Hook‐basal‐body assembly state dictates substrate specificity of the flagellar type‐III secretion system

The assembly of the bacterial flagellum is orchestrated by the secretion of distinct early and late secretion substrates via the flagellar‐specific type‐III secretion system (fT3SS). However, how the fT3SS is able to distinguish between the different (early and late) substrate classes during flagellar assembly remains poorly understood. In this study, we investigated the substrate selectivity and specificity of the fT3SS of Salmonella enterica at different assembly stages. For this, we developed an experimental setup that allowed us to synchronize hook‐basal‐body assembly and to monitor early and late substrate secretion of fT3SSs operating in either early or late secretion mode, respectively. Our results demonstrate that the fT3SS features a remarkable specificity for only the substrates required at the respective assembly stage. No crosstalk of substrates was observed for fT3SSs operating in the opposing secretion mode. We further found that a substantial fraction of fT3SS surprisingly remained in early secretion mode. Our results thus suggest that the secretion substrate specificity switch of the fT3SS is unidirectional and irreversible. The developed secretion substrate reporter system further provides a platform for future investigations of the underlying molecular mechanisms of the elusive substrate recognition of the T3SS.

stood. In this study, we investigated the substrate selectivity and specificity of the fT3SS of Salmonella enterica at different assembly stages. For this, we developed an experimental setup that allowed us to synchronize hook-basal-body assembly and to monitor early and late substrate secretion of fT3SSs operating in either early or late secretion mode, respectively. Our results demonstrate that the fT3SS features a remarkable specificity for only the substrates required at the respective assembly stage. No crosstalk of substrates was observed for fT3SSs operating in the opposing secretion mode. We further found that a substantial fraction of fT3SS surprisingly remained in early secretion mode. Our results thus suggest that the secretion substrate specificity switch of the fT3SS is unidirectional and irreversible. The developed secretion substrate reporter system further provides a platform for future investigations of the underlying molecular mechanisms of the elusive substrate recognition of the T3SS.

K E Y W O R D S
flagella biosynthesis, motility, protein transport, substrate specificity switching, type-III secretion classified into two classes: early (rod/hook-type proteins) and late (filament-type proteins) (Minamino & Macnab, 1999). The fT3SS is composed of five core components (FliPQR, FlhBA) and an associated cytoplasmic ATPase complex (FliHIJ). FliPQR and FlhB form a unique helical pore complex that is primarily located within the periplasm above the inner membrane and conducts unfolded proteins into the narrow secretion channel (Johnson et al., 2019;Kuhlen et al., 2018Kuhlen et al., , 2020. The transmembrane domains of FlhA surround the pore complex in the inner membrane, while the large C-terminal domains of nine FlhA subunits form a nonameric ring underneath the pore in the cytoplasm (Abrusci et al., 2012;Terahara et al., 2018). FlhB is the socalled switch protein of the fT3SS that is involved in substrate specificity switching from early to late secretion substrates according to the flagellar assembly state. Substrate specificity switching was shown to require autocatalytic cleavage of the large C-terminal domain of FlhB (FlhB C ) at a highly conserved NPTH motif into the subdomains FlhB CN and FlhB CC (Ferris et al., 2005;Fraser et al., 2003;Minamino & Macnab, 2000;Monjarás Feria et al., 2015;Sorg et al., 2007;Zarivach et al., 2008). An autocleavage-deficient mutant (FlhB N269A ) is unable to secrete late secretion substrates (filament-type proteins) and hook-length is not controlled, resulting in the formation of polyhooks (Fraser et al., 2003). The switch in substrate specificity is induced after hook assembly is completed, presumably by an interaction of the molecular ruler protein FliK with FlhB C . When the hook reaches its final length of approximately 55 nm in Salmonella, FliK resides temporarily in the secretion channel through brief interactions between its N-terminal domain and the hook subunits (FlgE) and the hook cap (FlgD) (Minamino et al., 2009;Moriya et al., 2006). This pause in secretion places the C-terminal domain of FliK in close proximity to cleaved FlhB C , thereby allowing the interaction between the two domains and subsequent conformational rearrangements of the substrate docking platform to render the secretion apparatus competent for late substrate secretion (Ferris & Minamino, 2006;Kinoshita et al., 2017Kinoshita et al., , 2020Minamino, 2018;Minamino et al., 2020).
Here, we aimed to investigate the remarkable selectivity and specificity of the fT3SS in respect to early and late substrate recognition and secretion before and after hook completion. We therefore developed a synchronized secretion reporter platform that allowed us to analyze the substrate secretion profile of fT3SSs operating in early (before hook completion) or late secretion mode (after hook completion). We found that the fT3SS specifically secretes only the substrates required for HBB assembly either before and after hook completion, respectively. These results suggest that the fT3SS secretion substrate specificity switch is unidirectional.

| Establishment of a synchronized secretion reporter system
Flagellar assembly is tightly regulated and organized in a spatiotemporal manner, yet, assembly of multiple flagella in a single cell (e.g. in case of peritrichously flagellated bacteria such as Salmonella) is not synchronized (Chevance & Hughes, 2008;Macnab, 2003). A single Salmonella enterica (S. enterica) cell assembles on average 4-8 flagella (Aldridge & Hughes, 2002). Since the various flagella are in different assembly stages, the individual fT3SSs have to select the correct substrate proteins that are required at this particular stage of assembly out of the large pool of cytoplasmic proteins ( Figure S1a,b).
Thus, each individual fT3SS is able to discriminate between different substrate types (i.e. early and late) that are needed at different stages of its assembly (Macnab, 2003). We hypothesized that the substrate specificity of the fT3SS could be explained by two distinct mechanisms. First, the fT3SS might be able to switch its substrate recognition ability (i.e. its substrate specificity) between the different substrate types either reversibly or irreversibly. Second, the fT3SS might simply start with the ability to recognize and secrete early-type substrates (i.e. rod/hook-type) and subsequently, after termination of hook growth, enlarge its substrate recognition spectrum to also include late substrates (i.e. filament-type) in addition to the early substrates (rod/hook-type). In order to investigate these possibilities, we developed an experimental setup that allowed us to synchronize hook-basal-body (HBB) assembly at various stages, corresponding to fT3SSs operating in early or late mode. This allowed us to monitor (early and late) substrate secretion in a homogenous population ( Figure S1c) in contrast to previous studies Stafford et al., 2007). We defined the early mode as secretion of substrates during assembly of the HBB structure, and the late mode as secretion of substrates after HBB completion and substrate specificity switching of the fT3SS. In order to obtain a homogeneous population of HBBs with their fT3SSs operating in the early secretion mode, we utilized a hook-deletion mutant strain (ΔflgE).
A hook-deficient mutant is unable to induce the substrate specificity switch, which is thought to be mediated by an interaction between the molecular ruler FliK and the switch protein of the fT3SS, FlhB, upon hook completion (Ferris et al., 2005;Hirano et al., 1994;Williams et al., 1996).
The main challenge in obtaining a homogenous population of cells with all their fT3SSs operating in late secretion mode lies within the fact that conditions would need to be found, where all flagella would have completed HBB formation (and therefore have concluded the switch from early to late secretion mode). To overcome this problem, we synchronized flagellar assembly by placing the flagellar master regulator flhDC under control of an anhydrotetracycline (AnTc)-inducible promoter (P tetA -flhDC) (Karlinsey et al., 2000). We induced flhDC expression by addition of AnTc for 120 min, before we removed the inducer and continued bacterial growth for 90 min (pulse flhDC induction). We reasoned that this procedure allowed for enough time to complete HBB assembly of all initiated flagellar complexes (and, presumably, allow enough time for the substrate specificity switching to occur in all fT3SSs). To verify HBB completion and the switch in substrate specificity, we visualized assembled hooks We chose to label the hook as a marker for completed HBB assembly as it is the first structure protruding from the cell envelope during flagellar assembly. In order to determine the number of hooks that were assembled during a specific timeframe post induction of flhDC, we utilized a mutant of the hook subunit FlgE that harbored a surfaceexposed, accessible cysteine substitution (FlgE S171C ). This enabled us to stain the hook with different dyes at various times of the experiment using fluorophore-coupled maleimide dyes. Since an individual hook is made of ∼120 subunits of FlgE (Macnab, 2003), binding of the maleimide dye resulted in quantifiable fluorescent foci. During the initial 120 min pulse induction of flhDC (FlhDC+), we labeled the hooks with DyLight™ 550 Maleimide (t 1 , Alexa550) ( Figure 1a). After removal of the inducer, the cells were further incubated for 90 min to allow for HBB completion of all initiated complexes, and any newly formed hooks were masked by addition of Maleimide-PEG 2 -Biotin (Biotin-Mal). We masked with Biotin-Mal to be able to distinguish between hooks that assembled during flhDC induction and those that assembled after the time given for HBB completion. Finally, we labeled hooks that formed 90 min after removing the inducer with DyLight™ 488 Maleimide (t 2 , Alexa488). We reasoned that, if no new HBBs were detected after the final Alexa488 labeling, all HBBs in the analyzed population of cells would have had enough time to reach an identical assembly state during the 90 min incubation after inducer removal. To demonstrate the difference between synchronized and nonsynchronized HBB assembly, we continuously added AnTc to a control sample, resulting in constant induction of flhDC (Figure 1b). F I G U R E 1 Development of an experimental setup to synchronize flagellar assembly. The reporter strain harbors an inducible flagellar master regulator (P tetA -flhDC) and a surface-exposed, accessible cysteine residue within the hook subunit (FlgE S171C ) for labeling with fluorophore-coupled maleimide dyes. FlhDC expression was pulse-induced for 2 hr (a) or constantly induced (b). Top panels: Schematic of the experimental timeline indicating the induction time of FlhDC (FlhDC+), washing steps, the presumed time of hook-basal-body (HBB) completion, and the dyes used as well as the timepoints (t 1 , t 2 ) of sample acquisition for quantification of hooks. Middle panels: Representative fluorescent microscopy images of samples acquired at t 2 . We note that for pulse flhDC induction, a general cytosolic background fluorescence is observed in the Alexa488 channel. Scale = 2 µm. Bottom panels: Quantification of the number of hooks per cell (ℕ 0 ) at t 1 (magenta) and t 2 (grey) labeled with DyLight™ 550 Maleimide (Alexa550) and DyLight™ 488 Maleimide (Alexa488), respectively. Single cells are represented by individual data points (the number of analyzed cells are for pulse FlhDC induction; t 1 , n = 110, t 2 , n = 149 and for constant FlhDC induction; t 1 , n = 101, t 2 , n = 137). The average number of hooks per cell ± SD per timepoint is indicated. SD, standard deviation Fluorescent foci of labeled hooks were quantified after flhDC induction (t 1 , hooks labeled with Alexa550) and after HBB completion (t 2 , hooks additionally labeled with Alexa488) (Figure 1). On average, individual cells displayed 6.6 ± 2.2 (pulse AnTc induction) and 6.5 ± 2.2 (constant AnTc induction) hooks at time point t 1 , that is 120 min after the initial induction of flhDC. We next compared the number of HBBs per cell at t 2 for samples where we used pulse or constant induction of flhDC. We observed a substantial reduction in the number of hooks per cell (2.3 ± 1.1) for pulse-induced cells at t 2 , presumably because the cells continued to divide after inducer removal but stopped the synthesis of new HBBs (Figure 1a). In contrast, we observed on average 6.2 ± 2.6 hooks per cell at t 2 under conditions of constant induction of flhDC, which demonstrates that the cells continuously produced new HBBs during the duration of the experiment (Figure 1b).
Importantly, for pulse-induced cells we were unable to detect any newly synthesized hooks 90 min after inducer removal (t 2 , hooks additionally labeled with Alexa488). This suggested that incubation for 90 min after removing the inducer of flhDC expression allowed for enough time to complete assembly of all previously initiated HBB complexes. We note that HBBs that completed assembly during the 90 min incubation period could not be included in the analysis because of the addition of Biotin-Mal. However, control results indicate that the total number of hooks per cell did not differ substantially when masking was not performed ( Figure S2). For pulse-induced cells, only about 10% newly labeled HBBs were detected 90 min after inducer removal (t 2 , hooks additionally labeled with Alexa488) ( Figure S2a). This suggested that only a minor fraction of HBBs initiated and completed assembly during the Biotin-Mal masking timeframe. These results are consistent with the previously reported short half-life of the FlhDC complex (Claret & Hughes, 2000) and verified that we were able to control the number of hooks per cell by controlling transcription of flhDC. However, we note that we might underestimate the total number of HBBs due to restraints given by the diffraction limit of the used microscope. Next, we examined whether all HBB complexes had performed the switch to late secretion mode (i.e. able to secrete filament-type substrates). Therefore, we simultaneously visualized epitope-tagged hook subunits (FlgE 3×HA ) using anti-HA immunostaining (anti-HA-Alexa Fluor 555 [Alexa555]) and the filament (FliC T237C ) using a fluorophore-coupled maleimide dye (DyLight™ 488 Maleimide

F I G U R E 2
[Alexa488]), respectively. Negative effects of the HA epitope tag on hook polymerization were excluded by performing swimming motility assays and monitoring the switch from early to late substrate secretion using a previously described Class 3 promoter reporter plasmid (P motA -luxCDABE) ( Figure S3a,b) (Brown et al., 2008;Goodier & Ahmer, 2001). We reasoned that all HBBs, which had successfully switched to late secretion mode would have a filament attached, while those that remained in early secretion mode would have no filament attached. We utilized the same experimental setup as outlined above for Figure 1, but performed the labeling only after the time given for HBB completion (Figure 2a). We observed a surprisingly large fraction of hooks (21%) without attached filaments (Figure 2b,c). No unattached filaments were observed, which likely excludes mechanical shearing of the flagellar filaments. To support our result, we performed intentional mechanical shearing of filaments by passing the culture multiple times through a syringe needle (Renault et al., 2017;Turner et al., 2012) before labeling the hook and filament in order to evaluate the frequency of hook detachment. We found that the ratio of hook-attached filaments and afilamentous hooks was independent of the induced shearing force ( Figure S3c).
We thus concluded that a certain fraction of HBBs is defective in late substrate secretion for unknown reasons and are therefore unable to polymerize filaments. However, because no new HBBs were assembled after the time given for HBB completion and the majority of HBBs had switched to late secretion mode, we decided to employ this experimental setup in order to investigate the substrate specificity of fT3SS operating in early or late secretion modes.

| The fT3SS is highly specific in substrate recognition and secretion
The above-described setup allowed us to analyze next which substrates are recognized and secreted by the fT3SS operating in either early or late secretion mode. In order to investigate the secretion of early and late substrates, we used reporter strains with the following main properties: (a) the flagellar master regulator was placed under an AnTc-inducible promoter (P tetA -flhDC) to facilitate synchronization of fT3SS assembly as described above; (b) a deletion of the hook (early mode, ΔflgE) or the hook-filament junction (late mode, ΔflgKL) in order to ensure similar secretion levels within the synchronized population (we reasoned that filament assembly might slow-down the secretion of late substrates); and (c) individual early and late substrates were induced from a chromosomally encoded arabinoseinducible promoter (P araB -substrate) to uncouple each substrate from the transcriptional hierarchy and to increase substrate expression, which facilitated detection of secreted substrate protein by Western blot analysis (Figure 3a and Figure S1c). After pulse flhDC induction for 120 min (FlhDC+) and the HBB completion period of 90 min, we washed the cells again to remove all substrates that had been secreted into the supernatant up to this point ( Figure 3b). Afterwards, we induced expression of individual early or late substrates by addition of 0.2% arabinose (ara; substrate+) and acquired samples after 0, 30, 60, and 90 min to monitor their secretion profile over time.
As expected, early rod/hook-type substrates (we analyzed as representative substrates the hook cap protein FlgD and the molecular ruler protein FliK) were predominantly secreted in strains locked in the early secretion mode, but surprisingly, also to some extent in strains operating in the late secretion mode (Figure 3c, left panel).
In contrast, late filament-type substrates (we analyzed the hookfilament junction protein FlgK and the filament cap protein FliD as representative substrates) were exclusively secreted under late secretion mode conditions (Figure 3c, right panel).
These observations suggest that the fT3SS enlarges its substrate recognition spectrum upon switching from early to late secretion mode (while presumably being able to continuously secrete early substrates in addition to late substrates, although early substrate secretion was substantially reduced). However, we note that we observed a minority of hooks without attached filaments in the previous fluorescent labeling experiment (Figure 2b,c), which might also account for the minimal secretion of early substrates in late secretion mode fT3SSs.
We therefore designed an experimental setup that allowed us to specifically block secretion through late-mode fT3SSs by plugging their secretion pore. To this end, we introduced a FliC-GFP fusion protein into our strains of interest. It had previously been shown that the T3SS is incapable of unfolding GFP, and thus GFP fusions to secretion substrates are not secreted (Akeda & Galán, 2005;Singer et al., 2014). The late substrate and filament subunit FliC on the other hand is unfolded and injected into the secretion channel by the fT3SS. We reasoned that a FliC-GFP fusion protein would therefore allow us to plug the secretion pore only in fT3SSs which had switched their substrate specificity to the late secretion mode ( Figure 4a): FliC is recognized, unfolded, and injected into the secretion channel by a late mode fT3SS, while the bulky, folded GFP remains at the pore entrance. Thereby the secretion of any other substrate would be blocked. However, HBBs that persisted in early secretion mode would not be affected and we hypothesized that under these conditions the observed secretion of early substrates in the late secretion mode strain would not be altered. If, however, the early substrates were indeed secreted by an fT3SS that had switched to late secretion mode, their secretion would also be abolished by the FliC-GFP plug.
As shown in Figure 4b,c, expression of the FliC-GFP plug substantially decreased, but did not entirely abolish, secretion of the late substrate FlgK. This could be explained by the fact that the plug also inhibits FlgM secretion from the cell, which subsequently prevents These results demonstrate a remarkable specificity for early or late substrates of fT3SSs that are operating in either early or late secretion mode, respectively. Residual early substrate secretion in the late mode strain is likely due to a fraction of fT3SSs that remained in early secretion mode and the artificial increase of substrate availability by overexpression via the P araB promoter.

| D ISCUSS I ON
T3SSs are able to selectively secrete different classes of secretion substrates at different assembly stages of the macromolecular F I G U R E 3 Specificity of early and late substrate secretion in fT3SSs operating in early or late secretion mode. (a) Schematic representation of the main features of the used secretion reporter strains. Strains were deleted for the hook (ΔflgE) or the hook-filament junction (ΔflgKL) proteins to obtain a population of cells locked in early or late mode of secretion, respectively. The flagellar master regulator flhDC was placed under an anhydrotetracycline-inducible promoter (P tetA -flhDC) and individual early and late substrates were expressed from the chromosomally encoded P araB promoter (P araB -substrate). (b) Schematic of the experimental timeline indicating the time of FlhDC induction (FlhDC+), washing steps, the presumed time for hook-basal-body (HBB) completion and the induction of substrates from the P araB promoter by addition of arabinose (ara). Samples were acquired 0, 30, 60, and 90 min after substrate induction. (c) Secreted and cellular levels of early (left) and late substrates (right) in early and late secretion mode strains. The induced substrate from the P araB promoter is displayed on the top left-hand side of each representative Western blot. A protein size marker in kDa is indicated in the center of each Western blot. DnaK served as a cell lysis and loading control and was only detected in the cellular fraction. Individual proteins were detected using polyclonal (α-FlgD, α-FliK, α-FlgK, α-FliD) or monoclonal (α-DnaK) antibodies machine. However, the mechanism of this remarkable substrate selectivity remains unclear. We investigated here in particular the question of substrate selectivity of the flagellar T3SS. We reasoned that two different mechanisms could account for the previously observed sequential secretion of (early then late) substrates via the fT3SS. Firstly, it appears possible that the substrate specificity switch results in conformational rearrangements within the export apparatus that change the substrate selectivity of the fT3SS completely from one substrate class to another: that is before the switch, the fT3SS recognizes only early substrates, and after the presumed hook-type to filament-type substrate secretion by suppressing the interaction of late substrate-chaperone complexes with FlhA C before hook completion (Inoue et al., 2021). Secondly, it also appears possible that conformational rearrangements within the export apparatus simply enlarge the substrate selectivity spectrum of the fT3SS so that it additionally recognizes and secretes a broader range of substrates: that is, after the switch, the fT3SS recognizes and secretes late substrates in addition to early substrates. Such a switch mechanism intuitively appears to be simpler mechanistically but might also waste more cellular resources as F I G U R E 4 fT3SSs operating in late mode specifically secrete late substrates. (a) Schematic representation of the mode of action of the FliC-GFP approach in order to plug fT3SSs operating in late secretion mode. (b and d) Secreted and cellular protein levels of the late substrate FlgK (b) and the early substrate FliK (d) in early and late mode strains expressing the FliC-GFP plug. (c and e) Quantification of three independent biological replicates of the exemplary secretion assays shown in panels b and d, respectively. A protein size marker in kDa is indicated on the right-hand side of each representative Western blot. DnaK served as a cell lysis and loading control and was only detected in the cellular fraction. Individual proteins were detected using polyclonal (α-FlgK, α-FliC, α-FliK) or monoclonal (α-DnaK) antibodies early substrates would continue to be secreted, although no longer required for flagellar assembly post-switch.
Accordingly, in order to investigate these possibilities, we developed a synchronized secretion substrate reporter setup that allowed us to test secretion of the different substrate classes in fT3SSs operating in either early or late secretion mode (Figures 1 and 2). This approach differs from other studies which investigated late substrate secretion in a heterogenous population Stafford et al., 2007).
Using the synchronized reporter setup, we found that fT3SSs operating in early mode were able to secrete only early-type substrates, whereas strains where the fT3SSs operated in late mode predominantly secreted late-type substrates ( Figure 5). Surprisingly, we observed in the late-mode strains also a low level of early substrate secretion (Figure 3). Importantly, however, our initial experiments characterizing the synchronized reporter setup revealed that about 20% of the fT3SSs assembling a hook structure apparently did not switch to late secretion mode because these HBBs did not assemble a filament (Figure 2). Since we frequently observed individual cells, which displayed both filament-attached HBBs and HBBs without a filament, we concluded that not a general lack of late secretion substrate expression accounts for the defect in filament assembly. We speculate that during the HBB construction process a certain probability exists for assembly mistakes that render the fT3SS unable to undergo the conformational changes needed for the substrate specificity switch and thus complete hook assembly. This conclusion appears to be reasonable since the correct spatiotemporal assembly of complex, multicomponent macromolecular machines such as the flagellum is nontrivial. Hence, bacteria evolved sophisticated regulatory mechanisms in order to maximize the probability of a productive assembly, such as hierarchical gene expression of various substrates needed at different stages of assembly, different secretion substrate classes, as well as checkpoints coupling the assembly state to gene expression or substrate selectivity.
Accordingly, we conclude that the residual secretion of early substrates in the synchronized late mode strain is likely due to a fraction of fT3SSs that were unable to perform the switch to late secretion mode. Further evidence supports the conclusion that a minor fraction of fT3SSs remained in early secretion mode. We found that secretion of late-type substrates, but not early-type substrates, was prevented in the late mode strain by expressing a flagellin (late substrate)-GFP fusion protein, which specifically blocks secretion after the switch (Figure 4). This suggests that the switch to late mode secretion is irreversible since the residual early substrate secretion under late mode conditions was due to a minor fraction of fT3SSs that were unable to switch to late mode.
In summary, our results support a model of the substrate selectivity mechanism where the fT3SS undergoes a complete, unidirectional (i.e. from early-to late-type substrates) and irreversible (i.e. fT3SS in late secretion mode do not, or very infrequently, revert to early secretion mode) switch in substrate specificity ( Figure 5). Such a mechanism, where after the switch, the fT3SS is unable to secrete early-type substrates would facilitate selection and secretion of only those substrates required at a specific assembly stage under conditions where multiple flagella are assembled simultaneously but not synchronously within one cell.
Finally, the developed synchronized secretion substrate reporter strain provides a platform that will promote future investigations of the molecular mechanisms that underlie the elusive substrate recognition and substrate specificity switch of the fT3SS.

| Synchronized secretion assay
An overnight culture of FliC phase-locked bacteria was inoculated 1:100 in LB supplemented with 100 ng/ml AnTc and incubated at 37℃, 180 rpm for 2 hr. Cells were washed with 1×PBS after centrifugation and resuspended in fresh LB after an additional centrifugation step. Incubation was pursued for 1.5 hr, and cells were washed again with 1×PBS as described before. The bacterial cell number was adjusted equally to OD 600 = 0.8 in all mutants according to the optical density measured at 600 nm (OD 600 ). The individual substrates from the P araB locus were induced with 0.2% arabinose. A total amount of 1.9 ml culture was harvested at the indicated times. Cellular and supernatant fractions were collected by centrifugation (13,000×g, 3 min, 4℃). Proteins were precipitated by addition of 10% TCA and resuspended in 2×SDS sample buffer with 2% β-mercaptoethanol for subsequent protein analysis by SDS-PAGE and Western blotting.

| Maleimide fluorophore labeling of the hook
In order to visualize the assembled hook structures, we used strains harboring a cysteine substitution within the hook protein FlgE (FlgE S171C ). The assay was performed as described for the synchronized secretion assay above, and the experimental timeline is presented in Figure 1 and Figure S2. Fifteen minutes before the first incubation step ended, 500 µl culture was transferred into a microcentrifuge tube. Cells were pelleted at 2,500×g for 3 min and After washing, the cells were applied to a homemade flow-cell and visualized at 100× magnification by epifluorescence microscopy (t 1 ). The remaining sample was centrifuged again, resuspended in fresh LB, and incubated for 75 min. The cells were pelleted, resuspended in 1×PBS, and 1 mM Maleimide-PEG 2 -Biotin (Thermo Fisher Scientific, A39261) was added for masking newly formed F I G U R E 5 Schematic model of substrate selectivity of the fT3SS. The fT3SS features two distinct secretion modes during assembly: (a) early and (b) late mode. The ability to secrete substrates (↑) or not (×) is indicated. (a) The early secretion mode is defined as the ability of the fT3SS to secrete early substrates (blue) before hook completion. The early substrates include, among others, the hook-basal-body (HBB) components and the molecular ruler FliK. Inlay panel: Schematic representation of the fT3SS displaying the FliPQR-FlhB complex (shades of green) and the nonameric FlhA C ring (orange). FlhB C (light green) undergoes autocatalytic cleavage into the subdomains FlhB CN and FlhB CC before incorporation into the export apparatus. FlhB CC provides binding sites for early substrates, while binding sites for late substrates on FlhA C are occluded during HBB assembly. The early mode exclusively secretes early, but not late substrates. (b) Upon HBB completion, a switch in substrate specificity occurs (indicated by a yellow star), which initiates the late mode. Inlay panel: Conformational rearrangements of the substrate docking platform formed by FlhB C and FlhA C result in a highly specific and irreversible switch in substrate selectivity. The fT3SS switches from early to late substrate secretion by prohibiting early substrate recognition and concurrently providing binding sites for late substrates (magenta) in complex with their cognate chaperones (light magenta) on FlhA C . Conformational rearrangements are indicated by a change in shape and color (FlhB CC , light green to dark pink; FlhA C , orange to dark purple). IM, inner membrane; OM, outer membrane; PG, peptidoglycan HBBs. The cells were incubated for 10 min, subsequently pelleted again, and resuspended in 1×PBS. The wash step was repeated, and cells were resuspended in fresh LB. Incubation was pursued for 75 min. The cells were pelleted, resuspended in 1×PBS, and 10 µg/ ml DyLight™ 488 Maleimide (Thermo Fisher Scientific, 46602) was added to label newly formed HBBs. The sample was incubated for 10 min and fixation as well as visualization was performed as described above (t 2 ).

| Simultaneous hook and filament labeling
For labeling the hook and filament structures simultaneously, we used a strain harboring a 3×HA epitope tag inserted into the hook protein (FlgE 3×HA ) and a previously described cysteine mutation on the filament subunits (FliC T237C ) for labeling with fluorophorecoupled maleimide dyes. The basic assay was performed as described above for the synchronized secretion assay. For labeling of the hook and filament, 500 µl culture was transferred into a microcentrifuge tube after the last incubation step. The cells were pelleted at 2,500×g for 3 min and resuspended in 1×PBS. 10 µg/ml anti-HA-Alexa555 (Thermo Fisher Scientific, 26183-A555), and DyLight™ 488 Maleimide (Thermo Fisher Scientific, 46602) was added for sufficient labeling of both structures for 30 min with continuous shaking. Subsequently, the cells were washed twice with 1×PBS and fixed with PFA at a final concentration of 2% for 10 min at room temperature. After washing, the cells were applied to a homemade flow-cell and visualized at 100× magnification by epifluorescence microscopy. To exclude shearing of flagellar filaments due to harsh handling, samples were treated with different roughness as described before (Renault et al., 2017;Turner et al., 2012). Before labeling, the culture was passed 0, 10 (mild shearing), or 30 times (strong shearing) through a 22-gauge needle to cause mechanical shearing of flagellar filaments.

| Motility assay
Swimming motility assays were performed by inoculating 1 µl overnight culture in a 0.3% soft agar plate. The plate was incubated for 5 hr at 37℃, and the diameter of the swimming halo was measured using ImageJ. Values were normalized to the wild-type strain.

| Luciferase assay
Analyzed strains harbored the pRG19 plasmid encoding for a transcriptional fusion of the luxCDABE operon to the Class 3 promoter of motA (P motA -luxCDABE) (Brown et al., 2008;Goodier & Ahmer, 2001), allowing to estimate the timing of switching from early to late secretion mode. An overnight culture was diluted 1:100 in 200 µl LB supplemented with 12.5 µg/ml chloramphenicol in a white walled 96-well plate. Luminescence and OD 600 were measured for 6 hr in 15 min intervals in a Synergy H1 microplate reader (BioTek).

ACK N OWLED G M ENTS
The authors thank Tohru Minamino for providing them with poly-

CO N FLI C T O F I NTE R E S T
The authors declare no competing interests.

AUTH O R CO NTR I B UTI O N S
AG and ME conceived and designed the study. AG, MH, and SH performed the experiments. AG, MH, SH, and ME analyzed the data. ME contributed reagents, materials, and funding. AG and ME wrote the paper. All authors edited the paper.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.