Type III secretion system effectors form robust and flexible intracellular virulence networks

INTRODUCTION Infections with many Gram-negative pathogens, including Escherichia coli, Salmonella, Shigella, and Yersinia, rely on the injection of effectors via type III secretion systems (T3SSs). The effectors hijack cellular processes through multiple mechanisms, including molecular mimicry and diverse enzymatic activities. Although in vitro analyses have shown that individual effectors can exhibit complementary, interdependent, or antagonistic relationships, most in vivo studies have focused on the contribution of single effectors to pathogenesis. Citrobacter rodentium is a natural mouse pathogen that shares infection strategies and virulence factors with the human pathogens enteropathogenic and enterohemorrhagic E. coli (EPEC and EHEC). The ability of these pathogens to colonize the gastrointestinal tract is mediated by the injection of effectors via a T3SS. Although C. rodentium infects 31 effectors, the prototype EPEC strain E2348/69 translocates 21 effectors. RATIONALE The aim of this study was to test the hypotheses that, rather than operating individually, the T3SS effectors form robust intracellular networks that can sustain large contractions and that expanded effector repertoires play a role in distinct disease phenotypes and host adaption. RESULTS We tested the effector-network paradigm by infecting mice with >100 C. rodentium effector mutant combinations. First, using machine learning prediction algorithms, we discovered additional effectors, NleN and NleO. We then sequentially deleted effector genes from two distinct starting points to reach sustainable endpoints, which resulted in strains missing 19 unrelated effectors (CR14) or 10 effectors involved in the modulation of innate immune responses in intestinal epithelial cells (IECs) (CRi9). Moreover, we deleted Map and EspF, which target the mitochondria and disrupt tight junctions. Unexpectedly, all strains colonized the colon and activated conserved metabolic and antimicrobial processes in the IECs while eliciting distinct cytokine and immune cell infiltration responses. In particular, although infection with C. rodentium Δmap/ΔespF failed to induce secretion of interleukin-22 (IL-22), CR14 and CRi9 triggered heightened secretion of IL-6 and granulocyte-macrophage colony-stimulating factor (GM-CSF) and of IL-22, interferon-γ (IFN-γ), and IL-17 from colonic explants, respectively. Nonetheless, infection with CR14 or CRi9 induced protective immunity against secondary infections. Although Tir, EspZ, and NleA are essential, other effectors exhibit context-dependent essentiality in vivo. Moreover, C. rodentium expressing the effector repertoire of EPEC E2348/69 failed to efficiently colonize mice. We used curated functional information and our in vivo data to train a machine learning model that predicted values for colonization efficiency of previously uncharacterized mutant combinations. Notably, a mutant with a low predicted value, lacking only nleF, nleG8, nleG1, nleB, and espL, failed to colonize. CONCLUSION Our analysis revealed that T3SS effectors form robust networks, which can sustain substantial contractions while maintaining virulence, and that the composition of the effector network contributes to host adaptation. Alternative effector networks within a single pathogen triggered markedly different immune responses yet induced protective immunity. CR14 did not tolerate any further contraction, which suggests that this network reached its robustness limit with only 12 effectors. As the robustness limits of other effector networks depend on the contraction starting point and the order of the deletions, machine learning models could transform our ability to predict alternative network functions. Together, this study demonstrates the robustness of T3SS effector networks and the ability of IECs to withstand drastic perturbations while maintaining antibacterial functions. The work of T.I.R. and J.S.C. was funded by the CRUK Centre grant with reference no. C309/A25144. N.W. was supported by PhD fellowships from the Manna Center Program for Food Safety and Security at Tel Aviv University and the Edmond J. Safra Center for Bioinformatics at Tel-Aviv University. The work of E. N.-B. was funded by a PhD FPU grant from the Spanish government (FPU2017/04179). A. R.-P. was partially funded by project PID2019-106960GB-I00 (AEI/FEDER, Spain) and by InGEMICS-CM, FSE/FEDER, Comunidad de Madrid Project B2017/BMD-3691. L.A.F. is supported by grant BIO2017-89081-R (AEI/MICIU/FEDER, European Union). G.F. is supported by an MRC program grant (MR/R02671/), a Royal Society grant (IC160080), and a Wellcome Investigator Award (107057/Z/15/Z). Peerreview