Intracellular S. flexneri infection causes shigellosis in humans. Although most intracellular bacteria reside in vacuoles, S. flexneri, like Burkholderia spp., live freely in the host cytosol, inevitably exposing their LPS to caspase-11/4. Wild-type (WT) mice, unlike Casp11−/− mice, survived B. thailandensis infection8 (Fig. 1a, Extended Data Fig. 1a). Mice are increasingly being used as a surrogate host for S. flexneri. Unexpectedly, both WT and Casp11−/− mice succumbed to lethal S. flexneri infection (Fig. 1a) and tolerated similarly the low-dose challenge (Extended Data Fig. 1a). Given the absence of caspase-11-mediated protection, we assayed non-canonical inflammasome activation upon S. flexneri infection. Casp1−/− immortalized bone marrow-derived macrophages (iBMDMs) were used to avoid interference by the canonical inflammasome. Although B. thailandensis and S. Typhimurium ΔsifA induced Casp11-dependent GSDMD cleavage and pyroptosis8, S. flexneri triggered little pyroptosis (Fig. 1b) despite a higher infection efficiency (Extended Data Fig. 1b). In epithelium-derived human SiHa and A431 cells, S. flexneri, unlike S. Typhimurium ΔsifA, also did not activate the caspase-4–GSDMD pyroptosis pathway (Fig. 1b, Extended Data Fig. 1c, d). Purified LPS from S. flexneri was highly pro-pyroptotic (Extended Data Fig. 1e). Thus, S. flexneri evaded caspase-11/4-mediated pyroptosis.
The guanylate-binding protein (GBP) family promotes the release of LPS from intracellular bacteria and its presentation to caspase-11/4 (refs. 9,10). IpaH9.8, a Shigella ubiquitin-ligase T3SS effector, targets multiple GBPs for degradation11,12,13,14. A 2013 report proposed that S. flexneri uses the T3SS effector OspC3 to target caspase-4 but, notably, not caspase-11 (ref. 15). We examined whether IpaH9.8, OspC3 or another factor underlies evasion of pyroptosis by S. flexneri using SiHa, A431 and iBMDM cells (Fig. 1c, Extended Data Fig. 1d, f, g). Infection with ΔipaH9.8, compared to WT bacteria, caused negligibly increased pyroptosis. By contrast, ΔospC3 induced extsensive pyroptosis with evident GSDMD cleavage, which was diminished by re-expression of OspC3 in the bacteria or deletion of CASP4/11 deletion in the host cells. Deletion of all seven GBPs from A431 cells affected pyroptosis during early but not late infection (Extended Data Fig. 1h). This is consistent with the notion that GBPs, having little LPS-binding activity (Extended Data Fig. 1i), are not absolutely required for bacteria-induced caspase-4 activation. Thus, S. flexneri requires OspC3 to evade LPS-stimulated pyroptosis.
OspC3 expression in host cells blocked the induction of pyroptosis by S. flexneri ΔospC3, S. Typhimurium and even LPS alone (Extended Data Fig. 1j), suggesting that it has a bacteria-independent function. OspC3 co-immunoprecipitated with the p20/p10 form of caspase-4(C258A) (protease-deficient; C/A hereafter) in 293T cells (Extended Data Fig. 2a). The interaction did not cause p20–p10 dissociation, in contrast to earlier findings15. OspC3 also co-immunoprecipitated with inactive p20- and-p10-unprocessed caspase-4/11 (Fig. 2a, Extended Data Fig. 2b). OspC3 did not affect the proteolytic activity of caspase-4/11-p20/p10 (Extended Data Fig. 2c–e). Purified OspC3 also did not inhibit LPS-induced activation of pro-caspase-4 to cleave GSDMD, but it blocked pyroptosis when electroporated into cells (Extended Data Fig. 2f, g). Thus, hijacking of caspase-4/11 by OspC3 involves a cell-dependent mechanism.
In 293T cells, OspC3 induced slower migration of caspase-4-p10 on an SDS gel (Fig. 2b, Extended Data Fig. 2a). Caspase-4/11-p30, co-expressed with OspC3 in Escherichia coli, exhibited a marked shift on a native gel (Fig. 2c), indicating a post-translational modification (PTM). Electrospray ionization–mass spectrometry (ESI–MS) identified a 524-Da modification, which was located to 314RDSTMGSIF322 within caspase-4-p10 by collision-induced dissociation (CID)–MS (Fig. 2d, Extended Data Fig. 2h, i). MS/MS detected fragment ions with mass-to-charge ratios of 136.06, 348.07 and 428.04, matching the mass of adenine, AMP and ADP, respectively (Extended Data Fig. 2j). This reminded us of ADP-ribosylation, in which ADP-ribose (ADPR) from nicotinamide adenine dinucleotide (NAD+) is usually transferred to serine, arginine, asparagine, aspartate, glutamate or glutamine. Although the OspC3-induced PTM is 17 Da smaller than ADP-ribosylation, NAD+ enabled recombinant OspC3 to modify caspase-4/11-p30 by 524 Da (Fig. 2e, Extended Data Fig. 2k, l). In S. flexneri infection, the 314RDSTMGSIF322 peptide and the corresponding caspase-11 peptide showed the 524-Da modification in an ospC3-dependent manner (Extended Data Fig. 2m). Consistently, caspase-4-p30-C/A from cells infected with WT S. flexneri but not ΔospC3 had a mobility shift similar to that in the in vitro assay (Fig. 2f). Thus, OspC3 catalyses an NAD+-mediated PTM on caspase-4/11.
Electron-transfer/higher-energy collision dissociation (EThcD)–MS showed that Arg314 and Arg310 in caspase-4 and -11, respectively, harboured the modification (Fig. 3a, Extended Data Fig. 3a). Replacing these residues with lysine or asparagine abolished the modification (Fig. 3b, Extended Data Fig. 3b–e). Quantitative high-performance liquid chromatography (HPLC)–MS analyses of the reaction (Extended Data Fig. 4a) revealed that one molecule of free nicotinamide (Nam) was released upon modification of one molecule of caspase-4 by one molecule of NAD+. Thus, the OspC3-catalysed PTM may contain an initial ADP-ribosylation and an additional 17-Da mass reduction reaction.
Fourteen NAD+ analogues or derivatives were assayed in OspC3 modification of caspase-4 (Extended Data Fig. 4b). NAD+ fragments (ADPR, cyclic-ADPR (cADPR) and nicotinamide mononucleotide (NMN)), α-NAD+, nicotinic acid adenine dinucleotide (NAAD+), nicotinamide adenine dinucleotide phosphate (NADP+) and NADPH were inactive. By contrast, NADH, thio-NAD+ and thio-NADH, altered in the Nam part, supported the 524-Da modification (Fig. 3c, Extended Data Fig. 4b, c). Deamino-NAD+, biotin-NAD+, ε-NAD+ or nicotinamide guanine dinucleotide (NGD+) allowed modifications that preserved the mass difference between the cognate analogue and NAD+. These confirm the transfer of ADPR to caspase-4/11 with Nam being the leaving group. Indeed, OspC3-modified caspase-4 was recognized by an anti-ADP-ribose antibody16 (Fig. 2f, Extended Data Fig. 4d).
NGD+-mediated modification also had a 17-Da mass reduction from the ‘GDP-ribosylation’. A non-specific pyrophosphohydrolase, NUDT16 (ref. 17), removed an AMP from OspC3-modified caspase-4 (Extended Data Fig. 4d, e). These data suggest that the 17-Da loss occurs on the phosphoribosylated arginine. We performed stable isotope labelling by amino acids in cell culture (SILAC), using 13C6,15N4–l-arginine to label Flagcaspase-4-p20/p10 expressed alone or with OspC3 in 293T cells. MS detected a 523-Da (not 524-Da) increase on a caspase-4 Arg314-containing peptide (Fig. 3d, Extended Data Fig. 5a). The 1-Da change suggests that one Nω atom in ADP-ribosylated arginine is removed via internal deamination, explaining the 17-Da reduction. Consistent with this, free NH3/NH4+ was detected in the modification reaction (Fig. 3e).
The above analyses predict that a nucleophile adjacent to the ADP-ribosylated arginine guanidino performs the deamination. Studies of ADP-ribosylation-based elimination18 suggest that the ribosyl-2′-OH could be a candidate nucleophile. β-2′-Deoxy-2′-H-NAD+ (2′-H-NAD+) and 2′-fluoro-NAD+ were assayed in the OspC3-catalysed modification (Extended Data Fig. 5b). 2′-Fluoro-NAD+, which is incompetent for canonical ADP-ribosylation19, could not support OpsC3 modification of caspase-4. OspC3 could use 2′-H-NAD+; notably, the modification was merely the transfer of 2′-deoxy-ADPR without further deamination (Fig. 3f, Extended Data Fig. 5c). The remaining puzzle is the atom on which the initial ADP-ribosylation occurs. Both Nω and Nδ in arginine can accept ADPR from NAD+. Ninhydrin could bond simultaneously with the two Nω in native arginine, as noted with Arg314 in unmodified caspase-4 (Extended Data Fig. 5d, e). For canonical arginine Nω-ADP-ribosylation (Rab4a by ExoS20), the modified arginine resisted conjugation by ninhydrin (Extended Data Fig. 5f). For 2′-H-NAD+-mediated modification, the 2′-deoxy-ADP-ribosylated Arg314 could react with ninhydrin (Extended Data Fig. 5e). We propose that OspC3 modifies Arg314/Arg310 of caspase-4/11 by two steps (Extended Data Fig. 5g). First, the arginine Nδ (rather than Nω) performs nucleophilic substitution of the Nam in NAD+. Second, the ribosyl-2′-OH of ADPR initiates a deamination to remove one Nω, forming an oxazolidine ring. We designate this arginine ADP-2′-imine-ribofurano[1′,2′:4,5]oxazolidination modification as ADP-riboxanation (Fig. 3g), catalysed by arginine ADP-riboxanase activity in OspC3.
S. flexneri harbours ospC1, ospC2 and a pseudogene ospC4 in the ospC3 locus. OspC1, OspC2 and OspC3 (>60% sequence identity; Extended Data Fig. 6a) share a C-terminal ankyrin-repeat domain (ARD) and an N-terminal (N) domain (Fig. 4a). ΔospC1 or ΔospC2 caused no increase in pyroptosis in infected cells (Extended Data Fig. 1f). OspC1/2 could not block cytosolic LPS-induced pyroptosis (Extended Data Fig. 6b). Purified OspC1/2 barely modified caspase-4 (Fig. 4b). Notably, the ARD of OspC3, but neither OspC1/2 nor their ARDs, readily co-immunoprecipitated with caspase-4-p30 (Fig. 4c, Extended Data Fig. 6c). Replacing the ARD in OspC3 with that of OspC1/2 diminished its caspase-4-modification and pyroptosis-blocking activity (Fig. 4b, Extended Data Fig. 6b). Conversely, chimeric proteins with the OspC1/2 N-domain and OspC3 ARD were highly active. Thus, the ARD of OspC3 determines caspase-4/11 recognition; OspC1/2 use their ARDs to target other host proteins for ADP-riboxanation.
Random mutagenesis identified Phe141, Phe186, Glu192, Glu326 and His328 in OspC3 (Fig. 4a, Extended Data Fig. 6a) as essential for ADP-riboxanating caspase-4/11 and blocking pyroptosis (Extended Data Fig. 6d–g). Another D177A mutation supported ADP-ribosylation but blocked subsequent deamination (Fig. 4d, e). Although OspC3(D177A)-modified caspase-4 was sensitive to ADP-ribosylarginine hydrolase (ADPRH), WT OspC3-catalysed ADP-riboxanation resisted demodification by ADPRH and other known host ADP-ribosylhydrolases (Extended Data Fig. 6h, i). Thus, hijacking of caspase-4/11 by ADP-riboxanation is more advantageous to bacterial virulence.
OspC3 blocked LPS-induced caspase-4 autoprocessing (Fig. 4f). This was recapitulated by mutations of Arg314 that also inhibited infection or LPS-induced pyroptosis (Fig. 4f, Extended Data Fig. 7a, b). Mutations of Arg310 in caspase-11 had the same effect (Extended Data Fig. 7c–e). OspC3 could also ADP-riboxanate already activated caspase-4/11 (Fig. 2b); the modified caspase-4/11-p20/p10, like their Arg314/Arg310 mutants, failed to target GSDMD (Fig. 4g, h, Extended Data Fig. 7f, g) owing to structural interference with the GSDMD-binding exosite21. Arg314/Arg310, which are conserved in caspases (Extended Data Fig. 7h), coordinate substrate P1 aspartate; caspase-4/11-p20/p10 R314/R310 mutants could not cleave the peptide substrate (Extended Data Fig. 7i). Thus, ADP-riboxanation blocked caspase-4/11 activation and cleavage of their substrate.
We used the inactive OspC3 E192A/H328A mutant (EH/AA) (Extended Data Fig. 6d–g) and assessed the function of caspase-11 ADP-riboxanation in Shigella infection. WT mice survived S. flexneri ΔospC3 infection, and this effect was reversed by complementation with OspC3 WT but not the EH/AA mutant (Fig. 5a). Accordingly, mice infected with ΔospC3 alone or ΔospC3 expressing OspC3 EH/AA had lower bacterial burdens than mice infected with WT OspC3-expressing strain (Fig. 5b). Unlike WT mice, Casp11−/− mice succumbed equally to S. flexneri WT and ΔospC3, which replicated to a similarly high level. Notably, S. flexneri ΔospC3-infected mice produced much more anti-Shigella IgG than WT bacteria-infected mice at the 10% LD50-normalized dose (the burden of ΔospC3 was not higher than WT bacteria at 24 h after infection), but this effect was abolished in Casp11−/− and Gsdmd−/− mice (Fig. 5c, Extended Data Fig. 8a). ΔospC3-infected mice were more resistant to lethal S. flexneri re-infection, and this increased resistance was also absent in Casp11−/− mice (Fig. 5d). Such effects occurred at multiple inoculation doses (Extended Data Fig. 8b, c). These findings suggest that caspase-11-mediated pyroptosis has an intrinsic function of activating humoral immunity and also highlight the importance of OspC3-catalysed ADP-riboxanation for evasion of caspase-11-mediated pyroptosis by Shigella.
Development of a Shigella vaccine has been challenging. ΔicsA and ΔguaBA mutants are being developed as live-attenuated vaccines22. Compared to ΔicsA and ΔguaBA, S. flexneri ΔospC3 induced a higher level of anti-Shigella IgG (Extended Data Fig. 8d). Deletion of ospC3 on either ΔicsA or ΔguaBA background further boosted antibody production and conferred better protection from WT S. flexneri re-challenge (Fig. 5e, f). The better immunization of ΔicsAΔospC3 over ΔicsA also occurred at lower or higher inoculation doses in both C57BL/6 and BALB/c mice (Extended Data Fig. 8e–h). These findings are valuable for Shigella vaccine development, although with the limitation of the mouse model.
A BLAST search identified 27 OspC homologues in diverse bacteria, including Vibrio, Salmonella, Erwinia and Chromobacterium (Extended Data Fig. 9a). Homology of their catalytic domains and ARDs to those of OspC3 ranges from 99% to 56% and from 100% to 27%, respectively. Certain homologues readily ADP-riboxanated caspase-4/11 and blocked LPS-induced pyroptosis, and this effect was abolished by the corresponding EH/AA mutations (Extended Data Fig. 9b, c). For homologues that could not modify caspase-4/11, replacing their ARDs with that of OspC3 enabled the modification (Extended Data Fig. 9c). CopC from Chromobacterium violaceum, a deadly bacterium that causes hepatic abscesses in humans, could ADP-riboxanate caspase-4 and inhibit LPS-induced pyroptosis, less potently than OspC3 (Extended Data Fig. 10a–c). CopC could also modify caspase-4/11 during infections, and C. violaceum ΔcopC showed decreased replication in infected mouse liver (Extended Data Fig. 10d–f). Thus, OspC-like ADP-riboxanases are widely used by bacteria for various functions, including blocking pyroptosis.
In summary, Shigella uses OspC3 to modify caspase-11/4 and thereby thwart the inflammasome/pyroptosis-mediated defence. This differs from known bacterial inflammasome-modulating strategies that are self-alterations or indirect, such as inhibition of the Pyrin inflammasome by Yersinia YopM23. Future studies will uncover other inflammasome/pyroptosis-targeting effectors. OspC3 catalyses arginine ADP-riboxanation; the activity is shared by the OspC family in bacteria. Arginine ADP-riboxanation might also exist in eukaryotes, and could be identified by mining ADP-ribosylome proteomic data.