 14 , our structures and biochemical data show that the FliH APAR more strongly interacts with the FliM-FliN complex than with FliN alone,  ARTICLE suggesting that the FliM-FliN complex is the physiologically relevant binding partner for FliH. Our structures suggest a partial model for the subtype-specific assembly of the T3SS sorting platforms: the heterotypic interaction between SPOA domains within a given T3SS subtype functions as an adaptor for ATPase and its regulator through interaction with the APAR peptide (Fig. 8). However, a number of questions remain regarding the higher-order architecture of the sorting platform in situ. We hypothesize that the puncta formed by SpaO in vivo represent the high-molecular weight sorting platforms described by Lara-Tejero et al. 4 . Diepold et al. have quantified the stoichiometry and dynamics of these puncta in Yersinia, showing them to possess B22 copies of the SpaO homologue per punctum and to be in dynamic exchange with the cytoplasm 11 . In contrast, the recent tomographic reconstruction of Shigella injectisomes by Hu et al. revealed the presence of only six SpaO homologue-dependent pods of density beneath the injectisome, and their localization was OrgB homologue independent 10 . Taken together with our findings, these results suggest that there may be two subpopulations of SpaO in vivo: one stably associated with the injectisome, and a second dynamic population in exchange with the cytoplasm, requiring the SPOA1,2-APAR interaction to form high-molecular weight, perimembranous sorting platforms. Recent analyses of FliI ATPase dynamics by Bai et al. suggest a similar two-population model, which they hypothesize functions to deliver secretion substrates to the assembling flagella 16 .
What might be the mechanism for sorting platform targeting, and how might this factor into T3SS machine function? Perhaps APAR binding to the SPOA1,2 scaffold induces conformational changes in OrgB, InvC and/or the amino terminus of SpaO, which facilitates interaction with the membrane integral components of the T3SS. Alternatively, the SPOA1,2-APAR assembly might function simply by inducing proximity between sorting platform components. Intriguingly, the FliH APAR region has previously been shown to interact with the membrane integral export gate protein FlhA 17 , suggesting that OrgB/FliH may function as a hub, bridging the ATPase, export gate and SpaO/FliM-FliN. How the sorting platform component OrgA-which lacks a clear flagellar homologue-factors into complex assembly remains to be determined. Similarly, the implications of sorting platform assembly for substrate recruitment, dechaperoning and secretion remain unclear. Our structures will support the precise interrogation of sorting platform interactions in the biomechanics of secretion, and the necessity of the SPOA1,2-APAR interaction makes it a novel target for the design of anti-virulence compounds.

Methods
Bioinformatics. Sequence alignments were performed using Clustal Omega 18 or M-COFFEE 19 . Secondary-structure and disorder predictions were performed using the PSIPRED server 20 .
Molecular biology. PCR was performed using OneTaq (New England Biolabs), Phusion (New England Biolabs) or PfuTurbo (Agilent) as per manufacturer guidelines with oligonucleotides purchased from Integrated DNA Technologies. All mutations or gene fusions were created by overlap extension PCR. Gene sequences from S. typhimurium were PCR amplified from the T3SS-competent strains SB300 (wild type; gift from J. Galán) or SB1741 (3 Â FLAG::SpaO, silent SpaO L79 CTG to L79 CTA variant; gift from J. Galán) 4 . The T4 lysozyme (C54T, C97A) sequence was obtained from Addgene plasmid 18111. An additional mutation (D20N) in T4 lysozyme was made to decrease toxicity in E. coli 21 , and the terminal three residues were mutated to alanines to decrease conformational entropy. Standard molecular biology protocols were followed to clone sequences of interest into modified pCDFduet or pETduet vectors for expression in E. coli or pBAD for expression in S. typhimurium. Restriction enzymes and Quick Ligase (New England Biolabs) were used as per manufacturer specifications. Salmonella genomic mutants were produced using homologous recombination from SacB-expressing suicide plasmids 22 . All SpaO and OrgB mutants were prepared on the SB1741 background. FliM and FliN mutants were prepared on the SB300 background.
Protein expression and purification. Constructs were transformed into BL21(DE3)Gold E. coli for heterologous expression and protein expression induced mostly as described 23 . Specifically, bacteria were grown to an optical density at 600 nm of 0.5-0.6 at 37°C in LB medium, the cultures were cooled to 18°C, induced with 250 mM isopropyl-b-D-thiogalactoside and grown overnight at 18°C. Selenomethionine (SeMet)-substituted protein was produced in the methionine auxotrophic E. coli B834(DE3) grown in methionine-free media supplemented with SeMet 23 . Uniformly labelled 15 N/ 13 C or 2 H/ 15 N/ 13 C protein samples were produced by overexpression in isotopically enriched minimal media. Deuterium oxide, 15 N-ammonium chloride, and 13 C-glucose were obtained from Cambridge Isotope Labs.
After induction overnight at 18°C, cells were harvested by centrifugation and For purification under denaturing conditions, guanidinium chloride was added to the lysate to a final concentration of 6 M. The post-extraction lysate was clarified by centrifugation at 30,000g for 15 min at 4°C and loaded onto NiNTA resin in a batch at 25°C. Still at 25°C, the resin was washed with denaturing wash buffer (8 M urea, 500 mM NaCl, 20 mM Tris-Cl pH ¼ 8.0 and 30 mM imidazole-Cl pH ¼ 8.0) and eluted in denaturing elution buffer (8 M urea, 200 mM NaCl, 20 mM Tris-Cl pH ¼ 8.0 and 360 mM imidazole-Cl pH ¼ 8.0). The elution was supplemented with 5 mM EDTA and 5 mM dithiothreitol (DTT), and the protein refolded by dialysis against 200 mM NaCl, 20 mM Tris-Cl pH ¼ 8.0 and 1 mM DTT (3-4 changes, dialysis time of 24 h total, 4°C). For T4 lysozyme fusions, Hepes-Na pH ¼ 7.0 was substituted for Tris-Cl pH ¼ 8.0. Insoluble material was removed by centrifugation or filtration and affinity tags were removed by cleavage with HRV 3C protease.
Affinity-purified proteins were further purified by ion-exchange chromatography using an AKTA FPLC and the following columns (GE Healthcare): T4 lysozyme fusions were purified by cation exchange on a SourceS column; all other constructs were purified by anion exchange on a SourceQ column. For cation-exchange chromatography, proteins were loaded in a batch in 10 mM Hepes-Na pH ¼ 7.0, 50-100 mM NaCl and eluted by a NaCl gradient (from 0 to 1,000 mM) in the same buffer. For anion exchange, proteins were loaded in a batch in 20 mM Tris-Cl pH ¼ 8.0, 50-100 mM NaCl and eluted by a NaCl gradient (from 0 to 1,000 mM) in the same buffer.
Prior to crystallography, ion-exchange-purified proteins were further purified by gel filtration chromatography on a Superdex 75 column (GE Healthcare SpaO(232-297) was concentrated to 8 mg ml À 1 and crystallized with 35% PEG400, 200 mM calcium acetate and 100 mM sodium acetate pH ¼ 5.0. Crystals were cryoprotected in the mother liquor. Microseeding was performed to enhance crystal uniformity and diffraction. Briefly, crystals to be seeded were harvested in precipitant solution and vortexed in a microfuge tube with a small stir bar for B60 s. The slurry of microseeds was serially dilluted (5-10-fold steps) in precipitant solution and five selected microseed precipitant mixtures were mixed with fresh protein as in a normal hanging-drop experiment.
Structure determination. Data were collected at the National Synchrotron Light Source (Brookhaven National Laboratory) beamline X29A at a temperature of À 173°C using the following X-ray wavelengths: 0.979 Å for SeMet crystals, 1.075 Å for native crystals. Diffraction data sets were indexed and integrated in iMOSFLM 24 and scaled and reduced with AIMLESS 25 . Data sets were truncated at I/sI42.0, and all sets were determined to have a CC 1/2 40.7 in the outermost resolution shell 26 (Tables 1 and 2).
Structures were refined and validated in Phenix (Tables 1 and 2 15 N/ 13 C-labelled SpaO(140-297) was co-refolded with an excess of unlabelled thioredoxin::OrgB(2-30). The thioredoxin solublization tag was cleaved off by overnight incubation with HRV 3C protease. Protease and affinity tags were removed on NiNTA resin and the SpaO-OrgB complex was separated from the majority of free thioredoxin by Superdex 75 gel filtration chromatography. The final concentration of the protein complex was 0.2 mM in 10 mM citrate buffer at pH 5.6 supplemented with 10% v/v deuterium oxide, 100 mM NaCl and 2 mM dithiothreitol.
The NMR data were collected on Bruker 600, 800 and 900 MHz AVANCE spectrometers equipped with TCI/TXI CryoProbes at 20°C for the apo-SpaO and 30°C for the APAR-bound forms. For resonance assignments of apo-SpaO, transverse relaxation-optimized triple-resonance 31 experiments including trHNCO, trHN(CA)CO, trHNCA, trHN(CO)CA, trHNCACB and trHN(CO)CACB were acquired at 600 and 900 MHz. A three dimensional 15 Nedited nuclear Ö verhauser enhanced spectroscopy-heteronuclear single quantum coherence spectrum with 100 ms mixing time was also acquired at 900 MHz. To assign APAR-bound SpaO, a suite of conventional backbone experiments 32 were acquired at 600 and 800 MHz.
The data were processed in Topspin 2.1 spectra and analysed using the Autolink module in CARA 1. 5 (ref. 33). In both apo-SpaO and its complex with APAR, we were able to successfully assign 495% of the backbone resonances. The heteronuclear chemical shifts were analysed using the TALOS þ 34 database to predict the secondary structure of the protein. The weighted CSDs were calculated from amide proton (H) and nitrogen chemical shifts ( 15 N) using the following equation: CSD ¼ O((DdH) 2 þ ((Dd 15 N)/5) 2 ) Co-affinity purification assay. For co-affinity purification of the SpaO-OrgB-InvC complex (Figs 3 and 5), the proteins indicated were co-expressed and purified under native conditions as described above. For the SPOA-APAR::lysozyme pull-down experiment (Fig. 6), the indicated SPOA-containing proteins were Ni-affinity purified under native conditions, their affinity tags were removed by overnight incubation with HRV protease 3C and they were further purified by anion-exchange chromatography (as above). APAR::lysozyme fusions were separately purified under denaturing conditions and were subjected to cationexchange chromatography after refolding (as above). One mg of hexahistidinetagged APAR::lysozyme fusion protein was mixed with 2 mg of the indicated SPOA-containing protein in 0.2 M NaCl and 20 mM Tris-Cl pH ¼ 8.0 (final volume 4 ml) and incubated on ice for 2 h. The mixture was passed twice over 2 ml of NiNTA resin, washed with 8 ml wash buffer (200 mM NaCl, 20 mM Tris-Cl pH ¼ 8.0, 5% v/v glycerol and 30 mM imidazole-Cl pH ¼ 8.0) and then eluted in 3.5 ml elution buffer (200 mM NaCl, 20 mM Tris-Cl pH ¼ 8.0, 5% v/v glycerol and 360 mM imidazole-Cl pH ¼ 8.0).
In vitro secretion assay. S. typhimurium of the indicated genotype were grown for 6 h at 37°C in LB medium with NaCl supplemented to a final concentration of 0.3 M. Cells were pelleted by centrifugation at 3,400g for 0.5-1 h and the supernatants were 0.22-mm filtered. Secreted proteins were precipitated from the filtered supernatants with 15% trichloroacetic acid overnight at 4°C. The precipitate was pelleted by centrifugation at 3,400g for 1 h at 4°C, resuspended in ice-cold acetone and transferred to a microfuge tube. After 0.25 h on ice, the precipitate was harvested by centrifugation at 16,000g for 0.75 h at 4°C and resuspended in 0.2 M Tris-Cl pH ¼ 8.0 and 0.2 M NaCl to neutralize any residual acid before the addition of SDS-polyacrylamide gel electrophoresis loading buffer. For plasmid complementation analysis, S. typhimurium were electroporated with SpaO sequences cloned into the pBAD vector and expression was induced with 0.01% arabinose for the entire duration of the experiment.
Fluorescence microscopy. S. typhimurium were grown as for the in vitro secretion assay. Cells were harvested by centrifugation, washed three times in PBS and fixed overnight with 4% formaldehyde in PBS at 4°C. Cells were again washed three times in PBS, counterstained with 4,6-diamidino-2-phenylindole (Sigma-Aldrich) and 10 mM Nile Red (Sigma-Aldrich), and immobilized on poly-L-lysine (Sigma-Aldrich)-coated coverslips. Covers were mounted in Prolong Diamond (Life Technologies) and sealed with nail polish. Slides were imaged on a DeltaVision Image Restoration Microscope with a Â 100 objective (Applied Precision). Images were deconvoluted in Softworx (Applied Precision) and processed identically in ImageJ (NIH) and Photoshop (Adobe).