The pod components of the Shigella T3SS sorting platform accommodate multiple copies of Spa33 (SctQ) Sean K Whittier, Shoichi Tachiyama, Samira Heydari, Wendy L Picking, Jun Liu, William D Picking, Leigh Knodler Abstract The bacterial type III secretion system (T3SS) uses a membrane-embedded injectisome assembly to export effector proteins into host cells. While atomic-level structural details have been revealed for much of the T3SS apparatus, the model of the cytoplasmic sorting platform remains largely low-resolution. A central structural element of the sorting platform is the so-called “pod” protein, SctQ, which anchors the sorting platform to the inner membrane via interaction with the adaptor protein SctK, and connects to the central ATPase via the spoke protein SctL. SctQ proteins also interact with alternatively translated homodimers of their C-terminal SPOA2 domains. Low resolution electron density maps have provided an outline of the sorting platform architecture, and fluorescence microscopy studies have suggested a 1:4:2 SctK:SctQ:SctL stoichiometry. While there are experimental and AlphaFold structures of individual components and complexes of the sorting platform pod, there is currently no model for the pod structure that adequately fits the electron density or accounts for the proposed stoichiometry. Here we use AlphaFold to generate a model of the upper portion of the Shigella pod complex in which two copies of the SctQ protein, Spa33, bind the adaptor protein MxiK, with each copy of Spa33 bound to an alternately translated SPOA2-SPOA2 domain. We show through mutation of energetically critical interface residues, predicted by computational mutant scanning, that both Spa33 binding sites on MxiK are required for T3SS activity in Shigella flexneri, as well as binding of Spa33 to the SPOA2-SPOA2 homodimer. We find that this model fits well to the upper two-thirds of the pod electron density, albeit in a manner that places the protein components slightly closer to the inner membrane than traditionally presented. Further, cryogenic electron tomography shows Spa33/MxiK interface mutant sorting platforms are destabilized and lack clear density for the SctL spoke protein MxiN and central ATPase Spa47, while Spa33/SPOA2-SPOA2 interaction mutants are further disrupted and lack clear lower pod density. The density for the lower portion of the sorting platform pod, thought to arise from Spa33 and MxiN, remains unmodeled, largely due to the inability of AlphaFold to confidently predict the structure of MxiN or MxiN complexes. However, we show that, allowing for potential conformational rearrangements, the pod can accommodate three full-length copies of Spa33 in a manner that fills some of the lower electron density. How a potential fourth copy of Spa33 could occupy the remaining density with MxiN remains unclear. The type III secretion system (T3SS) is a critical virulence factor used by numerous gram-negative pathogens to inject effector proteins into host cells. These effectors are secreted through the type III secretion apparatus, or injectisome, a large macromolecular assembly comprising an extracellular needle complex, a periplasmic basal body, and a cytoplasmic sorting platform. The sorting platform is formed by six, radially symmetric “pods” which anchor a central ATPase in position to power secretion through the injectisome. Each pod is thought to consist of three main protein components, SctK, SctQ, and SctL, present in a 1:4:2 stoichiometry, respectively. Here we use AlphaFold modeling and site-directed mutagenesis to develop a structural model for the Shigella flexneri pod that accounts for multiple copies of SctQ that fits within the the experimental electron density of the sorting platform. Introduction The type III secretion system (T3SS) is used by gram-negative bacterial pathogens to secrete protein effectors into eukaryotic host cells []. These effectors are secreted by the type III secretion apparatus (T3SA), or injectisome, the architecture of which is conserved among phylogenetically distant pathogens []. The T3SA can be divided into three distinct portions (see Fig SA in ): a hollow needle-tip assembly that extends from the outer membrane to contact host cells, basal body embedded in the inner and outer membranes, and a cytoplasmic sorting platform that is responsible for substrate recognition and driving secretion [–]. Given its role in human disease, elucidating the structure of the T3SA is critical from the standpoint of developing drugs that inhibit its function or vaccines that may prevent severe disease. The T3SA is also a complex protein nanomachine, and the study of its structure and function may allow us to glean basic biophysical and biochemical principles that govern large biomolecular assemblies. Perhaps the least understood portion of the T3SA is the cytoplasmic sorting platform [], this highly dynamic macromolecular complex is thought to be transiently assembled, with some individual protein components in frequent exchange with the cytosol [,]. Its overall organization is understood: a central ATPase (SctN in unified nomenclature []) is connected via radially symmetric spokes (SctL) to six “pods”, formed by SctQ, that then connect to membrane-bound ring of SctD via the adaptor protein, SctK [,]. Fluorescence microscopy experiments indicate the presence of one copy of SctK, two copies of SctL, and up to four copies of SctQ per pod [,]. Furthermore, pod proteins play an active role in secretion, shuttling effector proteins form the cytosol to the export gate []. The SctQ protein is comprised of independently folding N- and C-terminal domains [], SctQN and SctQC. The structures of the C-terminal domains of SctQ proteins from Shigella (Spa33) and Salmonella (SpaO) have been shown experimentally to be a dimer of two “surface presentation of antigens" (SPOA) domains, SPOA1 and SPOA2, and NMR data show that SpaOC binds to a peptide from the SctL spoke protein OrgB [–]. Furthermore, SctQ genes feature an internal translation start site upstream of SPOA2 []. These alternatively translated SPOA2 domains form a SPOA2-SPOA2 homodimer, SctQshort, which associates with the full-length SctQ protein and may be required for T3SS function. Recent insight into the pod structure has come from AlphaFold modeling and in vivo crosslinking of the SctQ complexes with SctK, SctL, and SctQshort in the Salmonella system [,]. Those results indicate that SctQ interacts with SctK via its N-terminal domain, and with SctL through SctQC. Additional work reported the interaction between SctQ and SctQshort and showed that SctK simultaneously binds to SctL, SctQ, and SctQshort. While those results have greatly expanded our understanding of the organization of the pod, a single SctK:SctQ:SctL complex is not sufficient to fill the electron density map from cryo-electron tomography (cryo-ET) []. Furthermore, fluorescence microscopy of the Yersinia injectisome indicates that each pod should contain as many as four copies of SctQ and two copies of SctL []. A full model of the sorting platform, therefore, should include multiple copies of SctQ and SctL, fit well to the electron density map, and retain the molecular interactions revealed in experimentally verified AlphaFold models. In this work, we present a model for the upper two-thirds of the Shigella sorting platform pod. We show that AlphaFold predicts a complex of the Shigella SctK protein MxiK binding two copies of the SctQ protein Spa33 with high confidence and demonstrate, through mutation of critical interface residues, that each of these binding sites on MxiK is required for T3SS function in Shigella flexneri. Additionally, we find that mutations to the AlphaFold-predicted Spa33/Spa33short interface drastically reduce or eliminate T3SS activity, indicating that Spa33short is a critical component of the pod, congruent with results from other type III secretion systems [,]. The AlphaFold-generated complex of MxiK with two copies of full-length Spa33 and two copies of Spa33short fits well into the upper portion of pod density in cryo-ET maps of the S. flexerni sorting platform, although the protein components are shifted “upward” toward the inner membrane relative to where they previously have been shown to reside in the cryo-ET density map (Fig SA in ). This positioning is somewhat non-canonical, but it fits the density well and explains the curious hole seen between regions commonly attributed to MxiK and Spa33 (Fig 3c). Finally, cryo-ET structures of pod complex mutants retain some pod density but lack density for the SctL spoke MxiN and central ATPase, Spa47. The presence of pod density in these mutant injectisomes suggests that a stable pod structure is required for recruitment of the ATPase. It may also suggest a possible alternative model comprising four copies of Spa33 that may explain additional density at the bottom third of the pod not accounted for by the MxiN dimer, as will be presented in the Discussion. Results Spa33 binds MxiK at two sites AlphaFold predicts a similar structure as reported in the Salmonella T3SS when asked to predict the multimer structure of Spa33 and MxiK in a 1:1 stoichiometry (Fig SBFig SBa in ) []. The resulting model shows Spa33N binding to a hydrophobic patch on the surface of MxiK. However, there is another, similarly sized hydrophobic patch near this binding site on the surface of MxiK (Fig SBFig SBb in ) that is not predicted to be occupied by either copy of the MxiG cytoplasmic domain. Further, a 1:1 Spa33:MxiK interaction model does not fully account for observed electron density in this region of the sorting platform. Therefore, we used AlphaFold to predict the structure of a 2:1 Spa33:MxiK complex. Here, the predicted model shows that one copy of Spa33N binds to the same site as in the 1:1 complex but now has the second copy of Spa33N binding to the adjacent, unoccupied hydrophobic patch on MxiK (). AlphaFold accurately predicts the structure of each component protein domain, as indicated by pLDDT scores (Fig SBFig SBc in ), and is highly confident of the overall complex structure, with low PAE scores across all residues (). pTM and ipTM scores of the predicted complex are 0.82 and 0.79, respectively, indicating that the prediction is likely that of the “true” structure. This 2:1 SctQ:SctK interaction is not unique to Shigella, as AlphaFold predicts similar 2:1 complex for SpaO/OrgA in Salmonella and YscQ/YscK and PscQ/PscK in Yersinia and Pseudomonas (Fig SC in ). PAE values for these complexes are low, but with higher error for the second SpaO interaction in Salmonella, which may indicate a reduced binding affinity at this site. The predicted complex indicates that Spa33 binds to each site of MxiK using the same surface, specifically the region between residues 64–86, which form helices 3 and 4 of Spa33N and the turn between them ( and ). Computational alanine scanning of each Spa33 interface with MxiK reveals large energetic binding contributions from the same group of hydrophobic residues in Spa33N, with F67 and F70 predicted to be hot-spot residues (ΔΔG > 2.0 kcal/mol) (), and Y64 and L79 near the hot-spot cutoff []. To confirm that this surface is involved in MxiK binding, we made perturbing mutations at each of these positions, expressed the mutant proteins in a Δspa33 (Spa33Null) strain of S. flexneri, and characterized secretion phenotype using contact mediated hemolysis [] and Congo red (CR)-induced secretion []. The replacement of each residue with aspartic acid, which places a formal charge in the hydrophobic interface, largely eliminates hemolytic activity for F67D, F70D, and L79D, with only F70D retaining a small level of hemolysis above that of the uncomplemented Spa33Null strain (). The Y64D mutant, however, retained wild-type hemolysis capability. These results are mirrored in the case of CR-induced secretion. F67D, F70D, and L79D all had no detectable IpaB or IpaD secretion, while Y64D secreted wild-type quantities of each (, Fig SD in ). Single alanine substitutions did not greatly reduce hemolytic activity for any of these positions, however the simultaneous replacement of F67 and F70 with alanine resulted in a null phenotype (Fig SBFig SBd in ). Western blots were used to confirm that none of these mutants affected expression levels of either full-length Spa33 or Spa33short (Fig SE in ). The two binding sites on MxiK, as mentioned above, are largely hydrophobic. Computational alanine scanning of MxiK interface residues reveals a handful of hot-spot residues that dominate energetically. At the first binding site, which is homologous to the interaction described in Salmonella, one hot-spot residue is predicted, L46, with three other residues, L27, L33, and I39, having ΔΔG > 1 kcal/mol upon mutation to alanine (). The second binding site is energetically dominated by F16, with F16A giving a predicted ΔΔG > 3.5 kcal/mol. Three other residues have ΔΔG > 1 kcal/mol, with L12 have the largest predicted contribution of these. As with Spa33, we confirmed these interfaces by mutating energetically critical residues, expressing the mutants in a ΔmxiK (MxiKNull) S. flexneri strain, and determining contact-mediate hemolysis and CR-induced secretion. Single aspartate substitutions of binding site 1 residues L27, L33, and I39 had a minor impact on translocon pore formation, with each mutant having about 75% wild-type hemolysis activity. L46D, on the other hand, had no detectable hemolysis (). Additionally, a double alanine mutant, L33A/F46A, was only 10% as active as wild-type MxiK (). L27D, L33D, and I39D had larger impact on induced secretion, with each secreting 50% or less than wild-type. F46D and L33A/F46A each had no detectable IpaB or IpaD secretion (, Fig SD in ). Two residues at binding site 2 were mutated next (L12 and F16). Aspartate substitutions L12D and F16D were modestly perturbing, causing 40% and 65% loss of hemolytic capability relative to wild-type, respectively (), while a simultaneous substitution of each residue with aspartate resulted in a null hemolysis phenotype (). Again, mutant effects on induced secretion were greater than those seen in hemolysis. Both L12D and F16D showed ~95% reduction in secretion compared to wild-type (, Fig SD in ). The double mutant L12D/F16D had no visible IpaB or IpaD bands after CR-induction (, Fig SD in ). As with Spa33, western blots show that these mutations do not impact MxiK expression levels (Fig SE in ). Spa33 binding of the Spa33short homodimer is required for T3SS function The spa33 gene features an internal translation start site just upstream of the SPOA2 region [,,]. This results in an alternately translated SPOA2 protein product, which forms a SPOA2-SPOA2 homodimer, Spa33short []. Recent work using AlphaFold with the Salmonella Spa33 homologue, SpaO, has shown that the SPOA2 homodimer binds to the full length SpaO protein between the N- and C-terminal domains []. In the case of Spa33, AlphaFold predicts a similar complex with the Spa33short homodimer binding along the flexible linker between Spa33N and Spa33C (, Fig SF in ). The confidence of the AlphaFold structure is high, with low PAE values between domains (Fig SF in ). Computational alanine scanning of the Spa33/Spa33short complex reveals two predicted hot-spot residues, F219 and Y221, and several other residues predicted to have ΔΔG > 1 kcal/mol (). The locations of these residues indicate two critical areas of interaction between Spa33 and the SPOA2 homodimer. The first of these involves F219 and Y221, along with I224, which insert into a hydrophobic pocket in Spa33C (). The second area of interaction is defined by two salt-bridges formed between E249 and E285 in Spa33short and K12 and K5, respectively, in the N-terminal peptide of Spa33N (). Again, we experimentally confirm the importance of these interactions through mutation and phenotypic characterization of T3SS activity using contact-mediated hemolysis. Because of their sequential proximity, F219 and Y221 were mutated together. Mutation of the F219/Y221 positions or either of the salt-bridge residues in Spa33short resulted in a greater than 50% loss of hemolytic function, relative to wild-type. Simultaneous mutation of F219, Y221, and E285, however, eliminates T3S activity (). Western blots of S. flexneri overexpressing these mutants show levels of both full-length Spa33 and Spa33short comparable those of wild-type Spa33 (Fig SE in ), indicating that these mutations do not affect protein expression. These results indicate that binding of Spa33 with Spa33short is required for a fully functional T3SS in Shigella. Fitting model of upper pod region to experimental electron density map The above results lead us to propose a model for the upper portion of the SP pod in which a single copy of MxiK is bound to two copies of Spa33, each of which is complexed with a Spa33short homodimer (, ). If this model is valid, it should fit within experimentally obtained electron density maps of the Shigella SP. EM density fitting was performed by using ChimeraX [] to simulate EM density of our model at a resolution of 13 angstroms and then fit the simulated density to a density map from a focused refinement of the sorting platform structure by maximizing the cross correlation between them. As a sanity check for model fitting, we used AlphaFold to include two copies of the MxiG cytoplasmic domain (MxiGC) bound to MxiK, as well as a short peptide from MxiN bound to each copy of Spa33C. The fit model should position MxiGC close to the inner membrane and the MxiN peptide near the spoke density. The best fit of this model to the experimental electron density map is shown in , panels d-f. This fit gives a correlation between simulated and experimental electron density maps of 0.74, and places MxiGC near the inner membrane and the MxiN peptide in the density region close to the spoke. Importantly, the position of each protein component is shifted upward toward the membrane relative to the current placements in most models of the SP [,] (Fig SA in ). The region of density commonly designated as arising from MxiK is occupied by two copies of Spa33N in this model, with MxiK and the MxiGC domains occupying a common region of density closest to the inner membrane. Attempts to fit our model into alternative positions in the density map give poor correlation values and do not visually appear to fit the density well. Positioning the model with MxiK in the region of density typically thought to contain SctK gives a correlation value of 0.41 (Fig SG in ). The model presented in leaves the lower third of the pod density unoccupied. This density must arise, at least in part, by a dimer of the SctL spoke protein MxiN, which must be able to connect the central ATPase to Spa33C. Unfortunately, AlphaFold does not predict the structure of the MxiN dimer with high confidence and will often predict different structures for different seeds or when the Spa33C binding partner is present (Fig SH in ). We are, therefore, unable to model this portion of the density. Assuming, however, that there is a bend in the MxiN dimer, such as that shown in the AlphaFold structure in Fig SHc in , it is possible to connect the Spa47 ATPase to Spa33C. shows a hypothetical conformation of the MxiN dimer that connects Spa33C to Spa47. Here, each MxiN monomer extends downward from Spa33C until it curves toward Spa47 around residue 57 and forms an extended coiled-coil alpha helix. This is not to propose that this is the actual conformation of MxiN, but only to show that the position of Spa33C in is not too distal to attach to the spoke. Cryo-EM structures of pod mutant injectisomes To understand how interfacial mutants affect the T3SA, we used cryo-ET to compare in-situ structures of mutant injectisomes to a non-secreting T3SA in which the Spa47 ATPase is inactive (Spa47K165A) []. Minicells of Δspa33 or ΔmxiK S. flexneri complemented with spa33F67A/F70A, spa33F70D, spa33F219D/Y221A, or mxiKL46D were prepared and visualized by cryo-ET. Sorting platforms were then compared through their averaged structures. Prior to averaging, individual injectisomes in tomograms were classified according to the presence or absence of either extracellular needles or any indication of sorting platform density. Fewer than 1% of injectisomes had extracellular needles, indicating they are largely non-secreting. Classification of subtomograms based on sorting platform density is more difficult than the presence of needles. Here, we classified injectisomes with any visual indication of sorting platform density, no matter how faint, as sorting platform positive (SP+), and those without as sorting platform negative (SP-). Pod mutant injectisomes were less likely to be SP+ than the Spa47K165A mutant (88% SP+), ranging from 65% SP+ (Spa33F70D) to 37% (Spa33F219D/Y221A). Subtomogram-averaged structures also indicate that pod mutant injectisomes are inactive. The basal body of all mutant structures, and Spa47K165A, appear in a “closed” conformation, with no visible inner rod assembly and the presence of a septum structure blocking the channel (). Unlike Spa47K165A, which has a fully intact sorting platform, none of the averaged pod mutant structures have any density for Spa47 or the MxiN spokes. Interestingly, some pod density is apparent for each mutant sorting platform. Going by the model presented here in , all Spa33/MxiK interaction mutants have regions of upper pod density corresponding to MxiK and Spa33, as well as density for the lower third of the pod. The density of the Spa33short interaction mutant Spa33F219D/Y221A is less resolved in the lower third of the pod. These data suggest that mutation of the Spa33/MxiK interface or of the Spa33/Spa33short interface destabilizes the pod structure such that, while pods may still transiently form, they are unable to form spokes or recruit the Spa47 ATPase. Because of a sixfold radial symmetry operation performed on the data during processing, we cannot state precisely how many pods are present in the mutant sorting platforms, but the weaker pod density relative to Spa47K165A suggests fewer than the full complement of six pods. Incorporation of additional copies of Spa33 into pod structure The presence of significant density for the lower third of the pod without visible spokes in the Spa33/MxiK interaction mutant injectisomes is intriguing because this region of density would be expected to arise, at least in part, from the spoke protein MxiN. One possibility is that MxiN is present on these mutant pods but is bound in a more compact form (similar to Fig SHa in ) when the ATPase is absent. However, as shown in , when the MxiN dimer is extended toward Spa47 it does not fill the density of the lower third of the pod. It is, therefore, likely that other protein molecules are present in this region. Given that previous studies have indicated the presence of four copies of Spa33 per pod, it is almost certain that a substantial portion of the lower pod density arises from additional copies of Spa33. When asked to predict the multimer structure of two copies of Spa33, AlphaFold produces a structure in which the Spa33C domain of each Spa33 monomer is bound to the other monomer in manner nearly identical to the Spa33short homodimer interaction (Fig SI in ), although the PAE values for this complex are much higher than those of the Spa33/Spa33short structure. Replacement of one of the Spa33short dimers in , with and additional copy of Spa33 gives a model for the pod that fits the upper density in the same orientation but now has an additional Spa33N domain available to fill a portion of the lower pod region (). This AlphaFold model has the lower Spa33N domain tightly complexed with the Spa33C domain of the other Spa33 molecule, a position that does not fit the density below very well () and may block the interaction of Spa33C with MxiN from below. However, this lower copy of Spa33N can fill some unoccupied density in the lower pod if it is rotated by approximately 180o and partially disengaging from Spa33C (). This reorientation may be facilitated by a more energetically favorable interaction with MxiN. Incorporation of a fourth copy of Spa33 by similar replacement of the remaining copy of Spa33short introduces a second copy of Spa33N into the lower pod density (). Again, AlphaFold has this Spa33N domain tightly complexed with Spa33C of the Spa33 molecule above and it does not fit into the lower pod density. While it may be possible to reorient this domain by simple rotation or translation to fit into the nearby lower pod density (arrow in ), this would leave nearly all the linker between the N- and C-terminal domains occupying a region of space in which there is no electron density. Alternatively, the fourth copy of Spa33 may be entirely contained within the lower pod region. But while it is possible to place an additional Spa33 N- and C-terminal domain into the lower density, doing so leaves little room for MxiN (Fig SA0 in ). Discussion Here we have used AlphaFold together with computational alanine scanning and in vivo mutant fitness characterization to create a model of the Spa33/MxiK portion of the S. flexneri sorting platform pod in which MxiK binds two copies of Spa33 via Spa33N, and each copy of Spa33 is complexed with a Spa33short homodimer between Spa33N and Spa33C (). This complex fits quite well into the upper portion of the experimental electron density map of the pod from cryo-electron tomography (cryo-ET). Notably, our proposed positioning of MxiK and Spa33 in the density map places these proteins closer to the inner membrane than is typically represented in sorting platform models (, Fig SA in ) [,,,]. Here, the two Spa33N domains occupy density that is generally shown as arising from MxiK, and MxiK shares the membrane-proximal density with MxiGC. However, this placement of the model gives better cross correlation values than any alternate positioning and visually fits the density map the best. Furthermore, our model explains the curious “donut hole” near the center of the pod electron density, as it arises from the space at the center of the Spa33/Spa33short tetramer. The four Spa33C/Spa33short domains fill the large section of density beneath the Spa33N domains and the Spa33C domains are positioned properly to bind the N-terminal peptide of MxiN from the region below. Finally, although we don’t have a good model of the MxiN homodimer structure, it is possible for MxiN to connect to Spa33C at its position in our model of the upper pod (). The 2:1 Spa33:MxiK ratio in the upper pod is also congruent with the rapid exchange of Spa33 into and out of the sorting platform. FRAP experiments in Yersinia indicate that SctQ proteins are frequently cycling in and out of the sorting platform [], and this is likely the case in all T3SS. In the case of a 1:1 Spa33:MxiK ratio, the disengagement of a Spa33 molecule would temporarily eliminate a pod and potentially destabilize the sorting platform. The presence of a second copy of Spa33 also binding to MxiK would allow for one copy of Spa33 to exchange out without eliminating the pod connecting the spoke to the inner membrane. While this model of the Spa33/MxiK portion of the pod fits the density well and accounts for two copies of Spa33, it leaves the lower portion of the pod density unmodeled and the two remaining copies of Spa33 believed to be present in the pod unaccounted for. Our cryo-ET data shows the presence of electron density in the lower pod region in the absence of density for the spoke protein MxiN. Furthermore, placement of a hypothetical model of MxiN that connects the Spa47 ATPase to Spa33C still leaves large regions of the lower pod density unfilled. Given the experimentally determined stoichiometry, the lower pod is very likely formed by additional copies of Spa33 together with two copies of MxiN. Unfortunately, we have no good model for MxiN, either as a monomer, a homodimer, or in complex with Spa33. AlphaFold does not fold MxiN reliably or with high confidence and predicts only a small portion of its N-terminus to interact with Spa33C. Attempts to incorporate additional copies of Spa33 into the lower density are, therefore, speculative. A third copy of Spa33 can be fit into the pod density by replacing one copy of Spa33short in the upper pod model with the C-terminal domain of the third Spa33 molecule. This arrangement places a Spa33N domain in the lower pod density. AlphaFold places this Spa33N in complex with the C-terminal domain of the Spa33 above it, an orientation that doesn’t fill the nearby density well and potentially interferes with the MxiN interaction with Spa33C. However, if we allow for rotation of this Spa33N. a rearrangement potentially induced by the presence of MxiN, it can fill the nearby density and frees Spa33C to interact with MxiN. In a similar manner, replacement of the remaining Spa33short in the upper pod model with the C-terminal domain of a fourth copy of Spa33 positions a second Spa33N near the lower pod density. Again, the AlphaFold model has this domain in complex with the domain above it and not completely in the experimental density map. While some rotation and/or translation of this domain may position it into the lower pod map, doing so would still leave the linker between Spa33N and Spa33C in a region where there is no electron density. Alternatively, the fourth copy of Spa33 may be located entirely within the lower pod density. A fourth copy of both Spa33N and Spa33C can be placed inside the lower pod density, but doing so leaves little room for MxiN. It is therefore unclear how a fourth copy of Spa33 is incorporated into the pod. One possibility is that a fourth copy of Spa33 is present only as Spa33N. Spa33N has been shown to be produced independently through transcriptional slippage [], and the absence of the added C-terminal domain would leave more room for MxiN. The presence of an alternately translated N-terminal domain together with three full-length copies of Spa33 would also satisfy the reported stoichiometry of four copies per pod, as that number was determined using Spa33 constructs that were N-terminally tagged with a fluorophore. And while this transcriptional slippage product is not required for type III secretion [], that does not preclude its presence in the wild-type sorting platform. That said, without an accurate model for MxiN or MxiN/Spa33 complexes, experimental structures are likely required to adequately account for all copies of Spa33 in the sorting platform pod. Materials and methods AlphaFold predictions and computational alanine scanning All predicted structures were made using the AlphaFold 3 server (aphafoldserver.com) or ColabFold []. For all proteins and protein complexes, the top five AlphaFold models are essentially identical, with the exception of MxiN, the structure of which AlphaFold does not accurately or consistently predict. The top-ranked structure of each AlphaFold job was relaxed with Amber to correct residual steric clashes. Relaxed AlphaFold multimer structures were used as inputs for BUDE alanine scanning mutagenesis via the BAlaS server to identify mutationally sensitive interface residues (https://pragmaticproteindesign.bio.ed.ac.uk/balas/) []. Strains, plasmids, and mutagenesis Δspa33 and ΔmxiK S. flexneri were obtained from C. Lesser (Massachusetts General Hospital, Cambridge, Massachusetts, USA) and A. Allaoui (Universite Libre de Bruxelles, Bruxelles, Belgium), respectively. Δspa33 and ΔmxiK strains were complemented by transformation with pUC18-derived pWPSF4 plasmid containing wild-type or mutant spa33 and mxiK genes []. Mutants were created using overlap extension PCR and cloned into pWPSF4 using New England Biolabs HiFi Assembly Master Mix (NEB, E2621S). Defibrinated sheep red blood cells were purchased from Colorado Serum Company, Denver, Colorado. Contact-mediated hemolysis, CR-induced secretion, and western blotting Δspa33 and ΔmxiK S. flexneri strains were transformed with pWPSF4 vectors carrying WT or mutant spa33 or mxiK genes and grown overnight on selective tryptic soy agar plates. The next morning colonies were pickied, transferred to 10 mL tryptic soy broth (TSB). For contact-mediated hemolysis [], cultures were grown at 37o C to an OD600 of ~0.6, centrifuged, and cell pellets resuspended in a volume of PBS to give an OD600 of 20. Three mL of defibrinated sheep red blood cells were diluted with 40 mL PBS, centrifuged, and resuspended in 3 mL of cold PBS. For each of five replicates, 50 μL of each Shigella sample were mixed with 50 μL of washed red blood cells in a single well in a round bottom 96 well plate, centrifuged for 10 minutes at 2000xg, and incubated for one hour at 37o C. 100 μL of PBs was then added to each sample and mixed thoroughly. Samples were then centrifuged again for 10 minutes at 2000xg. 100 μL of each sample supernatant was then removed and the absorbance at 545 nm was recorded. All A545 values were normalized to the average A545 of all WT samples. For CR induction [], cultures were grown at 37o C to an OD600 ~ 1.0, centrifuged, and cell pellets resuspended to an OD600 = 20 (approx.. 500 μL). CR was added to each cell suspension to a final concentration of 0.025% and cell suspensions incubated at 37o C for 20 minutes. Bacteria were then pelleted by centrifugation and supernatants aspirated to new tubes. Trichloroacetic acid was added to each supernatant sample to a concentration of 10% and samples incubated on ice for 30 minutes. Precipitated protein was pelleted by centrifugation at 14000xg for 15 minutes at 4o C. Precipitate pellets were washed twice with ice cold acetone. After removing acetone, pellets were dried by placing tubes in a 95o C heat block for 10 minutes. Dry pellets were resuspended in 40 μL PBS, 10 μL 5X SDS-PAGE loading buffer was then added, and samples boiled for 10 minutes. 20 μL of each sample was then separated on a 10% SDS-PAGE gel and blotted onto a nitrocellulose membrane. Each blot was treated with mouse serum containing anti-IpaB and anti-IpaD IgG antibodies, washed, and then incubated with anti-mouse secondary antibody labeled with Alexa Fluor 633. Bots were imaged on an iBright 1500 gel imaging station and band intensities quantified using ImageJ software. To monitor protein expression with western blotting, a polyhistidine tag was added to the N-terminus of all MxiK variants, and to the N- and C-termini of all Spa33 variants in pWPSF4 and transformed into a ΔmxiK or Δspa33 S. flexneri strain. T3SS expression was induced by growing all cells in 20 mL TSB at 37o C overnight with shaking at 200 rpm. The following morning, cell cultures were centrifuged for 10 minutes at 2000xg. Media supernatants were discarded, and cell pellets resuspended in 20 mL of histidine binding buffer (20 mM Tris, 500 mM NaCl, 15 mM imidazole, pH 8.0). Cells were lysed by brief sonication and lysate clarified by centrifugation at 30,000 xg for 30 minutes. Clarified lysate was passed over 1 mL of Ni-NTA resin, which was then washed with 20 column volumes of binding buffer. Captured his-tagged proteins were eluted in 3 mL of elution buffer (binding buffer with 500 mM imidazole). 25 μL of each eluted protein was mixed with 5 μL of 6X SDS-PAGE loading buffer and boiled for 5 minutes. 10 μL of each boiled sample was separated on a 12% SDS-PAGE gel. Gels were blotted onto a nitrocellulose membrane, blocked, and then incubated with mouse-derived anti-his IgG for 1 hour. Blots were then washed and incubated with an anti-mouse IgG labeled with Alexa Fluor 633 for 1 hour, washed again, and imaged on an iBright 1500 gel imager. Preparation of Shigella minicells for cryo-electron tomography (cryo-ET) The sample preparation of Shigella minicells for cryo-ET approach was followed previous description [,,].Shigella mutants were streaked onto Congo red plates and incubated at 37 °C overnight. An isolated single colony was inoculated into 10 mL of tryptic soy broth (TSB) and grown overnight at 37 °C with shaking. To generate minicells, 2 mL of the overnight culture was inoculated into 200 mL of TSB and grown at 37 °C with shaking to late log phase. Kanamycin (50 μg/mL), ampicillin (100 μg/mL), and spectinomycin (100 μg/mL) were supplied into all bacterial plates and media for bacterial growth. Minicells were purified by sequential centrifugation steps. The bacterial cultures were centrifuged twice at 2,000 x g for 10 min, followed by centrifugation of the supernatant at 20,000 x g for 10 min. Bacterial pellets were resuspended in 1mL of PBS, and then the resuspension was centrifuged at 2,000 x g for 5 min. The supernatant was then centrifuged at 15,000 x g for 5 min. The final bacterial pellet was resuspended in PBS. Prior to vitrification, 10 nm gold tracer solution (Aurion) was added to the purified minicell sample. Then, 5 μL of the mixture was deposited on glow-discharged cryo-electron microscopy (cryo-EM) grids. To create a thin layer of cryo-fixation specimens, filter paper was placed on the cryo-EM grids, and specimens were then frozen in liquid ethane and propane mixture using a homemade gravity plunger and Leica EM GP2 plunger. Data collection using cryo-ET The frozen specimens on cryo-EM grids were clipped by C-clip rings and C-clips (Thermo Fisher) and then loaded into a Titan Krios G2 microscope at Yale University. The microscope is equipped with a K3 summit direct detection camera and BioQuantum energy filter (Gatan). For imaging, the microscope was operated with 300kV acceleration voltage in almost -170 °C conditions. SerialEM was used to operate the microscope and acquire images []. For data collection by cryo-ET, the magnification for image recording was set at 42,000x, corresponding to 2.15Å physical pixel size, and the target defocus was set at ~-4.9 μm. To collect tilt series images, the dose-symmetric scheme in FastTomo script was used to tilt the stage to cover an angle ranging from -48°to +48°in 3° increments []. The total electron dose for each tilt series was almost 65e-/Å2, and the dose fraction mode was used to collect 10 frames for each tilt angle image. Cryo-ET data processing and subtomogram averaging Image drifting caused by electron beam exposure during image acquisition was corrected using MotionCor2 []. IMOD was then used to generate image stacks and to trace 10 nm fiducial beads to align all images in each tilt series [,].Gctf was used to estimate defocus values [], and then IMOD was used for contrast transfer function (CTF) correction. All tilt series were binned by a factor of 6 using the binvol function in IMOD []. Tomo3D was used to reconstruct 6x6x6 binned tomograms using the simultaneous iterative reconstruction technique (SIRT) [,]. 693, 186, 167, and 97 tomograms were reconstructed from Spa33F67A/F70A and Spa33F70D, MxiKL46D, and Spa33F219D/Y221A mutants, respectively. From these tomograms, 1863, 527, 528, and 142 injectisomes were selected from Spa33F67A/F70A and Spa33F70D, MxiKL46D, and Spa33F219D/Y221A mutants, respectively. For subtomogram averaging, 6x6x6 binned tomograms using weighted back projection (WBP) were reconstructed by Tomo3D, and i3 software was used to align the injectisome from each mutant [,]. For further structural analysis, unbinned subtomograms were extracted from tomograms based on the aligned position of the injectisome and subsequently binned by factors of 4 and 2 using the binvol function in IMOD. To compare structural phenotypes among the mutants, all injectisomes were aligned together using large masks. In-situ structures of the injectisome in each mutant were then determined based on the alignment positions of the injectisomes. Finally, each structure was refined using binned 2 subtomograms. For the fitting the predicted model, C6 symmetry was applied to a structure of the sorting platform structure of the wild type determined by binned 2 subtomograms to refine the structure. The sorting platform structure determined by unbinned subtomograms was then refined by focused refinement. Funding Statement The author(s) declare that financial support was received for the research and/or publication of this article. This work is supported by grant R01AI172097 (WLP) from National Institute of Allergy and Infectious Diseases (NIAID). Cryo-ET data were collected at Yale Cryo-EM Resources funded in part by NIH grant 1S10OD023603-01A1 (JL). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.