PGM activity in human cells is carried out by the enzyme PGM1. PGM1 deficiency leads to defects in glycogen storage, and to several disorders resulting from insufficient protein glycosylation (Stojkovic et al., 2009; Tegtmeyer et al., 2014). To test for the presence of this enzyme in C. trachomatis inclusions we expressed PGM1 with either an amino-terminal (Flag-PGM1) or a carboxy-terminal (PGM1-Flag) Flag tag in HeLa cells, which were further infected with C. trachomatis L2. The cells were fixed 24 h later and stained with a mouse antibody against the Flag tag and a rabbit antibody against the inclusion protein Cap1. An irrelevant Flag-tagged protein (Flag-CymR) was included as a negative control, and Flag-tagged chlamydial glycogen synthase (Flag-GlgA), as a positive control (Gehre et al., 2016). Flag-CymR, Flag-PGM1 and PGM1-Flag had a similar distribution ( Figure 1A ). Only very few dots could be observed in the inclusion lumen. In places where a more substantial overlap with the inclusion was observed, it was in areas positive for Cap1 staining, suggesting a cytoplasmic distribution adjacent to the inclusion membrane (see for instance the PGM1-Flag staining in Figure 1A ).In contrast, as previously reported, we could detect Flag-GlgA in the inclusion lumen, presumably because it is co-transported with host glycogen into the inclusion lumen (Gehre et al., 2016). One caveat of this experiment is that Flag-GlgA was expressed at higher level compared to the other three constructs, which facilitates its detection in the inclusion lumen. Altogether, these observations indicate that PGM1 is not translocated in the inclusion lumen. We next tested whether silencing PGM1 expression had any effect on bacterial development. Cells were infected 24 h after siRNA transfection with two different siRNA (sipgm1_1 and sipgm1_2), and the resulting progeny was measured. The efficacy of each of the siRNA against PGM1 was validated by western blot ( Figure 1B ). None of the two siRNAs tested had a significant effect on the progeny collected 48 hpi ( Figure 1C ). We concluded from this series of experiments that host PGM1 does not reach the inclusion lumen, at least not in detectable amount, and is not essential for the generation of infectious bacteria. Therefore, it is unlikely that the conversion of Glc1P into Glc6P in the inclusion lumen relies on the host PGM activity by PGM1.

CT295 and CT815 are two C. trachomatis proteins that belong to the eggNOG cluster of orthologous groups COG1109 [ Figure 2A (Huerta-Cepas et al., 2016)]. Members of this large group of bacterial, archaeal, and eukaryotic proteins display PGM activity and related functions such as phosphomannomutase or phosphoglucosamine mutase activities (Stiers et al., 2017). CT295 (gene annotated as mrsA_1) and CT815 (mrsA_2) homologs are found among all known chlamydiae. These proteins represent separate monophyletic clades in COG1109 and are related to proteins from diverse bacterial taxa ( Figure 2B ). CT295 and CT815 are only distantly related with each other and thus do not originate from a recent gene duplication event ( Figure 2C ).

To test if CT295 and/or CT815 displayed PGM activity we expressed them in Escherichia coli with a carboxy-terminal histidine tag. However, CT295-HIS was not expressed by E. coli, and after testing the orthologous protein in different Chlamydiaceae we chose the C. caviae ortholog, CCA00344, which displayed 59% amino acid identity with CT295. CCA00344-HIS and CT815-HIS were purified from E. coli extracts ( Figure 3 and Figure S1A ). As a positive control we also produced the E. coli phosphomannomutase CpsG, because our initial attempts to produce the E. coli PGM failed, and CpsG turned out to display good PGM activity in vitro. To test the ability of these recombinant proteins to convert Glc1P into Glc6P we first incubated them with Glc1P in the presence of glucose 1,6-diPhosphate (Glc1,6diP), a cofactor for the PGM reaction in E. coli (Ray and Roscelli, 1964). The samples were incubated for 20 min at 37°C, before stopping the reaction by boiling for 5 min. Sugars present in the samples were analyzed by high performance anion exchange chromatography with pulsed amperometric detection (HPAEC-PAD) ( Figure 3 and Figure S1B ). Incubation of Glc1P in the presence of CCA00344-HIS led to its full conversion into Glc6P, while only low levels of Glc6P were detected in the presence of CT815-HIS ( Figure 3A ). Interestingly we noted that while we confirmed that CpsG-HIS required the presence of Glc1,6diP to convert Glc1P into Glc6P, CCA00344-HIS did not, indicating that the chlamydial enzyme evolved to bypass co-factor requirement ( Figure 3 and Figure S1C ). Also, Glc6P appeared as two peaks (a and b in Figure 3A ), which correspond to the stereoisomers α-Glc6P and β-Glc6P. This indicated that CCA00344-HIS favored the conversion of α-Glc6P into β-Glc6P, a property that we did not observe in CT815 nor CpsG ( Figure 3 and Figure S1C, D ).

To circumvent the lack of expression of CT295 in the bacterial system, we turned to a mammalian expression system. We cloned CT295 and CT815 with an amino-terminal Flag tag in a mammalian expression plasmid and expressed it in HEK293T cells. Flag-CymR expression plasmid served as a negative control. One day after transfection the cells were lysed, Flag-tagged proteins were immobilized using beads covalently conjugated to anti-Flag antibodies, and tested for their ability to convert Glc1P into Glc6P. The reaction products were analyzed as described above. Glc6P was only detected in the samples incubated with Flag-CT295 loaded beads, not with Flag-CT815 nor with the negative control ( Figure 3B ). These data demonstrated the ability of CT295 to convert Glc1P into Glc6P.

When we monitored the reverse reaction, supplying Glc6P and following the appearance of Glc1P, very low amounts of Glc1P were detected after incubation with CCA00344-HIS or CpsG-HIS, and none with CT815-HIS ( Figure 3C and Figure S1D ). This indicated that, in vitro, the conversion of Glc1P into Glc6P was more favorable than the reverse reaction. To confirm that CT295 was nonetheless capable of catalyzing the reverse reaction, we made use of human fibroblasts deficient for PGM1. One of the phenotypes associated with PGM1 deficiency is a defect in protein glycosylation, due to the inability to convert Glc6P into Glc1P to feed UDP-Glc synthesis and thereby protein glycosylation (Tegtmeyer et al., 2014). As a consequence, several glycosylated proteins like the cell-surface glycoprotein intercellular adhesion molecule 1 (ICAM-1) are hardly expressed at the cell surface in PGM1 deficient cells (Tegtmeyer et al., 2014). ICAM-1 expression can thus be used as a read-out for PGM activity. As expected, ICAM-1 was hardly detected at the surface of PGM1 deficient fibroblasts ( Figure 3D ). Expression of Flag-CymR had a marginal effect on ICAM-1 expression. In contrast, as expected, transfection of the wild-type human PGM1 restored ICAM1 expression in the PGM1 deficient fibroblasts. Transfection of CT295 also restored ICAM-1 expression, confirming its ability to catalyze the conversion of Glc6P into Glc1P. Transfection of CT815 also resulted in increased ICAM-1 expression, but to a lesser extent, supporting the hypothesis that while CT815 displays some PGM activity, it is likely that it uses a different sugar than glucose as a favorite substrate ( Figure 3D ).

Altogether, these data demonstrate that CT295, and its ortholog in C. caviae, are bona fide PGMs.

If CT295 is the PGM that converts Glc1P into Glc6P in the inclusion lumen it needs to be secreted out of the bacteria. We have shown that C. trachomatis glycogen metabolism enzymes possessed secretion signals recognized by the T3S machinery of S. flexneri (Gehre et al., 2016). To explore if this was also true for CT295, we tested for the presence of a functional T3S signal within its amino-terminus using the heterologous secretion assay in S. flexneri as described previously (Subtil et al., 2001). Briefly, we fused these amino-terminal sequences to the reporter protein calmodulin-dependent adenylate cyclase (Cya). The chimeric proteins were then expressed in S. flexneri ipaB (constitutive T3S) or mxiD (deficient in T3S) null strains. When we analyzed the culture supernatant versus the bacterial pellet, we found the Cya reporter in the supernatant of ipaB cultures ( Figure 4A ). The same expression pattern was observed for an endogenous Shigella T3S substrate, IpaD. Conversely, we found the cAMP receptor protein (CRP), a non-secreted protein, exclusively in the bacterial pellet. This control shows that Cya detection in the supernatant was not due to bacterial lysis. Finally, the Cya reporter was not recovered in the supernatants of the mxiD strain, demonstrating that its secretion was dependent on the T3S system.

We transformed C. trachomatis with an expression plasmid constitutively expressing the mCherry protein, as well as CT295 with a carboxy-terminal 3xFlag tag under the control of an inducible promoter, as an initial attempt using the endogenous promoter showed no protein expression. Upon induction with anhydrotetracycline (aTc) a weak band was detected by western blot on whole cell lysates with anti-Flag antibodies at the expected size ( Figure 4B ). By immunofluorescence, anti-Flag antibodies displayed a weak signal with punctate distribution. CT295-Flag was detected in areas also positive for mCherry or DAPI, indicating that only intra-bacterial CT295-Flag was detected. Thus, this approach could not demonstrate the secretion of CT295-Flag in the inclusion lumen, possibly because it is not sensitive enough to detect the enzyme diluted in this large volume.

Since C. trachomatis GlgA is detected both in the inclusion lumen and in the cytoplasm (Lu et al., 2013) we considered the possibility that CT295 could be translocated in the host cytoplasm. We did not detect CT295-Flag in the cell cytoplasm at 24 hpi ( Figure 4C ) nor at 40 hpi ( Figure 4D ). Bacteria expressing GlgA-Flag under the control of its endogenous promoter, and constitutively expressing the green fluorescent protein (GFP) were used for comparison. As expected, a cytoplasmic staining was observed for GlgA-Flag, although only in ~ 10% (n=150 infected cells from 2 independent experiments) of the infected cells, possibly due to low sensitivity of the assay ( Figure 4D ).

As an alternative approach to test for CT295 presence in the host cytoplasm we tested whether infection could restore PGM activity in the cytoplasm of PGM1 deficient fibroblasts. Indeed, if CT295 reached the cytoplasm, it might lead to at least a partial recovery of ICAM-1 glycosylation. Cells transfected with Flag-PGM1 served as a positive control. We observed that ICAM-1 expression, which was undetectable at the surface of PGM1 deficient fibroblast, remained absent from the surface of cells infected for 40 h with C. trachomatis ( Figure 4E ). One pitfall of this experiment is that C. trachomatis L2 takes up UDP-glucose from the host cytoplasm to accumulate glycogen in the inclusion lumen (Gehre et al., 2016). Therefore, the presence of a bacterial PGM in the cytoplasm might not be sufficient to restore ICAM-1 glycosylation. We thus also used the plasmid-less L2(25667R), which is less likely to deplete the host UDP-glucose stores, as it does not accumulate glycogen in the inclusion (Carlson et al., 2008). Infection with L2(25667R) did not restore ICAM-1expression at the cell surface either ( Figure 4E ). These experiments suggest that CT295 is translocated directly into the inclusion lumen, without an intermediate cytoplasmic step.

We next asked whether the secretion of a PGM was a conserved feature among Chlamydiaceae. Previous studies have reported glycogen accumulation in C. suis and C. muridarum inclusions, but the polysaccharide was not observed in the inclusions of more phylogenetically distant Chlamydia species such as C. caviae or C. pneumoniae. We verified this feature by performing periodic acid-Schiff staining (PAS) on cells infected with these different species. Inclusion staining was observed only on cells infected with C. trachomatis, C. suis and C. muridarum ( Figure 5A ). Two amino-terminal amino-acids (W23L24) are conserved in the orthologs of CT295 in all chlamydiae, and the preceding amino acids show low levels of conservation ( Figure 5B ). To test whether these different peptides also function as T3S signals in other species than C. trachomatis we constructed chimeras between the amino-terminal 22 amino-acids of the different CT295 orthologs and the Cya reporter gene and tested their secretion when expressed in the S. flexneri ipaB strain. The chimera made from the amino-terminal segment of C. suis and C. muriarum PGMs were recovered in the supernatant, but not those originating from C. caviae or C. pneumoniae. Secretion was dependent on the T3S system since the chimera were not recovered in the supernatant when expressed by the S. flexneri mxiD strain ( Figure 5C ). Thus, the accumulation of glycogen in the inclusion of Chlamydiaceae correlates with their ability to secrete a phosphoglucomutase.

HeLa and HEK293T cells (ATCC) were cultured in Dulbecco’s modified Eagle’s medium (DMEM) with Glutamax (Gibco™ #31966), supplemented with 10% (v/v) fetal bovine serum (FBS). Fibroblast CDG_0072 deficient in PGM1, a kind gift from Pr. Hudson Freeze (La Jolla, CA), were grown in DMEM (Gibco™ # A1443001) supplemented with 1 g/L glucose, 584 mg/mL L-glutamine, 110 mg/mL sodium pyruvate, and 10% FBS. Cells were routinely checked for absence of mycoplasma contamination by PCR. C. trachomatis LGV serovar L2 strain 434/Bu (ATCC), the plasmid-less strain LGV L2 25667R (Peterson et al., 1990) and GFP-expressing L2 (L2^IncDGFP) (Vromman et al., 2014) were purified on density gradients as previously described (Scidmore, 2005). Other Chlamydia strains used were C. suis SWA107, a kind gift from Nicole Borel (Zürich, Switzerand), C. pneumoniae CWL029, C. muridarum MoPn and C. caviae GPIC. The Shigella flexneri ipaB and mxiD strains are derivates of M90T, the virulent wild-type strain, in which the respective genes (ipaB and mxiD) have been inactivated (Allaoui et al., 1993). The Escherichia coli strain DH5α was used for cloning purposes and plasmid amplification. Both S. flexneri and E. coli strains were grown in Luria-Bertani (LB) medium.

Protein sequences of CT295 and CT815 were analyzed with eggNOG v4.5.1 (Huerta-Cepas et al., 2016). They are both members of COG1109 at the last universal common ancestor (LUCA) taxonomic level (including all domains of life). At the bacteria-specific eggNOG clustering, CT295 and CT815 belong to two distinct EggNOGs, ENOG4107QSU and ENOG4107QJF. To identify closely related sequences, we used eggNOG-mapper v1.0.1 (Huerta-Cepas et al., 2017) with the bacteria optimized database using the “–database bact” option and default settings on a broad selection of published chlamydial genomes. We downloaded all COG1109 sequences (n=4,829; http://eggnogapi45.embl.de/nog_data/text/fasta/COG1109), added all chlamydial ENOG4107QSU (n=104) and ENOG4107QJF (n=69) sequences and aligned them using the COG1109 hidden markov model (http://eggnogapi45.embl.de/nog_data/file/hmm/COG1109). To reduce redundancy we clustered the alignment with the cd-hit program (Fu et al., 2012) at 80% sequence identity over at least 90% of the shorter sequence (“-aS 0.9 -c 0.8”). We performed de novo multiple sequence alignment of the clustered sequences (n=2,869) with MAFFT 7.222 (Katoh et al., 2002) using the “–localpair” and “–maxiterate 1000” parameter and trimmed the alignment with trimAl “-gappyout” (Capella-Gutiérrez et al., 2009). Only sequences with a gap-rate < 50% were retained (n=2,829). We performed model testing under the empirical LG model (Le and Gascuel, 2008), and with the empirical mixture models C10 to C60 (best model: C60+LG+G+F) and maximum likelihood tree reconstruction with IQ-TREE 1.6.2 (Minh et al., 2013). We inferred support values from 1,000 ultrafast bootstrap replicates (Hoang et al., 2018) with the “-bnni” option for bootstrap tree optimization and from 1,000 replicates of the SH-like approximate likelihood ratio test (Guindon et al., 2010).We next extracted the sister clade of CT295 (n=93) and used it as an outgroup for a more detailed analysis of CT295 related proteins. Sequence alignment, trimming, and initial phylogenetic reconstruction (best model: C60+LG+G+F) was performed as described above. We then used the resulting best tree as a guide tree for the posterior mean site frequency (PMSF) model (Wang et al., 2018) for improved site heterogeneity modeling under C60+LG+G+F and inferred 100 non-parametric bootstraps. Phylogenetic trees were visualized and edited using the Interactive Tree Of Life v4 (Letunic and Bork, 2019).

Primer sequences and cloning methods are described in Figure S2 , all constructs were verified by sequencing. The gene ct815 was cloned as follows: the gene was amplified with primers 1 and 4, and with primers 2 and 3. PCR products were mixed, heated to 65°C for 10 minutes, and annealed at 37°C for 30 minutes. The mix was then cloned into the pET28a previously digested by NcoI and XhoI restriction enzymes. Chimeras to test for the presence of T3S signal were constructed like in (Subtil et al., 2001), using the first 22 codons of the gene of interest. Flag-GlgA is described in (Gehre et al., 2016) and the negative control, CymR, that corresponds to a bacterial repressor for gene expression from Pseudomonas putida (Mullick et al., 2006) was constructed through Gateway cloning like Flag-PGM1 and PGM1-Flag. For GlgA, cloning was performed in pBOMB4 (Bauler and Hackstadt, 2014) using the endogenous promoter of the ctl0167 gene, and the construct was transformed into the LGV L2 25667R strain. The CT295-Flag construct was built by overlap PCR and inserted by Gibson reaction into p2TK2(spect)-SW2-mCh(Gro)-NotI-3xFlag-IncDTerm, a kind gift of Isabelle Derré. Briefly, a fragment containing the Tet repressor, the Tet promotor, and the 5’ part of ct295 was amplified by PCR using p2TK2(spect)-SW2-mCh(Gro)-Tet-IncV-NotI-3xFlag-IncDTerm as template (Stanhope et al., 2017). A second fragment covering the whole ct295 sequence was amplified using C. trachomatis serovar D as template. The two fragments were joined by overlap PCR and inserted by Gibson reaction into p2TK2(spect)-SW2-mCh(Gro)-NotI-3xFlag-IncDTerm digested with NotI and NdeI. The resulting plasmid allows for constitutive expression of mCherry under the control of the groESL promoter and terminator and the inducible expression of the CT295-Flag under the control of the aTc-inducible promoter. It was stably transformed in C. trachomatis L2 as described (Wang et al., 2011).

E. coli BL21 (DE3) was used to produce recombinant CCA00344, CT815, and CpsG harboring a polyhistidine tag at the C-terminal part of the protein. Bacteria were grown at 37°C to an optical density at 600 nm of 0.9 in 2 liters of LB with 30 µg/mL of kanamycin under shaking. 0.5 mM of isopropyl-β-D-thiogalactopyranoside (IPTG) was added and incubation was continued for 18 h at 16°C. Bacteria were harvested by centrifugation, and proteins were purified from clarified lysate by affinity chromatography on Ni²⁺-nitrilotriaceta-agarose (Qiagen) in 20 mM Tris-HCl 500 mM NaCl pH 8.0 buffer and by size exclusion chromatography on Superdex 200 HL 16/600 column (GE Healthcare) in 20 mM Tris-HCl 500 mM NaCl pH 7.0 buffer. All purified proteins were stored at 4°C.

HEK293T cells were used for production of Flag-tagged constructs. Cells were transfected with the indicated plasmids using the JetPrime kit (Polyplus Transfection) according to the manufacturer recommendations. Twenty-four hours post transfection, cells were lysed for 15 min at 4°C in lysis buffer [0.05 M Tris, 0.15 M NaCl, 5% Glycerol and 0.5% Nonidet P-40, pH = 7.5 supplemented with protease inhibitors (Roche)]. Samples were centrifuged at 12 000 xg for 10 min, the supernatants were collected and diluted 1:10 in lysis buffer except that Nonidet P-40 was brought to 0.05%. A fraction of the lysates was kept for analysis of the expression of the proteins by western blot. Anti-flag beads (Sigma) washed in the 0.05% Nonidet P-40 lysis buffer were added to the lysates and incubated for 2h at 4°C. The beads were then washed three times in 10 mM Tris-HCl, 0.5mM MgSO4 (pH 8.0) and used immediately for the PGM activity assay.

HIS-tagged CCA00344, CT815 or CpsG (0,1 mg/mL) were incubated for 20 minutes at 37°C with 0.2 mg/mL of glucose-1,6-diphosphate and either 0.66 mg/ml Glc1P or 0.33 mg/ml Glc6P in 10 mM Tris-HCl, 0.5 mM MgSO₄ buffer (pH 8.0) and then boiled for 5 minutes to stop the reaction. For the reactions using Flag-tagged proteins 0.66 mg/ml Glc1P was added on the beads in 10 mM Tris-HCl, 0.5mM MgSO4 (pH 8.0) in a 0.1 ml reaction volume, and incubated for 20 min at 37°C. The reaction was then stopped by boiling the samples for 5 min. Reaction products were analyzed by high performance anion exchange chromatography with pulsed electrochemical detection (Dionex, model ICS3000). The samples were injected into a CarboPAC-PA1 column pre-equilibrated for 20 minutes with 88.8% of 100 mM NaOH (eluant A) and 11.2% of 100 mM NaOH 720 mM NaOAc (eluant B). After injection, a gradient run (flow rate of 0.350 mL.min^-1) was performed as follows: 0-2 min 88.8% A + 11.2% B, 2-20 min 88.8% A + 11.2% B – 65% A + 35% B, 20-35 min 65% A + 35% B – 40% A + 60% B, 35-37 min 40% A + 60% B – 100% B. The data displayed are representative of at least 2 independent enzymatic assays. Different columns were used in this study; therefore, retention times were calibrated for each using pure Glc1 and Glc6P.

Cells were transfected with the indicated plasmids using the JetPrime kit (Polyplus Transfection) according to the manufacturer recommendations. Twenty-four hours post transfection, cells were fixed with 2% (w/v) paraformaldehyde (PFA) 2% (w/v) sucrose in PBS for 30 minutes at room temperature, then washed with PBS twice. For the panel shown in Figure 1 , a 4% PFA 4% sucrose solution was used for fixation. Cells were blocked with 10 mg/mL bovine serum albumin (BSA) for 30 minutes before being stained with antibodies anti-ICAM-1 coupled to APC (Invitrogen 17.0549.42) at dilution 1/1000. Cells were then washed twice with PBS, permeabilized with 0.05% saponin in PBS and DNA was stained with Hoechst 33342 (Molecular Probes). For the panel shown in Figure 4C the cells were permeabilized after ICAM-1-APC staining to probe for Flag (rabbit, Sigma F7425) or Hsp60 (rabbit, homemade) followed with A488-conjugated secondary antibodies. For the panel shown in Figures 1 , 4B , the cells were permeabilized in 0.05% saponin in PBS and stained with mouse anti-Flag M2 antibody (Sigma) and homemade rabbit anti-Cap1 antibody, followed with species-specific secondary antibodies. Images were acquired on an Axio observer Z1 microscope equipped with an ApoTome module (Zeiss, Germany) and a 63× Apochromat lens. Images were taken with an ORCA-flash4.OLT camera (Hamamatsu, Japan) using the software Zen. All images shown are representative of observations made on at least 3 biological replicates, in which more than 50 cells were observed each time.

For periodic-acid-Schiff (PAS) stain cells were fixed in 4% PFA 4% sucrose in PBS for 30 min at room temperature, permeabilized for 5 min in 0.05% saponin in PBS with 1 mg/ml BSA, and staining was performed as described (Schaart et al., 2004). Briefly, cells were incubated in 1% periodic acid (Sigma) for 5 min. Thereafter coverslips were put in tap water for 1 min, quickly rinsed in mQ-H₂0 and then applied to Schiff reagent (Sigma) for 15 min at room temperature. Afterwards the coverslips were rinsed again in mQ-H₂0, incubated in tap water for 10 min followed by an incubation step in PBS for 5 min. Periodic acid oxidizes the vicinal diols in sugars such as glycogen to aldehydes, which now in turn react with the Schiff reagent to give a purple colour. Coverslips infected with C. caviae and C. pneumoniae were further stained with home-made rabbit antibody against C. muridarum GroEL, that cross-reacts with the orthologous proteins in these species, followed with Alexa488-conjugated secondary antibodies. Images were acquired using transmission light or green fluorescence excitation on an Axio Vert.A1 microscope (Zeiss, Germany) and an LD A-Plan 40x lense. Images were taken with an AxioCam ICc 1 camera using the software Zen. All images shown are representative of observations made on at least 2 biological replicates, in which more than 50 inclusions were observed each time.

For siRNA transfection 50 000 cells were plated at J0 in a 24-well plate and immediately mixed with Lipofectamine RNAiMAX (Invitrogen) following the manufacturer’s recommendation, using 10 nM of siRNA (see Figure S2 for sequences). Cells treated with siRNA for 24 h (J1) were infected with L2^IncDGFP bacteria at a MOI = 0.2. Transfection of siRNAs was repeated on J2. On J3 (48 hpi), cells were detached, lysed using glass beads and the supernatant was used to infect new untreated cells plated the day before (100 000 cells/well in a 24-well plate), in serial dilution. The next day, 3 wells per condition with an infection lower than 30% (checked by microscopy) were detached, fixed and analyzed by flow cytometry to determine the bacterial titer as described (Vromman et al., 2014).

Chimera were constructed as described previously (Subtil et al., 2001) with the primers and templates displayed in Figure S2 . The sequence of all chimera was verified by sequencing. Analysis of secreted proteins was performed as described previously (Subtil et al., 2001), except that the ipaB or mxiD Shigella flexneri were co-transformed with the pMM100 plasmid that codes for the lacI repressor (selection with tetracyclin) (Jaumouillé et al., 2008) and with the indicated Cya chimera (selection with ampicillin). This was done to avoid the very strong expression of some of the chimeras in the absence of lacI repressor. Briefly, 1 ml of a 30°C overnight culture of S. flexneri ipaB or mxiD transformed with different Cya chimeras was inoculated in 15 ml of LB broth and incubated at 37°C for 1 h. Expression of the Cya chimeras was induced by adding 10 µM IPTG for all chimera except PGMtracho/Cya which was induced at 100 µM IPTG. Bacteria were harvested by centrifugation 3 h later and the supernatant was filtered through a Millipore filter (0.2 μm). To precipitate the proteins 1/10 (v/v) of trichloroacetic acid was added to the supernatants. The pellet fraction and the supernatant fraction (concentrated 40-fold compared to the pellet fraction) were resuspended in sample buffer for SDS-PAGE, transferred to Immobilon-P (PVDF) membranes and immunoblotted with the proper primary antibodies diluted in 1X PBS containing 5% milk and 0.01% Tween-20. Primary antibodies used were mouse anti-cya, rabbit anti-CRP and rabbit anti-IpaD antibodies generously given by Drs N. Guiso, A. Ullmann and C. Parsot, respectively (Institut Pasteur, Paris), mouse anti-Flag M2 (Sigma) and anti-MOMP (Argene). Goat anti-mouse and anti-rabbit IgG-HRP (GE Healthcare) were used at 1:10000 dilution. Blots were developed using the Western Lightning Chemiluminescence Reagent (GE Healthcare). The data displayed are representative of at least three independent experiments.